Awm April2019 4april PDF

4/5/2019 Velocity of Ecological System Trajectory
Velocity of Ecological System Trajectory

A new method for dimension reduction and ecological regime shift
detection
Jessica L. Burnett
N.B. Price, A.J. Tyre, C.R. Allen, D.G. Angeler, & C.R. Allen
2019.04.06
ﬁle://localhost/Users/jessicaburnett/Documents/GitHub/velocity_presentations/awm_april2019/awm_april2019.html#1 1 1/27
/ 27
Ecosystems are Complex

(and complicated)
- high dimensional
- many ( ∞ ?) interactions
- non-linear
- non-ergodic (open)
- dynamic
/ 27

(and complicated)
- high dimensional
- many ( ∞ ?) interactions
- non-linear
- non-ergodic (open)
- dynamic
- difficult to model mathematically
/ 27

Image credit: coolkeywest.com
/ 27
The Holy Grail of Ecology
forecasting change in time to prevent or

mitigate undesirable consequences
/ 27
The Holy Grail of Ecology

Predicting undesirable change
Image credit: The Ocean Agency
/ 27
Ecological Regime Shifts

what? a persistent change in the structure or
functioning of a system
how? loss of negative feedback(s) maintaining the

system
goal? predict in time to prevent

/ 27
Methods for Detecting Regime Shifts

> 70 (!) methods proposed in literature
∼
1
3
suitable for multivariable data
few explicitly handle noisy data
/ 27
Regime Shifts Detection Methods

Some multivariable methods include:
variance index (max eigenvalue of covariance matrix)
Principal Coordinates Analysis
/ 27
Regime Shifts Detection Methods

Some multivariable methods include:
Fisher Information
ﬁle://localhost/Users/jessicaburnett/Documents/GitHub/velocity_presentations/awm_april2019/awm_april2019.html#1 1010/27
/ 27
Some Multivariable Regime Shifts Detection

Methods
Some multivariable methods include
Fisher Information
Velocity
/ 27
Toy System
x normal(μ = 25, σ = 5) for t > 50
x normal(μ = 100, σ = 50) for t ≥ 50
/ 27
Velocity, Δs
Δt
the linear speed of a system's trajectory in phase space
/ 27
Step 1: Calculate s , 'distance travelled'

 X

 2
st = ∑(xit − xit−1 )
⎷
i=1
/ 27
Step 2: Calculate v , t he linear speed of s

Δs
v =
Δt
/ 27
Velocity, Toy System
/ 27
Potential Limitations of v
v increases with increasing e ect size
/ 27
v increases with increasing e ect size
/ 27
Smoothing Noisy Data Before Calculating v
[1] Price NP, JL Burnett. R Package for regularized differentiation.

github.com/natbprice/tvdiff
/ 27
v Identi es Potential Periodicities in

Certain Regimes
/ 27
Empirical System: Paleodiatoms
Data published in T. Spanbauer et al. 2014. Prolonged Instability Prior to a

Regime Shift. PLoS One.
/ 27
Empirical System: Paleodiatoms
/ 27
Distance travelled s & Velocity v identify

these turnover events
/ 27
Conclusions: v is Simple, Intuitive

Reduction of high dimensional data
Capable of handