Thesis AndreasBachmann 2001

GIS-based Wildland Fire Risk
Analysis
Dissertation
zur
Erlangung der naturwissenschaftlichen Doktorwürde

(Dr. sc. nat.)
vorgelegt der
Mathematisch-naturwissenschaftlichen Fakultät
der
Universität Zürich
von
Andreas Bachmann
von
Amden SG
Begutachtet von
Prof. Dr. Robert Weibel

Prof. Dr. Domingos X. Viegas
Dr. sc. techn. Britta Allgöwer
Zürich 2001
Die vorliegende Arbeit wurde von der Mathematisch-naturwissenschaftlichen
Fakultät der Universität Zürich auf Antrag von Prof. Dr. R. Weibel und
Prof. Dr. K. Brassel als Dissertation angenommen.
iii
Abstract
This dissertation deals with issues that arise from analysing the spatial patterns
of wildland fire risk.
The expanding wildland-urban interface especially in fire prone vegetation
types has lead to an increased susceptibility of communities. Additionally, the
restructuring in the agricultural sector in most industrial nations yields in the
abandoning of large areas that cannot be longer profitably cultivated. Wildland
fires cannot be suppressed completely and the management is subject to limited
resources. It is thus important to analyse the spatial distribution of wildland
fire risk in order to delineate and prioritise particular susceptible areas. The
proposed wildland fire risk analysis allows explicit consideration of the manifold
spatial input data and the spatial behaviour of the process itself.
Wildland fire management is seen as a risk management process and special
interest is given to the risk analysis step. The risk analysis framework is based on
methods of quantitative risk analysis as it is applied in technical risk engineering.
The main actors of the framework are the scenarios and the objects at risk. A
scenario is given by a starting point of a fire in a given area of interest. Objects
at risk can be any entities that are of interest for the managing organisation.
For every object-scenario-relation the probability of an impact and the expected
impact is determined. These can then be summed up for either an object or a
scenario resulting in individual, respectively collective risk values. The explicite
distinction of risk values for single objects as well as for scenarios allows a
detailed analysis of the risk situation. This, in turn, can support the planning
of more effective and efficient management activities.
An important aspect was the study of uncertainties in the input data and
their propagation throughout the risk analysis. The uncertainty propagation
within the most important applied model, Rothermel’s model for wildland fire
behaviour, was analysed with a first order Taylor series. In the given study
case, it could be shown that the uncertainties of the input data produced very
large uncertainties in the results with variation coefficients larger than 1. The
uncertainties of the input data stem mainly from their inherent, natural, spatial
and temporal variablity.
The wildland fire risk analysis was tested in the area of Southern Switzerland
with special emphasis on the post fire risk of debris flows.
iv
Zusammenfassung
Die vorliegende Dissertation beschäftigt sich mit der räumlichen Analyse des
Waldbrandrisikos1
In den letzten Jahren hat sich die Wahrnehmung von Waldbränden dahinge-
hend verändert, dass sie nicht mehr länger nur als Bedrohung angesehen werden.
Vielmehr versucht man ihre Bedeutung mit systemtheoretischen Ansätzen zu er-
fassen. Dies bedingt jedoch eine detailliertere räumliche Analyse, welche nicht
nur das Brandverhalten sondern auch die Zusammenhänge zwischen Ausbruchs-
orten und gefährdeten Objekten berücksichtigt. Wir betrachten Waldbrandma-
nagement als einen Risikomanagementprozess, in welchem die Risikoanalyse ein
wichtiger Schritt ist.
Das in dieser Arbeit vorgestellte Rahmenwerk zur quantitativen Waldbrand-
risikoanalyse ermöglicht eine derart vertiefte Betrachtungsweise. Da bis anhin in
der Waldbrandforschung entsprechende Ansätze zur Risikoanalyse fehlten, wur-
den diese aus dem Bereich des technischen Risiko-Ingenieurwesens übernommen
und den spezifischen Eigenschaften von Waldbränden angepasst. Als Kern wird
die sogenannte Risikomatrix verwendet, welche alle Szenarien mit sämtlichen
Objekten verknüpft. Mit Hilfe dieser Technik ist es möglich sowohl das Risiko
von Szenarien (“Risikoquellen”) als auch das Risiko von gefährdeten Objekten
(“Risikoempfänger”) zu bestimmen. Die explizite Unterscheidung des Risikos
von Szenarien und von Objekten erlaubt eine detailliertere Analyse der Risiko-
situation. Dies wiederum, ermöglicht die Planung von effektiveren und effizien-
teren Management-Massnahmen.
Da die Resultate der Waldbrandrisikoanalyse Grundlage zur Beurteilung des
Risikos bilden, ist es wichtig, Unsicherheiten abzuschätzen, bzw. zu quantifizie-
ren. Die Unsicherheitsfortpflanzung wurde im Rothermel-Modell für Waldbrand-
verhalten genauer untersucht. Dazu wurde die Methode der Taylor-Reihe erster
Ordnung verwendet. Es konnte gezeigt werden, dass die Unsicherheiten der Ein-
gangsvariablen zu grossen Fehlern in den Resultaten führen, wobei Variations-
koeffizienten grösser als 1 beobachten wurden. Die Unsicherheiten der Eingangs-
variablen sind vorallem bedingt durch eine inhärente, natürliche, räumliche und
zeitliche Variabilität.
Anhand einer Versuchsstudie in einer Region im Süden der Schweiz wird der
praktische Einsatz der vorgestellten Waldbrandrisikoanalyse demonstriert. Als
gefährdete Objekte wurden die Einzugsgebiete potentieller Murgänge verwen-
det.
1 Der im Englischen gebräuchliche Begriff “wildland fire” wird hier einfachheitshalber mit
“Waldbrand” übersetzt und umfasst sämtliche Feuer in einer natürlichen Umwelt. D.h.,
Steppen-, Busch- oder Grasbrände werden ebenfalls unter diesem Begriff subsummiert.
Acknowledgements
Dr. Britta Allgöwer for her great encouraging support, for all the many com-
ments and discussions, and for the opportunity to work for the GIS of the Swiss
National Park.
Profs. R. Weibel and K. Brassel for offering the opportunity of working in a
stimulating environment and for supporting this thesis.
Prof. D. X. Viegas for acting as an external reviewer.
Marco Conedera of the Sottostazione, Bellinzona, for his competent information
and suggestions and for providing access to the database used in the case study.
Othmar Wigger and Dani Wirz for their competent and obliging Unix admin-
istration all day and night, for the help-full programming advices and for the
nifty tips and tricks.
Daria Martinoni and Mats Bader for their support in mathematical problems
and Heiri Leuthold and Michel Hermann for the good times and stimulating
discussions in the K-8.
Contents
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
1 Introduction 1
1.1 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Wildland Fire and Management 5

2.1 Wildland Fire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Fire Behaviour . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Wildland Fire Occurrence . . . . . . . . . . . . . . . . . . 9
2.1.3 Wildland Fire Effects . . . . . . . . . . . . . . . . . . . . 9
2.2 Wildland Fire Management . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Wildland Fire Management Activities . . . . . . . . . . . 12
2.2.3 Existing Wildland Fire Managements Approaches . . . . 13
2.3 Risk Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.2 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.1 Hazard . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4.2 Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Risk Management . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.6 Geographical Information Systems and Wildland Fire Management 19
3 Quantitative Risk Analysis 25

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.1 Basic Elements . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.2 Relationship between Objects at Risk and Scenarios . . . 26
3.2.3 Individual Risk of an Object . . . . . . . . . . . . . . . . 27
3.2.4 Collective Risk of a Scenario . . . . . . . . . . . . . . . . 27
3.2.5 Probability and Frequency . . . . . . . . . . . . . . . . . . 28
3.3 Uncertainty in Risk Analysis . . . . . . . . . . . . . . . . . . . . 30
3.3.1 Uncertainties and Errors . . . . . . . . . . . . . . . . . . . 30
3.3.2 Error Models . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.3 Error Propagation in Modelling . . . . . . . . . . . . . . . 31
viii Contents
4 A Consistent Wildland Fire Risk Terminology Is Needed! 35

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 “Risking” Redefinitions . . . . . . . . . . . . . . . . . . . . . . . 36
4.3 Quantitative Wildland Fire Risk . . . . . . . . . . . . . . . . . . 37
4.3.1 Calculating Risk . . . . . . . . . . . . . . . . . . . . . . . 38
4.3.2 Research Integration . . . . . . . . . . . . . . . . . . . . . 39
4.4 An Analytical Framework . . . . . . . . . . . . . . . . . . . . . . 41
4.5 Need for a Robust Terminology . . . . . . . . . . . . . . . . . . . 42
5 Framework For Wildfire Risk Analysis 47

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.2 Wildfire Risk - A Definition . . . . . . . . . . . . . . . . . . . . . 48
5.3 General Layout of the Framework . . . . . . . . . . . . . . . . . . 49
5.4 Risk Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.4.1 Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.4.2 Scenario And Fire Occurrence . . . . . . . . . . . . . . . . 51
5.4.3 Objects And Fire Effects . . . . . . . . . . . . . . . . . . 52
5.4.4 Relating Objects and Scenarios . . . . . . . . . . . . . . . 52
5.4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6 Uncertainty Propagation in Wildland Fire Behaviour Modelling 57

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.2 Fire Behaviour Modelling . . . . . . . . . . . . . . . . . . . . . . 58
6.2.1 Rothermel’s Model . . . . . . . . . . . . . . . . . . . . . . 58
6.2.2 Input Variables . . . . . . . . . . . . . . . . . . . . . . . . 59
6.3 Uncertainty Propagation . . . . . . . . . . . . . . . . . . . . . . . 60
6.3.1 Monte Carlo Simulation vs. Taylor Series . . . . . . . . . 60
6.3.2 Partial Derivatives of ros, sdr and ef w . . . . . . . . . . 61
6.4 Study Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.4.1 Test Site Malcantone . . . . . . . . . . . . . . . . . . . . . 62
6.4.2 Fuel Types . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.4.3 Fuel Moisture . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4.4 Terrain and Wind . . . . . . . . . . . . . . . . . . . . . . 64
6.4.5 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.5.1 Rate of Fire Spread . . . . . . . . . . . . . . . . . . . . . 65
6.5.2 Fire Spread Direction . . . . . . . . . . . . . . . . . . . . 66
6.5.3 Effective Wind Speed . . . . . . . . . . . . . . . . . . . . 67
6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
7 Wildland Fire Risk Analysis in Southern Switzerland 73

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.2.1 Wildland Fire Risk Analysis . . . . . . . . . . . . . . . . . 74
7.2.2 Impact Probability Function . . . . . . . . . . . . . . . . 75
7.2.3 Damage Function . . . . . . . . . . . . . . . . . . . . . . . 77
7.3 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
7.3.1 The Malcantone . . . . . . . . . . . . . . . . . . . . . . . 77
7.3.2 Objects: Catchment Areas of Potential Debris Flow . . . 78
Contents ix
7.3.3 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.3.4 Fire Behavior . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.4.1 Impact Probabilities . . . . . . . . . . . . . . . . . . . . . 82
7.4.2 Expected Damage . . . . . . . . . . . . . . . . . . . . . . 82
7.5 Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . 83
8 Synthesis 89
8.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
A Models used 103

A.1 Rothermel’s Model . . . . . . . . . . . . . . . . . . . . . . . . . . 104
A.2 Debris Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . . 124
x List of Figures
List of Figures
1.1 Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . 3
3.1 Components of quantitative risk analysis . . . . . . . . . . . . . . 26

3.2 Contributing factors of the individual risk of an object . . . . . . 27
3.3 Collective risk of a scenario . . . . . . . . . . . . . . . . . . . . . 28
4.1 Hypothetical study area showing relationships between objects

and scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Collective risk of a scenario . . . . . . . . . . . . . . . . . . . . . 40
4.3 Risk methodology and wildland fire research . . . . . . . . . . . . 41
4.4 Framework for a quantitative wildland fire risk analysis . . . . . 42
4.5 Example of expected damage to objects and by scenario in the
Malcantone region, Ticino, Switzerland . . . . . . . . . . . . . . . 43
5.1 Framework for the wildfire risk analysis . . . . . . . . . . . . . . 49

5.2 Risk matrix for one situation Es . . . . . . . . . . . . . . . . . . 51
5.3 Relationship between two scenarios and one object . . . . . . . . 53
6.1 Area of interest with the three fuel types . . . . . . . . . . . . . . 62

6.2 Results of fire behaviour modelling and uncertainty propagation 66
7.1 Risk matrix for one situation . . . . . . . . . . . . . . . . . . . . 75

7.2 Probability function depending on spread time . . . . . . . . . . 76
7.3 Monthly distribution of wildland fires in the Malcantone region . 80
7.4 Yearly ignition probabilities for the Malcantone Region . . . . . . 81
7.5 Yearly impact probabilities for catchment areas and for scenarios 83
7.6 Yearly expected damages for catchment areas and damages for
scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
List of Tables xi
List of Tables
2.1 Fire suppression interpretations of flame length . . . . . . . . . . 8

2.2 Number of wildland fires in 1996, distinguished by cause . . . . . 10
2.3 Burnt areas due to wildland fires in 1996, distinguished by cause 10
6.1 Fuel model properties . . . . . . . . . . . . . . . . . . . . . . . . 59

6.2 Values for the three fuel types in the Malcantone valley . . . . . 63
6.3 Properties with constant values for all three fuel types . . . . . . 63
6.4 Correlation-matrix for fuel model variables . . . . . . . . . . . . 64
6.5 Correlations between slope and aspect . . . . . . . . . . . . . . . 64
6.6 Rate of spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.7 Direction of maximum fire spread . . . . . . . . . . . . . . . . . . 67
6.8 Effective wind speed . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.9 Sensitivity analysis for the 13 NFFL and the 3 Ticino fuel types 68
7.1 Fire adaptiveness, rain intensities and derived return periods, and
yearly debris flow probabilities . . . . . . . . . . . . . . . . . . . 77
7.2 Objects categories and associated damage values . . . . . . . . . 78
7.3 Fuel types and their coverage in the study area . . . . . . . . . . 82
1
Introduction
1.1 Problem statement

Basically, the term “wildland fires” denotes fires that occur in natural environ-
ments such as forest stands, bush vegetation, and so on. They are in some
regions an ubiquitous phenomenon that regularly occurs during a well defined
season. Wildland fires can have a dominant influence on vegetation and in many
cases the vegetation itself is adapted to regular fires.
Although they occur in the “wildland”, wildland fires are often related to
human activities in two ways: in many regions they occur mainly because of
human interaction such as arson, negligence, incautious cultivation (burning
ridges) and so on. On the other hand, wildland fires “destroy” properties, forest
stands and sometimes even casualties are to be accounted for. Nevertheless, it
is too simplistic to reduce wildland fires to these aspects as they are part of the
ecosystem in which it is seldomly possible to establish fully deterministic causal
relationships. Furthermore, systems that are dominated by markedly slow pro-
cess such as tree growth run a high risk of not being adequately accounted for by
shortsighted management. For example, during decades (starting in the 1940s)
it was the common policy of American communities to suppress and fight every
fire. Public funding for fire fighting organisations experienced a steady augmen-
tation while the reduction of burnt areas occurred only up to the mid-1950s and
then stagnated. The experiment of total exclusion of fire led to high accumu-
lations of fuel and consequently to more extreme fire behaviour which boosted
the costs of fire fighting: a vicious circle.
Fighting large wildland fires is an arduous task and often they are not con-
trolled unless the environmental conditions are more favourable. Preventing
ignition of fires is also hard because, given dry and hot weather, they can start
at any place with burnable vegetation. To predict where the next spark will ig-
nite a fire is very difficult because in many cases this involves the consideration
of hardly quantifiable socio-economic factors such as land use code, arson, and
so on. Additionally, the process wildland fire itself is highly variable in time and
space and has several scales. For instance, the combustion process takes place
within a few seconds and centimetres whilst the smoke dispersion evolves at a
scale of some 10–100 km and weeks to months. The most crucial point is that
wildland fires are a dynamic process that interacts with its equally dynamic
2 Introduction
environment (e.g. wind field).

All this together makes it very hard to predict in a deterministic way the
results of any planned management activity. A promising way is to accept
this fact and use management procedures like risk management that explicitly
include the consideration of uncertain events.
1.2 Objectives
The aim of this work was to establish and test a framework that allows the
quantitative spatial analysis of wildland fire risk.
The framework has to consider the particularities of wildland fires (no fixed
starting point) and the wide variety of possible outcomes, e.g. positive impacts
of fires like fuel elimination and successional aspects as well as negative ones such
as destroyed properties or increased susceptibility of the soil to erosion. A crucial
point is the flexibility of the framework that allows a risk manager to “plug-in”
her/his own models and exploit her/his knowledge efficiently. Therefore, only
the risk relevant interfaces are specified and the examples given throughout this
work should serve as a guideline that illustrates the process of applying the
framework to a specific situation.
Using the framework as an integrative environment should help to structure
various research and management aspects of wildland fires.
As the analysis of wildland fire risk requires the handling of spatial data a
geographical information system (GIS) will be applied. By making use of the
GIS’s ability to manage, analyse, and visualise spatial data facilitates greatly
the task.
An important aspect within risk analysis is the attention of uncertainties
and error propagation. Risk analysis deals with probable, future events which
implicitly bear uncertainties. Additionally, the results of a risk analysis usually
imply extensive modelling with spatial data sets that are subject to various kinds
of uncertainties. The objective of this thesis is to demonstrate that uncertainty
propagation in wildland fire behaviour modelling is an aspect of major influence.
1.3 Structure of the thesis

The work is subdivided into three major parts (see figure 1.1). In the introduc-
tion a short review of general aspects of wildland fires and their management
are given. This is followed by a survey of the basic principles of quantitative
risk analysis with special emphasis on uncertainty handling.
Having prepared the field, four papers are included, forming the main part
of the thesis. The order is not chronological but rather governed by content
considerations. The first paper deals mainly with definitions of the fundamental
terms in wildland fire risk analysis. The next paper describes the general layout
of the framework including the formal relations of all elements. The third paper
focuses on the specific but very important aspect of uncertainty propagation in
fire behaviour modelling. And finally, a study is presented that illustrates the
application of the framework in a test area in Southern Switzerland.
The last chapter tries to summarise the various aspects studied throughout
this dissertation project.
1.3 Structure of the thesis 3
Introduction
2 3
Wildland Fire
Quantitative
and Management Ris nalysis
4 5 6

A Consistent Wildland

Framework for Uncertainty Propagation
Fire Ris erminology Wildfire Ris nalysis in Wildland Fire
is Needed! Behaviour Modelling
Papers
7
Wildland Fire Risk
Analysis in
Southern Switzerland
Conclusions
8
Synthesis
Figure 1.1: Structure of the thesis

2
Wildland Fire and

Management
2.1 Wildland Fire

Wildland fire is a multi-scale process in space as well as in time. The core
process, the combustion, takes place in an area of some centimetres within some
seconds. The next phenomenon that is of interest is the propagation of the fire
front at a scale of some 100 meters and minutes to hours. From the start of a
fire to its extinction the extent can reach up to 10’000 square kilometres that
may be burnt within weeks. Additionally, the spatial extent can be considerably
larger looking at indirect effects like smoke and soot plumes. After an actual
fire primary and secondary effects can persist over decades.
There are usually three main topics of wildland fire research:
fire behaviour,
fire occurrence, and
fire effects.
Fire behaviour research studies the physical combustion processes at different

scales and the interaction between the combustion and its environment. The
goal of fire occurrence research is to explain where, when and why a wildland fire
starts. It deals not only with physical preconditions for fires but also with the
socio-economical factors that foster fires such as arson or negligence. Research
into fire effects, finally, deals with primary and secondary impacts of wildland
fires on the landscape, physically and socio-economically.
2.1.1 Fire Behaviour

The behaviour of wildland fire at the landscape scale is the result of very lo-
calised physico-chemical combustion processes and subsequent highly variable
heat transfers. Chandler et al. (1983), Weber (1991), Pyne et al. (1996), Perry
(1998) give extensive introductions to these fundamental processes.
6 Wildland Fire and Management
Fuel
Most generally, fuel denotes combustible, organic particles (fuel particles) ar-
ranged in a three dimensional porous medium, the so called fuel bed. Fuel
particles are described by their chemical and physical properties, fuel beds by
the texture (spatial arrangement) and composition of fuel particles. The par-
ticles consist mainly of cellulose, hemicellulose, lignin, extractives and miner-
alic substances. Physical properties are size, shape, density, and heat content.
The most important parameter for the pyrolysis and combustion process is the
surface-to-area ratio of the particles. Large surfaces compared to a small vol-
ume favour the absorption of radiation and the volatilisation of pyrolysed gases.
Heat content is the energy released by complete combustion per unit quantity
and it does not vary much in organic fuel. Fuel beds are subdivided into vertical
layers, categorised according to ground, surface, and aerial or crown fuel. The
layers show a significantly different fire behaviour. Ground fuels consists of duff,
roots and (rotten) logs lying beneath the current year’s accumulation of fallen
material. Because the fuel is very compact and partially decomposed, fire will
spread slowly and typically smolder. Surface fuel is the most important layer as
most of the fires will start in it. It consists of dead leaves, needles, twigs, and
stems lying on the surface (litter ), living herbs and grass, shrubs and smaller
trees up to a height of around 2 m. Aerial or crown fuel consists of the foliage,
needles, twigs and smaller branches of trees and large shrubs. Generally, crown
fuel consists of living vegetation and thus has a considerable moisture content.
Fire will only spread if there are strong winds and/or sustained heat transfer
from a surface fire.
Physical and Chemical Processes

The behaviour of wildland fires is determined by the following loop back (Weber
1991, Perry 1998):
1. a localised ignition source causing the release of reactive gases and their
flaming combustion (chemical process: oxidation)
2. transfer of the heat by any available mechanism to unburnt fuel (physical

process)
3. absorption of heat by unburnt fuel until combustion starts
The process is sustainable if sufficient energy released in step 1 is transfered

to (step 2) and absorbed by (step 3) unburnt fuel particles.
When a fuel particle is being heated various constituents start evaporating.
At the beginning (up to 100°) mainly water is vaporised absorbing significant
amounts of energy. At 130° – 190° lignin and cellulose and at 200° hemicelluloses
start decomposing. At 280° the reaction becomes mostly exothermic reaching a
maximum at 320°. Pilot ignitions occur at the surface but only at approx. 500°
the charcoal at the surface starts gleaming allowing a self-sustaining reaction.
The energy produced by the oxidation can be transferred to unburnt fuel
particles by means of radiation, convection or conduction. Conduction plays an
important role in smoldering of rotten wood, while radiation and convection are
the key factors for the rate of spread. The heat transfer is very sensitive to the
local geometry of a fire front (heat source) and unburnt fuels. The distance from
2.1 Wildland Fire 7
unburnt fuel particles to a flame front can be considerably shortened if wind or

slope tilts the latter. The rate of spread is mainly depending on the wind speed,
but topography has also a great influence. Assuming constant and homogenous
conditions in a completely flat plain a fire starting from a single ignition point
would spread evenly in every direction. The flame front of this idealistic fire
would form a circle. In presence of wind on a arbitrarily tilted plain the circle
can be best approximated by a double ellipse (Anderson 1983).
Types of Wildland Fires

When a fire has passed the initial state of ignition, i.e. sufficient energy is
released to ignite more unburnt fuel particles, it reaches a steady state of forward
spread rate, burning through the ground fuel bed. This type of wildland fire
is called “surface fire”. Important measures of wildland fire behaviour are the
rate of spread, flame length and intensity. The rate of spread is measured
perpendicularly to the fire line. The fastest rates are measured at the head of a
fire, slower ones on the flanks and slowest on the back. If not stated explicitely
rate of spread denotes the maximal rate of spread at the head of a fire. Rate
of spread is highly influenced by wind and slope. The presence of abundant
fine fuel particles favours higher rates of spread as the heating up is very fast
compared to heavy fuels. Typical rates of spread for surface fires range from 0.3
m/min in litter with no fuel and no slope up to more than 300 m/min in dry,
short grass with high wind (Pyne et al. 1996)1 .
Intensity is heat energy release per unit time and is most commonly calcu-
lated for a unit length of the fire front. This measure is called fireline intensity,
resp. Byram’s intensity (Byram 1959). It is the product of the available heat
and the rate of spread.
I =h·w·r (2.1)
where h is heat per unit mass [kJ/kg], w is fuel load per unit area [kg/m2 ] and r
is rate of spread [m/s]. The fireline intensity, then, is the energy release of a unit
length of the fire front per time unit [kW/m] and it is not feasible to measure
it directly in the field. The reason for the popularity of this measure is its good
correlation with flame length, a property that has a great influence on the choice
of fire suppression activities. In table 2 the standard interpretations of various
flame length ranges are listed as they have been proposed first by Roussopoulos
and Johnson (1975) and are now included in official handbooks of the US Forest
Service, e.g. in the Fire Behavior Field Reference GUIDE (NWCG 1992).
Surface fires are fires that spread through the surface fuel layer. In some sit-
uations it is possible that the fire encroaches to the overstory fuel layer, torching
individual trees or causing a crown fire. The transition from surface to crown
fire depends on the vertical distance between ground and the base of crown
fuel, the foliar moisture content and the bulk density of crown fuel (van Wagner
1977, 1993). Crown fires are categorised in passive, active, and independent.
Passive crown fires depend on a sustained heat flux from a underlying surface
fire. The heat transfer laterally is too weak to support the horizontal spread
1 For comparison: In the USA (NWCG 1997) smoke-jumpers (initial attack parachute fire
fighters) have to pass a fitness test that requires a 1.5-mile run in 11 minutes or less (equivalent
to 200 m/min.) and a 110-pound pack out hike of 3 miles on level terrain in 90 minutes or
less (∼55 m/min with 50 kg). Ordinary fire fighters have to pass a 3-mile (4.8 km) hike with
45 pound (20 kg) in less than 45 minutes (for the arduous level), this is ∼100 m/min.
Flame Fireline inten- Interpretations

length [m] sity [kW/m]
< 1.2 < 350 Fires can generally be attacked at the head
or flanks by persons using handtools.
Hand line should hold the fire.
1.2 – 2.4 350 – 1750 Fires are too intense for direct attack on the
head by persons using handtools.
Hand line cannot be relied on to hold fire.
Equipment such as dozers, pumpers, and
retardant aircraft can be effective.
2.4 – 3.3 1750 – 3500 Fires may present serious control problems:
torching out, crowning, and spotting.
Control efforts at the fire head will probably
be ineffective.
3.3 < 3500 < Crowning, spotting, and major fire runs are
probable.
Control efforts at head of fire are ineffective.
Table 2.1: Fire suppression interpretations of flame length (NWCG 1992, values
converted to metric-units)
within the crown layer. This form represents rather a series of individual trees
torching than a moving flame front. At higher intensities an active crown fire
can be observed: it is still dependent on heat transfer from the surface fire, but
there is now a continuous flame front spreading through the crowns. The crown
fire becomes independent if enough heat is released and steep slopes or strong
winds ensure that the main portion of heat flux remains within the crown fuel.
If intensities become even greater the wildland fire can evolve into a fire
storm. This is the case when the heat of the fire initiates a strong convection
circle with strong “inflowing” winds (wind speeds of over 150 km/h have been
measured). Fires of this severity cannot be contained unless environmental
conditions change significantly.
Modelling Wildland Fire Behaviour

In general, modelling wildland fire behaviour seeks to address “the prediction
of the quantitative parameters which may be used to describe the spread of fire
through a fuel bed” (Perry 1998, p.224). One of the first attempts to model fire
behaviour was by Fons (1946). Generally, there exist three types of models that
try to describe wildland fire behaviour: physical, semi-empirical and empirical
models. There exist also other classification schemes (e.g. Chandler et al. 1983,
Weber 1991) but this is the most coherent one and is used by Perry (1998) who
gives an extensive overview of current approaches to wildland fire modelling.
Physical models are based on the analysis of the underlying physical and
chemical processes. Important models are presented by Albini (1986), Grishin
(1984), Weber (1989). The advantage of this sort of models is that they hold for
2.1 Wildland Fire 9
any set of input parameters because the physical laws are universally applicable.
A major disadvantage is the difficulty to measure key input parameters in the
field and thus applying and verifying the models in a practical test.
Empirical models rely on statistical descriptions of the relationship between
some measured input parameters and their corresponding output parameters
that describe some aspect of wildland fires. A good example is the McArthur
danger meter which predicts the rate of spread of Australian grassland fuel
based on temperature and humidity (McArthur 1966). This sort of models can
only be applied within the conditions they have been established for and a lot
of well documented fires are needed to support sound statistical analysis.
Semi-empirical models have usually a rationale based on physical laws but
the solution of the equations is accomplished by statistical analysis. This is the
case with the most widely known and used Rothermel model (Rothermel 1972)
which forms the basis for computer applications such as FARSITE (Finney1998),
BEHAVE (Andrews 1986, Burgan and Rothermel 1984), SPARKS (Schöning
1996). Additionally, there exists the public available ANSI-C-library fireLib
(Bevins 1996).
2.1.2 Wildland Fire Occurrence

Generally, fire occurrence researchers study where, when and why wildland fires
occur. As said before, we have to consider various scales ranging from regions
(e.g. Mediterranean countries) down to specific locations (in the neighbourhood
of waste-dumping-areas) and from specific decades (in the 60ties) to daytime
specifications (early afternoon).
Wildland Fires in the World

Wildland fires occur virtually anywhere on the globe. The most important as-
pect is the vegetation and climate that favours the building of a sufficiently large
fuel layer. These preconditions are given within a wide range of regions: from
boreal forests to Mediterranean ecosystems and semi-arid regions to grasslands
and savannas. Chandler et al. (1983) give an extensive overview on frequencies
and types of wildland fires all over the world. More recent statistics can be
found in Pyne et al. (1996) and increasingly on the internet (UN-ECE 1997,
NZFS 2001, NIFC 2001, CCFM 2001, Davies 1997). Besides the vegetational
and climatic preconditions wildland fire occurrence is bound to the presence of
ignition sources. An interesting aspect is the distinction between human and
natural causes for both the count and the amount of burnt area (see tables 2.2
and 2.3). In contrast to the Mediterranean countries, the numbers for Canada
show an overwhelming influence of lightning on the burnt areas. In boreal forests
large areas are sparsely populated and fires are likely to be detected only when
they have become already too large to be fought effectively. Furthermore, dur-
ing heavy lightning storms too big a number of fires are starting at the same
time.
2.1.3 Wildland Fire Effects

It is difficult to generalise the effects of wildland fires since they are influenced
by the fire behaviour, the pre- and post-fire state of the ecosystem and other
Country Total Human Natural Unknown

France 6’401 2’435 38% 225 4% 3’741 58%
Greece 1’508 470 31% 71 5% 967 64%
Italy 9’093 7’434 82% 72 1% 1’587 17%
Portugal 29’078 18’319 63% 582 2% 10’177 35%
Russian Federation 32’833 29’680 90% 2’290 7% 863 3%
Spain 16’772 13’181 79% 671 4% 2’920 17%
Australia (Victoria) 376 278 65% 98 26% 3 9%
New Zealand 3’735 3’637 97% 96 3% 2 0%
USA 115’166 99’606 86% 15’560 14% - -
Canada 6’200 2’848 46% 3’239 52% 113 2%
Table 2.2: Number of wildland fires in 1996, distinguished by cause (Sources:

UN-ECE 1997, NZFS 2001, NIFC 2001, CCFM 2001, Davies 1997)
Country Total Human Natural Unknown

France 114 59 52% 256 2% 52 46%
Greece 253 143 56% 10 0% 110 44%
Italy 238 180 76% 85 0% 58 24%
Portugal 830 - - - - - -
Russian Federation 23’119 - - - - - -
Spain 598 480 80% 29 5% 89 15%
Australia (Victoria) 142 22 16% 118 83% 1 1%
New Zealand - - - - - - -
USA 27’122 13’618 50% 13’504 50% - -
Canada 17’487 299 2% 17’187 98% 108 0%
Table 2.3: Burnt areas [km2 ] due to wildland fires in 1996, distinguished by
cause (Sources: UN-ECE 1997, NZFS 2001, NIFC 2001, CCFM 2001, Davies
1997)
independent environmental influences. Considering the spatial and temporal

variability of all these factors one can easily imagine the variety of possible fire
effects.
Nevertheless, the basic “impact” of fires on the main compartments of an
ecosystem can be described in general terms.
Soil A key factor is the moisture content of the soil affecting specific heat
and thermal conductivity. The direct thermal impact of a fire rarely reaches
deeper than 10 cm. A low intensity fire produces maximum temperatures at
the soil surface between 100°C and 250°C and generally, the litter and duff
layer will be reduced but rarely completely, i.e. bare soil might be uncovered
at few exposed places. High intensity fires show temperatures between 500°C
and 750°C at the surface and a complete removal of duff and litter leaving a
bare soil covered with ”white ash”. The temperature profile of the soil can be
a problem with respect to the changed microclimate: by removing canopy, duff
and litter layers and darkening with residual soot surface temperatures can be
increased significantly.
2.1 Wildland Fire 11
The chemical properties of soils are affected directly in the form of changes
of mineral structure induced through heat impact, e.g. decomposition of clay
minerals. Indirectly, the burning of organic matter produces inorganic residues
that may infiltrate and interact with the soil. The destruction of the vegetation
and the removal of the duff and litter layers can lead to increased erosion rates
either by water or wind.
Air The aerial emission of wildland fires consists of solid, liquid and gaseous
intermediate hydrocarbons and inorganic residues. Depending on the concen-
trations of noxious pollutants and the duration of exposure smoke imissions can
pose major health problems in the form of irritations of the respiratory system
and the eye. Beside the chemical imissions reduced visibility can be a major
problem for ground and air traffic, too. The dispersion of smoke depends on the
size and intensity of a fire and the atmospheric stability. A recent major event
was the conflagration in 1997 in Kalimantan and Sumatra where at the climax
an area of 3 Mio. km2 was covered by the haze vitiating 300 Mio. people during
several weeks (Heil 1998).
Vegetation There exists abundant literature on fire effects on the vegetation,

mostly on specific plant species and/or vegetation communities. For instance,
in the searchable online-database provided by the US forest service, over 27’000
scientific references are compiled, covering fire effect information for over 1’100
North American plants, animals and community types (USDA 2001).
Fires have long been part of various ecosystems which have developed differ-
ent strategies to adapt to their effects. Two main topics are fire tolerance and
regeneration. Fire tolerant species can survive the direct impact of flames by
means of fire-resistant foliage, thick bark that insulates and protects the cam-
bium and / or by shedding lower branches. Obviously, this strategy is successful
only below a certain intensity. The other strategy aims at fast regeneration in
the form of either resprouting or fast recolonisation of burnt areas. Resprouting
can start from buds protected by bark or buried in the ground. Noteworthy
is the adaptations of seeds, namely the dormancy and serotinous cones. Seeds
that remain viable in the soil, are usually wrapped with a seed coat that inhibits
the emergence of radicles and/or the uptake of water. For the germination it is
necessary that the seed coat is broken physically which can be accomplished by
the heat impact of a fire. A special case is the serotinous cone. This cone does
not release the mature seed because it is sealed with surface resin. During a fire
the resin melts and the cone releases the seed.
Wildlife Invertebrates that live mainly in the duff and litter layers are very
susceptible to fire. Vertebrates are usually only little affected by fires: either
they are able to evade or hide in safe places like water courses or holes. Smaller
animals such as rodents can survive in their underground nests as the soil is an
excellent insulator. Birds and flying insects usually escape fires easily, however
ground nesting species may loose eggs or brood.
In most cases the biggest effect is caused by a significantly changed habitat
after the fire.
Buildings and Structures Effects on buildings and other structures are not
core topics of the wildland fire research community. An exception is the struc-
ture ignition assessment model (SIAM) of Cohen et al. (1991). SIAM considers
structure and site characteristics and fire behaviour and is based on physical
principles. A more qualitative approach based on rules of thumb can be found
in the internet at www.firewise.org.
2.2 Wildland Fire Management

2.2.1 Introduction
The prevention and mitigation of wildland fires includes a wide spectrum of
activities. Knowing just a little about the complexity of ecosystems it is clear
that they cannot be limited to purely reactive measures such as fire fighting.
Wildland fire management starts much earlier and includes activities such as
fuel management, prescribed burning, educating and increasing awareness on
fire aspects, etc. Although usually not explicitely stated or acknowledged, fire
management must be seen as a part of a comprehensive environmental manage-
ment strategy whose goal is a sustainable use of natural resources in order to
improve the quality of living (Barrow 1999). Due to the fact that fires are not
continuous processes but rather erratic events in an ecosystem, risk manage-
ment procedures are applied as they allow the explicit consideration of random
events.
2.2.2 Wildland Fire Management Activities

Management activities can be classified into three groups depending on whether
they are applied before, during or after a fire.
Prevention Prevention primarily seeks to reduce the unplanned ignitions. As

this is only seldomly, if at all, feasible a second goal is to prevent the
development of extreme fires.
Reducing unplanned ignitions depends on a clear identification of causes
and circumstances. In the case of natural causes no promising strategies
do exist. Experiments in cloud-seeding to inhibit lightning storms have
not yet yielded in applicable methods (Fleagle and Businger 1980). Re-
ducing human caused ignitions can be achieved by means of education
and information, legislation and enforcement. Education and information
try to raise the awareness and knowledge of the public or specific target
groups (e.g. farmers traditionally burning wasteland and harvested fields).
Legislation and enforcement regulate and prescribe fire related behaviour.
Concerning legislation it is important to have a broad consideration not
only of nominal and functional law but also to integrate economical and
social circumstances, e.g. arson as a result of inappropriate land manage-
ment policies.
Controlling fire behaviour can be done by controlling the fuel, i.e. reducing
the amount of fuel or altering the fuel type and structure. The prevention
of excessive fuel accumulation (i.e., fuel reduction) can be done by phys-
ically removing fuel in laborious manual work or by using fire. Another,
2.2 Wildland Fire Management 13
complementary tool is the building of fuel breaks. Fuel breaks are strips
that contain no or sparse fuel. The width of the strips depends on the fuel
situation and on the anticipated fire behaviour. Additionally, fuel breaks
often serve as access routes and safety zones for fire fighting resources. A
variant of fuel breaks are so called greenbelts. The indigenous vegetation
is replaced with less flammable species. A driving force behind this sort
of fuel management is to split large areas of homogenous vegetation and
establish a diverse mosaic of fuel types.
Suppression Fire fighting is still a laborious task, despite the increasing mo-
torisation and the deployment of air-borne techniques. Unless the en-
vironmental conditions are favourable (i.e. abundant precipitation) the
containment of a wildland fire requires still a lot of handwork.
The idea of early detection and effective initial attacks is to extinguish
fires before they evolve to uncontrollable conflagrations. Early detection
of fires is achieved by (temporally) increased surveillance which in the
simplest form consists of (motorised) patrols and manned look out posts.
Automatic detection using video/infrared cameras from fixed or space/air-
borne platforms is gaining more interest, but there are still problems of
false alarms and large time lags.
If a fire is detected fast deployment of initial attacks is crucial. Topics of
interest are optimisation of access routes, air attacks with helicopters and
air-planes, and smoke jumping.
Mitigation The effects of wildland fires are eased primarily by economically
supporting private persons, communities and companies that have suf-
fered damage. As long as the effects are small this is mostly a field of
activity of private insurance companies. In the case of greater conflagra-
tion public and international funds will normally be activated. Besides
the financial mitigation, activities such as reforestation are carried out to
inhibit secondary effects.
2.2.3 Existing Wildland Fire Managements Approaches

Simard (1976) defines wildland fire management as:
The application of management, physical, and ecological principles
to the management of the wildland fire process so as to render the
impact of wildland fire on the natural resource base, the ecosystem,
and the environment consistent with the goals of the managing or-
ganisation. Wildland fire management includes the traditional fire
control related activities (suppression, detection, etc.) as well as
broader relationships between fire and wildland management, such
as prescribed fire and fuel management.
A more general approach is described by Barrow (1999):
The overall goal of environmental management is a sustainable use
of natural resources in order to improve the quality of human living.
While Simard stresses the “goals of the managing organisation”, Barrow
explicitly names the goal, i.e. “improve the quality of human living.” In both
cases it becomes clear that widland fire management is driven by the perception
and the needs of society. Usually, society is not a uniform, unanimous entity
but rather a conglomerate of various groups with presumably different goals.
This requires a transparent and democratic decision making process.
2.3 Risk Terminology

2.3.1 Introduction
The following discussion of risk terminology constitutes the theoretical basis for
chapter 4. Its intention is to clarify the concepts of risk by reviewing its different
notions in various situations such as daily life.
When looking through wildfire risk related literature one notices a great
confusion on the proper use of terminology and, due to that, the absence of a
comprehensive methodology. Further, models and computer simulations shoot
up like mushrooms; verification procedures, on the other hand, are a difficult
task. All this leads to the situation that comparisons of the various investiga-
tions carried out in this field may be impossible or at least debatable. Hence,
results may not be comparable and thus of no use for the improvement of wildfire
management. Moreover, the somewhat inconsiderate use of the various terms
danger, hazard and risk may result in misunderstandings that – in the worst
case – can have fatal consequences. Both, managers and researchers depend
on comprehensive, reliable communication facilities. Much of the success of
any management organisation or process depends on the profound knowledge
of the boundary conditions and predefined rules. Some reasons for the lack of
an overall approach to wildfire risk may result from the following facts:
The term risk is part of everyday life where, depending on the context, a
wide range of notions is assigned to it. Terminology is always to some extent
a linguistic and / or cultural issue. Every language has its own words and
associated meanings. For instance, the terms hazard and danger are the same
word in German (i.e. Gefahr ).
The phenomenon fire has as many aspects as people who are dealing with it:
Fire managers and fighters, environmentalists, foresters, house and land owners,
scientists, land planing organisations, etc. Based on their primary interests, each
of these communities has different notions of the term wildfire risk.
For the purposes of this thesis, we present a coherent wildfire risk terminol-
ogy that is based on the experiences and achievements in technical risk engineer-
ing (see also chapter 4). We consider the application of this sound methodology
and terminology appropriate to carry out quantitative risk analysis in the con-
text of wildland fire management. A further objective is to show what part
various topics of wildland fire research play in the context of wildland fire risk
analysis. We believe that this will open up new insights to the phenomenon
wildland fire as it reveals relationships between different factors which may not
be obvious at first sight. Finally, we recognise that a terminology is only then a
useful terminology when it is accepted and applied by a majority of the people
involved. We do not want to impose a new way of thinking, but we would like
to stimulate a discussion that will help to establish a consistent wildland fire
risk terminology and methodology.
2.4 Definitions 15
2.3.2 Scope
Definitions are constructed and valid only within a given scope (Seiffert 1997).
They combine words with notions of, sometimes, complex phenomena in a
unique way. Definitions are essential for the reliable communication between
involved people working on the same topic. They are used as abbreviations for
complex and difficult to explain matters. Definitions are never true or false, but
useful or not useful within the scope they are applied.
The scope of the anticipated terminology is the application of technical risk
engineering (in particular quantitative risk analysis) in the context of wildland
fire management and research, explicitly considering spatio-temporal compo-
nents. In the following, some of the known definitions for danger hazard and
risk are discussed. In a next step, we will then propose our definitions that are
tailored for the use within the mentioned scope.
2.4 Definitions
Danger Fire bound definitions of the term danger tend to be vague; they
reveal the difficulties of explaining a phenomenon that is mainly based on hu-
man perception. Wildland fires as a natural process are neither bad nor good;
only when human optics come into play they are valued. Hence, some definitions
recapitulate the components (physical preconditions and triggering factors [igni-
tion sources]) that lead to an (undesired) event, some even include management
activities into the definitions or refer to indices that express fire danger.
However, none of these definitions really tries to characterise the phenome-
non danger per se. Consulting a common dictionary, Webster’s College Dictio-
nary (1992), a more abstract but more appropriate approach can be found that
describes danger as a set of menacing circumstances:
(1) liability or exposure to harm or injury; risk; peril. (2) an instance

or cause of peril; menace. (...) Danger is the general word for
liability to injury or harm, either near at hand and certain, or remote
and doubtful.
In the following, two wildland fire standard glossaries are cited. One was
published by the Food and Agricultural Organisation of the United Nations
(further referred to as FAO 1986) and the other was edited by the Canadian
Committee on Forest Fire Management (Poulin et al. 1987, further referred to
as CCFFM 1987).
FAO (1986) defines fire danger as follows:
Fire Danger: The resultant, often expressed as an index, of both

constant and variable factors affecting the inception, spread, and
difficulty of control of fires and the damage they cause.
CCFFM (1987) states:
A general term used to express an assessment of both fixed and vari-

able factors of the fire environment that determines the ease of igni-
tion, rate of spread, difficulty of control, and fire impact.
Both definitions use the term danger to describe the set of preconditions
that influence wildland fires. However, in order to distinguish the contribution
of each factor it would be more appropriate not to mix them but to consider
them separately. E.g., the ease of ignition is relevant for the probability of occur-
rence while difficulty of control is related to the outcome. As said before, these
definitions do not explain the term danger but summarise what may influence
and describe fires; vaguely the possibility of an event taking place is expressed
as well. The same is true for the Glossary of Forestry Terms by the Ministry
of Forests (Province of British Columbia, Canada; further referred to as MOF
(1997)):
Fire danger: an assessment of both fixed and variable factors of
the fire environment, which determine the ease of ignition, rate of
spread, difficulty of control, and the fire impact.
2.4.1 Hazard
Although often used as a synonym for danger, definitions of the term hazard
add a more precise, more materialised notion by identifying hazard objects.
Eventually the relation to possible (human) harm is established as well. The
following first definition in Webster’s College Dictionary (1992) comes close to
the intention of materialising danger:
(1) something causing danger, peril, risk, or difficulty (...) (2) the
absence or lack of predictability; chance; uncertainty (...) (8) to take
or run the risk of a misfortune, penalty etc.
Both, the FAO and the CCFFM glossaries give very narrow definitions where
hazard is mainly related to the fuel complex and its properties. FAO (1986):
(1) (North America) A fuel complex, defined by volume, type con-
dition, arrangement, and location, that determines the degree both
of ease of ignition and of fire suppression difficulty. (2) (non-US
English speaking world) A measure of that part of the fire danger
contributed by fuels available for burning. Note: is worked out from
their relative amount, type, and condition, particularly their mois-
ture content.
Both variants define hazard as some sort of measure for a specific aspect
of fire behaviour related to fuels properties (“ease of ignition ... fire suppres-
sion difficulty and “fire danger”). The CCFFM glossary (1987) restricts the
definition even more. It concentrates on the description of “endogenous” fuel
properties and excludes the influence of “exogenous” properties like fuel mois-
ture and/or suppression activities explicitly:
A general term to describe the potential fire behaviour, without regard
to the state of weather-influenced fuel moisture content and / or
resistance to fireguard construction for a given fuel type. This may
be expressed in either the absolute (e.g. ‘cured grass is a fire hazard’)
or comparative (e.g. ‘clear cut logging slash is a greater fire hazard
than a deciduous cover type’) sense. Such an assessment is based
on physical fuel characteristics (e.g. fuel arrangement, fuel load,
condition of herbaceous vegetation, presence of ladder fuels).
2.4 Definitions 17
MOF (1997) defines fire hazard also as a function of the fuel properties:
Fire hazard: the potential fire behavior for a fuel type, regardless
of the fuel type’s weather-influenced fuel moisture content or its re-
sistance to fireguard construction. Assessment is based on physical
fuel characteristics, such as fuel arrangement, fuel load, condition of
herbaceous vegetation, and presence of elevated fuels.
In the domain of technical risk engineering, hazard is explained by a more

general concept. Allen (1992, p. 9) defines hazard as follows:
Hazard: a physical situation with a potential for human injury, dam-

age to property, damage to the environment or some combination of
these.
A hazard is not only the precondition for a specific process, as it is seen

by the FAO, CCFFM, and MOF glossaries, but it is the process itself. This
definition has the advantage that it is applicable to any process that can lead
to damage. Hence, according to this definition, wildland fire is the hazard as
are debris flows or avalanches.
2.4.2 Risk
The term risk is used by very different communities and in various situations.
Accordingly, a very wide range of notions is assigned to it. Nevertheless, risk
seems to be attributed to two meaning complexes: loss, harm and injury on
the one hand and chance and probability on the other hand. Most definitions
show difficulties, though, to clearly express this duality or they just include one
part into their explanations. Taken as a whole, the set of definitions given by
Webster’s College Dictionary (1992) comes close to the two components of risk:
(1) exposure to the chance of injury or loss; a hazard or dangerous

chance. (2) Insurance: (...) b. the degree of probability of such loss,
c. the amount that the insurance company may loose (...).
The following definitions presented by FAO and CCFFM are examples for
explanations that look at one part of risk only. According to FAO (1986) risk
is:
(1) The chance of fire starting, as affected by the nature and inci-
dence of causative agencies; an element of the fire danger in any
area. (2) Any causative agency.
The FAO definitions do not include the notion of damage but extend the
term risk with causative agency. Risk is in that sense not only an abstract
descriptive property but it becomes a special type of action. The same is true
for the definition given by CCFFM (1987):
The probability or chance of a fire starting determined by the pres-

ence and activities of causative agents (i.e. potential number of ig-
nition sources).
Both definitions represent a twofold cutback of the term risk: they neglect
the outcome (i.e. damage) as well as the preconditions and concentrate on the
triggering factors of a fire (ignition sources) only. MOF (1997) comes closest to
the dual approach towards risk:
Risk: the probability of an undesirable event occurring within a spec-

ified period of time. With regard to insect populations, risk involves
components to evaluate the likelihood of an outbreak, the likelihood
of trees being attacked (susceptibility) or the likelihood of trees be-
ing damaged (vulnerability). In fire prevention, risk involves those
things or events that cause fires to start (including the physical ig-
niting agents and people).
However, when defining fire risk, the same glossary restricts the definition
again to the aspect of occurrence probability and just cites the CCFFM glossary:
Fire risk: the probability or chance of fire starting determined by the

presence and activities of causative agents.
In wildland fire oriented literature only very few examples can be found that
take into account both aspects of risk probability and outcome. In contrast,
Hall (1992), coming from structural fire science, refers to fire risk as being the
product of probability and damage:
A measure of fire risk has two parties: (1) a measure of the expected
severity (e.g., how many deaths, injuries, dollars or damage per fire)
for all fires or for a particular type of fire, and (2) a measure of the
probability of occurrence of all fires or of that particular type of fire.
In general, a fire risk measure will be a product of an expected severity

term and a probability term or a sum of such products. Unfortunately, Hall’s
definition seems to have lacked the necessary support and / or diffusion in the
wildland fire research community.
In the area of technical risk engineering, we find the same rationale for the
term risk (e.g. Jones (1992), Gheorghe and Seiler (1996), Merz et al. (1995)).
In the most general case, risk comprises likelihood and outcome (often referred
to as damage or damage potential) and for quantitative risk analysis it is the
product of both (Jones 1992).
We would now like to give a general definition of the previously discussed
terms. In chapter 4 we specify them in greater details with respect to their
application in quantitative risk analysis.
Danger: Danger is an abstract concept of human perception. Danger per se

does not exist. It is defined by subjective human and societal perception
and assessment of factors (of the physical and non-physical environment)
that are considered harmful.
Hazard: A hazard is a process leading to undesirable outcomes.
Risk: Risk comprises the probability of an undesired event and the outcome of
it. An undesired event is a realization of a hazard.
2.5 Risk Management 19
For this work, we will neither use nor define the terms danger and fire danger
anymore, while the term hazard will be used as a synonym for the process itself
(i.e. wildland fire). Furthermore, what is usually thought of as hazard or
danger should be looked at as either a precondition (or a set of preconditions)
of a fire (e.g. fuel properties) or as a triggering factor (i.e. ignition cause).
Both elements are essential for a fire, but only a part of the puzzle. In the
anticipated GIS-based framework for quantitative wildland fire risk analysis,
however, these elements can and must be positioned and assessed according to
their contribution to fire risk.
Consequently, we will speak of risk (i.e.wildland fire risk ) alone and more
specifically of quantitative risk as this allows the embedding of it in a risk
management process, where risk analysis and also risk assessment are important
steps (Bärtsch 1998). Moreover, we use the value-free term outcome, as the
effects of a wildland fire do not necessarily have to be negative.
2.5 Risk Management

Risk management can be seen as a process consisting of the following steps
(Zogg 1981, Jones 1992, Bärtsch 1998):
hazard identification
risk analysis
risk assessment
activities / handling
controlling
Hazard identification marks the start of the risk management process: Un-
desired events are identified and their general behaviour is analysed. At this
stage, a comprehensive statement of facts on the hazard is compiled. The risk
analysis step aims at determining probabilities and impacts of the hazard re-
specting the given situation. Determination of probabilities and impacts can be
done quantitatively or qualitatively, but should be as objective as possible. In
chapter 3 we will discuss in greater detail the quantitative risk analysis. The
risk assessment step compares the obtained risk values with the management
goals and socio-economic constraints. Usually, the risk of a hazard is either
acceptable, non-acceptable or lies somewhere in between. In this latter case,
a cost-benefit-analysis is applied to determine measures that lower the risk.
Non-acceptable risks require counter-measures to be applied without consider-
ing costs. Depending on the risk level, management activities are evaluated
and applied. Finally, the whole process is controlled and reinitiated regularly
or upon special demand, e.g. dramatically changing environmental conditions.
2.6 Geographical Information Systems and

Wildland Fire Management
The functionality of GI Systems is generally grouped into 1) data input 2) data
management 3) data transformation 4) data output and 5) user interaction
20 Bibliography
(Burrough and McDonnell 1998, Jones 1997). Data denotes not only spatial
data but also attribute data. The integrative character of GIS and its applica-
tion in the field of decision support (Malczewski 1999), environmental modelling
(Goodchild et al. 1993) and risk analysis (Malczewski 1999, Hewitt 1993, Re-
jeski 1993, Wadge et al. 1993, von Braun 1993, Parrish et al. 1993) is widely
acknowledged.
GIS-technology is applied for various aspects of wildland fire management.
For a comprehensive list of the possible roles of GIS in the context of wildland
fire management refer to Salazar L. A. (1989), Chuvieco et al. (1997). A widely
used field of GIS application is the delineation of zones that require special
management activities, like zones suitable for prescribed burning (Chou 1992)
or priority zones for suppression (Gronlund et al. 1994, Van Wagtendonk 1990,
Eftichidis G. 1994, Chuvieco and Salas 1996). GIS functionalities are also in-
tegrated into operational decision support systems in the context of planning
and coordination of suppression activities (Wybo and Meunier 1993, Beck 1991,
Richardson 1993).
In this thesis the GIS ArcInfo will be used as a primary data handling,
modelling and analysing environment. This will facilitate the work as standard
functionality are readily available. Additional functionality can usually be im-
plemented with some script or macro language. For instance, for this work we
used the basic AML-scripts to model fire behaviour and fire spread time which
have been implemented by Schöning (1996).
Bibliography
Albini, F.A., 1986, Wildland fire spread by radiation - a model including fuel
cooling by natural convection. Combustion Science and Technology, 45:101–
113.
Allen, F.R., 1992, The management of risk to society from potential accidents.
Elsevier Applied Science, London.
Anderson, H. E., 1983, Predicting wind-driven wildland fire size and shape.
Research Paper INT-69, USDA Forest Service, Intermountain Forest and
Range Experiment Station.
Andrews, P. L., 1986, Behave: Fire behavior prediction and fuel modeling
system - burn subsystem, part 1. General Technical Report INT-194, USDA
Forest Service, Intermountain Research Station.
Barrow, C.J., 1999, Environmental management: principles and practice.
Routledge environmental management series, Routledge, London.
Bärtsch, A., 1998, Konzeption eines Waldbrandmanagements für die Region
Engiadina Bassa/Val Müstair und den Schweizerischen Nationalpark. Mas-
ter’s thesis, Deptarment of Geography, University of Zurich, Zurich, Switzer-
land.
Beck, J.A., 1991, GIS and modelling technologies for fire research and man-
agement in western australia. 11th Conference on fire and forest meteorology,
edited by Andrews P. L. Potts D. F., pp. 9–16, Society of American Foresters,
Missoula, MT.
Bibliography 21
Bevins, Collin D., 1996, fireLib user manual and technical reference. Inter-
mountain Research Station, Forest Service, U.S. Department of Agriculture.
URL http://www.montana.com/sem/public_html/firelib/firelib.html
Burgan, R. E. and Rothermel, R. C., 1984, Behave: fire behavior predic-

tion and fuel modelling system - fuel subsystem. General Technical Report
INT-167, USDA Forest Service, Intermountain Forest and Range Experiment
Station.
Burrough, Peter. A. and McDonnell, Rachael A., 1998, Principles of

Geographical Information Systems. Oxford University Press Inc., Oxford, GB.
Byram, G.M., 1959, Combustion of forest fuels. Forest Fire Control and Use,
edited by K.P. Davis, pp. 61–89, McGraw Hill, New York.
CCFM, Canadian Council of Forest Ministers, 2001, National forestry

database: Fire statistics 1996. Internet.
URL http://nfdp.ccfm.org/summary/broch_e/qf3e.htm
Chandler, C., Cheney, P., Thomas, P., Trabaud, L. and Williams, D.,
1983, Fire in forestry. Forest Fire Behavior and Effects, vol. I. John Wiley &
Sons, New York.
Chou, Y.H., 1992, Management of wildfires with a geographical information

system. International Journal of Geographic Information Systems, 6(2):123–
140.
Chuvieco, E. and Salas, J., 1996, Mapping the spatial distribution of for-
est fire danger using GIS. International Journal of Geographic Information
Systems, 10(3):333–345.
Chuvieco, Emilio, Salas, Francisco Javier and Vega, Cristina, 1997,

Remote sensing and GIS for long-term fire risk mapping. A review of remote
sensing methods for the study of large wildland fires, edited by Emilio Chu-
vieco, pp. 91–108, Departemento de Geografı́a, Universidad de Alcalá, Alcalá
de Henares, E, megafires project ENV-CT96-0256.
Cohen, J.D., Chase, R.A., LeVan, S.L. and Tran, H.C., 1991, A model for
assessing potential structure ignitions in the wildland/urban interface. 11th
Conference on fire and forest meteorology, edited by Andrews P. L. Potts D.
F., pp. 50–57, Society of American Foresters, Missoula, MT.
Davies, Chris, 1997, Analysis of fire causes on or threatening public land

in victoria 1976/77–1995/96. Research Report 49, Fire Management Branch,
Departement of Natural Resources and Environment, Victoria, AU.
Eftichidis G., Xanthopoulos G., Varela V., 1994, A preparedness plan-

ning frame for forest fire prevention. 2nd Int. Conference on Forest Fire Re-
search, pp. 105–112, Coimbra, Portugal.
FAO, (Food and Agricultural Organization), 1986, Wildfire manage-

ment terminology. No. 70 in Forestry Paper, United Nations, FAO, Forest
Resources Development Branch, Rome, Italy.
22 Bibliography
Fleagle, Robert G. and Businger, Joost A., 1980, An Introduction to

Atmospheric Physics, vol. 25. Academic Press, second edn.
Fons, W.L., 1946, Analysis of fire spread in light forest fuels. Journal of Agri-
cultural Research, 72:93–121.
Gheorghe, A. and Seiler, H., 1996, Was ist Wahrscheinlichkeit? Die Bedeu-
tung der Wahrscheinlichkeit beim Umgang mit technischen Risiken. Polypro-
jekt Risiko und Sicherheit Dokument Nr. 17, ETH (Federal Institute of Tech-
nology Zurich).
Goodchild, Michael F., Parks, Bradley O. and Steyaert, Louis T.

(Editors), 1993, Environmental Modeling with GIS. Oxford University Press,
New York, USA.
Grishin, A.M., 1984, Steady-state propagation of the front of a high-level

forest fire. Soviet Physics Doklady, 29:917–919.
Gronlund, A. G., Xiang, W.-N. and Sox, H., 1994, GIS, expert system
technologies improve forest fire management techniques. GIS World, Febru-
ary:32–36.
Hall, J. R. Jr., 1992, Fire risk analysis. Fire hazard and fire risk assessment,
edited by Hirschler M. M., pp. (21–10)–(21–18), American Society for Testing
and Materials (ASTM), Philadelphia, PA, papers presented at the symposium
of the same name, held in San Antonio, TX on 3 Dec. 1990.
Heil, Angelika, 1998, Fire-related transboundary haze and air pollution in

south-east asia 1997. Internet.
URL http://www.icsea.or.id/sea-span/SCIPOL2/STUDY423.htm
Hewitt, Mason J., 1993, Risk and hazard modelling. Environmental Modeling
with GIS, edited by Michael F. Goodchild, Bradley O. Parks and Louis T.
Steyaert, p. 317, Oxford University Press, New York, USA.
Jones, Christopher B., 1997, Geographical Information Systems and Com-

puter Cartography. Longman, Essex, GB.
Jones, D. A. (Editor), 1992, Nomenclature of hazard and risk assessment in the

process industries, Institution of Chemical Engineers, Rugby, Warwickshire,
UK.
Malczewski, Jacek, 1999, GIS and multicriteria decision analysis. John Wi-
ley & Sons, New York.
McArthur, A.G., 1966, Weather and grassland fire behaviour. Tech. rep.,
Commonwealth Australia, Department of National Development, Forest and
Timber Bureau, leaflet No 100.
Merz, H.A., Schneider, Th. and Bohnenblust, H., 1995, Bewertung

von technischen Risiken. No. 3 in Polyprojekt Risiko und Sicherheit, vdf
Hochschulverlag AG, Zürich.
Bibliography 23
MOF, Ministery of Forests, 1997, Glossary of forestry terms. province of

british columbia. Internet.
URL http://www.for.gov.bc.ca/pba/publctns/glossary/glossary.htm
NIFC, National Interagency Fire Center, 2001, Wildland fire statistics

1996. Internet.
URL http://www.nifc.gov/stats/wildlandfirestats.html
NWCG, 1997, Fitness and Work Capacity. National Wildfire Coordinating

Group, second edn.
NWCG, National Wildfire Coordinating Group, 1992, Fire behavior

field reference GUIDE. United States Department of Agriculture.
NZFS, New Zealand Fire Service, 2001, Fire statistics 1996. Internet.
URL http://www.fire.org.nz/more_info/statistics/PDFs/
Statistics.pdf
Parrish, David A., Townsend, Laura, Saunders, Jerry, Carney, Ger-

ald and Langston, Carol, 1993, U.s. epa region 6 coparative risk project:
Evaluating ecological risk. Environmental Modeling with GIS, edited by
Michael F. Goodchild, Bradley O. Parks and Louis T. Steyaert, pp. 348–352,
Oxford University Press, New York, USA.
Perry, G.L.W., 1998, Current approaches to modelling the spread of wildland

fire: a review. Progress in Physical Geography, 22(2):222–245.
Poulin, S., Merrill, D.F. and Alexander, M.E., 1987, Glossary of Forest
Fire Management Terms. National Research Council of Canada, Ottawa, CA,
fourth edn.
Pyne, Stephen J., Andrews, Patricia L. and Laven, Richard D., 1996,
Introduction to wildland fire. John Wiley & Sons.
Rejeski, David, 1993, GIS and risk: A three culture problem. Environmental
Modeling with GIS, edited by Michael F. Goodchild, Bradley O. Parks and
Louis T. Steyaert, pp. 318–331, Oxford University Press, New York, USA.
Richardson, A., 1993, GIS and imagery join forces to combat wildland fire.
13th Annual ESRI User Conference, pp. 429–434, Palm Springs, USA.
Rothermel, R. C., 1972, A mathematical model for predicting fire spread

in wildland fuels. General Technical Report INT-115, USDA Forest Service,
Intermountain Forest and Range Experiment Station.
Roussopoulos, P.J. and Johnson, V.J., 1975, Help in making fuel manage-
ment decisions. Research paper, USDA Forest Service, North Central Exper-
iment Station.
Salazar L. A., Nilsson C. V., 1989, Integrating geographic information

systems into fire management. 10th Conference on fire and forest meteorology,
edited by MacIver D. C. Auld H. Whitewood R., pp. 311–316, Society of
American Foresters, Ottawa, Canada.
24 Bibliography
Schöning, R., 1996, Modellierung des potentiellen Waldbrandverhaltens mit

einem Geographischen Informationssystem. MSc. Thesis, Department of Ge-
ography, University of Zurich, Zurich, Switzerland.
Seiffert, Helmut, 1997, Einführung in die Wissenschaftstheorie, vol. 4. Beck,
München.
Simard, A. J., 1976, Wildland fire management: the economics of policy al-
ternatives. Forestry Technical Report 15, Department of Fisheries and Envi-
ronment, Canadian Forestry Service.
UN-ECE, UN/ECE Timber Committee, 1997, Forest fire statistics 1995-
1997.
URL http://www.unece.org/trade/timber/ff-stats/ff-95-97.htm
USDA, U.S. Department of Agriculture, Forest Service, Rocky
Mountain Research Station, Fire Sciences Laboratory, 2001, Fire
effects information system. Online, accessed: 20.4.2001.
URL http://www.fs.fed.us/database/feis/
van Wagner, C.E., 1977, Conditions for the start and spread of crown fires.
Canadian Journal of Forest Research, 7:23–34.
van Wagner, C.E., 1993, Prediction of crown fire behavior in two stands of
jack pine. Canadian Journal of Forest Research, 23:442–449.
Van Wagtendonk, J.W., 1990, GIS applications in fire management and

research. Fire and the environment: ecological and cultural perspectives, edited
by S.C. Nodvin and T.A. Waldrop, vol. General Technical Report SE-69, pp.
212–214, USDA Forest Service, Knoxville, Tennessee.
von Braun, Margrit, 1993, The use of GIS in assessing exposure and reme-
dial alternatives at superfund sites. Environmental Modeling with GIS, edited
by Michael F. Goodchild, Bradley O. Parks and Louis T. Steyaert, pp. 339–
347, Oxford University Press, New York, USA.
Wadge, G., Wislocki, A.P. and Pearson, E.J., 1993, Spatial analysis in
GIS for natural hazard assessment. Environmental Modeling with GIS, edited
Weber, R.O., 1989, Analytical models for fire spread due to radiation. Com-
bustion and Flame, 78:398–408.
Weber, R.O., 1991, Modelling fire spread through fuel beds. Progress in En-
ergy and Combustion Science, 17:67–82.
Wybo, J.-L. and Meunier, E., 1993, Architecture of a decision support system
for forest fire prevention and fighting. 1993 Int. emergency management and
engineering conference: Research and applications, edited by J.D. Sullivan,
pp. 186–190, IEEE Computer Society, Arlington, Virginia.
Zogg, H. A., 1981, ”Zürich”-Gefahrenanalyse: Grundprinzipien. Tech. rep.,

Zürich Versincherungs-Gruppe.
3
Quantitative Risk Analysis
3.1 Introduction
Risk analysis deals with probable, future events. Consider the following state-
ment: “Most probably it will rain this afternoon”. The event consists of water
drops falling from the sky and the probability is expressed by the term “most
probably”. For the decision to take an umbrella along this information is suffi-
cient for most “decision makers”. For other decision processes it is necessary to
have more and more precise risk information. Quantitative risk analysis tries
to operationalise “probability” and “impact”. In order to determine risk values
scenarios have to be defined. Scenarios represent future, probable events and are
constructed according to the management organisation’s goals and needs. For
instance, “worst case” scenarios are constructed to study events that produce
the most unwanted consequences. Other scenarios are constructed to represent
“every day” events.
As already mentioned, risk consists of the probability p and the impact d of
an event. The most common operationalisation of risk is the multiplication of
probability and impact:
r =p·d (3.1)
To apply this general formula in a spatial context we will first define the
components of the risk analysis and then explain their relationships. The fol-
lowing description of the risk analysis components is largely based on Merz et al.
(1995).
3.2 Definitions
For a quantitative risk analysis we will assume a closed and well defined system.
Within this system we have scenarios Si which are the so called risk donors and
objects at risk Oj which are the risk acceptors.
26 Quantitative Risk Analysis
3.2.1 Basic Elements

Scenario Risk Donor
Every scenario Si represents a realisation of an event at a given location in the
area of interest and is described by the probability pi that it will occur.
Objects at Risk Risk Acceptors

An entity that will undergo modifications due to a realization of the hazard.
The decision of what entities are to be considered is made by the managing
organisation. Examples of objects are buildings, roads, and forest stands but
also nature reserves that host rare species. An object must have an impact
function that defines the magnitude of modifications with respect to the hazard’s
impact. A simple example of such a function could be:

4p · 4v if fire line intensity > 500 kW/m,
d= (3.2)
0 otherwise
If the fire line intensity is higher than the given threshold the tree will die
and the impact is determined by the reduced price 4p due to the poorer quality
of burnt wood and the reduced volume 4v due to the premature cutting.
3.2.2 Relationship between Objects at Risk and Scenarios

For every combination of Oi and Sj we have to determine the conditional indi-
vidual impact probability eij and the individual damage dij .
Sj
eij , dij
Oi
Figure 3.1: Components of quantitative risk analysis
eij = conditional, individual impact probability that expresses the probability

that object Oi is affected when scenario Si is known to have happened.
dij = is the potential damage that arises from an impact on object Oi .
If we know the probability pj that the scenario Sj will occur, we can deter-
mine the individual impact probability kij , the probability that an object Oi is
affected by a scenario Sj , by multiplication:
kij = eij · pj (3.3)

3.2 Definitions 27
Based on this basic relationship it is possible to determine the risk for both
the donor and the acceptor. The first is called the collective or scenario risk.
The latter is the individual risk.
3.2.3 Individual Risk of an Object

Most commonly, several scenarios Si contribute to the impact probability of
various objects Oj . From the object’s perspective it is sometimes not interest-
ing to know each scenario’s contribution but to determine the probability to be
affected by any scenario (figure 3.2). This is the so called the impact probabil-
ity ki of object Oi and, assuming m independent but mutually not exclusive
scenarios Sj , it is given by
m
Y m
X
ki = 1 − (1 − kij ) ≈ kij (3.4)
j=1 j=1
The approximation by summation holds only if the individual impact prob-

abilities kij are small.
Individual Risk of
an Object
Impact Probability Expected Damage

of Object
m
Σ kij Σ kij ⋅ dij

m
ki = di =
j =1 j =1
Individual Impact Probability

kij = pj ⋅ eij
Probability of Scenario Conditional Individual Individual Expected Damage

pj Impact Probability dij
eij
Figure 3.2: Contributing factors of the individual risk of an object
The expected impact di of an object Oi is a weighted sum of all individual

impacts dij . The weighting is done with the individual impact probabilities kij :
n
X
di = kij dij (3.5)
j=1
3.2.4 Collective Risk of a Scenario

The probability of a scenario pj is defined explicitly and denotes the likelihood
that it will occur. The expected impact dj is the weighted sum of all individual
expected impacts dij . The weighting is done with conditional individual impact
probabilities eij .
Xn
dj = eij dij (3.6)
i=1
Collective Risk
of a Scenario
Probability of Scenario Expected Damage

pj of Scenario
n
dj = Σ eij ⋅ dij
i =1
Conditional Individual Individual Expected Damage

Impact Probability dij
eij
Figure 3.3: Collective risk of a scenario
Additionally, we are often interested in the number of objects affected by a

scenario which is given by
Xn
kj = pj eij (3.7)
i=1
3.2.5 Probability and Frequency

The likelihood of the realisation of a scenario can be indicated by either a
frequency or probability statement and the time frame in which it holds. For
instance, “It will start raining with a chance of 90%” has a somewhat different
notion if it is referring to the next 12 hours or to the entire next year. In 1933,
Kolmogorov introduced probability theory with its three basic axioms. For
every event E in the universe Ω a number P (E), called probability, is defined
such that the following holds:
0≤ P (E) ≤1 , for every E (3.8)

P (Ω) =1 (3.9)
S X
P ( Ei ) = P (Ei ), for discrete, mutually exclusive events(3.10)
i i
Probability and frequency are related to each other by the so called frequency
interpretation. According to this interpretation the probability of an event is the
proportion of times m the said event occurs when the experiment is conducted
an abundantly large number of times n:
m
P (E) ≈ (3.11)
n
3.2 Definitions 29
From the basic axioms some general rules can be derived that are important
for this work:
P (Ē) = 1 − P (E) (3.12)

P (∅) = 1 − P (Ω) = 0 (3.13)
Where Ē is the complement of E and ∅ the empty set. If two events E and
F are not mutually exclusive, than the event E ∪ F , “either E or F occurs”, is
given by
P (E ∪ F ) = P (E) + P (F ) − P (E ∩ F ) (3.14)
where E ∩ F denotes ’both E and F occur’. Its probability is given by
P (E ∩ F ) = P (E) · P (F |E) (3.15)

= P (F ) · P (E|F ) (3.16)
where P (E|F ) is the conditional probability of E given F, i.e. the probability

that E occurs if F is known to have occurred. If E and F are independent,
P (E|F ) becomes P (E) and
P (E ∩ F ) = P (E) · P (F ) (3.17)
= P (F ) · P (E) (3.18)
Although the probability of an event denotes the “chance” or “likelihood”

it is not a proposition of the certainty of the probability itself. For instance, if
we have a dice and we want to throw a “3” in the first try the chances are quite
low, that is 1/6. Nevertheless, if we have a fair dice, we are very certain about
the value of 1/6. In other words, we are sure about the uncertain result of the
experiment. In section 3.3 we will give more insights on uncertainty.
3.3 Uncertainty in Risk Analysis
Uncertainty plays an important role in risk analysis. First, it is a central object
of study and secondly, it is an inherent part of it. As risk analysis deals with
future, uncertain events the first point is obvious. The second point is not
restricted to risk analysis but has to be considered in every situation where
models are applied.
3.3.1 Uncertainties and Errors

Generally, uncertainty can be viewed as “the absence of information that may
or not be obtainable” (Rowe 1994). In the context of quantitative risk analysis
we will distinguish between the uncertainty of the models and of the data. A
concise introduction to uncertainties can be found in Vĉkovski (1998).
Establishing a model aims at reducing the complexity of the system under
study. This is achieved by abstraction, i.e. by considering some features while
discounting less relevant ones. Model uncertainties are introduced by either the
simplification, the application of an inadequate model or both. The detection
of model uncertainties is accomplished through empirical verification. If the
uncertainties are substantial another model should be applied or the model
should be reformulated. We will not dwell further on problems arising from
models but concentrate on the data uncertainties instead.
Bandemer and Näther (1992) describe three types of data uncertainty: vari-
ability, imprecision and vagueness. Variability denotes the fact that an observed
entity does not show the identical state in comparable situations. This may be
due to inherent non-deterministic properties of the phenomenon (e.g. wind ve-
locity) or an incomplete understanding of the underlying system. Imprecision
stems from the impossibility of measuring to an arbitrary level of precision.
Besides the fundamental problem of measuring as known by the Heisenberg un-
certainty principle, there will be always practical limitations (e.g., in the form
of technical, financial, time or human resources constraints). The last group
of uncertainty is vagueness which occurs mainly in qualitative reasoning. All
types of uncertainty can and will occur simultaneously.
3.3.2 Error Models

Having acknowledged the presence of uncertainties there exist several ways of
dealing mathematically with them: interval mathematics, stochastics and fuzzy
set theory.
Interval Mathematics
The basic idea of interval mathematics is to represent the value of an entity with
a set of possible values. In the simplest case of <1 this is an interval (Bandemer
and Näther 1992).
A = [ a, a ] (3.19)
The interval is treated as a whole set and it is treated according to the rules
of interval mathematics.
3.3 Uncertainty in Risk Analysis 31
Stochastics
In stochastics the error ε is seen as a random variable that builds the differ-
ence between the true but unknown value b and the measured but uncertain
observation A.
A=b+ε (3.20)
Note that because ε is a random variable A becomes a random variable,

while b is deterministic.
In the most common case, the probability distribution function of ε will be
assumed to be a Gaussian function with an expected mean value of 0 and a spe-
cific standard deviation σ. However, any other probability distribution function
could be applied. If series of measurements exist it is possible to determine the
distribution statistically, otherwise the knowledge about the phenomenon and
the measurement process is used to choose an appropriate theoretical function.
Fuzzy Sets
In fuzzy set theory (Zadeh 1965) a value is gradually assigned to a set A accord-
ing to a given membership function mA . The membership function is usually
defined on the interval [0,1] and expresses the degree of acceptance that a given
x belongs to A. The two preceding techniques can be seen as special cases of
fuzzy sets (Bandemer and Näther 1992). Hootsmans (1996) gives an extensive
overview on various aspects of spatial fuzzy data sets.
3.3.3 Error Propagation in Modelling

Often it is desirable to know the uncertainties not only of data values but to
estimate the error propagation that occurs when uncertain data are used in
models. A comprehensive overview of quantitative attribute errors in spatial
modelling is given by Heuvelink (1998).
Taylor Series
The basic idea of a Taylor Series is to approximate a multivariate function f (·)
by a sum of a polynomial. In the case of a first order Taylor series we get
f (x1 , ..., xm ) = f (h1 , ..., hm ) (3.21)

Xm
+ {(xi − hi )fi0 (hi , ...hm )} + remainder
i=1
If we have chosen a Gaussian error model and replace xi with a random

variable Ai and hi with bi (Ai and bi as defined in (3.20)) it yields:
f (Ai , ...Am ) = f (bi , ..., bm ) (3.22)

Xm
+ {(Ai − bi )fi0 (bi , ...bm )} + remainder
i=1
Neglecting the remainder and assuming

E[Ai − bi ] = E[εi ] = 0 (3.23)
the expected value of (3.22) is given by
E[f (Ai , ..., Am )] ≈ (3.24)

" m
#
X
≈ E f (bi , ...bm ) + {(Ai − bi )fi0 (bi , ...bm )} = f (bi , ...bm )
i=1
The variance of (3.22) is given by
E (f (Ai , ..., Am ) − E[f (Ai , ..., Am )])2 ≈

(3.25)
 
m
!2
X
≈ E  f (bi , ...bm ) + {(Ai − bi )fi0 (bi , ...bm )} − f (bi , ...bm ) 
i=1
 ! m 
m
X X
{(Ai − bi )fi0 (bi , ...bm )} ·  (Aj − bj )fj0 (bj , ...bm ) 

= E
i=1 j=1
m
X m
X
ρij σi σj fi0 (bi , ...bm )fj0 (bj , ...bm )

= (3.26)
i=1 j=1
The expectation value of f (·) is the function value of the input means bi . The
resulting variance depends on standard deviations of Ai , correlations between
the input variables and the steepness of the function which is given by the
partial derivatives. The application of a Taylor series is limited to functions
that are continuously differentiable and to the assumption of a Gaussian error
model. Furthermore, when studying not only a single function but a connected
set of functions the deduction of the derivatives can be tedious. As long as the
function is sufficiently smooth and the errors are not too big, the application
of the first order Taylor series is straightforward. In the opposite case, the
method is likely to fail and grossly over- or under-predicting the uncertainty.
The main advantage lies certainly in the low computational costs involved. Once
the derivatives are available the re-computation of (3.26) with different input
values is straight forward.
Monte Carlo
The basic idea of the Monte Carlo method is to repeatedly compute a function
with input values that are randomly sampled from their joint distribution. The
results are stored and statistically analysed (Johnson 1987, Heuvelink 1998).
The Monte Carlo method puts no restriction and limitations to the model that
is analysed, treating it rather as a “black box”. On the other hand, the greatest
challenge is the generation of a large number of random samples. Especially
demanding is the case where many input variables are correlated. The ran-
dom sampling of one variable cannot be done independently from all others but
should be done simultaneously. This can be achieved by using a Cholesky de-
composition of the covariance matrix. In the case of spatial interaction more
Bibliography 33
sophisticated methods like conditional (Englund 1993, Alabert 1987, Gómez-

Hernández and Journel 1993) or autoregressive simulation have to be applied
(Whittle 1954, Hunter and Goodchild 1997).
Interval
When uncertainties are given in the form of intervals (see 3.3.2) the uncertainty
of a function has to be evaluated using interval arithmetics. The basic idea
is to replace all variables with an interval. For any arithmetic operator rules
have to be derived such that the resulting interval C contains all results of the
operation A A between all elements a ∈ B and b ∈ B (Moore 1979, Alefeld and
Claudio 1998, Vĉkovski 1998, Hugentobler 2000). The resulting intervals tend
to be over-estimated as they can contain more than all possible values. Interval
arithmetic is very useful in the field of computing to control roundoff errors.
Bibliography
Alabert, Francois, 1987, The practice of fast conditional simulations
through the lu decomposition of the covariance matrix. Mathematical Ge-
ology, 19(5):369–386.
Alefeld, G. and Claudio, D., 1998, The basic properties of interval arth-
metic, its software realization and some applications. Computers and Struc-
tures, 67:3–8.
Bandemer, Hans and Näther, Wolfgang, 1992, Fuzzy data analysis.
Kluwer Academic Publishers, Dordrecht, NL.
Englund, E., 1993, Spatial simulation: Environmental applications. Environ-
mental Modeling and GIS, edited by M.F. Goodchild, B.Q. Parks and L.T.
Steyaert, pp. 432–437, University Press: Oxford, New York.
Gómez-Hernández, J.J. and Journel, A.G., 1993, Joint sequential simu-
lation of multigaussian fields. Geostatistics Troia ’92, edited by A. Soares,
vol. 1, pp. 85–94.
Heuvelink, G.M., 1998, Error propagation in envrionmental modelling with
GIS. Taylor and Francis, London.
Hootsmans, Robert Maria, 1996, Fuzzy sets and series analysis for vi-
sual decision support in spatial data exploration. Ph.D. thesis, University of
Utrecht, Utrecht.
Hugentobler, Marco, 2000, Fortpflanzung von Unsicherheiten in dreiecks-
basierten digitalen Geländemodellen mit Intervallarithmetik. MSc.Thesis,
Dept. of Geography, University of Zurich.
Hunter, Gary J. and Goodchild, Michael F., 1997, Modeling the uncer-
tainty of slope and aspect estimates derived from spatial databses. Geograph-
ical Analysis, 29(1):35–49.
Johnson, M. E., 1987, Multivariate statistical simulation. John Wiley & Sons,
Inc., New York.
34 Bibliography
Kolmogorov, A. N., 1933, Grundbegriffe der Wahrscheinlichkeitsrechnung.

Springer Verlag, Berlin.
Moore, Ramon E., 1979, Methods and applications of interval analysis. SIAM
(Society for Industrial and Applied Mathematics), Philadelphia, USA.
Rowe, William D., 1994, Understanding uncertainty. Risk Analysis,
14(5):743–750.
Vĉkovski, Andrej, 1998, Interoperable and distributed processing in GIS. Tay-
lor & Francis, London.
Whittle, P., 1954, On stationary processes in the plane. Biometrika, 41:434–
449.
Zadeh, L.A., 1965, Fuzzy sets. Information and Control, 8:338–353.
4
A Consistent Wildland Fire

Risk Terminology Is
Needed!
Andreas Bachmann and Britta Allgöwer, Department of Geography, University

of Zurich∗
The success of any organization depends on a clear understanding of

terms, rules, limits, and conditions.
Wildland fire risk assessment takes into account not only the area
burned, but also the value of properties, such as homes in the
wildland/urban interface.
Wildland fire risk analysis focuses on two general areas: a fire s
probability of occurrence and its outcome.
When used in risk analysis, all impacts must be converted into
monetary terms to permit a collective comparison.
Risk analysis can integrate the fields of wildland fire research,
stimulating interdisciplinary work.
Keywords: Wildfire Risk, Risk Terminology, Quantitative Risk Analysis, Ge-

ographical Information Systems (GIS)
4.1 Introduction
Even a casual familiarity with the literature related to wildland fire risk will
reveal the inconsistent and confusing use of the terms “danger”, “hazard”, and
“risk”. Lack of clear definitions can be an obstacle to sound research and man-
agement. For example, computer models and simulations of “fire danger” or
∗ Published in Fire Management Today, vol.61(4), 2001. This article is an abbreviated
version of a paper published in the proceedings of the Joint Fire Science Conference and
Workshop, 15-17 June 1999, Boise, ID.
36 A Consistent Wildland Fire Risk Terminology Is Needed!
“fire hazard”, if based on fuzzy definitions, are difficult to validate; a compari-

son of research results can be impossible if different researchers are really talking
about different things. Moreover, unless firefighters on the fireline base their ac-
tions on precisely understood conditions, the consequences might be fatal. The
success of any organization depends on a clear understanding of terms, rules,
limits, and conditions.
Quantitative risk analysis in the context of wildland fire management re-
quires a solid basis in a sound terminology that is understood and shared
throughout the wildland fire community. First, let’s look at another form of
risk management – the field of technical risk engineering– for terminological
options and alternatives to our current definitions (see sidebar). Then let’s see
if they can be usefully applied to wildland fire research.
New Definitions
Based on definitions used in technical risk engineering, we propose to
redefine the following terms used in wildland fire risk analysis:
Danger: Obsolete. Danger is an abstract concept based on human
perception. Danger per se does not exist. It is defined by subjec-
tive human and societal perceptions and assessments of events and
outcomes that are considered harmful. The concept is useless for
wildland fire research and management.
Hazard: A process with undesirable outcomes.
Wildland fire hazard: A wildland fire with undesirable outcomes.
Risk: The probability of an undesired event and its outcome. An

undesired event is a realization of a hazard.
Wildland fire risk: The probability of a wildland fire occurring at a
specified location and under specific circumstances, together with its
expected outcome as defined by its impacts on the objects it affects.
4.2 “Risking” Redefinitions

Current wildland fire glossaries illustrate problems with the definitions of “fire
danger”, “fire hazard”, and “fire risk”. The field of technical risk engineering
offers alternatives.
Fire Danger Current glossaries define the term “fire danger” very broadly.
For example the National Wildfire Coordinating Group’s Glossary of Wildland
Fire Terminology (NWCG 1992) defines fire danger as the “ sum of constant
danger and variable danger factors affecting the inception, spread, and resistance
to control, and subsequent fire damage.” This definition conflates fire danger
“factors” that are really quite different. For example, a fire’s inception has to
do with the probability of its occurrence, whereas resistance to control is related
to the fire’s outcome. It makes more sense to treat each factor separately to
determine its relative contribution to the risk of fire. The term “danger” is too
abstract and subjective to be useful for scientific research.
4.3 Quantitative Wildland Fire Risk 37
Fire Hazard Current glossaries define the term ”fire hazard” very narrowly.
For example, the Canadian Interagency Forest Fire Centre’s Glossary of Forest
Fire Management Terms (CIFFC 1999) defines fire hazard as “potential fire
behavior” based on “physical fuel characteristics (e.g. fuel arrangement, fuel
load, condition of herbaceous vegetation, presence of ladder fuels).” In this
definition, hazard is a set of fuels preconditions for wildland fire behavior.
By contrast, in the domain of technical risk engineering, hazard is more
broadly defined as “a physical situation with a potential for human injury,
damage to property, damage to the environment or some combination of these”
(Allen 1992). The broader definition has the advantage of applicability to any
process that can lead to damage. In this case, the hazard is wildland fire itself,
not some subset of preconditions that might or might not lead to fire damage.
Fire Risk The term risk generally has two distinct areas of meaning: (1)
chance and probability; and (2) loss, harm, and injury (Landau 1999). Cur-
rent wildland fire glossaries focus on the former area, neglecting the former.
For example, the United Nations Food and Agriculture Organizations’ Wildfire
management Terminology (FAO 1986) defines fire risk as “ The chance of fire
starting, as affected by the nature and incidence of causative agencies.” This
definition focuses on fire cause (the probability of ignition), ignoring fire effects
(the likelihood of damage).
By contrast, in the field of technical risk engineering, risk comprises both
the likelihood and the outcome of an event (Jones 1992, Gheorghe and Nicolet-
Monnier 1995, Merz et al. 1995). For example, Hall (1992) refers to structural
fire risk as both “a measure of the expected severity (e.g., how many deaths,
injuries, dollars or damager per fire)” and “ a measure of the probability of
occurrence.” Hall’s definition has the advantage of including both components
of fire risk – probability and damage.
New Definitions We base our wildland fire risk analysis and assessment on
proposed new definitions:
We drop the term “fire danger” as useless for wildland fire research and
management.
We use the term “fire hazard” as a synonym for the process of wildland
fire itself.
We make “fire risk” our central term. Specifically, we address “quanti-

tative wildland fire risk,” embedding the concept of risk in a risk man-
agement process, where risk analysis and risk assessment are important
steps (Bärtsch 1998). Moreover, we use the value-free term “outcome” to
describe fire effects, which are not always negative.
4.3 Quantitative Wildland Fire Risk

For a quantitative analysis, we must operationalize the term risk by describ-
ing the mathematical relationships between probability and outcome and by
defining indicators that can be used to measure their value. Generally, risk
r is defined as the product of the probability p and the expected outcome d

(Kumamoto Hiromitsu 1996, Jones 1992, Merz et al. 1995), as follows:
r =p·d (4.1)
The probability p, according to the axioms of Kolmogorov (1933), is defined

for a given time period; for example, the occurrence probability of a wildland
fire in the next year is 0.8. Kolmogorov’s axioms do not prescribe how prob-
abilities are determined but define properties and calculation rules that have
to be satisfied. For example, a method resulting in an index of fuel moisture
with an arbitrary scale can be transformed into an ignition probability with
any appropriate function. The expected outcome d is a measure or description,
such as “100 acres [40 ha] of highly productive timber stands burned down” or
“increased biological diversity after disturbance.”
Risk deals with future events, which cannot be predicted in a deterministic
way. Scenarios must be constructed to represent possible realizations of a hazard
(in this case, wildland fire). The scenarios define all relevant preconditions and
causes of an event and thus enable the quantitative determination of risk.
Risk can be seen from two perspectives, subjective and collective. The sub-
jective perspective is from a single object exposed to risk, the so-called risk
acceptor. The object might be affected by many scenarios of presumably differ-
ent types of hazards. The collective perspective is from a particular scenario,
the so-called risk donor. Each scenario might affect many risk acceptors. The
two perspectives both play a role in assessing risk. For example, although the
collective risk might be accepted, the subjective might not be, either because the
damage is too great or because the probability of one object is too high. This
case can occur when individual outcomes and / or probabilities are unequally
distributed.
Two problems are specific to wildland fire risk analysis: spatial location and
impact indicators. The first problem is identifying the location. Whereas most
technical hazards have fixed locations (for example, a road or power plant),
wildland fires can, in principle, start in any location covered by combustible
vegetation. In assessing wildland fire risk, analysts must account for an infinite
number of potential locations.
The second problem is selecting relevant impact indicators for various types
of affected objects, such as timber or endangered-species habitat. Merz et al.
(1995) maintain that damages should, whenever possible, be quantified to per-
mit their discussion and assessment. Taking the problems of spatial location and
impact indicators into account, we define wildland fire risk as the probability of
a wildland fire to occur at a specified location and under given circumstances
and its expected outcome as defined by the impacts on the affected objects.
4.3.1 Calculating Risk

Let’s apply our definition to a given study area (fig. 4.1) with a certain number
of objects Oi and scenarios Sj . For every relation between a scenario Sj and an
object Oi the individual probability of impact eij and the individual expected
impact dij at the object is calculated. The risk of a given scenario, i.e. the
collective wildland fire risk, is then:
4.3 Quantitative Wildland Fire Risk 39
Sj
eij , dij
Oi
Figure 4.1: Hypothetical study area showing relationships between objects and
scenarios. For every relation between a scenario Sj and an object Oi , the in-
dividual probability of impact eij and the individual expected impact dij at
the object can be calculated. The collective risk of the scenario can then be
computed.
n
X
rj = pj · eiij · dj (4.2)
i=1
Figure 4.2 maps the fundamental relations in the collective risk of a scenario.
The collective risk is the product of probability pj is the probability that a fire
will start at a given location and expected outcome dj . The expected outcome is
a weighted sum of the impacts of all objects. This method takes into account not
only the area burned, but also valuable properties, such as buildings – expecially
important in the urban wildland interface, where the value of burned timber is
often negligible compared to the value of destroyed homes (Alexandrian 1996).
The weighting is done with the impact probability – that is the probability
that a fire will reach and affect an object. The impact probability is determined
through fire spread. After the fire perimeter of a given scenario is known, an
individual impact probability of 1 will be assigned to all objects within the
perimeter. Objects lying outside the perimeter will then have an individual
impact probability of 0 and thus won’t contribute to the sum of impacts.
In a comprehensive risk analysis, several scenarios are usually constructed
to reflect all relevant cases. The risk for the whole area is then the sum of all
individual scenario risk.
4.3.2 Research Integration

Wildland fire risk analysis focuses on two general areas: a fire’s probability and
its outcome. These are the focal points for the three main wildland fire research
ares – fire effects, fire behavior, and fire occurrence (fig. 3). Risk analysis can
thus stimulate interdisciplinary approaches.
Probability of Occurrence Flame depends on three factors (Pyne et al.

1996): a fuel, a heat source, and oxygen. Oxygen can usually be taken for
granted as a constant. Fuel is the essential precondition; the heat source is
Collective Risk
of a Scenario
pj ⋅ dj
Expected Outcome
Probability of Scenario of Scenario
pj n
dj = Σ eij ⋅ dij
i =1
Individual Individual
Impact Probability Expected Outcome
eij dij
Figure 4.2: Collective risk of a scenario, the product of the probability pj of a fire
starting at a given location and its expected outcome dj . The expected outcome
is a weighted sum of the impacts of the fire on all objects. The weighting is
done through the impact probability – that is, the probability that a fire will
reach and affect an object.
the immediate cause. Fire cause and fuel conditions are the subject of fire
occurrence research and fire behavior research, respectively. Most fires have
human cause, so most fire occurrence research focuses on human activities.
Fuel complex research has two main components: fuels classification and fuel
moisture estimation.
When used in wildland fire risk analysis, any approach to fire occurrence
of fuels complexes should deliver a probability as a result. The probability of
occurrence can be expressed as:
pj = piginition · pprecondition (4.3)
where pignition expresses the probability that any cause starts the wildland
fire and pprecondition is the probability that the fuel complex and fuel moisture
permit a fire to start. The product of both probabilities assures that if one of
each is zero, the occurrence probability is zero too.
Outcome The extent of the outcome is determined by a weighted sum of the

impacts on each affected object, with the weighting done through impact prob-
ability (eq.2). In determining impact probability, we are especially interested
in the rate of spread (Viegas et al. 1998) and the ease of suppression. Acces-
sibility, now a standard functionality in geographic information systems, can
give a useful measure of suppression effectiveness, permitting the overall per-
formance of firefighting organization in a given region to be integrated into the
4.4 An Analytical Framework 41
Methodology
Wildfire Risk
Risk
Probability Outcome
of Occurrence
Research
Wildfire
Wildfire Wildfire Wildfire

Occurrence Behavior Effects
Figure 4.3: Risk methodology and wildland fire research. Through its focus
on fire outcome and probability of occurrence, quantitative fire risk assessment
integrates each of the three major areas in wildland fire research.
risk analysis. The impact of wildland fires on objects is at the core of fire effects
research, which focuses on such areas as tree mortality (Ryan 1998), erosion
potential (Marxer et al. 1998), and structural ignition (Cohen et al. 1991). A
fire’s impact on an object is a function of fire behavior and the susceptibility of
the object to flame.
When used in wildland fire risk analysis, all impacts must be converted into
monetary terms to permit a collective comparison. Monetary conversion is often
difficult for impacts such as erosion potential or destruction of natural resources
such as wildlife habitat. However, economists are increasingly exploring ways
to assign value to nonmarket objects (Cában 1998).
4.4 An Analytical Framework

Based on the terminology presented here, Schöning et al. (1997), Bachmann and
Allgöwer (1998) developed a framework for analyzing the spatial distribution of
wildland fire risk using a geographic information system. In addition to sce-
narios and objects, “situations” are used to capture all risk-relevant parameters
(fig. 4.4), such as weather, fuels, precipitation, and holiday activities. Situations
help to group scenarios. In any given area, a wildland fire might start at an
infinite number of locations. Some spatial discretization (usually rasterizing) is
needed to obtain a finite number of scenarios.
Based on fire occurrence research methodology, a probability of ignition is
assigned for each location in the study area, given a particular situation. The
probability is then determined for each object affected under a given scenario,
using appropriate fire behavior models. Finally, based on fire effects research,
the amount of damage suffered by each object is estimated under each scenario.
The resulting parameters are combined in a risk matrix depicting the relations
Scenarios Situations Objects
Fire Occurrence Fire Behavior Fire Effects
Risk Matrix
Risk caused by Total Risk for each Risk for each

each Scenario Situation Object
Figure 4.4: Framework for a quantitative wildland fire risk analysis. Scenar-
ios, objects, and situations capture all risk-relevant factors. Each is determined
through methodologies associated with the three main areas of wildland fire re-
search (fire occurrence, fire behavior, and fire effects). The resulting parameters
are combined in a risk matrix permitting calculation of risk characteristics for
scenarios, objects, and the situation as a whole.
among all scenarios and objects for a given situation. The matrix permits
the calculation of risk characteristics pertaining to scenarios, objects, and the
situation as a whole. For example, figure 4.5 shows the expected damage by
scenario and object (buildings).
4.5 Need for a Robust Terminology
A robust wildland fire risk terminology can support a rigorous risk analysis. Risk
analysis, in turn, can integrate the fields of wildland fire research, stimulating
interdisciplinary work. In operational use, risk analysis can help fire managers
better understand how the various aspects of fire occurrence, fire behavior, and
fire effects fit together to form the totality of wildland fire.
For more information, please contact Andreas Bachmann, Spatial Data Han-
dling Division, Department of Geography, University of Zurich, Winterthurerstr.
190, CH-8057 Zuerich, Switzerland, 41-1-635-52-52 (tel.), 41-1-635-68-48 (fax),
bachmann@geo.unizh.ch (email).
Bibliography 43
Figure 4.5: Example of expected damage to objects and by scenario in

Malcantone-Ticino, Switzerland, using a quantitative wildland fire risk anal-
ysis.
Acknowledgments
The work reported in this article was funded by the Swiss National Science
Foundation (Project-Nr. 21-50842.97) and by the Ministry for Education and
Science (EC-Project INFLAME Nr. ENV4-CT98-0700, BBW Nr. 97.0182-2).
Bibliography
Alexandrian, D., 1996, A new method of fire danger mapping in the for-
est urban interface. Wildfire Management: Systems, Models & Techniques
(Workshop), Athens.
Bachmann, A. and Allgöwer, B., 1998, Framework for the assessment of
wildfire risk. 3rd International Conference on Forest Fire Research / 14th
Conference on Fire and Forest Meteorology, edited by D.X. Viegas, vol. 2,
pp. 2177–2190, Luso (P).
land.
Cában, A.G., 1998, The importance of nonmarket values in land and fire
management planning. Proceedings of the III International Conference on
Forest Fire Research, 14th Conference on Fire and Forest Meteorology, edited
by D.X. Viegas, vol. 2, pp. 2191–2208, Coimbra, Portugal.
44 Bibliography
CIFFC, (Canadian Interagenxy Forest Fire Centre), 1999, Glossary

of fire management terms. CIFFC, Winnipeg.

Gheorghe, A. and Nicolet-Monnier, M., 1995, Integrated regional risk

assessment, vol. 2 of Environmental science and technology library; vol. 4.
Kluwer, Dordrecht, NL.

UK.

Kumamoto Hiromitsu, Henley Ernest J., 1996, Probabilistic Risk Assess-

ment and Management for Engineers and Scientist. IEEE, New York, 2nd
edn.
Landau, S.I. (Editor), 1999, The new international Webster’s student dictio-
nary of the English language. Trident Press International, New York, NY.
Marxer, P., Conedera, M. and Schaub, D., 1998, Postfire runoff and soil
erosion in sweet chestnut forests in south switzerland. Proceedings of the III
International Conference on Forest Fire Research, 14th Conference on Fire
and Forest Meteorology, edited by D.X. Viegas, vol. 2, pp. 1317–1331, Coim-
bra, Portugal.

NWCG, (National Wildfire Coordinating Group), 1992, Glossay of

wildland fire terminology. PMS 205/NFES 1832, National Interagency Fire
Center, Boise, ID.
Bibliography 45
Ryan, K.C., 1998, Analsysis of the relative value of morphological variables

in predicting fire-caused tree mortality. Proceedings of the III International
Conference on Forest Fire Research, 14th Conference on Fire and Forest Me-
teorology, edited by D.X. Viegas, vol. 2, pp. 1511–1526, Coimbra, Portugal.
Schöning, R., Bachmann, A. and B.Allgöwer, 1997, GIS-based framework
for wildfire risk assessment. Final report for the European research project
minerve 2, Department of Geography, Division of Spatial Data Handling,
University of Zürich.
Viegas, D.X., Ribeiro, P.R. and Cruz, M.G., 1998, Characterisation of the
combustibility of forest fuels. Proceedings of the III International Conference
on Forest Fire Research, 14th Conference on Fire and Forest Meteorology,
edited by D.X. Viegas, vol. 1, pp. 467–482, Coimbra, Portugal.
5
Framework For Wildfire

Risk Analysis

of Zurich∗
Summary
This paper presents a conceptual framework for a comprehensive analysis and
assessment of wildfire risk in order to support a well balanced forest fire man-
agement. The term wildfire risk is used here as a measure of the hazard posed
by wildfire, expressing both the probability of damage to be caused, and the
expected extent of that damage. Therefore, in the analysis of wildfire risk the
probability of a fire to occur at any location, and its impact on objects (both
man-made and natural) must be considered. To do so, the knowledge about fire
occurrence, fire spread and fire effects must be combined. Usually, these three
topics are treated separately in wildfire research. The proposed framework aims
at linking (existing) fire occurrence modelling and methods for the assessment
of fire effects by models for fire behaviour and propagation. As the analysis
requires the handling of large data sets which are mostly spatially referenced,
the use of a Geographical Information System (GIS) offers great advantages.
5.1 Introduction
One important part of every wildfire management (which is generally spoken a
special case of risk management) is a thorough analysis of the risk situation in
a given region. Ideally, the risk analysis will depict the critical areas so that
effective prevention and presuppression activities can be planned. To do so, the
analysis has to consider the spatial nature of the phenomenon wildfire. The
proposed framework explicitly permits the determination of spatial patterns of
∗ Published in the Proceedings of the III International Conference on Forest Fire Research,
14th Conference on Fire and Forest Meteorology, Vol. II, pp.2177–2190, held in Luso (Coim-
bra), Portugal 16–20 November 1998.
48 Framework For Wildfire Risk Analysis
wildfire risk at a high resolution. Its methodology is based on well known tech-
nical risk analysis and combines three main topics of wildfire research: wildfire
occurrence, effects and behaviour.
The application of the framework involves the handling of large amounts of
data, most of which being spatially referenced. Thus, the use of a Geographical
Information System (GIS) offers great advantages not only in the handling but
also for analysing and visualising this data (Chuvieco and Salas 1996).
5.2 Wildfire Risk - A Definition

To clarify the term wildfire risk and to prevent misunderstandings, we provide
a short definition. The first part, i.e. wildfire, we use in this paper to describe
any fire occurring in a more or less natural environment. It may be fuelled by
living or dead vegetation. The term includes subterms like forest fire, bush fires
etc.
The definition of the second part is derived from definitions used in the
technical risk analysis domain. This is done mainly because in wildfire research
there seems to be no common agreement on the notion of risk. The definitions
served by FAO (1986) † are an affirmation of the vagueness and the misunder-
standings associated with this term. A very comprehensive definition of risk
given by Allen (1992) serves as a starting point:
RISK: The likelihood of specified undesired events occurring within a

specified period or in specified circumstances arising from the
realisation of a specified hazard.
HAZARD: A physical situation with a potential for human injury, damage
to property, damage to the environment or some combination
of these.
As we can see, the term risk consists of two elements: The probability of an
event and the potential damage caused by this event. In order to be applied
in the context of a quantitative risk analysis, a common operationalisation is
(Gheorghe and Seiler 1996):
Risk(R) = P robability(P ) · Amountof Damage(D) (5.1)

Applied to wildfire risk the following definition will be used in this paper:
WILDFIRE RISK: The probability of a wildfire to occur at a specified

location and under given circumstances (later on called situa-
tions), and its expected damage on endangered objects.
The term “fire danger” given by the FAO (1986) comes somehow close to
that definition, while the term “fire risk” in one of two definitions clearly denotes
only the probability of a fire to occur.
† Fire Danger: The resultant, often expressed as an index, of both constant and variable
danger factors affecting the inception, spread, and difficulty of control of fires and the damage
they cause.
Fire Hazard: (1) [North American] A fuel complex, defined by volume, type condition, arrange-
ment, and location, that determines the degree both of ease of ignition and of fire suppression
difficulty.
5.3 General Layout of the Framework 49
Two specific points have to be considered in great detail when applying

risk analysis methods to the environmental hazard posed by wildfires. Most
technical hazards are fixed in space, e.g. the location of a power plant is given
by spatial co-ordinates, or limited to some well defined locations, e.g. roads.
This is clearly not the case with wildfires which, on principle, can start at any
point that is covered by combustible vegetation. The other crucial point is
the selecting of relevant damage indicators for various types of affected objects,
e.g. power plant vs. endangered species habitat. In the opinion of Merz et al.
(1995), damages should whenever possible be quantified, in order to enable their
discussion and assessment.
5.3 General Layout of the Framework

The key elements of the framework are scenarios, situations and endangered
objects (see Figure 5.1).
Scenarios Situations Objects
Fire Occurrence Fire Behaviour Fire Effects
Risk Matrix
Risk caused by Total risk for each Risk for each

each scenario situation object
Figure 5.1: Framework for the wildfire risk analysis
Scenarios represent potential fire events. A scenario describes the ignition

at a specific location during a defined situation, e.g. the start of a wildfire at
point XY with strong south winds.
Situations define relevant states of the risk analysis parameters like weather,
fuel, human behaviour, etc., e.g. a spring weekend with strong southern winds.
Objects cover all natural or artificial entities which are considered endangered
by wildfire. Examples: settlements, traffic ways, forest stands, vegetation with
protective function (avalanches, erosion, etc.). The objects considered may vary
in different risk analysis contexts. An object may also depend on its properties
and/or representation of the current situation (e.g., the susceptibility of traffic

ways may be a function of traffic volume).
Using the methods of fire occurrence research, a probability must be assigned
for each scenario to occur in the considered time span (i.e., the probability of
ignition for each location in the study area, given a situation). The probability
is then determined for each object to be affected by a given scenario. This is
accomplished using appropriate fire behaviour models. Finally, the amount of
damage each object will suffer given a scenario must be estimated. This is the
issue of fire effects research. For the actual risk analysis, these parameters are
combined in a risk matrix which depicts the relations between all scenarios and
all objects for a given situation. The matrix permits the calculation of risk
characteristics pertaining to scenarios, objects and the situation as a whole.
5.4 Risk Matrix

The core of the framework is the risk matrix ( Figure 2 ). The general layout,
described by Merz et al. (1995), has been adapted to the specific risk of wildfires.
Every row represents an object, while the scenarios form the columns. Each cell
therefore represents the relation between a scenario (i.e. risk donor) and the
object (i.e. risk acceptor). For every object the risk values (probability and
expected damage) of all scenarios posed on the object can be accumulated (last
column). In the same way, the influence of every scenario on all objects may be
calculated (last row). Finally, the risk values for the entire area are summed up
in the lowest right cell. A risk matrix is constructed for every situation so that
it is possible to compare different situations against each other.
Where:
Es The situation for which the risk analysis is performed.
Sj Scenario (potential fire event defined by an ignition location and the
situation).
Oi Object (any man-made or natural entity considered as endangered by
wildfire).
pj Conditional probability that scenario Sj takes place, given situation
Es . Can be determined using methods of fire occurrence modelling
(e.g. by statistical analysis of past fire events).
tij Time for fire to spread from ignition point to object Oi , given scenario
Sj . Can be determined using fire behaviour and propagation models.
f () Function to transform spread time into spread probability.
kij Probability that object Oi is affected by scenario Sj .
Φ() Function to estimate damage suffered by object Oi based on fire be-
haviour.
dij Damage caused at object Oi under the assumption that the fire reaches
the object, given scenario Sj .
ki Probability that object Oi will be affected by any fire.
di Expected damage caused by all fires at object Oi .
kj Expected number of objects that will be affected by scenario Sj .
dj Expected damage caused by scenario Sj at all object in its burned area.
k Expected total number of objects affected by wildfires, given situation
Es .
d Expected total damage caused by wildfires, given situation Es .
5.4 Risk Matrix 51
Scenarios object
Situation
Es S1 S2 . . . Sj . . . Sm related risk
p1 p2 . . . pj . . . pm values
O1
O2
...
Oi (
eij = f Fire Spread ) j=m
Objects
ki = ∑ kij
kij = pj ⋅ eij j =1
( )
j=m
dij = Φ Iij di = ∑ kij ⋅ dij

j =1
...
On
scenario i=n
kj = ∑ kij
i=n
K = ∑ ki
related i =1 i =1
risk i=n i=n
values dj = ∑ kij ⋅ dij D = ∑ di

i =1 i =1
Figure 5.2: Risk matrix for one situation Es
5.4.1 Situation
A situation Es defines the global conditions that affect the risk situation for the
entire area of interest. This includes mainly the definition of the input param-
eters for the fire behaviour modelling (fuel description, topographic character-
istics, climatic conditions, etc.), but also the efficiency of wildfire management
activities (e.g. available resources, organisational structures of fire brigades,
etc.) or special fire occurrence patterns (e.g. fire works at the celebration of
national days) may be considered.
Depending on the goals of the managing organisation, situations may be
derived from an analysis of historical data, or they may be constructed in order
to study worst cases.
5.4.2 Scenario And Fire Occurrence

The term scenario Sj denotes a possible event, i.e. a possible realisation of a
wildfire. It is defined by a spatial position of the outbreak and the probability
of its occurrence (pj ). Theoretically, an infinite number of outbreak positions
exists in a given area of interest. In order to be able to implement the framework,
space has to be discretised using a standard tessellation like a grid consisting
of square cells. The cell size defines the spatial resolution of the entire analysis
and should be as small as possible. Practical limiting factors are the availability
of spatial highly resolved input data and/or computational shortcomings, e.g.
assuming a study area of 10x10 km2 and a cell size of 50x50 m, then a 40’000
scenarios have to be considered.
For each scenario the probability pj for its occurrence in a certain time period
must be provided. Therefore, a method is required which permits the calculation
of the probability for a fire starting in a given unit of space and under conditions
defined by a specific situation. There exists a variety of methods to estimate the
occurrence probability analysing historical wildfire events (e.g., Martell et al.
1987, Marcozzi. et al. 1994, Bolognesi 1994). To take full advantage of the
proposed framework the statistical analysis should also consider variables that
influence the local pattern of fire occurrence (e.g. slope, distance from road,
etc.). An interesting approach is to analyse “spatial” variables using a logistic
regression (Chou 1992).
5.4.3 Objects And Fire Effects

As mentioned above, objects cover all natural or artificial entities which are
considered endangered by wildfire. The choice of what to include in the risk
analysis depends on the goals of the persons or organisations that are involved
and/or those which initiated it. The object must be static (i.e. a “moving”
object like a car can not be considered, whereas a street fulfils this condition)
and it has to lay within the area of interest. Every object (or object class)
has some properties that describe generally the monetary damage (e.g. value
of building) based on the sensitivity to a wildfire impact (e.g. heat resistance
of ceiling). Additionally, every object (or object class) has a corresponding
function Φ() (Figure 5.2) that relates the given local fire behaviour with the
given object’s properties in order to calculate the expected damage.
The definition and estimation of the fire effects on an object is one of the
most critical points (Chandler et al. 1983). As long as objects denotes buildings
or other human-made valuables, it should be possible to express the damage in
monetary terms, based for example on insurance taxation (in Switzerland every
building has to be insured). As for “natural” objects like natural reserves, it be-
comes very difficult to estimate the damage (e.g. How are the aesthetic aspects
of a forest accounted for in a region which is highly dependent on tourism?).
Relating Objects and Scenarios
5.4.4 Relating Objects and Scenarios

Theoretical Aspects
The main factors have now been introduced and we turn our attention to their
dependencies. We would like to assign to each scenario (risk donor) all objects
(risk acceptor) that are affected by it. It is clear that for wildfires it is not
sufficient to establish the connection on the basis of distance measuring. As
shown in Figure 5.3, scenario A is much nearer to the object “House”, but due
to the river the scenario will not actually affect it. So, the best way to establish
the connection between scenarios and objects is to consider the fire behaviour,
i.e. to model the spread time of a fire from the ignition point of a given scenario
to the object.
The next step is to set a threshold upon which the objects can be chosen to
be affected by a specific scenario. Alexandrian (1996) suggests that every start-
ing point which is in the range of ”about three hours” has to be considered,
which may be adequate for his specific study area of wildland urban interface.
5.4 Risk Matrix 53
A
House
e r
Riv
Figure 5.3: Relationship between two scenarios and one object
Considering the uncertainties resulting from the fire behaviour and spread mod-
elling and problems to define a crisp threshold, it seems more appropriate to use
an exponential function to relate the predicted spread time to the probability
kij
kij = eβ·tij (5.2)

The parameter tij denotes the fire spread time for an object and a given
scenario. For the coefficient β, an appropriate value needs to be assigned based
on knowledge or assumptions about the fire propagation behaviour and effects
of fire suppression activities in the study area (Bachmann 1996).
The choice of an exponential function is motivated by an analogy with po-
tential models used in quantitative Geography (e.g., Yeates and Garner 1976,
Rich 1980, Geertman and Ritsema van Eck 1995).
Fire Behaviour and Spread Modelling

As for every scenario-object pair the probability kij has to be known, every
spread time tij has to be calculated using appropriate fire behaviour and prop-
agation models. In other word, for every scenario the fire propagation for the
given situation has to be modelled in order to know the time lag involved for
every object to be hit by the fire front. As the number of scenarios is quite large
it is clear that the practical realisation is questionable. On the other hand,
the number of objects will normally be much smaller. Schöning et al. (1997)
describe a method called fire spread backtracking which permits the calcula-
tion of spread time starting at the objects, and thus reduces the computational
expenses significantly. The proposed method assumes that (1) the local fire
behaviour can be calculated neglecting spatial and temporal dependencies and
(2) that the spread pattern is described with a geometric model. Both assump-
tions are fulfilled if one follows the current wildfire modelling practice using the
Rothermel model (Rothermel 1972) in combination with a spread simulation
based on Anderson s double ellipse (Anderson 1983).
The backtracking is done by first calculating the local spread parameters (i.e.
direction and velocity of maximum spread). Second, the direction of maximum
spread has to be rotated by 180°. Finally the spread simulation can be started
at the object, and for any point the time can be obtained.
As the aim of the risk matrix is to give a comprehensive view of the risk
situation in an area, the simplifications based on the above assumptions can be
tolerated.
5.4.5 Results
The risk matrix permits the summing up for a distinct object, respectively for
a distinct scenario, and for the entire matrix itself. Although wildfire risk may
be presented as the product of the probability and the expected damage (Equa-
tion 5.1), for a detailed analysis it is more sensible to look at them separately.
The values are in any case only valid for the given situation and the specified
time period.
The summarised risk values for objects are denoted by ki and di . ki repre-
sents the probability that a specified object Oi is affected by any scenarios Sj
given Situation Es . di denotes the expected damage of object Oi due to these
scenarios.
For a specified scenario Sj , kj stands for the expected number of objects
affected by Sj . dj represents the sum of the expected damage of all affected
objects.
While the risk values of scenarios and objects may be visualised in the form of
a map, e.g. identifying areas that, if wildfires start in their boundaries, “cause”
high damage to the surrounding objects (obviously areas cannot “cause” the
damage).
The entire risk matrix K denotes the total expected number of objects that
are affected by any wildfire in the study area, and D is the total expected
damage.
5.5 Outlook
The technical realisation of the framework has been tested in a small study area
using the commercial GIS ARC/INFO (Schöning et al. 1997). It was shown that
a GIS is a good platform to carry out such an analysis and that the theoretical
concept holds.
In order to apply the framework in the wider context of wildfire management,
approaches have to be developed that cover the consideration of factors influ-
encing the risk situation, e.g. suppression activities, changing landuse patterns
or more efficient fire fighting techniques.
Further emphasis need to be placed on the uncertainties of input data and
on error propagation in the course of the analysis.
Acknowledgement
This research project is funded by the Swiss National Science Foundation
(Nr. 050842.97). Preliminary work was carried out in the context of the EC-
funded project MINERVE (EC Nr. EV5VCT-0570).
Bibliography 55
Bibliography
Alexandrian, D., 1996, Fuel management. Wildfire Management: Systems,
Models & Techniques (Workshop), Algosystems S.A., Athens, Athens.
Bachmann, A., 1996, GIS-Analyse für ein Waldbrandmanagementsystem. Ein-

satz von Geographischen Informationssystemen und Fernerkundung in der
Umweltanalyse, edited by K. Brassel, K.I. Itten and E.Schmitt, vol. 31 of
Geoprocessing Series, pp. 49–52, Departement of Geography, University of
Zurich, Zurich, Switzerland.
Bolognesi, R., 1994, Nx-log: Idées directrices. Neige et avalanches, 63:22–24.
Sons, New York.

140.
Systems, 10(3):333–345.

Geertman, S. C. M. and Ritsema van Eck, J. R., 1995, GIS and models
of accessibility potential: an application in planning. International Journal of
Geographic Information Systems, 9(1):67–80.
nology Zurich).
Marcozzi., M., Bovio., G, Mandallaz., D and Bachmann., P, 1994, In-

fluenza della meteorologia sull’indice di pericolo degli incendi boschivi nel
cantone ticino. Schweizerische Zeitschrift für Forstwesen, 194(3):183–199.
Martell, D.L., Otokul, S. and Stocks, B.J., 1987, A logistical model

for predicting daily people caused forest fire occurence in ontario. Canadian
Journal of Forestry Research, 17:394–401.
56 Bibliography

Rich, D.C., 1980, Potential models in human geography, vol. 26 of Concepts
and Techniques in Modern Geography. University of East Anglia, Norwich.
Yeates, M.H. and Garner, B.J., 1976, The North American City. New York.
6
Uncertainty Propagation in
Wildland Fire Behaviour
Modelling

of Zurich∗
Keywords: uncertainty propagation, sensitivity analysis, environmental

modelling, wildland fires, spatial decision making.
Acknowledgements
This work is funded by the Swiss National Science Foundation (Project-Nr. 21-
50842.97) and by the Federal Office of Science and Education (EC-Project IN-
FLAME Nr. ENV4-CT98-0700, BBW Nr. 97.0182-2)
Abstract
Rothermel’s model is the most widely used fire behaviour model in wildland
fire research and management. It is a complex model that considers 17 input
variables describing fuel type, fuel moisture, terrain and wind. Uncertainties in
the input variables can have a substantial impact on the resulting errors and
have to be considered, especially when the results are used in spatial decision
making. In this paper it is shown that the analysis of uncertainty propagation
can be carried out with the Taylor series method. This method is computation-
ally cheaper than Monte Carlo and offers easy to use, preliminary sensitivity
estimations.
∗ Accepted for publication in the International Journal of Geographical Information Sci-
ence, 16(2), 2002.

58 Uncertainty Propagation in Wildland Fire Behaviour Modelling
6.1 Introduction
Due to changing rural land use practices, land abundance on the one side and a
growing urban wildland interface on the other side, wildland fires increasingly
threaten buildings, infrastructure and other valuables (Alexandrian 1996). The
main goal of wildland fire management is to reduce the negative impacts of fire to
society under the constraint of limited resources. One important step to achieve
this is a risk analysis (BUWAL 1999). The goal of a risk analysis is to describe
the probabilities and outcomes of a certain spectrum of events. In the case of
the natural hazard “wildland fire” Bachmann and Allgöwer (1998) propose a
framework for the quantitative spatial analysis of wildland fire risk that allows
the generation of individual and collective risk maps. An important factor to
be analysed is the fire behaviour as expressed in spread rate and direction.
A complete risk analysis requires the evaluation of various scenarios and
situations, each involving several steps of modelling and data acquisition. The
results being statistically analysed and summarised are then ready for assess-
ment considering the societal context (Pyne et al. 1996, Miller et al. 1999). It is
important that the decision makers know not only the results of the risk analysis
but that they are also aware of inherent uncertainties and errors resulting from
it.
Within the wildland fires research community great effort is dedicated to
the investigation and improvement of fire behaviour modelling and its benefit
for operational use. In this context, all relevant input variables (i.e., fuels, cli-
matic conditions, topography and wind) require special consideration. However,
sensitivity analysis and uncertainty propagation for the various fire behaviour
models and their input variables are rarely performed (see Salvador et al. (2001
in press) as counter-example). The main goal of this paper is to evaluate which
uncertainty propagation methodology is the most appropriate for fire behaviour
modelling in the context of risk analysis. Additionally, we try to determine the
contribution of the input variables’ uncertainties to the errors of the most im-
portant results of fire behaviour modelling: rate of fire spread, spread direction
and effective wind speed.
6.2 Fire Behaviour Modelling

6.2.1 Rothermel’s Model
In 1972, Rothermel from the United States Department of Agriculture (USDA)
laid the basis for the most popular and still widely used fire behaviour model. In
the wildland fire research and management community, the model is a “quasi”
standard. Rothermel’s fire spread model (Rothermel 1972, Albini 1976) allows
the calculation of the behaviour of a surface fire (i.e., fires that mainly affect
ground vegetation and that pass rapidly through a stand). It assumes that
all input variables are homogeneous for a sufficiently small area and time pe-
riod. The main outputs of the model are the rate of fire spread (ros [m/s]), the
direction of maximum spread (sdr [º]) and effective wind speed (ef w [m/s]).
Furthermore, fire line intensity, heat release per area and flame length can be
derived. The latter group of outputs is of great interest for the analysis of im-
pacts due to wildland fires. Rothermel’s model was developed from a theoretical
6.2 Fire Behaviour Modelling 59
base; i.e., it represents the ratio between the heat flux received from a source
and the heat required for ignition by the potential fuel. The corresponding basic
equations have not been solved analytically but experimentally. Given a hori-
zontal plane and assuming no wind and homogenous fuels, a fire would spread
evenly in any direction with exactly the same rate of spread. The shape of a
fire front in such an idealistic situation would be a circle. Rothermel’s model
considers the influence of the wind and of the terrain by an addition of weighted
wind and terrain vectors. The wind vector is given by the wind direction and
the wind speed at mid-flame height. Slope (i.e. steepest gradient) and aspect
(i.e. exposition or direction of steepest gradient) give the terrain vector. The
resulting vector indicates the direction of maximum spread and its length the
relative influence on the rate of spread. The effective wind speed denotes the
wind speed along the direction of maximum spread. In the presence of slope
and wind the shape of a fire front is no longer a circle but is best estimated
by a double-ellipse which is defined by the rate of spread, spread direction and
effective wind speed (Anderson 1982).
6.2.2 Input Variables

The Rothermel’s model requires numerous input variables that can be divided
into four groups: fuel type, fuel moisture, topography and wind.
Particle Properties
Particle Classes Fuel Load [kg/m2 ] Surf.-to-Vol. [m2 /m3 ]
0.0 < 0.6 cm w0d1 measured svd1 measured
Dead 0.6 < 2.5 cm w0d2 measured svd2 constant
2.5 < 7.6 cm w0d3 measured svd3 constant
Live Herbs w0lh measured svlh constant
Woody w0lw measured svlw constant
Fuel bed properties
Depth [m] d measured
Moisture of Extinction [%] mx measured
Particle density [kg/m3 ] ρ constant
Total mineral content [%] st constant
Eff. mineral content [%] se constant
Heat content [kJ/kg] heat constant
Table 6.1: Fuel model properties
The mathematical description of the structure and texture of a fuel type (i.e.,
vegetation type prone to burn) is a fuel model. The fuel model that is used with
Rothermel’s model consists of several variables that describe the properties of
the particles (fuel loading and surface-to-volume-ratio). Additional variables
characterise the fuel bed as whole (Rothermel 1972, Burgan and Rothermel
1984). In Table 6.1 the set of variables needed to define a fuel type is listed
(showing variable names and indicating additionally which ones are measured
or held constant). The acquisition of fuel data in the field by measuring all
the required variables is very laborious and time consuming, especially if so-
phisticated sampling point distribution methods are applied (Brown et al. 1982,
Anderson 1982, Allgöwer et al. 1998). For large areas and sufficiently homo-
geneous fuels it is common to apply secondary data like vegetation maps (e.g.,
derived from aerial or space borne data) and to assign fuel models using photo
guides (Fischer 1982).
Besides this rather static description of the fuel, the fuel moisture exhibits
a more dynamic behaviour. For each particle class the moisture content is de-
scribed as the percentage of water content compared to oven dry weight (variable
names: md1 , md2 , md3 , mlh , and mlw ; Subscripts according to the schema of
table 6.1). The amount of water in dead fuel particles ranges from 1-2% in very
dry conditions up to 30-35%. Live fuel moisture content is generally between
50 and 300% (Pyne et al. 1996).
As already mentioned, the terrain is introduced (independently of the fuel)
by slope [º] (slp) and aspect [º] (asp) and wind is described as wind speed [m/s]
(wsp) and wind direction at mid-flame height (wdr).
6.3 Uncertainty Propagation

The main goal of this study is to evaluate uncertainty propagation methods that
can be applied for wildland fire risk analysis. Within the proposed risk analysis
framework we are primarily interested in the uncertainties of the intermediate
modelling steps and results. I.e, the results of fire behaviour modelling are used
for subsequent fire spread simulations. To study the propagation of uncertainties
it is sufficient to determine them “locally”. In this context, “locally” denotes
the estimation of the impact of variations of input variables with respect to any
fixed (nominal) input values. In comparison, global methods can be used to
determine uncertainty and sensitivity measures over the whole range of input
values (Saltelli et al. 2000).
6.3.1 Monte Carlo Simulation vs. Taylor Series

Rothermel’s fire behaviour model consists of about 80 equations. If used to its
full extent, it requires 17 input variables: Eight describing the fuel type, five for
fuel moisture, two for wind and terrain each. Most of the input variables are
continuously distributed in space – and many of them are correlated. At first,
the complexity of the Rothermel’s model suggests the application of Monte Carlo
simulations (Johnson 1987) as this does not require any conditions for the model
itself but uses it as a black box. Any desired level of accuracy of the method can
be achieved by increasing the number of runs. Thus, it could be used to generate
the “ground truth” against which other methods are verified. Yet, the problem
with Monte Carlo simulations is the generation of adequate random input data
set realisations that consider the joint distribution of all input variables. In
the case of fire behaviour modelling, spatial and multivariate correlations of 17
input variables have to be considered. There are several methods that allow the
consideration of correlated input variables (conditional simulation), e.g. joint
sequential Gaussian simulation. But their computational load is usually very
demanding. This adds up with the fact that with changing input variables
the whole Monte Carlo simulation has to be rerun. We decided to apply the
6.3 Uncertainty Propagation 61
analytical method of first order Taylor Series (Heuvelink 1998). At a first glance,
the derivation of around 80 equations looks inhibiting, but the great advantage
of this method is that the implementation work has to be done once and can then
be applied to any input data set without great effort. By applying a first order
Taylor series an approximation error is introduced that can be considerably large
if the variation of the input uncertainty is large compared to the “smoothness” of
the function. Precondition for the application of the Gaussian error propagation
is (a) that the Rothermel’s model is continuously differentiable and (b) that all
variables have errors that follow a Gaussian distribution with an expectation
value of zero.
Except for two equations that are not continuous functions over the whole
range of values, the Rothermel’s model is continuously differentiable with re-
spect to all input variables. The first exception is the calculation of a moisture
damping coefficient that uses the ratio between fuel moisture and moisture of
extinction. If the fuel moisture content is higher than the moisture of extinction,
the combustion processes cannot take place and the rate of spread is zero. In
the context of wildland fire risk analysis, this limitation is not a problem as fires
that burn under such conditions are usually not of great interest: They burn
but slowly and with very low intensities which means that they can easily be
suppressed and the chance for developing into a fire storm are very limited.
The second exception is a function that implements an upper limit for wind
speed.At higher wind speeds it is very probable that a fire evolves into a crown
fire and that spotting occurs (i.e., burning or hot particles are blown over hun-
dreds of meters which can start new fires when falling down into combustible
vegetation). This is clearly a situation that violates the limitations for the
application of Rothermel’s model which was designed for surface fires first of
all.
During all computations these two limitations have been tested and no case
was found where either fuel moisture or wind speed were critical.
6.3.2 Partial Derivatives of ros, sdr and ef w

17 derivations of rate of spread had to be derived for the variables w0d1 , w0d2 ,
w0d3 , w0lh , w0lw , svd1 , md1 , md2 , md3 , mlh , mlw , d, mx, wsp, wdr, asp and slp.
The variance of the rate of spread σ 2 can then be approximated by (Heuvelink
1998)
n X
X n
σ2 ≈ ρij fi0 fj0 σi σj (6.1)
i j
σi is the standard deviation and fi0 is the partial derivative of rate of spread
with respect to the variable i = w0d1 , w0d2 , ..., slp. ρij is the correlation be-
tween the variables i and j. Similarly, for direction of maximum spread and
effective wind speed the partial derivatives with respect to the following 11
input variables were derived: w0d1 , w0d2 , w0d3 , w0lh , w0lw , svd1 , d, wsp,
wdr, asp and slp. As the Rothermel model is a system of around 80 equa-
tions, it is beyond the scope of this paper to reproduce all relevant equa-
tions. The complete collection of formulae can be obtained from the authors
(http://www.geo.unizh.ch/gis/research/edmg/fire/unc.html).
6.4 Study Case

6.4.1 Test Site Malcantone
A small area in the Malcantone valley in Southern Switzerland was chosen as
test site (see figure 6.1). The Malcantone is representative for regions with
winter wildland fires. The main fire season is during March and April, before
the onset of vegetation and when the meteorological conditions are favourable
for drying the dead leaves from the precedent fall (Conedera et al. 1996).
Figure 6.1: Area of interest with the three fuel types, DHM-Data: ©Swiss
Federal Office of Topography 1995
6.4.2 Fuel Types

The fuel types were taken from Harvey (1996) who carried out a field study
during the fire season in late winter 1995. In order to optimise the number
of sampling points, four strata were defined based on a forest vegetation map.
On each stratum a regular raster with random origin was laid such that the
number of sampling points was equal for all of the strata. At every point
the fuel was collected and analysed according to the methodology of Brown
et al. (1982). Thorough cluster and discriminant analysis provided three fuel
types based on the averaging of the values of all sample points of each stratum.
Chestnut (Castanea sativa) is the most frequent fuel model covering an area
of 3.57 km2 (57%) and is very typical for the lower parts of the Malcantone
and the southern Alps in general. Frequently burned areas are characterised
by various fern (Polypodium) and broom species (Genista) – vegetation types
that are prone to fires. This fuel type covers about 5% of the forested areas.
6.4 Study Case 63
The third fuel type with coniferous species covers the smallest area but has the
biggest variation in its input variables.
The mean and standard deviations values of the sampled fuel properties are
shown in table 6.2 for each fuel type. Neither of them had dead fuel particles
of the size class 2.5-7.6 cm. The standard deviation within each fuel type class
is used as an estimator for the unknown uncertainty. The rest of the variables
(table 6.3) were held constant for all of the fuel types according to Andrews
(1986).
Coniferous Burned Areas Chestnut

mean σ mean σ mean σ
Loading, dead part., <0.6cm (w0d1 ) 0.95 0.28 0.95 0.21 0.53 0.13
Loading, dead part., 0.6-2.5cm (w0d2 ) 0.03 0.05 0.05 0.05 0.07 0.08
Live, woody particles (w0lw ) 0.01 0.03 0.13 0.18 0.02 0.03
Surface-to-volume-ratio, <0.6cm (svd1 ) 7215 3406 10851 2220 10383 1422
Depth of fuel bed (d) 0.07 0.06 0.35 0.18 0.1 0.06
Moisture of Extinction (mx) 32 13 16 2 18 4
Table 6.2: Values for the three fuel types in the Malcantone valley
Property Unit Value

Loading, dead particles, 2.5-7.6 cm (w0d3 ) [kg/m2 ] 0.0
surface-to-volume-ratio, 0.6-2.5 cm (svd2 ) [m−1 ] 357
surface-to-volume-ratio, 2.5-7.6 cm (svd3 ) [m−1 ] 98
surface-to-volume-ratio, herbs (svlh ) [m−1 ] 4920
surface-to-volume-ratio, woody (svlw ) [m−1 ] 4920
total mineral content (st) [%] 5.5
effective mineral content (se ) [%] 1.0
particle density (ρ) [kg/m3 ] 513
heat content (heat) [kJ/kg] 18600
Table 6.3: Properties with constant values for all three fuel types
One advantage of analytical error propagation method is that correlations

between the variables can be explicitely integrated. Thus, the correlation coef-
ficients between all the variables in table 6.2 have been calculated for each fuel
type based on the original sampling data as shown in table 6.4.
6.4.3 Fuel Moisture

The base moisture content was assumed to be 8% and was corrected for height
and aspect according to the Fire Behaviour Field Reference Guide (NWCG
1992). We assumed as standard deviation 0.8%. These values were assigned to
all three classes of dead fuel particles. For the living fuel particles values were
taken from the literature (150% mean and 15% standard deviation) (NWCG
1992, Rüegsegger 1996).
Coniferous w0d1 w0d2 w0lw svd1 depth mx

w0d1 1.000
w0d2 0.678 1.000
w0lw 0.161 0.172 1.000
svd1 0.403 -0.085 -0.063 1.000
depth 0.224 0.128 0.702 -0.242 1.000
mx -0.199 0.001 -0.177 0.079 -0.615 1.000
Burned Areas w0d1 w0d2 w0lw svd1 depth mx
w0d1 1.000
w0d2 -0.170 1.000
w0lw 0.550 -0.336 1.000
svd1 0.280 -0.175 -0.106 1.000
depth 0.439 -0.076 0.733 -0.082 1.000
mx 0.590 0.010 0.393 0.119 -0.135 1.000
Chestnut w0d1 w0d2 w0lw svd1 depth mx
w0d1 1.000
w0d2 -0.334 1.000
w0lw -0.072 0.022 1.000
svd1 0.261 -0.388 -0.356 1.000
depth 0.025 0.029 0.832 -0.321 1.000
mx 0.189 0.061 0.017 -0.124 -0.282 1.000
Table 6.4: Correlation-matrix for fuel model variables
6.4.4 Terrain and Wind

Slope and aspect were derived from a digital terrain model (cell size: 25 m) of
the Swiss Federal Office of Topography (SFOT). The error in height (σz ) lays
between three and five meters (SFOT 2000). The slope angle was calculated in
ARC/INFO which usesp the method of Horn (1981). The standard deviation of
the slope is given as σz 3/(4 · cellsize) . Aspect (clockwise angle from north)
was also calculated in ARC/INFO. The standard deviation could be derived in
similar ways as shown for slope, but we assumed σ=10 degrees for all cells as
a study of Wise (1998) showed only weak spatial variation. Correlations have
been calculated between slope and aspect for each fuel type area separately
(table 6.5).
Coniferous Burned Areas Chestnut

ρslp,asp 0.587 0.100 0.033
Table 6.5: Correlations between slope and aspect
Due to the lack of reliable data, wind speed and direction were set to constant
values for the whole area of interest. The authors are well aware of the fact that
this would not be sufficient for realistic wildland fire behaviour analysis. Since
6.5 Results 65
the main interest of this study lays on the application of uncertainty propagation
and not on the expectation values, it is assumed acceptable. Wind speed was
assigned an expectation value of 2.3 m/s ( σ= 1.0 m/s) and wind direction was
set to 0◦ (wind coming from the north, σ= 45 ◦ ). However, these values are not
completely arbitrary but are characteristic for the main fire season in the study
area; they correspond to the mean values for March and April (Rüegsegger
1996).
6.4.5 Procedure
The test site (2.5 x 2.5 km) was chosen so that all three fuel types are represented
in order to have a wider variety of input values. However, not the whole area
is covered with fuel so that in the end 6324 cells have contributed to the study.
All data were rasterised to a cell size of 25 meters. In a first step ros, sdr
and ef w were calculated for all cells based on the local input variable values.
Then, all partial derivatives for ros, sdr and ef w were calculated for all cells and
finally summed up according to equation 6.1 to obtain the variance. Finally, the
partial derivatives together with the standard errors of the input variables were
analysed statistically in order to find the most important error contributors. It
is noteworthy that the performed local sensitivity analyses are preliminary as
the linearisation due to the Taylor method is likely to fail if the model is non-
linear and the uncertainties of the input values are large. For a deeper analysis
of sensitivities global methods should be used (Turány and Rabitz 2000).
6.5 Results
When looking at the results, the range of error values strikes. However, the
scope of this paper is not to explain how and why the results from fire behaviour
modelling may vary to the extent it was observed during this study. The results
rather show, that this is a very critical point and that in fire behaviour modelling
much more emphasis should be put on the assessment of input data and results.
6.5.1 Rate of Fire Spread

In figure 2a, the three fuel types can be distinguished very clearly based on
the different rates of spread. The fuel type on the burned areas produces the
highest rate of spread with an average of 0.241 m/s, or ca. 860 m/h which
is not really fast. The remaining, within fuel type variation, reflects mainly
the influence of the terrain. This is due to the fact that slope, aspect and fuel
moisture are the only remaining input parameters besides the fuel types that
are not constant in the area of interest. The spatial pattern of the coefficients
of variation (figure 2b) shows only some influence of fuel characteristics. While
the coniferous stands have quite uniformly high values, the burned areas and
chestnut types exhibit patterns which do not differ greatly.
In table 6.6, for each fuel model the mean value, standard deviation (σ) and
the coefficient of variation (cv) are listed. It is important to note that σ and
cv denote the averages of all individual, estimated σi , resp. cvi (i=1..number
of cells) within each fuel type. They do not represent the standard deviation,
resp. coefficient of variation, of the individual rosi . Generally, the standard
Figure 6.2: Results of fire behaviour modelling: a) rate of spread, c) spread

direction (deviation of wdr) and e) effective wind speed. Results of uncertainty
propagation: coeff. of var. b) of rate of spread and f) effective wind speed and
d) the standard deviation for spread direction. (black=lowest, white=highest
value)
errors for the rate of spread are very high which is mainly a result of the great
uncertainties in the input values. This is especially true for the coniferous stands
with, nearly exclusively, the highest input uncertainties in the fuel variables.
Generally, only very few variables contribute to a significant part of the error.
The main contributors are fuel bed depth d, wind speed wsp and the surface-
to-volume-ratio of the finest dead fuel particles svd1 .
ros [m/s] mean σ cv main error contributors (% of σ)

Coniferous 0.026 0.063 2.45 svd1 (41%), d (28%), w0d1 &svd1 (16%)
Burnt areas 0.241 0.383 1.81 wsp (56%), d (13%), wdr (9%)
Chestnut 0.108 0.127 1.21 wsp (42%), d (34%), w0d1 (12%)
Table 6.6: Rate of spread
6.5.2 Fire Spread Direction

The direction of maximum fire spread is basically a function of wind and ter-
rain. As the the wind is held constant, the spatial pattern in figure 2c reflects
6.6 Conclusions 67
again primarily the terrain. For better interpretability the spread direction is
visualised as absolute deviation from South (direction where the wind is blowing
to, white = small, black = high deviation). The standard deviation (figure 2d)
shows also some influence of the terrain: The errors are especially great on the
south facing, steep slopes. The most sensitive input parameters are for all three
fuel types wind direction followed by wind speed (see table 6.7).
sdr [◦ ] mean σ main error contributors (% of σ)

Coniferous 194.9 78.7 wdr (79%), svd1 (10%), wsp (5%)
Burnt Areas 193.0 93.2 wdr (66%), wsp (26%), slp (3%)
Chestnut 183.6 53.0 wdr (82%), wsp (13%), asp (4%)
Table 6.7: Direction of maximum fire spread
6.5.3 Effective Wind Speed

Effective wind speed denotes the relative contribution of wind speed to the rate
of spread. This leads to the low values on steep, south facing slopes (figure 2e).
The standard error (figure 2f) seems to be higher where the effective wind speed
is smaller (white values) which is proved by a correlation coefficient of -0.97. As
listed in table 6.8 the main contributor of the estimated error is the uncertainty
of the wind speed.
efw [m/s] mean σ cv main error contributors (% of σ)

Coniferous 1.813 1.211 0.70 wsp (84%),wdr (10%), slp (3%)
Burned Areas 1.715 1.415 0.94 wsp (79%), wdr (17%), slp (2%)
Chestnut 2.272 1.028 0.47 wsp (91%), wdr (8%), slp (1%)
Table 6.8: Effective wind speed
6.6 Conclusions
The application of first order Taylor series for uncertainty propagation is very
helpful for analysing the contributions of the input variables’ uncertainties. The
experiences gained in the test case are promising. The analysis of uncertainties
for such a complex model like the Rothermel’s model will help to gain more
insight into the environmental process wildland fire. However, further investi-
gations have to be carried out to verify that the approximation errors of the
Taylor series approach are negligible. In the case of local uncertainties this
could be done by comparing them to the results of a Monte-Carlo-Simulation.
For the verification of the sensitivities global methods could be applied (see
Sobol’ (1993) and Saltelli et al. (2000)).
In the chosen test case the most important result is that the “errors” which
are to a big part due to a natural variation lead to a big error in the results.
Another result is that usually only a few input variables make up for the biggest
part of the error. Although, the results cannot be extrapolated, we liked to
refer to the results of a study presented by Salvador et al. (2001 in press).
They carried out a global sensitivity analysis for the same Rothermel model in
shrub vegetation of North-Eastern Spain and found that all variables had some
noticeable effect.
The predominant influence of few variables has been clearly demonstrated
in the case of direction of maximum fire spread (sdr) and effective wind speed
(ef w) where wsp and wdr are the key uncertainty contributors. It is not likely
that the input uncertainties of a wind field will get smaller in future as the
uncertainties are an inherent property of it. Thus, ways have to be found to
integrate this natural variation adequately into fire behaviour modelling and
analysis.
In the case of rate of spread, it can be shown that the big errors (cv > 1) stem
from the high input uncertainties. The three fuel types have been compared with
the 13 standard NFFL fuel types (Anderson 1982). These fuel types cover the
most widely found fuel situations in the USA. They have been constructed for
operational use during fire fighting activities when rapid assessment of the fuel
situation is needed. In order to compare the various fuel types a no-wind and
no-slope situation is assumed. The uncertainties of the all input variables are
equal to 10% of the variables values. Table 6.9 shows the relative contribution
of w0d1 , svd1 and d which are found, with some exceptions,to be the most
important contributors to the error of rate of spread.
Fuel Type cv w0d1 svd1 d others

NFFL 1 0.17 28 41 23
NFFL 2 0.20 41 36 16
NFFL 3 0.11 43 11 31
NFFL 4 0.15 25 17 21 w0lh 12 mlh 20
NFFL 5 0.13 13 14 26 w0lw 15 mlw 29
NFFL 6 0.14 15 9 26 w0lw 20 mlw 27
NFFL 7 0.15 47 12 19
NFFL 8 0.47 32 53 9
NFFL 9 0.41 49 41 7
NFFL 10 0.24 31 36 17
NFFL 11 0.17 47 11 28
NFFL 12 0.21 35 21 27
NFFL 13 0.26 29 35 21
Coniferous 0.39 53 38 8
Burned Areas 0.24 32 49 13
Chestnut 0.27 45 41 11
Table 6.9: Sensitivity analysis for the 13 NFFL and the 3 Ticino fuel types
A still open question is that of the causes and extent of uncertainties in the
fuel input variables. From the limited experience (unfortunately literature on
this topic is very scarce) of the authors, we believe that fuel is a highly variable
phenomenon and that coefficients of variation of 10% in many cases are rather
Bibliography 69
optimistic. To overcome this, more research activities are needed that focus
on the analysis and description of the spatial variability of fuel parameters.
This information could then be used to produce more ’realistic’ input data
sets using stochastic simulation (Ripley 1987). It is interesting to note that
stochastic modelling techniques are already being used for fire spread simulation.
Catchpole et al. (1989) used Markov chains, cellular automata are applied by
Wolfram (1984), Gonçalves and Diogo (1994), Clarke and Olsen (1996). In
all studies input variables are considered in a completely deterministic way
while the fire spread process is simulated as a stochastic process. It would be
interesting to compare this with the results of the combinations “stochastic input
data - deterministic spread simulation” and “stochastic input data - stochastic
spread simulation”.
For the application of Rothermel’s model in the context of wildland fire
risk analysis the results show that it was definitely worth analysing uncertainty
propagation. The considerable errors in the test case cannot be neglected and
have to be further analysed. In any case, they should not be used as a pretext
for not applying Rothermel’s model. In contrary, they represent additional
information about the variation of wildland fire behaviour and can be used for
improving decision in the context of wildland fire risk analysis.
Generally, we would like to revoke the fact that the Rothermel model is
a semi-empirical model with all its limitations that are inherent in models.
Specifically, the parameterisation of the model is only feasible by reducing and
concentrating on some aspects of the real world process.
Moreover, additional research has to be carried out on how to communicate
the uncertainties to the respective research and user communities.
Bibliography
Albini, F. A., 1976, Estimating wildfire behaviour and effects. General Tech-
nical Report INT-30, USDA Forest Service, Intermountain Forest and Range
Experiment Station.
Alexandrian, D., 1996, A new method of fire danger mapping in the for-
(Workshop), Athens.
Allgöwer, B., Harvey, S. and Rüeggsegger, M., 1998, Fuel models for
switzerland: Description, spatial pattern, index for torching and crowning.
3rd International Conference on Forest Fire Research / 14th Conference on
Fire and Forest Meteorology, edited by D.X. Viegas, vol. 2, pp. 2605–2620,
Luso (P).
Anderson, H. E., 1982, Aids to determining fuel models for estimating fire be-
havior. General technical report, USDA Forest Service, Intermountain Forest
and Range Experiment Station.
70 Bibliography

pp. 2177–2190, Luso (P).
Brown, James K., Oberheu, Rick D. and Johnston, Cameron M., 1982,
Handbook for inventorying surface fuels and biomass in the interior west. Gen-
eral Technical Report INT-129, USDA Forest Service, Intermountain Forest
and Range Experiment Station Ogden, UT.

Station.
BUWAL, Bundesamt für Umwelt, Wald und Landschaft, 1999, Risiko-

analyse bei gravitativen Naturgefahren: Methode. No. 107/1 in Umwelt-
Materialien, Naturgefahren, BUWAL, Bern.
Catchpole, E. A., Hatton, T. J. and Catchpole, W. R., 1989, Fire

pread through nonhomogeneous fuel modelled as a markov process. Ecological
Modelling, 48:101–112.
Clarke, K. C. and Olsen, G., 1996, Refining a cellular automaton model

of wildfire propagation and extinction. GIS and environmental modeling:
progress and research issues, edited by Parks B. O. Goodchild F. M., Steyaert
L. T., pp. 333–338, GIS World Books.
Conedera, Marco, Marcozzi, M., Jud, B., Mandallaz, D., Chatelain,

F., Frank, Carmen, Kienast, F., Ambrosetti, P. and Corti, G., 1996,
Incendi boschivi al Sud delle Alpi: passato, presente e possibili sviluppi futuri.
Rapporto di lavoro del Programma Nazionale di Ricerca ”Mutamenti climatici
e catastrofi naturali”, PNR 31, vdf Hochschulverlag, Zürich.
Fischer, W. C., 1982, Photo guide for appraising downed woody fuels in mon-
tana forests: Grand fir-lach-douglas-fir, western hemlock, western hemlock-
western redcedar, and western redcedar cover types. General Technical Report
Station.
Gonçalves, P.P. and Diogo, P.M., 1994, Forest fire propagation modeling:
a new methodology using cellular automata and geographic information sys-
tems. 2nd Int. Conference on Forest Fire Research, pp. 209–219, Coimbra,
Portugal.
Harvey, S., 1996, Brandgutdaten in der Waldbrandmodellierung. Diplomar-

beit, Universität Zürich Irchel.

Horn, B.K.P., 1981, Hill shading and the reflectance map. Proceedings of the
Institute of Electrical and Electronics Engineers, 69, pp. 14–47.
Bibliography 71
Inc., New York.
Miller, Carol, Landres, Peter B. and Alback, Paul B., 1999, Evalu-
ating risks and benefits of wildland fire at landscape scales. The Joint Fire
Science Conference and Workshop, edited by Greg E. Gollberg, vol. 1, pp. 78–
87, University of Idaho and the International Association of Wildland Fire,
Grove Hotel, Boise, Idaho.
Rüegsegger, M., 1996, Entscheidungsunterstützung im Waldbrandmanage-
ment mit GIS; Strategien und ausgewählte Beispiele. Diplomarbeit, Univer-
sität Zürich Irchel.
Ripley, Brian D., 1987, Stochastic Simulation. Wiley Series in Probability
and Mathematical Statistics, John Wiley & Sons.
Saltelli, A., Chan, K. and Scott, M. (Editors), 2000, Sensitivity Analysis.
Wiley series in probability and statistics, John Wiley & Sons, England.
Salvador, R., Piñol, J., Tarantola, S. and Pla, E., 2001 in press, Global
sensitivity analysis and scale effects fo a fire propagation model used over
mediterranean shrublands. Ecological Modelling.
SFOT, Swiss Federal Office of Topography, 2000, Product Description
for DHM25.
URL http://www.swisstopo.ch/en/digital/dhm25.htm
Sobol’, I.M., 1993, Sensitivity estimates for nonlinear mathematical models.
Mathematical modeling and computational experiment, 1(4):407–414.
Turány, Tamas and Rabitz, Herschel, 2000, Local methods. Sensitivity
Analysis, edited by A. Saltelli, K. Chan and M. Scott, Wiley series in proba-
bility and statistics, John Wiley & Sons, England.
Wise, S.M., 1998, The effect of GIS interpolation errors on the use of digi-
tal elevation models in geomorphology. Landform monitoring, modelling, and
analysis, edited by Stuart N. Lane, Keith S. Richards and Jim H. Chandler,
pp. 139–164, John Wiley & Sons.
Wolfram, S., 1984, Cellular automata: a model of complexity. Nature,
311(4):419–424.
7
Wildland Fire Risk Analysis

in Southern Switzerland

of Zurich∗
and
Marco Conedera
WSL Sottostazione Sud delle Alpi
PO BOX 57, via Belsoggiorno 22
6504 Bellinzona
Phone +41 91 821 52 31, Fax +41 91 821 52 39
email: marco.conedera@ wsl.ch
Abstract
Wildland fires creates primary and secondary effects in the area where they
occur. To assess and validate the role of wildland fires in an area both issues
are important. In Southern Switzerland, the loss in timber value due to fires
can be neglected, whereas the destruction or partial damage of the protective
functions of the affected forests may have major economic consequences as the
probability for secondary effects such as debris flows, erosion, landslides, etc.
increases. As objects of interest catchment areas for potential debris flows have
been considered. In Southern Switzerland, the most critical time period for fires
is March/April. This is a problem for catchment areas that are hit by a large
fire because of the limited recovering time before the heavy summer rainfalls.
A GIS-based framework for spatial wildland fire risk analysis was applied and
special emphasis was put on the post fire risk of debris flows. In the so called
risk matrix fire occurrence, fire behavior and fire effects have been combined in
order to calculate the spatial distribution of wildland fire risk.
∗ accepted for publication in Proceedings of the Fire Conference 2000, held in San Diego,
CA, on Nov.27–Dec.1, 2000 (full paper review).

74 Wildland Fire Risk Analysis in Southern Switzerland
Keywords: Wildland fire risk, risk analysis, fire occurrence, fire behavior, fire
effects, debris flow.
7.1 Introduction
In Southern Switzerland, the primary impacts of wildland fires are minimal.
Timber itself has little economical value and properties are only rarely affected
seriously by wildland fires. This is due to the moderate intensities of the surface
fires that usually prevail in this region. In most cases fire suppression activities
can protect properties and infrastructure very effectively. Yet, because of the
rugged terrain secondary impacts are much more important. The destruction
of the vegetation and the denudation of the soil significantly increase the risk
of erosion, land slides and debris flows. The goal of this study was to carry out
a wildland fire risk analysis with special emphasis on the post fire risk of debris
flows. There is strong evidence that heavy rainfalls throughout the summer
provoke devastating debris flows affecting not only the burned surface but also
objects of interest in the vicinity.
7.2 Methods
7.2.1 Wildland Fire Risk Analysis
The GIS-based framework for wildland fire risk of Bachmann and Allgöwer
(1998) allows the quantitative determination of the spatial pattern of probability
and damage of wildland fires. The core of the framework consists of the so called
risk matrix which was proposed originally by Merz et al. (1995) for technical risk
analysis. Schöning et al. (1997) adapted it to the special requirements of the
natural hazard wildland fire. In technical risk analysis the term risk incorporates
two aspects: the probability and the outcome of an undesired event (Jones 1992).
For a quantitative analysis it is usually operationalized by building the product
of both parts (see equation 7.1).
r =p·d (7.1)
p denotes the probability or frequency of an event within a given time period.
d describes the impact of the event and is usually expressed in monetary terms
for comparability reasons. For this study we define the term wildland fire risk
as the probability of a wildland fire to start at a specified location and under
specific circumstances, together with its expected outcome as defined by its
impacts on the objects it affects (Bachmann and Allgöwer 2001).
The key elements of the risk matrix are scenarios (risk donors), objects
(risk acceptors) and situations. A scenario represents a possible starting point
of a wildland fire and is defined by a location and a probability that the fire
starts. Objects are any entities, e.g. properties, power lines, etc. that are
of interest for the managing organization. Situations define the general fire
behavior environment, i.e. fuel types, moisture contents, wind and terrain.
For each scenario-object pair the conditional impact probability eij and the
expected impact dij has to be calculated (see figure 7.1). The impact probability
is the probability that a fire starting at scenario Si reaches object Oj . It is
7.2 Methods 75
Scenarios Individual Risk

S1 S2 Si Sn-1 Sn of an Object
... ...
p1 p2 pi pn-1 pn
O1
O2
...
Σ kij
n
eij kj =
Objects
i=1
Oj kij = pi ⋅ eij
Σ kij ⋅ dij
n
dij dj =
i=1
...
Om-1
Om
Σ ki
n
Σ eij
m
Collective ki = pi ⋅ k=
j=1 i=1
Risk of
Σ pi ⋅ di
n
Σ eij ⋅ dij
m
a Scenario d=
di =
j=1 i=1
Figure 7.1: Risk matrix for one situation
primarily a function of fire behavior and spread time. The expected impact dij
describes the impact that results from a fire starting at scenario Si . The risk
matrix allows the calculation of risk values for scenarios and objects as well as
for the whole area of interest by summing up the individual impact probabilities
and expected impacts. For an object we get the probability of being affected
by any fire in its neighborhood (kj ) and the expected damage (dj ). For each
scenario the expected number of objects that will be affected by it (ki ) and the
total expected damage (di ) can be calculated. The overall sum of the matrix
gives risk values for the whole area of interest.
An important task is the choice of an appropriate impact probability and
damage function. The framework does not impose the application of a pre-
defined function but allows users to ’plug in’ their own tailored methods. The
only requirements are that the impact probability function returns a probability
while the damage function returns values on a ratio scale. We will first present
the impact probability function and afterwards define the damage function.
7.2.2 Impact Probability Function

An ideal function for the conditional impact probability eij should reflect the
time a fire needs to reach an object. I.e., fires starting at ’far away’ scenarios are
less likely to reach an object than fires that are very close to it. Nevertheless,
euclidean distance is not a good indicator as it disregards the spatial variability
of wildland fires. A better solution is to calculate the spread time and transform
it with a heuristic approach into probabilities. Schöning et al. (1997) proposed

an exponential function, however it is difficult to validate the parameters of the
function. For this study, we chose the following procedure to overcome this
shortcomings.
The foundation is a very detailed database of wildland fires events in Switzer-
land which is operated by the Swiss Federal Research Institute for Forest, Snow
and Landscape (WSL). Based on records of past fires in the Malcantone region
between 1980 and 1998 a histogram of spread times was derived. This histogram
was then used to build an empirical inverse cumulative distribution (see figure
7.2).
0.8
0.6
Probability
0.4
0.2
0
0 720 1440 2160 2880 3600 4320
Time [min]
Figure 7.2: Probability function depending on spread time (Source: Database

of WSL)
For every object-scenario pair the spread time was calculated using Rother-
mel’s model (Rothermel 1972, Albini 1976) and a raster based implementation
of Anderson (1983) model for the general shape of a fire line. The spread time
was then transformed into a probability by looking up the correspondent value
in the inverse cumulative distribution.
Using this method we are probably overestimating the area of potential sce-
narios as in most cases a wildland fire will not spread unhindered for most of
the time. By omitting the influence of fire fighting activities the spread simula-
tion will underestimate spread time. On the other hand the spread times in the
database are actually the time span between alarm arrival and the extinction of
a fire. This way we miss the time between the ignition and the alert. Comparing
these two shortcomings two observations can be made. The under prediction of
spread time is a more serious problem for larger fires for which we can assume
longer periods of fire fighting. The second shortcoming, ignoring the detection
lag, has a strong influence on small fires. If we assume a uniform detection lag
for all fires its relative influence is greater for fires with a shorter duration.
The derivation of a impact probability function based on past fires is elegant,
because the spread times include not only the fire behavior but also fire fighting
activities. It should be possible to analyze the effectiveness of a fire fighting
organization in the risk analysis.
7.3 Case Study 77
7.2.3 Damage Function

The damage function considers properties of the object itself and the local fire
behavior in the neighborhood of the object. In this study, primary damage, i.e.
loss of timber, is not considered. But we analyze the post fire risk of a debris
flow. In this sense the damage function has the following form
dij = pdf · ddf (7.2)

pdf is the probability that the debris flow occurs and ddf is the damage of the
debris flow. In this study we assumed that the probability of a post fire debris
flow is depending on how fast the vegetation covers the soil and the probability
of heavy rain events during the summer time. Hofmann et al. (1998) found
that frequent fires and short time lapses since the last fire lead to well adapted
vegetation stands which recover very fast, thus reducing the risk of a debris
flow. Subsequently, the vegetation is classified into three classes according to
the adaptiveness to fire (see table 7.1). The more adapted the vegetation is the
more soil is covered in the first growing season. For bare soils Zimmermann
et al. (1997) found a threshold of 83 mm for 24h rain events to trigger debris
flows. The more soil is covered the heavier rain storms are needed to trigger a
debris flow. The probabilities for rain events are derived from the return period
charts for regional weather stations (Zeller et al. 1980).
Class Adaptiveness Rain Intensity Return pdf

[mm/24h] Period [y]
1 no fire adapted species 83 3 0.333
2 few fire adapted species 150 8 0.125
3 fire adapted species 200 20 0.050
Table 7.1: Fire adaptiveness, rain intensities and derived return periods, and
yearly debris flow probabilities
Although debris flows can threaten human life we only considered structural
damage. Characteristic values for structural damage of debris flows on various
object categories could be found in Borter (1998, p.120). Medium intensity
debris flows (height of deposition ≈ 1 m) were assumed. The damage values
have been assigned to every catchment area depending on the object category its
debris flow affects. For settlements we assumed that an area of 0.5 ha is affected
and for roads we assumed uniformly a spilled section of 25 m (see table 7.2).
7.3 Case Study

7.3.1 The Malcantone
The study area (76.3 km2 ) consists of the Malcantone region which is situated
in the pre-alps of Southern Switzerland. Vegetation wise, it belongs to the wide
spread chestnut belt of the Southern Alps. The general climate is defined by
relatively dry winters and wet summers.
Object Category Specific Damage Base Unit Applied Damage ddf

[CHF ] [CHF ]
Settlement core 1750 000.− 0.1 ha 8750 000.−
Cantonal roads 560.− 1m 140 000.−
Community roads 260.− 1m 60 500.−
Trails, Forest Roads 95.− 1m 20 375.−
Table 7.2: Objects categories and associated damage values (Source: Borter
(1998))
7.3.2 Objects: Catchment Areas of Potential Debris Flow

Debris flow is a process which is defined by the properties of the catchment
area (proliferation of debris) and the local geometry of the stream bed. The
catchment areas have been defined using a procedure that is published by the
Swiss Agency for the Environment, Forests and Landscape (Heinimann et al.
1998). The procedure allows the detection of debris flow stream sections based
on a digital elevation model (DEM,in raster format). In a first step the stream
network is calculated. Afterwards, debris proliferating areas in the vicinity of
stream sections are selected. Finally, potential debris flow sections are selected
by comparing the local slope with a marginal slope.
Stream Network For extracting the stream network we applied the common
hydrological approach from Mark (1984) and Jenson and Domingue (1988):
1. Elimination of depressions in the DEM
2. Calculate flow directions
3. Calculate flow accumulation
4. Choose a threshold
Heinimann et al. (1998) proposes a general threshold of 0.25 ha and addi-

tional processing of stream sections with a Strahler number of 1 (Strahler 1957).
To eliminate misclassified stream sections a threshold value of -2 on the hori-
zontal curvature (rate of change of surface perpendicular to aspect) is imposed.
Selecting Potential Debris Proliferating Areas Debris proliferating ar-

eas are adjacent to a stream section and are only considered up to a distance
of 250 meters. The slope along the flow directions on these areas does never
fall below a marginal value. This marginal slope depends on the amount of
vegetation coverage. As we are interested in post fire risk we assumed generally
vegetation-less areas applying a marginal slope of 20°. Using these two con-
straints a buffer around the stream network was generated that represents the
potential proliferating areas.
7.3 Case Study 79
Stream Process The occurrence of a debris flow is a function of the steepness

of a stream section γ [%] and its catchment area A [km2 ] . Using equation 7.3
a marginal slope α is calculated. If the actual slope is steeper a debris flow will
occur provided a sufficient run off.
α = 0.32 · A · γ −0.2 (7.3)
Simulating Range of Debris Flows For every potential debris flow stream
section a marginal slope is calculated using
f = 0.2 · A−0.26 (7.4)
with A being the catchment area [km2 ]. The debris flow proceeds as long as
it does not fall below this threshold.
Applying this methodology many catchment areas are extracted that are of
minor interest because they occur in ”unproductive” regions of the back country.
For this analysis only catchment areas have been selected that produce a debris
flow striking a road or a settlement. Furthermore, in order to minimize com-
puting time sample points were chosen every 100 m using a stratified sampling
within the selected catchment areas. This method resulted in 372 sample points
ensuring the inclusion of every catchment area. Every sample point represents
at least an area of 1 ha which is big enough to trigger a debris flow.
For the application within the risk matrix the sample points had to be
weighted in order to avoid overestimations for catchment areas with more than
1 sample point. E.g., if a catchment has an area of 3 ha and it is represented by
3 sample points, every sample point will trigger the same debris flow and thus
the damage of the debris flow would be incorporated 3 times in the risk matrix.
To prevent this every sample point was weighted with 1/n, n being the number
of sample points per single catchment area. After the calculation the risk values
of the sample points were summed up for each catchment area.
7.3.3 Scenarios
As we do not know where the next wildland fire will start, we have to define
scenarios that represent possible realizations of fires. In this study a scenario
defines a starting point of a fire and the probability that it starts. They were
constructed by overlaying the study area with a regular grid with a cell size of
25 m. Every cell containing burnable fuels is considered a scenario resulting in
a total number of 106’435.
The next step was to define for every scenario the probability that the fire
will start. To achieve this the historical database of the Sottostatzione was
analyzed (Conedera et al. 1996). In a first step temporal patterns have been
analyzed using descriptive statistics and Chi-square tests. In a second step the
spatial patterns of the ignition point have been analyzed (Langhart et al. 1998).
The time period from 1 January 1980 up to 31 December 1998 was considered.
A total of 154 fires occurred in this period in the Malcantone region with
an average burned area of 6.7 ha. Most fires in the Malcantone region are
due to human causes. In 56% of all cases the cause is evidence based human
(arson, negligence, etc.). In 43% the actual cause is unknown, but assumed to
be human. Only 2 fires occurred due to a natural cause (i.e. lightning).
The only temporal pattern that significantly deviates from a uniform distri-
bution is the monthly distribution of the number of fires (see figure 7.3). In the
considered period 30 fires occurred in March and 35 in April (42% of all fires).
Although there is a big variation of the number of fires (standard deviation =
1.84) we define for the Malcantone region the expected number of fires to occur
in the months March and April to be 3.42. If we simply assumed a uniform spa-
tial distribution each scenario would have a probability of 0.00003214 to start a
fire.
20 Unknown
Human
Natural
Number of Fires
15
10
0
1 2 3 4 5 6 7 8 9 10 11 12
Month
Figure 7.3: Monthly distribution of wildland fires in the Malcantone region

(1980-1998)
In order to determine the spatial distribution of fire ignition probability

various independent variables were tested with a Chi-square test. The only
significant variable was distance to a road or a settlement. Fuel type and solar
inclination (calculated as angle between the incoming sun vector and the vector
standing perpendicular on the terrain surface) showed the same distribution
for ignition locations as for the whole area of interest. Hence in the logistic
regression only the euclidean distance to a road or settlement (in meters) was
included. The resulting regression parameters were used to calculate for every
cell the probability of an inception of a fire.
p = (1 + exp(0.4980 + 0.00139d))−1 (7.5)
The spatial distribution was finally weighted in order to get the overall frequency
of around 3.4 fires per year (see figure 7.4).
7.3.4 Fire Behavior

Fire behavior was calculated using Rothermel’s model (Rothermel 1972, Albini
1976) and the following input datasets and parameters have been used.
Fuel The most important fuel type is chestnut litter. Sweet chestnut [Castanea
sativa Mill.] is with 32% the most important fuel type (see table 7.3), followed by
grass (29%) and areas without any fuel (13%, mostly settlements and roads).
Seven fuel types have been used to describe the fuel characteristics. Three
originate from a field study, carried out in 1995 (Harvey 1996, Allgöwer et al.
7.3 Case Study 81
Figure 7.4: Yearly ignition probabilities for the Malcantone Region
1998). For the remaining areas NFFL fuel types (Anderson 1982) were assigned
based on a manual classification of a vegetation and a forest stand map from
the Forest Service of Ticino.
Fuel Moisture The fuel moisture content for the dead particles was corrected
for altitude, slope and aspect according to the NWCG (1992). Base fuel moisture
content was assumed 10% for the smallest, 11% for the medium and 12% for the
largest dead particles. These values have been adopted from an experimental
fire that was carried out in March 1998 (Marxer and Conedera 1999).
Topography We used the digital elevation model of the Swiss Federal Office
of Topography with a cell size of 25 meters. The topography ranges from 237 m
to 1920 m above sea level. Especially in the northern parts we find very steep
terrain. Slope ranges between 0°and 64°. Slope and aspect have been derived
using the GIS ArcInfo.
Fuel Type Description Area [ha] Area [%]

0 no fuel 1066 13.8
NFFL 1 grass 2276 29.5
NFFL 5 brush 566 7.3
NFFL 6 dormant brush 22 0.3
NFFL 8 closed timber litter 1065 13.8
31 cultivated coniferous forest 176 2.3
32 frequently burned areas 79 1.0
33 chestnut 2470 32.0
Table 7.3: Fuel types and their coverage in the study area
Wind Wind speed and wind direction have been analyzed for each historical
fire event at the nearest available weather station of Lugano (Source: database
of the Swiss Meteorological Institute [SMI]). The mean value for wind speed is
3.0 m/s and the most frequent wind direction was north which is a very typical
Foehn situation, with strong, warm and dry winds coming down from the main
Alpine arc. These values have been applied uniformly to every cell in the area
of interest. This is clearly a critical point as it does not reflect the spatial and
temporal variation of a wind field.
7.4 Results
7.4.1 Impact Probabilities
The impact probabilities of the individual catchment areas range from 0.0002 to
0.22 (see figure 7.5), i.e. in the worst case there is a chance of 22% per year that
this catchment area will be affected by a fire. The impact probabilities of the
scenarios range from 0 to 0.00006 and denote the probability that a fire starts
in this cell and that the subsequent fire reaches a catchment area. The spatial
pattern of the impact probabilities shows the influence of the fire behavior and
the fire occurrence probability. E.g., in the northern parts of the study area,
the absence of roads results in low impact probabilities.
7.4.2 Expected Damage

Looking at the resulting map for damage (see figure 7.6) one has to differentiate
between scenario damage (damage caused by a risk donor) and object damage
(damage affecting a risk acceptor).
In the case of the scenario damage one only looks at the damage that may
occur in case a fire starts at that particular scenario (cell). With scenarios being
the risk donor(s), it is possible to calculate the expected damages (expressed
as monetary values) without considering whether or not the events are taking
place (i.e., fire occurrence probability). In our study area the expected damage
values vary considerably from 0 to 1.4 Mio. CHF. A damage of 0 means that
fires starting in these scenarios never reach any catchment areas whereas a
7.5 Discussion and Conclusions 83
Figure 7.5: Yearly impact probabilities for catchment areas and for scenarios
value 1.4 Mio CHF means that more than one catchment area is reached. The
spatial pattern of the expected damage reflects the distribution of post fire risk:
Scenarios close to the catchment areas that threaten settlements have much
higher values than those in the back country.
The object damage is the total sum of all scenario damages weighted by the
impact probability and fire occurrence probability. In the study area object
damage values were found that range from 5.5 CHF up to 36’585 CHF. These
values can be considered as the estimated annual depreciation for the objects
at risk.
7.5 Discussion and Conclusions

The study clearly shows the applicability of the framework. It considers the
spatial variation of all input variables which is very important for such a com-
plex analysis. The same advantage is valid for the results. By investigating
the various probabilities and damages separately, managers are given the op-
Figure 7.6: Yearly expected damages for catchment areas and damages for sce-
narios
portunity for a sophisticated analysis and assessment of a given risk situation.

Moreover, the modular structure of the framework allows the joint consideration
of various hazards thus enabling a comprehensive and integrated environmental
risk analysis.
Nevertheless, great care must be used when interpreting the results. Due
to the lack of data on occurrence of post fire debris flows it is not possible to
validate the analysis entirely. In addition, an important subject not treated in
this study is the analysis of uncertainty and error propagation. Bachmann and
Allgöwer (2000) studied the error propagation of the Rothermel model in the
same region and found that e.g. for rate of spread the expected values have a
mean coefficient of variation of 1.21 (chestnut litter fuel type). This huge error
is mainly a result of the spatial variability of the input variables.
Finally, much more emphasis could be put on the modeling of the range of
the debris flows. However, it is probably faster and more reliable to carry out
a field study as the path of a debris flow can be influenced by small landscape
Bibliography 85
features which are obvious to see in the field but which are not properly modeled
with the applied digital elevation model.
Acknowledgments
This work is funded by the Swiss National Science Foundation (Project-Nr. 21-
50842.97) and by the Federal Office of Science and Education (EC-Project IN-
FLAME Nr. ENV4-CT98-0700, BBW Nr. 97.0182-2)
The authors owe special thanks to Dr. Lorenza Re from the Istituto di
Scienze della Terra (Ticino, Switzerland) for her help and information on debris
flow events in the study area.
Bibliography
Experiment Station.
Luso (P).

pp. 2177–2190, Luso (P).
Bachmann, A. and Allgöwer, B., 2000, Error propagation in wildfire be-

haviour modelling. Accuracy 2000: Proceedings of the 4th International Sym-
posium on Spatial Accuracy Assessment in Natural Resources and Environ-
mental Sciences, edited by G.B.M. Heuvelink and M.J.P.M. Lemmens, pp.
23–30, Delft University Press, Delft (NL), Amsterdam, NL.
Bachmann, A. and Allgöwer, B., 2001, A consistent wildland fire risk ter-
minology is needed! Fire Management Today, 61(4):28–33.
URL http://www.fs.fed.us/fire/planning/fmt/fmt-pdfs/fmt61-4.pdf
Borter, Patricio, 1998, Risikoanalyse bei gravitativen Naturgefahren: Fall-

beispiele und Daten. No. 107/II in Umwelt-Materialien, Naturgefahren, Bun-
desamt für Umwelt, Wald und Landschaft (BUWAL).
86 Bibliography

Heinimann, Hans R., Hollenstein, Kurt, Kienholz, Hans, Krumme-
nacher, Bernhard and Mani, Peter, 1998, Methoden zur Analyse und
Bewertung von Naturgefahren. No. 85 in Umwelt-Materialien, Naturgefahren,
Bundesamt für Umwelt, Wald und Landschaft (BUWAL).
Hofmann, C., Conedera, M., Delarze, R., Carraro, G. and Giorgetti,
P., 1998, Effets des incendies de forêt sur la végétation au sud des alpes
suisses. Mitt. Eidgenöss. Forsch.anst. Wald Schnee Landsch., 73(1):1–90.
Jenson, S.K. and Domingue, J.O., 1988, Extracting topographic structure
from digital elevation data for geographic information system analysis. Pho-
togrammetric Engineering and Remote Sensing, 54(11):1593–1600.
UK.
Langhart, R., Bachmann, A. and Allgöwer, B., 1998, Temporal and
spatial patterns of wildfire occurrence (canton of grison, switzerland). 3rd
International Conference on Forest Fire Research / 14th Conference on Fire
and Forest Meteorology, edited by D.X. Viegas, vol. 2, pp. 2279–2292, Luso
(P).
Mark, D.M., 1984, Automated detection of drainage networks form digital
elevation models. Cartographica, 21:168–178.
Marxer, P. and Conedera, M., 1999, Prometheus s.v. fire experiment in
switzerland (s.antonino, 28 march 1998). International Forest Fire News,
20:93–96.

Bibliography 87
Strahler, A.N., 1957, Quantitative analysis of watershed geomorphology.

Transactions of the American Geophysical Union, 8:913–920.
Zeller, Jürg, Geiger, Heinz, Röthlisberger, Gerhard and Bosshard,
Walter, 1980, Starkniederschläge des schweizerischen alpen- und alpen-
randgebietes: Wallis/tessin. Tech. Rep. 5, WSL, Swiss Federal Institute for
Forest, Snow and Landscape, Birmensdorf, Switzerland.
Zimmermann, Markus, Mani, Peter and Gamma, Patrick, 1997, Mur-
ganggefahr und Klimaänderung - ein GIS-basierter Ansatz. NFP31 Schluss-
bericht, vdf Hochschulverlag AG an der ETH, Zürich.
8
Synthesis
8.1 Discussion
Goals The principal goal of this thesis was to develop a framework that allows
the spatial quantitative analysis of wildland fire risk. In a first step much em-
phasis was put on the specification of the risk analysis framework. Besides this
more conceptual work the importance of dealing with uncertainties has been
another important objective. Finally, we wanted to show how the proposed
wildland fire risk analysis can be applied in a study case using GIS.
Framework The idea of risk analysis emerged while we were searching to

utilise wildland fire behaviour modelling for decision support and we found
that there was quite a large number of publications already available on this
topic. Unfortunately, in the wildland fire research and management community
there is a lack of a generally accepted, consistent risk terminology and analysis
methods. Besides, newer fire management policies view wildland fires not any
longer as only a “threat” but are based on more holistic ecosystem approaches.
These require a deepened analysis of wildland fires that consider not only the
local fire behaviour, but also the interactions between fire ignition locations and
possible objects at risk in a comprehensive way. In this thesis we proposed a
framework for wildland fire risk analysis that overcomes these shortcomings.
The most important step was the adaption of sound risk engineering concepts
to the wildland fire context.
The concept and terminology of risk as they are already applied in risk engi-
neering offer great advantages for the analysis of wildland fires (see section 2.3
and chapter 4). Two aspects are mainly interesting:
1. Risk consists of a probability and an outcome.
2. There are two risk perspectives: the one of the scenario (risk donor) and
the other of the object at risk (risk acceptor).
By explicitly considering probabilities and outcomes separately the risk sit-
uation can be studied more precisely. For instance, the risk can be high because
either the probability or the outcome is high. In most cases the management
strategies to reduce the risk will be completely different. The same is true for
the two risk perspectives: The distinction of individual and collective risk allows
90 Synthesis
the consideration of a risk situation a single object is exposed to or a specific

area is posing to several objects. This enables a customised evaluation, planning
and execution of management activities. For example it could be possible to
decide if it is more efficient to apply an extensive fuel reduction strategy or to
enhance the “fire fitness” of some few objects at risk.
The wildland fire risk definitions given in this thesis offer the possibility to
integrate the three major topics of wildland fire research: Fire occurrence de-
livering information about the probability of fires is linked via fire behaviour
modelling with the fire effects on different objects of interest. We hope that
the given examples of structuring wildland fire research topics with respect to
quantitative risk analysis help introducing the proposed wildland fire risk ter-
minology. Additionally, this integrative synopsis is expected to stimulate inter-
disciplinary research. In the operational use it can be useful for fire managers
to see, how all the different aspects of wildland fires can be put together to
construct a more comprehensive understanding of this phenomenon.
Besides the concepts of risk we adapted quantitative risk analysis methods
to the natural hazard wildland fire (see chapter 3 and 5). Specifically, we had
to account the fact that a fire can start at any point in a given area of interest
which is in contrast to most technical hazards that have a fixed location (e.g.
power plants). The proposed framework is not stiff but rather flexible. There
are a few specifications that are required by the risk matrix, but essentially it
lies in the users responsibility to provide appropriate methods and data to fill
it. This flexibility allows the framework to study, for instance, the risk situation
of various types of objects, such as buildings or catchment areas of potential
post fire debris flows (see chapter 7). From a user’s point of view, the most
important task when applying the framework is to define an appropriate impact
probability and damage function
Uncertainties An important aspect in risk analysis is the consideration of in-

put data uncertainties and their effects on the results of models (see chapter 6).
We showed that the application of a first order Taylor series to Rothermel’s
model can be a computationally efficient way to estimate the influence of uncer-
tainties of the input data to the result. The computational efficiency is a crucial
point considering the application of this method within the risk analysis frame-
work. Alternative methods are based on Monte-Carlo-simulations which are
computationally very demanding. Especially, the generation of the input data
set that should represent a given spatial auto- and inter-correlation structure
can be cumbersome. Besides the good computational performance the Taylor
series approach has a drawback in the form of an approximation error which
can become a major problem. This is the case when the input errors are large
compared to the “smoothness” of the function.
The result of the practical study of uncertainty propagation in wildland fire
behaviour modelling showed that caution should be applied. In the case of rate
of spread, coefficients of variation greater than 1 have been observed. These large
uncertainties stem mainly from large input data uncertainties. Furthermore, we
showed that only few input variables account for the largest part of the results
errors. However, as the input variables uncertainties are not a “technical” mea-
sure problem, but rather represent their inherent, natural variability it seems
questionable to enhance traditional data acquisition efforts like field sampling.
8.2 Outlook 91
Instead, high resolution laser-scanning might be an interesting alternative.
Case Study In the case study we applied the framework to study the risk
situation for catchment areas for potential post fire debris flows. The impact
probability function is derived by analysing the spread times of historical fire
events. The damage function takes into account the potential damage of the
debris flows. The extraction of the catchment areas and of debris flow paths
is based on a digital terrain model only. The modelling steps have all been
implemented and carried out within the GIS ArcInfo. The application of the
GIS-technology greatly facilitated the task by providing the basic functionalities
like handling and visualising the data sets.
The case study shows the need for a differentiated spatial analysis of wildland
fire risk. With the detailed results it is now possible to study the risk situation
at a very small scale and for instance, to prioritise specific areas very effectively
for different management activities.
8.2 Outlook
Framework Although the framework is verified in the sense of its formal
structure being sound, its application still lacks a thorough verification. To
do so, a cooperation would be necessary with a professional wildland fire man-
agement organisation that disposes over enough detailed and accurate spatial
information.
In this context it would be interesting to see, how the risk analysis is applied:
What spatial and temporal scale is the optimal one? The managing or-
ganisation has to balance the marginal utility of more detailed information
on the risk situation with the increased data acquisition and computing
expenses.
Impact probability functions: The calculation of the spread time is com-

putationally expensive and a purely geometric selection algorithm could
speed up the calculation very much. For instance, in areas with a homoge-
nous fuel and terrain situation one could apply directly a function similar
to that of Anderson (1983) taking into account only horizontal distance,
wind speed and wind direction.
The framework can not only analyse a given risk situation but the user
can also apply it to model the effectiveness of different management strate-
gies. To do so, it would be necessary to study the influences of several
activities on the input variables. For instance, we would have to develop a
model that reflects the changed fuel situation due to periodical prescribed
burnings.
The flexibility of the framework not only allows the consideration of differ-
ent types of objects at risk, but it is also possible to extend its functionality
to integrate different hazards. For each hazard type a new dimension will
be added and the original risk matrix becomes then a multidimensional
risk space.
92 Synthesis
The framework offers a flexible method to determine spatial risk but it does
not offer ready-to-use solutions. There are a few specifications that are required
by the risk matrix, but essentially it lies in the users responsibility to provide
appropriate methods and data to fill it. This makes the framework very flexible
to be adapted to other hazards and even be applied with multiple hazards. For a
successful adaption three requirements have to be accomplished: A method for
the determination of the spatial occurrence patterns, a method for the impact
probability and an impact function have to be provided.
Uncertainties The study of uncertainties offers still a lot of research topics.

In the context of wildland fire risk analysis, we see three main topics of interest:
Description and modelling of uncertainties in the input data
Analysing uncertainty propagation
Develop models that consider uncertainties

For a reasonable fire behaviour modelling it is necessary to know the spatial
and temporal variability of a natural phenomenon such as “fuel”. However, it
is not yet feasible to increase the density of sampling points to an arbitrarily
fine spatial and temporal scale. A solution is in our opinion, the application of
stochastic simulation (Ripley 1987) to generate “realistic” input data sets. To
do so, it is necessary to first measure the variability by determining spatial and
temporal correlation structures. This information can then be applied to choose
the appropriate stochastic methods to model the spatio-temporal distribution
of the input data.
The results of the uncertainty propagation in Rothermel’s model need to be
further analysed by applying a Monte-Carlo simulation. The application of the
first Taylor series involves approximation errors that are unknown and have to
be validated with a Monte-Carlo simulation. In any case, we would like to stress
the need for more caution with the application of Rothermel’s model.
In order to study the overall uncertainties of the risk analysis the study of
uncertainties in Rothermel’s model should be extended to every function/model
applied within the framework. A special task will be the analysis of uncertainty
propagation in fire spread simulation which is based on a shortest path algorithm
Dijkstra (1959). Of the various techniques of uncertainty propagation analysis
only a Monte Carlo based method can be envisioned to be applicable.
Finally, we think it is also worth looking closer at models that try to model
wildland fire as a stochastic process. Although, they do not help directly to
understand better the basic physical process involved, they might represent the
“chaotic” fire behaviour at a higher scale in a more “realistic way”.
The visualisation of uncertainties was not treated directly within this the-
sis, but collaboration took place with a project that studied the possibilities of
JAVA 3D and VisAD to present uncertainties. The work of Horisberger (2001)
produced some nice results in the form of an intuitive application∗ . The initial
visual impressions are promising, but the very large response lags of the appli-
cation are unsatisfactory. In order to become more sensitive to uncertainties, it
is important to have easy-to-use-tools that allow the interactive visualisation of
them.
∗ The application is available at http://www.geo.unizh.ch/~phorisb/sourcecode.html
Bibliography
Alabert, Francois, 1987, The practice of fast conditional simulations
through the lu decomposition of the covariance matrix. Mathematical Ge-
ology, 19(5):369–386.
Experiment Station.
Albini, F.A., 1986, Wildland fire spread by radiation - a model including fuel
cooling by natural convection. Combustion Science and Technology, 45:101–
113.
Alefeld, G. and Claudio, D., 1998, The basic properties of interval arth-
metic, its software realization and some applications. Computers and Struc-
tures, 67:3–8.
Alexandrian, D., 1996a, Fuel management. Wildfire Management: Systems,

Models & Techniques (Workshop), Algosystems S.A., Athens, Athens.
Alexandrian, D., 1996b, A new method of fire danger mapping in the for-
(Workshop), Athens.
Luso (P).
Bachmann, A., 1996, GIS-Analyse für ein Waldbrandmanagementsystem. Ein-

satz von Geographischen Informationssystemen und Fernerkundung in der
Umweltanalyse, edited by K. Brassel, K.I. Itten and E.Schmitt, vol. 31 of
Geoprocessing Series, pp. 49–52, Departement of Geography, University of
Zurich, Zurich, Switzerland.
94 Bibliography

pp. 2177–2190, Luso (P).
Bachmann, A. and Allgöwer, B., 2000, Error propagation in wildfire be-

haviour modelling. Accuracy 2000: Proceedings of the 4th International Sym-
posium on Spatial Accuracy Assessment in Natural Resources and Environ-
mental Sciences, edited by G.B.M. Heuvelink and M.J.P.M. Lemmens, pp.
23–30, Delft University Press, Delft (NL), Amsterdam, NL.
Bachmann, A. and Allgöwer, B., 2001, A consistent wildland fire risk ter-
minology is needed! Fire Management Today, 61(4):28–33.
URL http://www.fs.fed.us/fire/planning/fmt/fmt-pdfs/fmt61-4.pdf
Bandemer, Hans and Näther, Wolfgang, 1992, Fuzzy data analysis.

Kluwer Academic Publishers, Dordrecht, NL.
Barrow, C.J., 1999, Environmental management: principles and practice.

Routledge environmental management series, Routledge, London.

land.
Beck, J.A., 1991, GIS and modelling technologies for fire research and man-
agement in western australia. 11th Conference on fire and forest meteorology,
edited by Andrews P. L. Potts D. F., pp. 9–16, Society of American Foresters,
Missoula, MT.
Bevins, Collin D., 1996, fireLib user manual and technical reference. Inter-
mountain Research Station, Forest Service, U.S. Department of Agriculture.
URL http://www.montana.com/sem/public_html/firelib/firelib.html
Bolognesi, R., 1994, Nx-log: Idées directrices. Neige et avalanches, 63:22–24.
Borter, Patricio, 1998, Risikoanalyse bei gravitativen Naturgefahren: Fall-

beispiele und Daten. No. 107/II in Umwelt-Materialien, Naturgefahren, Bun-
desamt für Umwelt, Wald und Landschaft (BUWAL).
Brown, James K., Oberheu, Rick D. and Johnston, Cameron M., 1982,
Handbook for inventorying surface fuels and biomass in the interior west. Gen-
eral Technical Report INT-129, USDA Forest Service, Intermountain Forest
and Range Experiment Station Ogden, UT.

Station.
Burrough, Peter. A. and McDonnell, Rachael A., 1998, Principles of

Geographical Information Systems. Oxford University Press Inc., Oxford, GB.
Bibliography 95
BUWAL, Bundesamt für Umwelt, Wald und Landschaft, 1999, Risiko-

analyse bei gravitativen Naturgefahren: Methode. No. 107/1 in Umwelt-
Materialien, Naturgefahren, BUWAL, Bern.
Byram, G.M., 1959, Combustion of forest fuels. Forest Fire Control and Use,
edited by K.P. Davis, pp. 61–89, McGraw Hill, New York.
Cában, A.G., 1998, The importance of nonmarket values in land and fire
management planning. Proceedings of the III International Conference on
Forest Fire Research, 14th Conference on Fire and Forest Meteorology, edited
by D.X. Viegas, vol. 2, pp. 2191–2208, Coimbra, Portugal.
Catchpole, E. A., Hatton, T. J. and Catchpole, W. R., 1989, Fire
pread through nonhomogeneous fuel modelled as a markov process. Ecological
Modelling, 48:101–112.
CCFM, Canadian Council of Forest Ministers, 2001, National forestry
database: Fire statistics 1996. Internet.
URL http://nfdp.ccfm.org/summary/broch_e/qf3e.htm
Sons, New York.
140.
Systems, 10(3):333–345.
Chuvieco, Emilio, Salas, Francisco Javier and Vega, Cristina, 1997,
Remote sensing and GIS for long-term fire risk mapping. A review of remote
sensing methods for the study of large wildland fires, edited by Emilio Chu-
vieco, pp. 91–108, Departemento de Geografı́a, Universidad de Alcalá, Alcalá
de Henares, E, megafires project ENV-CT96-0256.
CIFFC, (Canadian Interagenxy Forest Fire Centre), 1999, Glossary
of fire management terms. CIFFC, Winnipeg.
Clarke, K. C. and Olsen, G., 1996, Refining a cellular automaton model
of wildfire propagation and extinction. GIS and environmental modeling:
progress and research issues, edited by Parks B. O. Goodchild F. M., Steyaert
L. T., pp. 333–338, GIS World Books.

96 Bibliography

Davies, Chris, 1997, Analysis of fire causes on or threatening public land
in victoria 1976/77–1995/96. Research Report 49, Fire Management Branch,
Departement of Natural Resources and Environment, Victoria, AU.
Dijkstra, E.W., 1959, A note on two problems in connection withs graphs.
Numerische Mathematik, 1(1):269–271.
Eftichidis G., Xanthopoulos G., Varela V., 1994, A preparedness plan-
ning frame for forest fire prevention. 2nd Int. Conference on Forest Fire Re-
search, pp. 105–112, Coimbra, Portugal.
Englund, E., 1993, Spatial simulation: Environmental applications. Environ-
mental Modeling and GIS, edited by M.F. Goodchild, B.Q. Parks and L.T.
Steyaert, pp. 432–437, University Press: Oxford, New York.
Fischer, W. C., 1982, Photo guide for appraising downed woody fuels in mon-
tana forests: Grand fir-lach-douglas-fir, western hemlock, western hemlock-
western redcedar, and western redcedar cover types. General Technical Report
Station.
Fleagle, Robert G. and Businger, Joost A., 1980, An Introduction to
Atmospheric Physics, vol. 25. Academic Press, second edn.
Fons, W.L., 1946, Analysis of fire spread in light forest fuels. Journal of Agri-
cultural Research, 72:93–121.
Geertman, S. C. M. and Ritsema van Eck, J. R., 1995, GIS and models
of accessibility potential: an application in planning. International Journal of
Geographic Information Systems, 9(1):67–80.
Gheorghe, A. and Nicolet-Monnier, M., 1995, Integrated regional risk
assessment, vol. 2 of Environmental science and technology library; vol. 4.
Kluwer, Dordrecht, NL.
nology Zurich).
Gómez-Hernández, J.J. and Journel, A.G., 1993, Joint sequential simu-
lation of multigaussian fields. Geostatistics Troia ’92, edited by A. Soares,
vol. 1, pp. 85–94.
Gonçalves, P.P. and Diogo, P.M., 1994, Forest fire propagation modeling:
a new methodology using cellular automata and geographic information sys-
tems. 2nd Int. Conference on Forest Fire Research, pp. 209–219, Coimbra,
Portugal.
Bibliography 97
Goodchild, Michael F., Parks, Bradley O. and Steyaert, Louis T.

(Editors), 1993, Environmental Modeling with GIS. Oxford University Press,
New York, USA.
Grishin, A.M., 1984, Steady-state propagation of the front of a high-level

forest fire. Soviet Physics Doklady, 29:917–919.
Gronlund, A. G., Xiang, W.-N. and Sox, H., 1994, GIS, expert system
technologies improve forest fire management techniques. GIS World, Febru-
ary:32–36.

Heil, Angelika, 1998, Fire-related transboundary haze and air pollution in

south-east asia 1997. Internet.
URL http://www.icsea.or.id/sea-span/SCIPOL2/STUDY423.htm
Heinimann, Hans R., Hollenstein, Kurt, Kienholz, Hans, Krumme-

nacher, Bernhard and Mani, Peter, 1998, Methoden zur Analyse und
Bewertung von Naturgefahren. No. 85 in Umwelt-Materialien, Naturgefahren,
Bundesamt für Umwelt, Wald und Landschaft (BUWAL).

Hewitt, Mason J., 1993, Risk and hazard modelling. Environmental Modeling
with GIS, edited by Michael F. Goodchild, Bradley O. Parks and Louis T.
Steyaert, p. 317, Oxford University Press, New York, USA.
Hofmann, C., Conedera, M., Delarze, R., Carraro, G. and Giorgetti,

P., 1998, Effets des incendies de forêt sur la végétation au sud des alpes
suisses. Mitt. Eidgenöss. Forsch.anst. Wald Schnee Landsch., 73(1):1–90.
Hootsmans, Robert Maria, 1996, Fuzzy sets and series analysis for vi-
sual decision support in spatial data exploration. Ph.D. thesis, University of
Utrecht, Utrecht.
Horisberger, Patrick, 2001, Visualisierung von Unsicherheiten in Wald-

brandmodellen auf Basis von Java, Java3D und VisAD. MSc.Thesis, Dep-
tartment of Geography, University of Zurich, Zurich, Switzerland.
Horn, B.K.P., 1981, Hill shading and the reflectance map. Proceedings of the
Institute of Electrical and Electronics Engineers, 69, pp. 14–47.
Hugentobler, Marco, 2000, Fortpflanzung von Unsicherheiten in dreiecks-

basierten digitalen Geländemodellen mit Intervallarithmetik. MSc.Thesis,
Dept. of Geography, University of Zurich.
98 Bibliography
Hunter, Gary J. and Goodchild, Michael F., 1997, Modeling the uncer-
tainty of slope and aspect estimates derived from spatial databses. Geograph-
ical Analysis, 29(1):35–49.
Jenson, S.K. and Domingue, J.O., 1988, Extracting topographic structure

from digital elevation data for geographic information system analysis. Pho-
togrammetric Engineering and Remote Sensing, 54(11):1593–1600.
Inc., New York.
Jones, Christopher B., 1997, Geographical Information Systems and Com-

puter Cartography. Longman, Essex, GB.

UK.

Kumamoto Hiromitsu, Henley Ernest J., 1996, Probabilistic Risk Assess-

ment and Management for Engineers and Scientist. IEEE, New York, 2nd
edn.
Landau, S.I. (Editor), 1999, The new international Webster’s student dictio-
nary of the English language. Trident Press International, New York, NY.
Langhart, R., Bachmann, A. and Allgöwer, B., 1998, Temporal and

spatial patterns of wildfire occurrence (canton of grison, switzerland). 3rd
International Conference on Forest Fire Research / 14th Conference on Fire
and Forest Meteorology, edited by D.X. Viegas, vol. 2, pp. 2279–2292, Luso
(P).
Malczewski, Jacek, 1999, GIS and multicriteria decision analysis. John Wi-
ley & Sons, New York.
Marcozzi., M., Bovio., G, Mandallaz., D and Bachmann., P, 1994, In-

fluenza della meteorologia sull’indice di pericolo degli incendi boschivi nel
cantone ticino. Schweizerische Zeitschrift für Forstwesen, 194(3):183–199.
Mark, D.M., 1984, Automated detection of drainage networks form digital

elevation models. Cartographica, 21:168–178.
Martell, D.L., Otokul, S. and Stocks, B.J., 1987, A logistical model

for predicting daily people caused forest fire occurence in ontario. Canadian
Journal of Forestry Research, 17:394–401.
Marxer, P. and Conedera, M., 1999, Prometheus s.v. fire experiment in

switzerland (s.antonino, 28 march 1998). International Forest Fire News,
20:93–96.
Bibliography 99
Marxer, P., Conedera, M. and Schaub, D., 1998, Postfire runoff and soil
erosion in sweet chestnut forests in south switzerland. Proceedings of the III
International Conference on Forest Fire Research, 14th Conference on Fire
and Forest Meteorology, edited by D.X. Viegas, vol. 2, pp. 1317–1331, Coim-
bra, Portugal.
McArthur, A.G., 1966, Weather and grassland fire behaviour. Tech. rep.,
Commonwealth Australia, Department of National Development, Forest and
Timber Bureau, leaflet No 100.

Miller, Carol, Landres, Peter B. and Alback, Paul B., 1999, Evalu-
ating risks and benefits of wildland fire at landscape scales. The Joint Fire
Science Conference and Workshop, edited by Greg E. Gollberg, vol. 1, pp. 78–
87, University of Idaho and the International Association of Wildland Fire,
Grove Hotel, Boise, Idaho.
MOF, Ministery of Forests, 1997, Glossary of forestry terms. province of

british columbia. Internet.
URL http://www.for.gov.bc.ca/pba/publctns/glossary/glossary.htm
Moore, Ramon E., 1979, Methods and applications of interval analysis. SIAM
(Society for Industrial and Applied Mathematics), Philadelphia, USA.
NIFC, National Interagency Fire Center, 2001, Wildland fire statistics

1996. Internet.
URL http://www.nifc.gov/stats/wildlandfirestats.html
NWCG, 1997, Fitness and Work Capacity. National Wildfire Coordinating

Group, second edn.
NWCG, National Wildfire Coordinating Group, 1992a, Fire behavior

NWCG, (National Wildfire Coordinating Group), 1992b, Glossay of

wildland fire terminology. PMS 205/NFES 1832, National Interagency Fire
Center, Boise, ID.
NZFS, New Zealand Fire Service, 2001, Fire statistics 1996. Internet.
URL http://www.fire.org.nz/more_info/statistics/PDFs/
Statistics.pdf
Parrish, David A., Townsend, Laura, Saunders, Jerry, Carney, Ger-

ald and Langston, Carol, 1993, U.s. epa region 6 coparative risk project:
Evaluating ecological risk. Environmental Modeling with GIS, edited by
Michael F. Goodchild, Bradley O. Parks and Louis T. Steyaert, pp. 348–352,
Oxford University Press, New York, USA.
Perry, G.L.W., 1998, Current approaches to modelling the spread of wildland

fire: a review. Progress in Physical Geography, 22(2):222–245.
100 Bibliography
Poulin, S., Merrill, D.F. and Alexander, M.E., 1987, Glossary of Forest
Fire Management Terms. National Research Council of Canada, Ottawa, CA,
fourth edn.
Rüegsegger, M., 1996, Entscheidungsunterstützung im Waldbrandmanage-
ment mit GIS; Strategien und ausgewählte Beispiele. Diplomarbeit, Univer-
sität Zürich Irchel.
Rejeski, David, 1993, GIS and risk: A three culture problem. Environmental
Modeling with GIS, edited by Michael F. Goodchild, Bradley O. Parks and
Louis T. Steyaert, pp. 318–331, Oxford University Press, New York, USA.
Rich, D.C., 1980, Potential models in human geography, vol. 26 of Concepts
and Techniques in Modern Geography. University of East Anglia, Norwich.
Richardson, A., 1993, GIS and imagery join forces to combat wildland fire.
13th Annual ESRI User Conference, pp. 429–434, Palm Springs, USA.
Ripley, Brian D., 1987, Stochastic Simulation. Wiley Series in Probability
and Mathematical Statistics, John Wiley & Sons.
Roussopoulos, P.J. and Johnson, V.J., 1975, Help in making fuel manage-
ment decisions. Research paper, USDA Forest Service, North Central Exper-
iment Station.
Rowe, William D., 1994, Understanding uncertainty. Risk Analysis,
14(5):743–750.
Ryan, K.C., 1998, Analsysis of the relative value of morphological variables
in predicting fire-caused tree mortality. Proceedings of the III International
Conference on Forest Fire Research, 14th Conference on Fire and Forest Me-
teorology, edited by D.X. Viegas, vol. 2, pp. 1511–1526, Coimbra, Portugal.
Salazar L. A., Nilsson C. V., 1989, Integrating geographic information
systems into fire management. 10th Conference on fire and forest meteorology,
edited by MacIver D. C. Auld H. Whitewood R., pp. 311–316, Society of
American Foresters, Ottawa, Canada.
Saltelli, A., Chan, K. and Scott, M. (Editors), 2000, Sensitivity Analysis.
Wiley series in probability and statistics, John Wiley & Sons, England.
Salvador, R., Piñol, J., Tarantola, S. and Pla, E., 2001 in press, Global
sensitivity analysis and scale effects fo a fire propagation model used over
mediterranean shrublands. Ecological Modelling.
Schöning, R., 1996, Modellierung des potentiellen Waldbrandverhaltens mit
einem Geographischen Informationssystem. MSc. Thesis, Department of Ge-
ography, University of Zurich, Zurich, Switzerland.
Bibliography 101

Seiffert, Helmut, 1997, Einführung in die Wissenschaftstheorie, vol. 4. Beck,

München.
SFOT, Swiss Federal Office of Topography, 2000, Product Description

for DHM25.
URL http://www.swisstopo.ch/en/digital/dhm25.htm
Simard, A. J., 1976, Wildland fire management: the economics of policy al-
ternatives. Forestry Technical Report 15, Department of Fisheries and Envi-
ronment, Canadian Forestry Service.
Sobol’, I.M., 1993, Sensitivity estimates for nonlinear mathematical models.

Mathematical modeling and computational experiment, 1(4):407–414.
Strahler, A.N., 1957, Quantitative analysis of watershed geomorphology.

Transactions of the American Geophysical Union, 8:913–920.
Turány, Tamas and Rabitz, Herschel, 2000, Local methods. Sensitivity

Analysis, edited by A. Saltelli, K. Chan and M. Scott, Wiley series in proba-
bility and statistics, John Wiley & Sons, England.
UN-ECE, UN/ECE Timber Committee, 1997, Forest fire statistics 1995-

1997.
URL http://www.unece.org/trade/timber/ff-stats/ff-95-97.htm
USDA, U.S. Department of Agriculture, Forest Service, Rocky

Mountain Research Station, Fire Sciences Laboratory, 2001, Fire
effects information system. Online, accessed: 20.4.2001.
URL http://www.fs.fed.us/database/feis/
van Wagner, C.E., 1977, Conditions for the start and spread of crown fires.
Canadian Journal of Forest Research, 7:23–34.
van Wagner, C.E., 1993, Prediction of crown fire behavior in two stands of
jack pine. Canadian Journal of Forest Research, 23:442–449.
Van Wagtendonk, J.W., 1990, GIS applications in fire management and

research. Fire and the environment: ecological and cultural perspectives, edited
by S.C. Nodvin and T.A. Waldrop, vol. General Technical Report SE-69, pp.
212–214, USDA Forest Service, Knoxville, Tennessee.
Vĉkovski, Andrej, 1998, Interoperable and distributed processing in GIS. Tay-

lor & Francis, London.
Viegas, D.X., Ribeiro, P.R. and Cruz, M.G., 1998, Characterisation of the
combustibility of forest fuels. Proceedings of the III International Conference
on Forest Fire Research, 14th Conference on Fire and Forest Meteorology,
edited by D.X. Viegas, vol. 1, pp. 467–482, Coimbra, Portugal.
102 Bibliography
von Braun, Margrit, 1993, The use of GIS in assessing exposure and reme-
dial alternatives at superfund sites. Environmental Modeling with GIS, edited
Wadge, G., Wislocki, A.P. and Pearson, E.J., 1993, Spatial analysis in
GIS for natural hazard assessment. Environmental Modeling with GIS, edited
Weber, R.O., 1989, Analytical models for fire spread due to radiation. Com-
bustion and Flame, 78:398–408.
Weber, R.O., 1991, Modelling fire spread through fuel beds. Progress in En-
ergy and Combustion Science, 17:67–82.
Whittle, P., 1954, On stationary processes in the plane. Biometrika, 41:434–
449.
Wise, S.M., 1998, The effect of GIS interpolation errors on the use of digi-
tal elevation models in geomorphology. Landform monitoring, modelling, and
analysis, edited by Stuart N. Lane, Keith S. Richards and Jim H. Chandler,
pp. 139–164, John Wiley & Sons.
Wolfram, S., 1984, Cellular automata: a model of complexity. Nature,
311(4):419–424.
Wybo, J.-L. and Meunier, E., 1993, Architecture of a decision support system
for forest fire prevention and fighting. 1993 Int. emergency management and
engineering conference: Research and applications, edited by J.D. Sullivan,
pp. 186–190, IEEE Computer Society, Arlington, Virginia.
Yeates, M.H. and Garner, B.J., 1976, The North American City. New York.
Zadeh, L.A., 1965, Fuzzy sets. Information and Control, 8:338–353.
Zeller, Jürg, Geiger, Heinz, Röthlisberger, Gerhard and Bosshard,
Walter, 1980, Starkniederschläge des schweizerischen alpen- und alpen-
randgebietes: Wallis/tessin. Tech. Rep. 5, WSL, Swiss Federal Institute for
Forest, Snow and Landscape, Birmensdorf, Switzerland.
Zimmermann, Markus, Mani, Peter and Gamma, Patrick, 1997, Mur-
ganggefahr und Klimaänderung - ein GIS-basierter Ansatz. NFP31 Schluss-
bericht, vdf Hochschulverlag AG an der ETH, Zürich.
Zogg, H. A., 1981, ”Zürich”-Gefahrenanalyse: Grundprinzipien. Tech. rep.,
Zürich Versincherungs-Gruppe.
Appendix A
Models used
104 Models used
A.1 Rothermel’s Model

Generally, the weighting parameters that are formulated in the original paper
of Rothermel (1972, p.29/30) have been simplified in this paper. This could be
done by assuming a constant particle density ρp for any size class and category.
NOTE: The original equation numbers are referenced in square brackets, i.e.
[R(27)] denotes equation (27) in the paper of Rothermel (1972). Similarly,
[Albini, p.89] points to page 89 in the publication of Albini (1976).
List of Symbols
w0d1 Fuel loading [kg/m2], dead fuel 0-0.6cm
w0d2 Fuel loading [kg/m2], dead fuel 0.6-2.5cm
w0d3 Fuel loading [kg/m2], dead fuel 2.5-7.5cm
w0lh Fuel loading [kg/m2], live herbaceous fuel
w0lw Fuel loading [kg/m2], live woody fuel
svd1 Surface to volume ratio [1/m], dead fuel 0-0.6cm
svd2 Surface to volume ratio [1/m], dead fuel 0.6-2.5cm
svd3 Surface to volume ratio [1/m], dead fuel 2.5-7.5cm
svlh Surface to volume ratio [1/m], live herbaceous fuel
svlw Surface to volume ratio [1/m], live woody fuel
md1 Fuel moisture content [%], dead fuel 0-0.6cm
md2 Fuel moisture content [%], dead fuel 0.6-2.5cm
md3 Fuel moisture content [%], dead fuel 2.5-7.5cm
mlh Fuel moisture content [%], live herbaceous fuel
mlw Fuel moisture content [%], live woody fuel
d Fuel bed depth [m]
ρp Particle density [kg/m3]
heat Particle low heat content [kJ/kg]
st Total mineral content [%]
se Effective mineral content [%]
mx Moisture of extinction, dead fuel [%]
wsp Wind Speed [m/s]
wdr Wind Direction [°]
slp Slope [rad]
asp Aspect [°]
ϑ Split angle between upslope direction and direction where the wind is
blowing to [rad]
ρb Bulk density [kg/m3]
β Packing ratio
βopt Optimal packing ratio
betar atio Ratio mean/optimal packing ratio
wn Net fuel loading
ηs Mineral damping coefficient
ηM Moisture damping coefficient
ξ Propagating flux ratio
A Auxiliary function
Γ Potential reaction velocity [1/s]
Γmax Maximum reaction velocity [1/s]
Ir Reaction intensity [kW/m2]
φs Slope factor
B, C, E Auxiliary functions
φw Wind factor
vx, vy Vector components
vl Amount of the sum of the wind and slope factor, i.e. |φ~s + φ~w |
A.1 Rothermel’s Model 105
Auxiliary Functions
swd1 = svd1 w0d1

swd2 = svd2 w0d2
swd3 = svd3 w0d3
swlh = svlh w0lh
swlw = svlw w0lw
swd = swd1 + swd2 + swd3
swl = swlh + swlw
swt = swd + swl
2 2 2 2 2
s2wt = svd1 · w0d1 + svd2 · w0d2 + svd3 · w0d3 + svlh · w0lh + svlw · w0lw(A.1)
sw2d = svd1 · w02d1 + svd2 · w02d2 + svd3 · w02d3 (A.2)

sw2l = svlh · w02lh + svlw · w02lw (A.3)
sw2t = sw2d + sw2l (A.4)
swmd = swd1 · md1 + swd2 · md2 + swd3 · md3 (A.5)

swml = swlh · mlh + swlw · mlw (A.6)
Characteristic Surface-to-volume ratio [R(71,72)]
s2wz
σ =
swt
2 2 2 2 2
svd1 · w0d1 + svd2 · w0d2 + svd3 · w0d3 + svlh · w0lh + svlw · w0lw
=
svd1 · w0d1 + svd2 · w0d2 + svd3 · w0d3 + svlh · w0lh + svlw · w0lw
Mean Bulk Density [R(74)]
w0d1 + w0d2 + w0d3 + w0lh + w0lw

ρb = (A.7)
d
Packing Ratios
Mean Packing Ratio [R(31,73)]
ρb
β = (A.8)
ρp
Optimal Packing Ratio [R(37)]
βopt = 8.8578 · σ −0.8189 (A.9)

106 Models used
Net Fuel Loading [R(60), adjusted by A.(p.88)]

st

wnd1 = w0d1 · 1 − (A.10)
100
st

wnd2 = w0d2 · 1 − (A.11)
100
st

wnd3 = w0d3 · 1 − (A.12)
100
st

wnlh = w0lh · 1 − (A.13)
100
st

wnlw = w0lw · 1 − (A.14)
100
[R(59)]
st sw2d

wnd = 1− · (A.15)
100 swd
st sw2l

wnl = 1− · (A.16)
100 swl
Mineral Damping Coefficient [R(62)]

−0.19
se

ηs = 0.174 · (A.17)
100
Moisture Damping Coefficient
Ratio of “Fine” Fuel Loadings, Dead/Living [Albini, p.89]
hnd1 = 0.20482 · w0d1 · exp(−452.76/svd1 ) (A.18)

hnd2 = 0.20482 · w0d2 · exp(−452.76/svd2 ) (A.19)
hnd3 = 0.20482 · w0d3 · exp(−452.76/svd3 ) (A.20)
hnlh = 0.20482 · w0lh · exp(−1640.42/svlh ) (A.21)
hnlw = 0.20482 · w0lw · exp(−1640.42/svlw ) (A.22)
hnd = hnd1 + hnd2 + hnd3 (A.23)
hnl = hnlh + hnlw (A.24)
W = hnd /hnl (A.25)
Moisture Content of “Fine” Dead Fuel [Albini, p.89]
hnmd = hnd1 · md1 + hnd2 · md2 + hnd3 · md3 (A.26)

M fdead = hnmd /hnd (A.27)
Moisture of Extinction of Living Fuel [R(88), Albini, p.89]
M fdead

M xlive = 2.9 · W · 1 − − 0.226 · 100 (A.28)
mx
Moisture Ratios [R(65,66)]

swml
rml = (A.29)
swl · M xlive
swmd
rmd = (A.30)
swd · mx
Moisture Damping Coefficients [R(64)]
ηM d = 1 − 2.59 · rmd + 5.11 · rm2d − 3.52 · rm3d (A.31)

ηM l = 1 − 2.59 · rml + 5.11 · rm2l − 3.52 · rm3l (A.32)
ηM = wnd · ηM d + wnl · ηM l (A.33)
Reaction Velocity
Maximum Reaction Velocity [R(36,68)]
0.16828 · σ 1.5
Γmax = (A.34)
290 700 + 0.5997 · σ 1.5
A [R(70), Albini p.88]
A = 340.53 · σ −0.7913 (A.35)
Potential Reaction Velocity [R(38)]

A
β β
Γ = Γmax · · exp A · 1− (A.36)
βopt βopt
Reaction Intensity [R(27,58), Albini, p.89]
Ir = Γ · heat · ηs · ηM (A.37)
Propagating Flux Ratio [R(42)]

√
exp (0.792 + 0.37597 · σ) · (β + 0.1)
ξ = (A.38)
192 + 0.0791 · σ
Heat Sink
Effective heating number [R(14,77)]
εd1 = exp(−452.76/svd1 ) (A.39)

εd2 = exp(−452.76/svd2 ) (A.40)
εd3 = exp(−452.76/svd3 ) (A.41)
εlh = exp(−452.76/svlh ) (A.42)
εlw = exp(−452.76/svlw ) (A.43)
Heat of Preignition [R(12,78)]
Qd1 = 581.5 + 25.957 · md1 (A.44)

Qd2 = 581.5 + 25.957 · md2 (A.45)
Qd3 = 581.5 + 25.957 · md3 (A.46)
Qlh = 581.5 + 25.957 · mlh (A.47)
Qlw = 581.5 + 25.957 · mlw (A.48)
108 Models used
Heat Sink [R(77)]
hskz = svd1 · w0d1 · εd1 · Qd1 + svd2 · w0d2 · εd2 · Qd2 + svd3 · w0d3 · εd3 · Qd3
+ svlh · w0lh · εlh · Qlh + svlw · w0lw · εlw · Qlw (A.49)
hskz
hsk = ρb · (A.50)
swt
Slope and Wind
Slope Factor [R(80)]
φs = 5.275 · β −0.3 · tan(slp)2 (A.51)
Wind Factor [R(79,82,83,84)]
B = 0.02526 · (σ · 0.3048)0.54 (A.52)

C = 7.47 · exp(−0.133 · (σ · 0.3048)0.55 ) (A.53)
E = 0.715 · exp(−0.000359 · 0.3048 · σ) (A.54)
−E
β
φw = C · (3.281 · 60 · wsp)B · (A.55)
βopt
Combined Slope and Wind Factor

vx = φs + φw · cos(ϑ) (A.56)
vy = φw · sin(ϑ) (A.57)
p
vl = vx2 + vy 2 (A.58)
Spread Direction
vy

sdr = arcsin (A.59)
vl
Effective Wind Speed [R(79)]

1/B
vl
C · (β/βopt )−E
ef w = (A.60)
196.85
Rate of Spread [R(52)]
Ir · ξ · (1 + vl)
ros = (A.61)
hsk
Derivatives of the Rothermel equations

Characteristic Surface to Volume Ratio
svd1 s2wt

0
σw0 d1
= · svd1 − (A.62)
swt swt
svd2 s2wt

0
σw0 d2
= · svd2 − (A.63)
swt swt
svd3 s2wt

0
σw0 d3
= · svd3 − (A.64)
swt swt
svlh s2wt

0
σw0 lh
= · svlh − (A.65)
swt swt
svlw s2wt

0
σw0 lw
= · svlw − (A.66)
swt swt
w0d1 s2wt

0
σsv d1
= · 2 · svd1 − (A.67)
swt swt
Mean Bulk Density
ρb 0w0d1 = 1/d (A.68)

ρb 0w0d2 = 1/d (A.69)
ρb 0w0d3 = 1/d (A.70)
ρb 0w0lh = 1/d (A.71)
ρb 0w0lw = 1/d (A.72)
ρb 0d = ρb /d (A.73)
Packing Ratios
Mean Packing Ratio
0
βw0 d1
= ρb 0w0d1 /ρp (A.74)
0
βw0 d2
= ρb 0w0d2 /ρp (A.75)
0
βw0 d3
= ρb 0w0d3 /ρp (A.76)
0
βw0 lh
= ρb 0w0lh /ρp (A.77)
0
βw0 lw
= ρb 0w0lw /ρp (A.78)
βd0 = ρb 0d /ρp (A.79)
Optimal Packing Ratio
βopt 0w0d1 = −7.25324297 · σ −1.8189 · σw0

0
d1
(A.80)
βopt 0w0d2 = −7.25324297 · σ −1.8189
· 0
σw0 d2
(A.81)
βopt 0w0d3 = −7.25324297 · σ −1.8189
· 0
σw0 d3
(A.82)
βopt 0w0lh = −7.25324297 · σ −1.8189
· 0
σw0 lh
(A.83)
βopt 0w0lw = −7.25324297 · σ −1.8189
· 0
σw0 lw
(A.84)
βopt 0svd1 = −7.25324297 · σ −1.8189
· 0
σsv d1
(A.85)
110 Models used
Ratio Mean to Optimal Packing

0 0
βw0d1 · βopt − β · βopt 0w0d1

β
= 2
(A.86)
βopt w0d1
βopt
0 0
· βopt − β · βopt 0w0d2

β βw0 d2
= 2
(A.87)
βopt w0d2
βopt
0 0
· βopt − β · βopt 0w0d3

β βw0 d3
= 2
(A.88)
βopt w0d3
βopt
0 0
· βopt − β · βopt 0w0lh

β βw0 lh
= 2
(A.89)
βopt w0lh
βopt
0 0
· βopt − β · βopt 0w0lw

β βw0 lw
= 2
(A.90)
βopt w0lw
βopt
0
−β · βopt 0svd1

β
= 2
(A.91)
βopt svd1
βopt
0
β βd0
= (A.92)
βopt d
βopt
Net Fuel Loading

st svd1 sw2d

wn0w0d1 = (1 − )· · 2 · w0d1 − (A.93)
100 swd swd
st svd2 sw2d

wn0w0d2 = (1 − )· · 2 · w0d2 − (A.94)
100 swd swd
st svd3 sw2d

wn0w0d3 = (1 − )· · 2 · w0d3 − (A.95)
100 swd swd
st svlh sw2l

wn0w0lh = (1 − )· · 2 · w0lh − (A.96)
100 swl swl
st svlw sw2l

wn0w0lw = (1 − )· · 2 · w0lw − (A.97)
100 swl swl
st w0d1 sw2d

wn0svd1 = (1 − )· · w0d1 − (A.98)
100 swd swd
Moisture Damping
Ratio of “Fine” Fuel Loadings, Dead/Living

0 0.20482 · exp(−452.76/svd1 )
Ww0d1
= (A.99)
hnl
0 0.20482 · exp(−452.76/svd2 )
Ww0d2
= (A.100)
hnl
0 0.20482 · exp(−452.76/svd3 )
Ww0d3
= (A.101)
hnl
0 −0.20482 · W · exp(−1640.42/svlh )
Ww0 lh
= (A.102)
hnl
0 −0.20482 · W · exp(−1640.42/svlw )
Ww0 lw
= (A.103)
hnl
0 452.76 · hnd1
Wsv = 2
(A.104)
d1
svd1 · hnl
Moisture Content of “Fine” Dead Fuel

exp(−452.76/svd1 ) hnmd

(M fdead )0w0d1 = · md1 − (A.105)
hnd hnd

(M fdead )0w0d2 = · md2 − (A.106)
hnd hnd

(M fdead )0w0d3 = · md3 − (A.107)
hnd hnd
(M fdead )0md1 = hnd1 /hnd (A.108)
452.76 · hnd1 hnmd

(M fdead )0svd1 = 2
· md1 − (A.111)
svd1 · hnd hnd
Moisture of Extinction of Living Fuel

M fdead (M fdead )0w0d1
(M xlive )0w0d1 0
= 2.9 · Ww0 d1
· 1− − 2.9 · W · (A.112)
mx mx
0

M fdead
(M fdead )w0d2
(M xlive )0w0d2 = 0
2.9 · Ww0 d2
· 1− − 2.9 · W · (A.113)
mx mx
0

M fdead
(M fdead )w0d3
(M xlive )0w0d3 = 0
2.9 · Ww0 d3
· 1− − 2.9 · W · (A.114)
mx mx

M fdead
(M fdead )0svd1
(M xlive )0svd1 = 0
2.9 · Wsv d1
· 1− − 2.9 · W · (A.115)
mx mx
hnd1
(M xlive )0md1 = −2.9 · (A.116)
hnl · M xlive
hnd2
(M xlive )0md2 = −2.9 · (A.117)
hnl · M xlive
hnd3
(M xlive )0md3 = −2.9 · (A.118)
hnl · M xlive
M fdead

(M xlive )0w0lh = 0
2.9 · Ww0 lh
1− (A.119)
mx
M fdead

(M xlive )0w0lw = 0
2.9 · Ww0 lw
1− (A.120)
mx
M fdead
(M xlive )0mx = 2.9 · W · (A.121)
mx2
Moisture Ratio Dead Fuel

svd1 swmd

rmd 0w0d1 = · md1 − (A.122)
swd · mx swd
svd2 swmd

rmd 0w0d2 = · md2 − (A.123)
swd · mx swd
svd3 swmd

rmd 0w0d3 = · md3 − (A.124)
swd · mx swd
w0d1 swmd

rmd 0svd1 = · md1 − (A.125)
swd · mx swd
swd1
rmd 0md1 = (A.126)
swd · mx
swd2
rmd 0md2 = (A.127)
swd · mx
swd3
rmd 0md3 = (A.128)
swd · mx
rmd 0mx = −rmd /mx (A.129)
112 Models used
Moisture Ratio Living Fuel

swml · svlh swml · (M xlive )0w0lh
svlh · mlh − −
swl M xlive
rml 0w0lh = (A.130)
swl · M xlive
swml · svlw swml · (M xlive )0w0lw
svlw · mlw − −
swl M xlive
rml 0w0lw = (A.131)
swl · M xlive
rml 0w0d1 = −rml · (M xlive )0w0d1 /M xlive (A.132)
rml 0md1 = −rml · (M xlive )0md1 /M xlive (A.135)
rml 0svd1 = −rml · (M xlive )0svd1 /M xlive (A.138)
rml 0mx = −rml · (M xlive )0mx /M xlive (A.139)
swlh
rml 0mlh = (A.140)
swl · M xlive
swlw
rml 0mlw = (A.141)
swl · M xlive
Moisture Damping Coefficient Dead Fuel

%1 = −2.59 + 10.22 · rmd − 10.561 · rm2d
ηM d 0w0d1 = rmd 0w0d1 · %1 (A.142)
ηM d 0w0d2 = rmd 0w0d2 · %1 (A.143)
ηM d 0w0d3 = rmd 0w0d3 · %1 (A.144)
ηM d 0md1 = rmd 0md1 · %1 (A.145)
ηM d 0md2 = rmd 0md2 · %1 (A.146)
ηM d 0md3 = rmd 0md3 · %1 (A.147)
ηM d 0svd1 = rmd 0svd1 · %1 (A.148)
ηM d 0mx = rmd 0mx · %1 (A.149)
Moisture Damping Coefficient Living Fuel

%1 = −2.59 + 10.22 · rml − 10.561 · rm2l
ηM l 0w0d1 = rml 0w0d1 · %1 (A.150)
ηM l 0w0d2 = rml 0w0d2 · %1 (A.151)
ηM l 0w0d3 = rml 0w0d3 · %1 (A.152)
ηM l 0w0lh = rml 0w0lh · %1 (A.153)
ηM l 0w0lw = rml 0w0lw · %1 (A.154)
ηM l 0md1 = rml 0md1 · %1 (A.155)
ηM l 0md2 = rml 0md2 · %1 (A.156)
ηM l 0md3 = rml 0md3 · %1 (A.157)
ηM l 0mlh = rml 0mlh · %1 (A.158)
ηM l 0mlw = rml 0mlw · %1 (A.159)
ηM l 0svd1 = rml 0svd1 · %1 (A.160)

ηM l 0mx = rml 0mx · %1 (A.161)
Total Moisture Damping Coefficient
ηM 0w0d1 = wn0w0d1 · ηM d + wnd · ηM d 0w0d1 + wnl · ηM l 0w0d1 (A.162)

ηM 0svd1 = wn0svd1 · ηM d + wnd · ηM d 0svd1 + wnl · ηM l 0svd1 (A.165)
ηM 0w0lh = wn0w0lh · ηM l + wnl · ηM l 0w0lh (A.166)
ηM 0w0lw = wn0w0lw · ηM l + wnl · ηM l 0w0lw (A.167)
ηM 0md1 = wnd · ηM d 0md1 + wnl · ηM l 0md1 (A.168)
ηM 0mx = wnd · ηM d 0mx + wnl · ηM l 0mx (A.171)
ηM 0mlh = wnl · ηM l 0mlh (A.172)
ηM 0mlw = wnl · ηM l 0mlw (A.173)
Reaction Velocity
Maximum Reaction Velocity
√ 2
%1 = σ/ 29700 + 0.5997 · σ 1.5
Γmax 0w0d1 = 0
−7496.874 · σw0 d1
· %1 (A.174)
Γmax 0w0d2 = −7496.874 · 0
σw0 d2
· %1 (A.175)
Γmax 0w0d3 = −7496.874 · 0
σw0 d3
· %1 (A.176)
Γmax 0w0lh = −7496.874 · 0
σw0lh · %1 (A.177)
Γmax 0w0lw = −7496.874 · 0
σw0 lw
· %1 (A.178)
Γmax 0svd1 = −7496.874 · 0
σsvd1 · %1 (A.179)
A0w0d1 = 0
−269.461389 · σw0 d1
/σ 1.7913 (A.180)
A0w0d2 = −269.461389 · 0
σw0 d2
/σ 1.7913 (A.181)
A0w0d3 = −269.461389 · 0
σw0 d3
/σ 1.7913 (A.182)
A0w0lh = 0
−269.461389 · σw0 lh
/σ 1.7913 (A.183)
A0w0lw = −269.461389 · 0
σw0 lw
/σ 1.7913 (A.184)
A0svd1 = −269.461389 · 0
σsv d1
/σ 1.7913 (A.185)
114 Models used
Potential Reaction Velocity

A
β β
%1 = · exp A· 1−
βopt βopt

β β
g = 1 + ln −
βopt βopt
0
0 1 β
gw0 d1
= −1 ·
(β/βopt ) βopt w0d1
Γ0w0d1 Γmax 0w0d1 + Γmax · A0w0d1 · g − A · gw0

0

= %1 · d1
(A.186)
0
0 1 β
gw0 d2
= −1 ·
Γ0w0d2 %1 · Γmax 0w0d2 + Γmax · A0w0d2 · g − A · gw0

0

= d2
(A.187)
0
0 1 β
gw0 d3
= −1 ·
Γ0w0d3 Γmax 0w0d3 + Γmax · A0w0d3 · g − A · gw0

0

= %1 · d3
(A.188)
0
0 1 β
gw0 lh
= −1 ·
(β/βopt ) βopt w0lh
Γ0w0lh Γmax 0w0lh + Γmax · A0w0lh · g − A · gw0

0

= %1 · lh
(A.189)
0
0 1 β
gw0 lw
= −1 ·
(β/βopt ) βopt w0lw
Γ0w0lw %1 · Γmax 0w0lw + Γmax · A0w0lw · g − A · gw0

0

= lw
(A.190)
0
0 1 β
gsv d1
= −1 ·
(β/βopt ) βopt svd1
Γ0svd1 %1 · Γmax 0svd1 + Γmax · A0svd1 · g − A · gsv

0

= d1
(A.191)
0
β 1
Γ0d = %1 · Γmax · A · · −1 (A.192)
βopt d
(β/βopt )
Reaction Intensity
Ir 0w0d1 heat · ηs · Γ0w0d1 · ηM + Γ · ηM 0w0d1

= (A.193)
Ir 0w0d2 Γ0w0d2 ηM 0w0d2

= heat · ηs · · ηM + Γ · (A.194)
Ir 0w0d3 Γ0w0d3 ηM 0w0d3

= heat · ηs · · ηM + Γ · (A.195)
Ir 0w0lh Γ0w0lh ηM 0w0lh

= heat · ηs · · ηM + Γ · (A.196)
Ir 0w0lw Γ0w0lw ηM 0w0lw

= heat · ηs · · ηM + Γ · (A.197)
Ir 0svd1 Γ0svd1 ηM 0svd1

= heat · ηs · · ηM + Γ · (A.198)
Ir 0d = heat · ηs · Γ0d · ηM (A.199)
Ir 0md1 = heat · ηs · Γ · ηM 0md1 (A.200)

Ir 0mlh = heat · ηs · Γ · ηM 0mlh (A.203)
Ir 0mlw = heat · ηs · Γ · ηM 0mlw (A.204)
Ir 0mx = heat · ηs · Γ · ηM 0mx (A.205)
(A.206)
Propagating Flux Ratio
√
xiz = exp((0.792 + 0.37597 · σ) · (β + 0.1))
xin = 192 + 0.0791 · σ
0

0 0.187985 · σw0 d1
√ 0
xiz = xiz · √ · (β + 0.1) + 0.792 + 0.37597 σ · βw0 d1
σ
xin0 = 0
0.0791 · σw0 d1
0 xiz 0 · xin − xiz · xin0

ξw0 d1
= 2
(A.207)
xin 0

0.187985 · σw0d2 √ 0
xiz 0 = xiz · √ · (β + 0.1) + 0.792 + 0.37597 σ · βw0d2
σ
xin0 = 0
0.0791 · σw0 d2

ξw0 d2
= 2
(A.208)
xin 0

0.187985 · σw0 √ 0
xiz 0 = xiz · √ d3
· (β + 0.1) + 0.792 + 0.37597 σ · βw0 d3
σ
xin0 = 0
0.0791 · σw0 d3

ξw0 d3
= 2
(A.209)
xin 0

0.187985 · σw0lh √ 0
xiz 0 = xiz · √ · (β + 0.1) + 0.792 + 0.37597 σ · βw0lh
σ
xin0 = 0
0.0791 · σw0 lh

ξw0 lh
= 2
(A.210)
xin 0

0.187985 · σw0lw √ 0
xiz 0 = xiz · √ · (β + 0.1) + 0.792 + 0.37597 σ · βw0 lw
σ
xin0 = 0
0.0791 · σw0 lw

ξw0 lw
= 2
(A.211)
xin 0

0.187985 · σsv
xiz 0 = xiz · √ d1
· (β + 0.1)
σ
xin0 = 0
0.0791 · σsv d1

ξsv = (A.212)
d1
xin2
0 √
xiz = xiz · 0.792 + 0.37597 · σ · βd0
ξd0 = xiz 0 /xin (A.213)
116 Models used
Heat Sink
εd1 = exp(−452.76/svd1 )
εd2 = exp(−452.76/svd2 )
εd3 = exp(−452.76/svd3 )
εlh = exp(−452.76/svlh )
εlw = exp(−452.76/svlw )
Qd1 = 581.5 + 25.957 · md1
Qd2 = 581.5 + 25.957 · md2
Qd3 = 581.5 + 25.957 · md3
Qlh = 581.5 + 25.957 · mlh
Qlw = 581.5 + 25.957 · mlw
hskz = svd1 · w0d1 · εd1 · Qd1 + svd2 · w0d2 · εd2 · Qd2 + svd3 · w0d3 · εd3 · Qd3
+svlh · w0lh · εlh · Qlh + svlw · w0lw · εlw · Qlw
ρb · hskz
hsk =
swt
svd1 hskz

0
hskzw0 d1
= · εd1 · Qd1 −
swt swt
0 (ρb 0w0d1 · hskz + ρb · hskzw00
d1
) · swt − ρb · hskz · svd1
hskw0 d1
= 2
(A.214)
swt
svd2 hskz

0
hskzw0 d2
= · εd2 · Qd2 −
swt swt
d2
hskw0 = (A.215)
d2
swt2
svd3 hskz

0
hskzw0 d3
= · εd3 · Qd3 −
swt swt
d3
hskw0 d3
= 2
(A.216)
swt
svlh hskz

0
hskzw0lh
= · εlh · Qlh −
swt swt
0 (ρb 0w0lh · hskz + ρb · hskzw0
0
lh
) · swt − ρb · hskz · svlh
hskw0 = (A.217)
lh
swt2
svlw hskz

0
hskzw0 lw
= · εlw · Qlw −
swt swt
0 (ρb 0w0lw · hskz + ρb · hskzw00
lw
) · swt − ρb · hskz · svlw
hskw0 = (A.218)
lw
swt2
0
hskm d1
= ρb · 25.957 · swd1 · εd1 /swt (A.219)
0
hskm d2
= ρb · 25.957 · swd2 · εd2 /swt (A.220)
0
hskm d3
= ρb · 25.957 · swd3 · εd3 /swt (A.221)
0
hskm lh
= ρb · 25.957 · swlh · εlh /swt (A.222)
0
hskm lw
= ρb · 25.957 · swlw · εlw /swt (A.223)
452.76

0
hskzsv d1
= w0d1 · εd1 · Qd1 · 1 +
svd1
ρb w0d1

0 0
hsksv d1
= · hskzsvd1 − hskz · (A.224)
swt swt
ρb 0d · hskz
hskd0 = (A.225)
swt
Slope and Wind
Slope Factor
%1 = −1.5825 · β −1.3 · tan2 (slp)
φs 0w0d1 = 0
%1 · βw0 d1
(A.226)
φs 0w0d2 = 0
%1 · βw0 d2
(A.227)
φs 0w0d3 = %1 · 0
βw0 d3
(A.228)
φs 0w0lh = %1 · 0
βw0 lh
(A.229)
φs 0w0lw = %1 · 0
βw0 lw
(A.230)
φs 0d = %1 · βd0 (A.231)
1 + tan2 (slp)
φs 0s lp = 2 · φs · (A.232)
tan(slp)
Wind
B
0.00718092
%1 =
σ 0.46
0 0
Bw0 d1
= σw0d1
· %1 (A.233)
0 0
Bw0 d2
= σw0 d2
· %1 (A.234)
0 0
Bw0 d3
= σw0 d3
· %1 (A.235)
0 0
Bw0 lh
= σw0 lh
· %1 (A.236)
0 0
Bw0 lw
= σw0 lw
· %1 (A.237)
0 0
Bsv d1
= σsv d1
· %1 (A.238)
C
%1 = −0.284267115 · exp(−0.06919 · σ 0.55 )/σ 0.45
0 0
Cw0 d1
= σw0 d1
· %1 (A.239)
0 0
Cw0 d2
= σw0 d2
· %1 (A.240)
0 0
Cw0 d3
= σw0 d3
· %1 (A.241)
0 0
Cw0 lh
= σw0 lh
· %1 (A.242)
0 0
Cw0 lw
= σw0 lw
· %1 (A.243)
0 0
Csv d1
= σsv d1
· %1 (A.244)
E
%1 = −0.000078221 · exp(−0.0001094 · σ)
0 0
Ew0 d1
= σw0 d1
· %1 (A.245)
0 0
Ew0 d2
= σw0 d2
· %1 (A.246)
0 0
Ew0 d3
= σw0 d3
· %1 (A.247)
0 0
Ew0 lh
= σw0 lh
· %1 (A.248)
0 0
Ew0 lw
= σw0 lw
· %1 (A.249)
0 0
Esv d1
= σsv d1
· %1 (A.250)
118 Models used
Wind Factor
wB = (196.86 · wsp)B
wL = ln(196.86 · wsp)
−E
β
bE =
βopt

β
bL = ln
βopt
φw 0w0d1 = 0
Cw0 d1
· wB · bE
0
+ C · wB · Bw0 d1
· wL · bE
(β/βopt )0w0d1

0
+ C · wB · bE · −Ew0 d1
· bL − E · (A.251)
(β/βopt )
φw 0w0d2 = 0
Cw0 d2
· wB · bE
0
+ C · wB · Bw0 d2
· wL · bE
(β/βopt )0w0d2

0
+ C · wB · bE · −Ew0 d2
· bL − E · (A.252)
(β/βopt )
φw 0w0d3 = 0
Cw0 d3
· wB · bE
0
+ C · wB · Bw0 d3
· wL · bE
(β/βopt )0w0d3

0
+ C · wB · bE · −Ew0 d3
· bL − E · (A.253)
(β/βopt )
φw 0w0lh = 0
Cw0 lh
· wB · bE
0
+ C · wB · Bw0lh
· wL · bE
(β/βopt )0w0lh

0
+ C · wB · bE · −Ew0 lh
· bL − E · (A.254)
(β/βopt )
φw 0w0lw = 0
Cw0 lw
· wB · bE
0
+ C · wB · Bw0 lw
· wL · bE
(β/βopt )0w0lw

0
+ C · wB · bE · −Ew0 lw
· bL − E · (A.255)
(β/βopt )
φw 0svd1 = 0
Csv d1
· wB · bE
0
+ C · wB · Bsv d1
· wL · bE
(β/βopt )0svd1

0
+ C · wB · bE · −Esv d1
· bL − (E · ) (A.256)
(β/βopt )
φw · B
φw 0wsp = (A.257)
wsp
0
β
βopt
φw 0d = φw · −E · d
(A.258)
(β/βopt )
Slope and Wind Vector Addition
vx0w0d1 = φs 0w0d1 + φw 0w0d1 · cos(ϑ)

0
vyw0 d1
= φw 0w0d1 · sin(ϑ) (A.259)
0 vx0w0d1 · vx + vy · 0
vyw0 d1
vlw0 d1
= (A.260)
vl
vx0w0d2 = φs 0w0d2 + φ0ww0d2 · cos(ϑ)
0
vyw0 d2
= φw 0w0d2 · sin(ϑ) (A.261)
0 vx0w0d2 · vx + vy · vyw0
0
d2
vlw0 d2
= (A.262)
vl
vx0w0d3 = φs 0w0d3 + φ0ww0d3 · cos(ϑ)
0
vyw0 d3
= φw 0w0d3 · sin(ϑ) (A.263)
0 vx0w0d3 · vx + vy · vyw00
d3
vlw0 d3
= (A.264)
vl
vx0w0lh = φs 0w0lh + φw 0w0lh · cos(ϑ)
0
vyw0 lh
= φw 0w0lh · sin(ϑ) (A.265)
0 vx0w0lh · vx + vy · 0
vyw0 lh
vlw0 lh
= (A.266)
vl
vx0w0lw = φs 0w0lw + φw 0w0lw · cos(ϑ)
0
vyw0 lw
= φw 0w0lw · sin(ϑ) (A.267)
0 vx0w0lw · vx + vy · vyw0
0
lw
vlw0 lw
= (A.268)
vl
vx0d = φs 0d + φw 0d · cos(ϑ)
vyd0 = φw 0d · sin(ϑ) (A.269)
vx0d · vx + vy · vyd0
vld0 = (A.270)
vl
vx0svd1 = φw 0svd1 · cos(ϑ)
0
vysv d1
= φw 0svd1 · sin(ϑ) (A.271)
0 vx0svd1
· vx + vy · 0
vysv d1
vlsv d1
= (A.272)
vl
vx0asp = −phiw · sin(ϑ)
0
vyasp = phiw · cos(ϑ) (A.273)
0 vx0asp · vx + vy · 0
vyasp
vlasp = (A.274)
vl
vx0slp = φs 0slp
0 vx0slp · vx
vlslp = (A.275)
vl
vx0wdr = −phiw · sin(ϑ)
0
vywdr = phiw · cos(ϑ) (A.276)
0 vx0wdr · vx + vy · vywdr
0
vlwdr = (A.277)
vl
vx0wsp = φw 0wsp · cos(ϑ)
0
vywsp = φw 0wsp · sin(ϑ) (A.278)
0 vx0wsp · vx + vy · 0
vywsp
vlwsp = (A.279)
vl
120 Models used
Spread Direction
0 0
0 vyw0 d1
· vl − vy · vlw0 d1
sdrw0 d1
= p (A.280)
vl · vl2 − vy 2
0 0
0 vyw0 d2
sdrw0 d2
= p (A.281)
vl · vl2 − vy 2
0 0
0 vyw0 d3
sdrw0 d3
= p (A.282)
vl · vl2 − vy 2
0 0
0 vyw0 lh
· vl − vy · vlw0lh
sdrw0 lh
= p (A.283)
vl · vl2 − vy 2
0 0
0 vyw0 lw
· vl − vy · vlw0 lw
sdrw0 lw
= p (A.284)
vl · vl2 − vy 2
0
0 −vy · vlsv d1
sdrsv d1
= p (A.285)
vl · vl2 − vy 2
vyd0 · vl − vy · vld0
sdrd0 = p (A.286)
vl · vl2 − vy 2
0 0
0 vywsp · vl − vy · vlwsp
sdrwsp = p (A.287)
vl · vl2 − vy 2
0 0
0 vywdr · vl − vy · vlwdr
sdrwdr = p (A.288)
vl · vl2 − vy 2
0
0 −vy · vlslp
sdrslp = p (A.289)
vl · vl2 − vy 2
0 0
0 vyasp · vl − vy · vlasp
sdrasp = p (A.290)
vl · vl2 − vy 2
Effective Wind Speed
vl
eL = ln
C · (β/βopt )−E
ef w
eA =
B 2 · C · (β/βopt ) · vl
(A.291)

0 0 β
ef ww0 d1
= eA · −Bw0 d1
· eL · C · βopt
·vl (A.292)

0 β
−B · Cw0 · ·vl
d1
βopt
0 β β
+ B · C · Ew0 · ln βopt
· ·vl
d1
βopt 0
β
+B · C · E · βopt
·vl
w0d1
0
β

+B · C · βopt
·vlw0 d1

0 0 β
ef ww0 d2
= eA · −Bw0 d1
· eL · C · βopt
·vl (A.293)

0 β
−B · Cw0 · ·vl
d1
βopt
0 β β
+B · C · Ew0 · ln βopt
· ·vl
d1
βopt 0
β
·vl
w0d1
0
β

+B · C · βopt
·vlw0d1

0 0 β
ef ww0 d3
= eA · −Bw0 d3
· eL · C · βopt
·vl (A.294)

0 β
−B · Cw0 · ·vl
d3
βopt
0 β β
· ·vl
d3
βopt 0
β
·vl
w0d3
0
β

+B · C · βopt
·vlw0 d3

0 0 β
ef ww0 lh
= eA · ( −Bw0 lh
· eL · C · βopt
·vl (A.295)

0 β
−B · Cw0 · ·vl
lh
βopt
0 β β
· ·vl
lh
βopt 0
β
·vl
w0lh
β 0
+B · C · βopt
·vlw0 lh
)

0 0 β
ef ww0 = eA · ( −Bw0 · eL · C · ·vl (A.296)
lw lw
βopt
0 β
−B · Cw0 · ·vl
lw
βopt
0 β β
· ·vl
lw
βopt 0
β
·vl
w0lw
β 0
+B · C · βopt
·vlw0 lw
)

0 0 β
ef wsv = eA · ( −Bsv · eL · C · ·vl (A.297)
d1 d1
βopt
0 β
−B · Csv · ·vl
d1
βopt
0 β β
+B · C · Esv · ln( βopt
) · ·vl
βopt 0
d1
β
·vl
svd1
β 0
+B · C · βopt
·vlsv d1
)
vld0 · (β/βopt ) + vl · E · (β/βopt )0d

ef wd0 = ef w · (A.298)
B · vl · (β/βopt )
0
0 vlslp
ef wslp = ef w · (A.299)
B · vl
0
0 vlasp
ef wasp = ef w · (A.300)
B · vl
0
0 vlwsp
ef wwsp = ef w · (A.301)
B · vl
0
0 vlwdr
ef wwdr = ef w · (A.302)
B · vl
122 Models used
Rate of Spread
cz1 = Ir 0w0d1 · ξ · (1 + vl)

0
cz2 = Ir · ξw0 d1
· (1 + vl)
0
cz3 = Ir ·ξ · vlw0 d1
0
cz1 + cz2 + cz3 − ros · hskw0
ros0w0d1 = d1
(A.303)
hsk
cz1 = Ir 0w0d2 · ξ · (1 + vl)
0
cz2 = Ir · ξw0 d2
· (1 + vl)
0
0
ros0w0d2 = d2
(A.304)
hsk
cz1 = Ir 0w0d3 · ξ · (1 + vl)
0
cz2 = Ir · ξw0 d3
· (1 + vl)
0
0
cz1 + c2 + cz3 − ros · hskw0
ros0w0d3 = d3
(A.305)
hsk
cz1 = Ir 0w0lh · ξ · (1 + vl)
0
cz2 = Ir · ξw0lh
· (1 + vl)
0
cz3 = Ir ·ξ · vlw0 lh
0
ros0w0lh = lh
(A.306)
hsk
cz1 = Ir 0w0lw · ξ · (1 + vl)
0
cz2 = Ir · ξw0lw
· (1 + vl)
0
cz3 = Ir ·ξ · vlw0 lw
0
ros0w0lw = lw
(A.307)
hsk
cz1 = Ir 0svd1 · ξ · (1 + vl)
0
cz2 = Ir · ξsv d1
· (1 + vl)
0
cz3 = Ir ·ξ · vlsv d1
0
cz1 + cz2 + cz3 − ros · hsksv
ros0svd1 = d1
(A.308)
hsk
cz1 = Ir 0d ·ξ · (1 + vl)
cz2 = Ir · ξd0 · (1 + vl)
cz3 = Ir ·ξ · vld0
cz1 + cz2 + cz3 − ros · hskd0
ros0d = (A.309)
hsk
cz1 = Ir 0md1 · ξ · (1 + vl)
0
cz1 · hsk − Ir · ξ · (1 + vl) · hskm
ros0md1 = 2
d1
(A.310)
hsk
cz1 = Ir 0md2 · ξ · (1 + vl)
ros0md2 = 0
(cz1 · hsk − Ir · ξ · (1 + vl) · hskm d2
)/hsk2 (A.311)
cz1 = Ir 0md3 · ξ · (1 + vl)
0
cz1 · hsk − Ir · ξ · (1 + vl) · hskm
ros0md3 = 2
d3
(A.312)
hsk
cz1 = Ir 0mlh · ξ · (1 + vl)

0
cz1 · hsk − Ir · ξ · (1 + vl) · hskm
ros0mlh = 2
lh
(A.313)
hsk
cz1 = Ir 0mlw · ξ · (1 + vl)
0
cz1 · hsk − Ir · ξ · (1 + vl) · hskm
ros0mlw = lw
(A.314)
hsk2
0
Ir · ξ · vlslp
ros0slp = (A.315)
hsk
0
Ir · ξ · vlasp
ros0asp = (A.316)
hsk
0
Ir · ξ · vlwsp
ros0wsp = (A.317)
hsk
0
Ir · ξ · vlwdr
ros0wdr = (A.318)
hsk
Ir 0mx · ξ · (1 + vl)
ros0mx = (A.319)
hsk
124 Models used
A.2 Debris Flow Model

Procedure
The following listing is an Arc Macro Language (AML) script which automatically
extracts
a stream network
adjacent areas that proliferate debris
start sections of debris flows within the stream network, and
debris flows paths along the stream network
It runs in the GRID module of ArcInfo. The one and only input dataset is a
depression-less (i.e. hydrological correct) digital terrain model in ArcInfo’s GRID
format.
A.2 Debris Flow Model 125
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/* Environmental Systems Research I n s t i t u t e , Inc .
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/* P r o g r a m : DEBRIS . AML
/* Purpose : d e l i n e a t e the d e b r i s flow prone areas .
/*
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/* U s a g e : DEBRIS FLOW
/* U s a g e : DEBRIS < r o u t i n e > { a r g s }
/*
/* R o u t i n e s : MAIN − m a i n i n t e r f a c e
/* STREAM − e x t r a c t t h e s t r e a m n e t w o r k
/* AREAS − s e l e c t a d j a c e n t a r e a s o f n e t w o r k
/* STREAM PROCESS − d e b r i s f l o w s t a r t i n g p o i n t s
/* WATERSHEDS − d e l i n e a t e w a t e r s h e d s
/* FLOWPATH − p a t h s o f d e b r i s f l o w
/* WRITE DF TABLE − w r i t e h o r i z o n t a l f a c t o r t a b l e f o r FLOWPATH
/* routine
/* WRITE VF TABLE − w r i t e h o r i z o n t a l f a c t o r t a b l e f o r AREAS
/* routine
/*
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/* Calls : −
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/* N o t e s : T h e p r o g r a m a s s u m e s a h y d r o l o g i c a l c o r r e c t dhm c o n t a i n i g
/* no s i n k s . A d d i t i o n a l l y , t h e u n i t o f h o r i z o n t a l and h e i g h t
/* v a l u e s i s assumed to be meters .
/*
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/* H i s t o r y : a n d i bachmann − 0 8 / 1 6 / 0 0 − O r i g i n a l c o d i n g
/* a n d i bachmann − 0 6 / 0 1 / 0 1 − s t r e a m l i n i n g , c l e a n i n g up . . .
/*==========================================================================
/*
&args r o u t i n e a r g l i s t : r e s t
/*
&severity &error &routine b a i l o u t
/*
/* Check arguments
&if ˆ [ n u l l % r o u t i n e % ] &then
&call % r o u t i n e %
&else
&call usage
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine MAIN
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/ * @ p a r a m MAIN < dhm > { c u r v a t u r e t h r e s h o l d } { s l p t h r e s h o l d }
/ * d e f a u l t s : c u r v a t u r e −2 , s l p : 2 0
&set dhm = [ e x t r a c t 1 [unquote % a r g l i s t % ] ]
&set c u r t h = [ e x t r a c t 2 [unquote % a r g l i s t % ] ]
&set s l p t h = [ e x t r a c t 3 [unquote % a r g l i s t % ] ]
&if [ n u l l %dhm% ] &then
&do
routines %a m l $ f u l l f i l e%
&return
&end
/*
/ * STREAM : e x t r a c t the stream network
&run % a m l $ f u l l f i l e % s t r e a m %dhm% s t r e a m % c u r t h %
/ * AREAS : extract p r o l i f e r a t i n g areas
&run % a m l $ f u l l f i l e % a r e a s %dhm% s t r e a m f l a c c f l d i r % s l p t h %
/ * STREAM PROCESS : e x t r a c t mudslide points in stream network
&run % a m l $ f u l l f i l e % s t r e a m p r o c e s s %dhm% s t r e a m f l a c c d f l t m p
/ * WATERSHEDS : extract watersheds
&run % a m l $ f u l l f i l e % w a t e r s h e d s %dhm% f l d i r f l a c c d f l t m p d f l s t a r t wsh
/ * FLOWPATH : c a l c u l a t e mudflow t r a j e c t o r i e s . . .
&run % a m l $ f u l l f i l e % f l o w p a t h %dhm% s t r e a m f l d i r f l a c c d f l s t a r t d e b r i s f l o w
/*
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine STREAM
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/ * @ p a r a m STREAM < dhm> < o s t r e a m > { c u r v a t u r e t h r e s h o l d }
/*
/* E x t r a c t a stream network from a g i v e n d i g i t a l h e i g h t model
/ * **********************************************************
/*
/* parse the args
/ * −−−−−−−−−−−−−−
&if [ n u l l %dhm% ] | | ˆ [ e x i s t s %dhm% −GRID] &then
&return &warning No DHM s p e c i f i e d o r DHM n o t f o u n d !
/*
&set s t r e a m = [ e x t r a c t 2 [unquote % a r g l i s t % ] ]
&if [ n u l l % s t r e a m % ] | | [ e x i s t s % s t r e a m % −GRID] &then
&return &warning Output−Grid % s t r e a m % a l r e a d y e x i s t s !
/*
&set c u r v t h r s h l d = [ e x t r a c t 3 [unquote % a r g l i s t % ] ]
&if [ n u l l % c u r v t h r s h l d % ] &then
&set c u r v t h r s h l d = −2
/*
/*
/*
/* d e c l a r e temporary g r i d s
/*
&set f l d i r = [scratchname −PREFIX f l d i r −DIR]
126 Models used
&set f l a c c = [scratchname −PREFIX f l a c c −DIR]

&set s t r e a m n e t = [scratchname −PREFIX s t r n e t −DIR]
&set s t r o r d e r = [scratchname −PREFIX s t r o r d e r −DIR]
&set c u r v = [scratchname −PREFIX c u r v −DIR]
&set c u r v h = [scratchname −PREFIX c u r v h −DIR]
/*
&set k i l l l i s t = % f l d i r % % f l a c c % % s t r e a m n e t % % s t r o r d e r % % c u r v % % c u r v h %
/*
/*
SETMASK OFF
/*
/* 1. Calculate flow direction
/ * −−−−−−−−−−−−−−−−−−−−−−−−−−−
% f l d i r % = FLOWDIRECTION ( %dhm% )
/*
/ * 2 . C o n s t r a i n t on a c c u m u l a t i o n a r e a ( H e i n i m a n n 1 9 9 8 , S . 1 2 3 , P k t . 1 ) :
/ * −−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/* c a t c h m e n t a r e a > 0 . 2 5 h a = 2 5 0 0 m2
&describe %dhm%
&set m i n p i x e l = 2 5 0 0 / ( % g r d $ d x % * % g r d $ d y % )
%f l a c c% = FLOWACCUMULATION(% f l d i r %)
%s t r e a m n e t% = CON(% f l a c c % > % m i n p i x e l % , % f l a c c %)
/*
/ * 3 . C o n s t r a i n t b a s e d on S t r a h l e r − n u m b e r and p l a n a r c u r v a t u r e :
/* The s t r e a m − s e g m e n t s w i t h S t r a h l e r − N u m b e r 1 h a v e t o b e c o n c a v e
/ * −−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
% s t r o r d e r % = STREAMORDER(% s t r e a m n e t % , % f l d i r % , STRAHLER)
/*
/* c u r v h = h o r i z o n t a l c u r v a t u r e , r a t e of change of s u r f a c e p e r p e n d i c u l a r
/* to aspect , unit : 1/ 100 [ z u n i t s ] , i . e . %
% c u r v % = CURVATURE(%dhm% , # , % c u r v h %)
% s t r e a m % = CON(% s t r o r d e r % = = 1 , ˜
CON(% c u r v h % < % c u r v t h r s h l d % , % s t r o r d e r % ) , % s t r o r d e r %)
/*
/ * c l e a n up
/ * −−−−−−−−
&do g & l i s t % k i l l l i s t %
&if [ e x i s t s %g% −GRID] &then
KILL %g% ALL
&end
/*
/*
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine AREAS
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/ * @ p a r a m AREAS < dhm> < s t r e a m > < o a r e a s > < o f l d i r > { s l p t h r s h l d }
/*
/* d e l i n e a t e c o n t r i b u t i n g areas
/*
/* Heinimann d e s c r i b e s 3 d i f f e r e n t t y p e s :
/* a ) steep neighbouring areas
/* b ) s l i d e a r e a s ( a s d e l i n e a t e d b y a n e x p e r t ) −> w e f o r g e t t h a t !
/* c ) rock areas −> w e f o r g e t t h a t !
/ * we c o n c e n t r a t e on a ) : t h i s i s v a l i d f o r t h e M a l c a n t o n e r e g i o n .
/ * maximum d i s t a n c e = 2 5 0 m
/ * s l o p e > 2 0 d e g ( a s we a s s u m e f i r e h a v i n g d e s t r o y e d t h e v e g e t a t i o n , we
/* d o n ’ t h a v e a n y a r e a s WITH v e g e t a t i o n ! )
/*
&set a r e a s = [ e x t r a c t 3 [unquote % a r g l i s t % ] ]
&set f l d i r c r = [ e x t r a c t 4 [unquote % a r g l i s t % ] ]
&set s l p t h r s h l d = [ e x t r a c t 5 [unquote % a r g l i s t % ] ]
/*
&if [ n u l l % s l p t h r s h l d % ] &then
&set s l p t h r s h l d = 2 0
/*
&set s t r e a m b u f = [scratchname −PREFIX s t r b u f −DIR]
&set f l d i r b l k = [scratchname −PREFIX f l b l k −DIR]
/*
&set k i l l l i s t = % s t r e a m b u f % , % f l d i r b l k %
/*
/* c r e a t e a f i l e f o r the v e r t i c a l f a c t o r f u n c t i o n
&set f i l e = [scratchname −PREFIX v f t a b l e −SUFFIX . d a t −FILE]
&run % a m l $ f u l l f i l e % w r i t e v f t a b l e % s l p t h r s h l d % % f i l e %
/*
SETMASK OFF
% s t r e a m b u f % = PATHDISTANCE(% s t r e a m % , # , %dhm%, # , # ,%dhm% , ”TABLE % f i l e % ” , ˜
% f l d i r b l k %, # , 2 5 0 )
/* f l d i r b l k i s used to c o r r e c t the f l o w d i r e c t i o n s !
/* t h e pathdistance−command i n c l u d e s p i x e l s t h a t don ’ t f l o w p r e d o m i n a n t l y i n t o
/* t h e s t r e a m n e t w o r k , b u t maybe i n t o a n e i g h b o u r i n g one t h a t was e x c l u d e d
/* j u s t b e f o r e . S t i l l i f the c o n d i t i o n of the s l o p e a n g l e i s f u l l f i l l e d , than
/ * t h e r e must a l e a s t some p a r t o f t h e f l o w come d i r e c t l y t h i s way .
/*
/* Transform bac klink−co des to f l o w d i r e c t i o n − c o d e s :
/*
/* f r o m PATHDISTANCE ( ) b a c k l i n k d i r e c t i o n c o d i n g
/* 6 7 8
/* 5 0 1
/* 4 3 2
/* t o FLOWDIRECTION ( ) c o d i n g . .
/* 32 64 128
/* 16 1
/* 8 4 2
% f l d i r c r % = CON( % f l d i r b l k % = = 0 , FLOWDIRECTION ( %dhm% ) , ˜
INT ( POW( 2 , % f l d i r b l k % − 1 ) ) )
/*
% a r e a s % = FLOWACCUMULATION(% f l d i r c r %)
/*
/ * c l e a n up
/ * −−−−−−−−
KILL %g% ALL
&end
/*
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine WRITE VF TABLE
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/ * WRITE VF TABLE < t h r s h l d > < f i l e n a m e >
/*
/ * The f i l e v e r t a n g l e . d a t l o o k s l i k e t h i s
/* > % s l p t h r s h l d % 1
/* > 90 1
/*
/ * & s e t f i l e = [ s c r a t c h n a m e −PREFIX v f t a b l e −SUFFIX . d a t − F I LE ]
&set v a l u e = [ e x t r a c t 1 [unquote % a r g l i s t % ] ]
&set f i l e = [ e x t r a c t 2 [unquote % a r g l i s t % ] ]
/*
&set i o = [open % f i l e % o p e n s t a t u s −WRITE]
&if % o p e n s t a t u s % > 0 &then
&return &error c o u l d n o t open % f i l e % : e r r o r c o d e = % o p e n s t a t u s %
/*
&set l i n e = [ q u o t e % v a l u e % 1 ]
&set x = [ w r i t e % i o % % l i n e % ]
&set l i n e = [ q u o t e 9 0 1 ]
&set x = [ c l o s e % i o % ]
/*
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine STREAM PROCESS
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/ * @ p a r a m STREAM PROCESS < dhm> < s t r e a m > < f l a c c > < o d f l t m p >
/* <dhm > dhm
/* <s t r e a m > s t r e a m n e t w o r k
/* <f l a c c > f l o w a c c u m u l a t i o n o f t h e c o n t r i b u t i n g a r e a s
/* <o d f l s t a r t > p o t e n t i a l s t a r t stream p i x e l s
/*
/* s l o p e along the f l o w path s h o u l d be g r e a t e r than
/* J = 0 . 3 2 * f l a c c ˆ ( −0 . 2 ) [ 8 . 2 ] J = b o u n d a r y s l o p e ( Eq . 1 )
/* w i t h f l a c c b e i n g t h e u p s l o p e c a t c h m e n t a r e a i n Km2
/* r e c a l c u l a t e f l o w accumulation , only p i x e l s of s t r e a m b u f s h o u l d be
/* i n c l u d e d , i . e . the p i x e l s t h a t are p o t e n t i a l l y d e b r i s p r o l i f e r a t i n g . . .
/ * s t r e a m p i x e l w i t h f l a c c < = 1 ha ( = 1 6 p i x e l s o f 2 5 x25m ) d o n ’ t s u p p o r t
/* m u d s l i d e s . . .
/* r a t i o n a l e : most i m p o r t a n t p a r a m e t e r s are
/* o the downflow slope
/* o the potential debris
/* o and t h e d o w n f l o w
/*
/* p r o c e e d i n g :
/* general :
/* 1 . s e l e c t s t r e a m − p i x e l s w i t h c a t c h m e n t a r e a >= 1 h a
/* 2 . c a l c u l a t e b o u n d a r y s l o p e f o r e v e r y s e l e c t e d s t r e a m − p i x e l ( Eq . 1 )
/* 3 . s e l e c t a l l p i x e l s w i t h a c t u a l s l o p e >= b o u n d a r y s l o p e
/*
&set f l a c c = [ e x t r a c t 3 [unquote % a r g l i s t % ] ]
&set s t r p r o c = [ e x t r a c t 4 [unquote % a r g l i s t % ] ]
/*
/* &set = [ s c r a t c h n a m e −PREFIX −DIR]
&set b n d s l p = [scratchname −PREFIX b s l p −DIR]
&set f l d i r d r o p = [scratchname −PREFIX d r o p −DIR]
&set f l d i r t m p = [scratchname −PREFIX f l d −DIR]
/*
&set k i l l l i s t = %b n d s l p%, %f l d i r d r o p%, %f l d i r t m p%
/*
&set m i n a r e a = 1 0 0 0 0 /* == 1 ha == 10 0* 10 0 m
/* number o f p i x e l s t h a t c o r r e s p o n d t o 1 ha
&describe % f l a c c %
&set m i n p i x e l = % m i n a r e a % / ( % g r d $ d x % * % g r d $ d y % )
&set k m f a c t o r = ( % g r d $ d x % * % g r d $ d y % ) / 1 0 0 0 0 0 0
/*
SETMASK % s t r e a m %
/*
/* S e l e c t i n g a l l stream p i x e l s with catchment area g t %m i n p i x e l % P i x e l s .
% b n d s l p % = CON( % f l a c c % >= % m i n p i x e l % , ˜
0 . 3 2 * POW( % f l a c c % * % k m f a c t o r % , −0 . 2 ) )
/*
% f l d i r t m p % = FLOWDIRECTION(%dhm% , % f l d i r d r o p %)
% s t r p r o c % = CON(% f l d i r d r o p % > 1 0 0 * % b n d s l p % , % f l d i r d r o p %)
/*
SETMASK OFF
/*
/ * c l e a n up
/ * −−−−−−−−
KILL %g% ALL
&end
128 Models used
/*
/*
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine WATERSHEDS
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/ * @ p a r a m WATERSHEDS < dhm> < f l d i r > < c o n t r a r e a s >
/* @param <d f l t m p > < o d f l s t a r t > < o w s h > { s l p t h r s h l d }
/* delineate watersheds
&set dhm = [ e x t r a c t 1 [unquote % a r g l i s t % ] ] / * dhm
&set f l d i r = [ e x t r a c t 2 [unquote % a r g l i s t % ] ] /* f l d i r c r
&set f l a c c = [ e x t r a c t 3 [unquote % a r g l i s t % ] ] /* areas
&set s o u r c e = [ e x t r a c t 4 [unquote % a r g l i s t % ] ] /* s t r p r o c
&set d e b r i s f l o w = [ e x t r a c t 5 [unquote % a r g l i s t % ] ] /* o d f l s t a r t
&set wsh = [ e x t r a c t 6 [unquote % a r g l i s t % ] ] /* o wsh
&set s l p t h r s h l d = [ e x t r a c t 7 [unquote % a r g l i s t % ] ] /* s l p t h r s h l d
/*
&if [ n u l l % s l p t h r s h l d % ] &then
&set s l p t h r s h l d = 2 0
/*
/* a g g r e g a t e a d j a c e n t stream p i x e l s in order to reduce w a t e r s h e d s
&set o n e = [scratchname −PREFIX o n e −DIR]
&set d f b u f = [scratchname −PREFIX d f b u f −DIR]
&set wsh tmp = [scratchname −PREFIX wstmp −DIR]
&set o n e = [scratchname −PREFIX o n e −DIR]
&set r e g = [scratchname −PREFIX r e g −DIR]
/*
&set k i l l l i s t = %o n e% , % d f b u f % , % w s h t m p % , % r e g %
/*
SETMASK % s o u r c e %
SETWINDOW % s o u r c e %
SETCELL % s o u r c e %
%o n e% = 1
% r e g % = REGIONGROUP( %o n e% , # , EIGHT )
/*
/* g e n e r a t e the v e r t i c a l f a c t o r lookup t a b l e
/* c r e a t e a f i l e f o r the v e r t i c a l f a c t o r f u n c t i o n
&set f i l e = [scratchname −PREFIX v f t a b l e −SUFFIX . d a t −FILE]
&run % a m l $ f u l l f i l e % w r i t e v f t a b l e % s l p t h r s h l d % % f i l e %
/*
SETMASK % f l a c c %
/*
% d f b u f % = PATHDISTANCE(% s o u r c e % , # , %dhm%, # , # ,%dhm% , ”TABLE % f i l e % ” , ˜
#,#,250)
SETMASK % d f b u f %
%w s h t m p% = WATERSHED(% f l d i r % , % r e g % )
%w s h% = c o n ( % w s h t m p % . c o u n t > = 1 6 , % w s h t m p %)
SETMASK OFF
/ * c o n v e r t i n g , n o w e e d − t o l e r a n c e s ( w e m i g h t u s e i t a g a i n i n GRID )
/ * w s h c v = GRIDPOLY (%w s h %)
/* update the s t r e a m p i x e l s
% d e b r i s f l o w % = % s o u r c e % + %w s h%
/*
/ * c l e a n up
/ * −−−−−−−−
KILL %g% ALL
&end
/*
&set x = [ d e l e t e % f i l e % ]
/*
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine FLOWPATH
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/ * @ p a r a m FLOWPATH < dhm> < s t r e a m > < f l d i r > < f l a c c > < d f l s t a r t > < o f l o w p a t h >
/* delineate watersheds
/*
&set dhm = [ e x t r a c t 1 [unquote % a r g l i s t % ] ] / * dhm
&set s t r e a m = [ e x t r a c t 2 [unquote % a r g l i s t % ] ] / * s t r e a m
&set f l d i r = [ e x t r a c t 3 [unquote % a r g l i s t % ] ] / * f l d i r
&set f l a c c = [ e x t r a c t 4 [unquote % a r g l i s t % ] ] / * a r e a s
&set s o u r c e = [ e x t r a c t 5 [unquote % a r g l i s t % ] ] / * d f l s t a r t
&set f l o w p a t h = [ e x t r a c t 6 [unquote % a r g l i s t % ] ] / * o f l o w p a t h
/*
/* &set = [ s c r a t c h n a m e −PREFIX −DIR]
&set r b n d s l p = [scratchname −PREFIX b s l p −DIR]
&set f l d i r d e g = [scratchname −PREFIX f d e g −DIR]
&set h d i f f = [scratchname −PREFIX h d i f f −DIR]
&set s t a r t z = [scratchname −PREFIX z s t r −DIR]
&set z d i f f = [scratchname −PREFIX z d i f f −DIR]
&set pg = [scratchname −PREFIX pg −DIR]
&set d f l t m p = [scratchname −PREFIX d b f l −DIR]
&set mask = [scratchname −PREFIX msk −DIR]
/*
&set k i l l l i s t = % r b n d s l p % , % f l d i r d e g % , % h d i f f % , % s t a r t z % , ˜
% z d i f f % , %p g% , % d f l t m p % , %m a s k%
/*
/* boundary s l o p e w i t h i n the stream
/* f = 0 . 2 a ˆ{−0 . 2 6 } [8 .3 ] , p . 127
&describe % f l a c c %
&set k m f a c t o r = ( % g r d $ d x % * % g r d $ d y % ) / 1 0 0 0 0 0 0
/*
% r b n d s l p % = 0 . 2 * pow ( % f l a c c % * % k m f a c t o r % , −0 . 2 6 )
/*
/* c a l c u l a t e the p a u s c h a l g e f a e l l e only f o r stream p i x e l s

SETMASK % s t r e a m %
/* r e c l a s s f l d i r
% f l d i r d e g % = c o n ( l o g 2 (% f l d i r % ) < 6 , ( l o g 2 (% f l d i r % ) + 2 ) * 4 5 , ˜
( l o g 2 (% f l d i r % ) − 6 ) * 4 5 )
/*
/* h o r i z o n t a l d i s t a n c e
% h d i f f % = p a t h d i s t a n c e ( % s o u r c e %, # , # ,% f l d i r d e g % , ˜
”FORWARD ZEROFACTOR=1”,#,#,#,% s t a r t z % , # , i n t (%dhm% ) )
/* v e r t i c a l d i s t a n c e
%z diff% = % s t a r t z % − %dhm%
/* r i s e
%p g% = %z d i f f% / %h diff%
/*
/* compare pg w i t h r b n d s l p
% d f l t m p % = c o n ( %p g% >= % r b n d s l p % , %p g%)
/* do i t a g a i n
/*
%m a s k% = c o n ( i s n u l l (% s o u r c e % ) , % d f l t m p % , % s o u r c e %)
SETMASK %m a s k%
&call w r i t e d f t a b l e
%f l o w p a t h% = p a t h d i s t a n c e ( % s o u r c e % , # , %dhm% , % f l d i r d e g % , ˜
”FORWARD ZEROFACTOR=1”,%dhm% , ”TABLE % d o w n f l o w t a b l e %” )
SETMASK OFF
/*
/ * c l e a n up
/ * −−−−−−−−
KILL %g% ALL
&end
/*
&set x = [ d e l e t e % d o w n f l o w t a b l e % ]
/*
&return
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
&routine WRITE DF TABLE
/ *−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
/*
/ * The f i l e d f t a b l e . d a t l o o k s l i k e t h i s
/ * > −90 1
/ * > −3 . 0 0 0 1 1
&set d o w n f l o w t a b l e = [scratchname −PREFIX d f t a b l e −SUFFIX . d a t −FILE]
&set i o = [open % d o w n f l o w t a b l e % o p e n s t a t u s −WRITE]
&if % o p e n s t a t u s % > 0 &then
&return &error c o u l d n o t open % d o w n f l o w t a b l e % : e r r o r c o d e = % o p e n s t a t u s %
/*
&set l i n e = [ q u o t e −90 1 ]
&set l i n e = [ q u o t e −3 . 0 0 1 1 ]
&set x = [ c l o s e % i o % ]
/*
&return
/ *−−−−−−−−−−−−
&routine USAGE
/ *−−−−−−−−−−−−
/* D i s p l a y usage f o r t h i s tool
routines %a m l $ f u l l f i l e%
&return &inform
/ *−−−−−−−−−−−
&routine EXIT
/ *−−−−−−−−−−−
/ * C l e a n u p a n d e x i t menu
&dv . mud$*
&if [show &thread & e x i s t s t o o l $ m u d ] &then
&thread &delete t o o l $ m u d
&return
/ *−−−−−−−−−−−−−−
&routine BAILOUT
/ *−−−−−−−−−−−−−−
&severity &error &ignore
/* & c a l l e x i t
&type : [ u p c a s e % a m l $ f i l e % ] : Routine : [upcase % r o u t i n e %] − Line : % a m l $ e r r o r l i n e %
&return &warning \
130 Models used
Curriculum vitae
Andreas Bachmann
born July 14, 1968, in Zurich
citizen of Amden SG, Switzerland
Education
1975 – 1981 Primary school in Zurich.
1981 – 1987 High school in Zurich, (Realgymnasium Rämibühl), concluded
with “Matura” exam type “B” (classical gymnasium).
1988 – 1995 Studies in geography at the university of Zurich. Minors in
environmental studies, biology, chemistry, geology and
mathematics.
1995 Diploma in geography with a thesis on
“Verdunstungsmodellierung mit GIS im unteren Thurgebiet”
(“GIS-based Modelling of the Evapotranporation in the Region
of the River Thur”) advised by Prof. K. Brassel and
Prof. H. Lang (ETH).
1995 – 1997 Project Collaborator in the European research project
MINERVE II (“Modélisation d’Incendie et Études de Risques
pour la Valorisation de L’Environment”), Deptartment of
Geography, University of Zurich
1995 – 2001 Collaborator of the GIS of the Swiss National Park.
1997 – 2001 Research assistant at the Department of Geography, University
of Zurich. Dissertation with a thesis on “GIS-based Wildland
Fire Risk Analysis” advised by Prof. R. Weibel and
Dr. B. Allgöwer.

Thesis AndreasBachmann 2001

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Thesis AndreasBachmann 2001

Загружено:

Авторское право:

Доступные форматы

GIS-based Wildland Fire Risk

Erlangung der naturwissenschaftlichen Doktorwürde

Prof. Dr. Robert Weibel

2 Wildland Fire and Management 5

3 Quantitative Risk Analysis 25

4 A Consistent Wildland Fire Risk Terminology Is Needed! 35

5 Framework For Wildfire Risk Analysis 47

6 Uncertainty Propagation in Wildland Fire Behaviour Modelling 57

7 Wildland Fire Risk Analysis in Southern Switzerland 73

A Models used 103

1.1 Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . 3

3.1 Components of quantitative risk analysis . . . . . . . . . . . . . . 26

4.1 Hypothetical study area showing relationships between objects

5.1 Framework for the wildfire risk analysis . . . . . . . . . . . . . . 49

6.1 Area of interest with the three fuel types . . . . . . . . . . . . . . 62

7.1 Risk matrix for one situation . . . . . . . . . . . . . . . . . . . . 75

2.1 Fire suppression interpretations of flame length . . . . . . . . . . 8

6.1 Fuel model properties . . . . . . . . . . . . . . . . . . . . . . . . 59

1.1 Problem statement

environment (e.g. wind field).

1.3 Structure of the thesis

Figure 1.1: Structure of the thesis

Wildland Fire and

2.1 Wildland Fire

 fire occurrence, and

Fire behaviour research studies the physical combustion processes at different

2.1.1 Fire Behaviour

Physical and Chemical Processes

2. transfer of the heat by any available mechanism to unburnt fuel (physical

3. absorption of heat by unburnt fuel until combustion starts

The process is sustainable if sufficient energy released in step 1 is transfered

unburnt fuel particles to a flame front can be considerably shortened if wind or

Types of Wildland Fires

Flame Fireline inten- Interpretations

Modelling Wildland Fire Behaviour

2.1.2 Wildland Fire Occurrence

Wildland Fires in the World

2.1.3 Wildland Fire Effects

Country Total Human Natural Unknown

Table 2.2: Number of wildland fires in 1996, distinguished by cause (Sources:

Country Total Human Natural Unknown

independent environmental influences. Considering the spatial and temporal

Vegetation There exists abundant literature on fire effects on the vegetation,

2.2 Wildland Fire Management

2.2.2 Wildland Fire Management Activities

Prevention Prevention primarily seeks to reduce the unplanned ignitions. As

2.2.3 Existing Wildland Fire Managements Approaches

2.3 Risk Terminology

(1) liability or exposure to harm or injury; risk; peril. (2) an instance

Fire Danger: The resultant, often expressed as an index, of both

CCFFM (1987) states:

A general term used to express an assessment of both fixed and vari-

In the domain of technical risk engineering, hazard is explained by a more

Hazard: a physical situation with a potential for human injury, dam-

A hazard is not only the precondition for a specific process, as it is seen

(1) exposure to the chance of injury or loss; a hazard or dangerous

The probability or chance of a fire starting determined by the pres-

Risk: the probability of an undesirable event occurring within a spec-

Fire risk: the probability or chance of fire starting determined by the

In general, a fire risk measure will be a product of an expected severity

Danger: Danger is an abstract concept of human perception. Danger per se

Hazard: A hazard is a process leading to undesirable outcomes.

2.5 Risk Management

fire occurrence, and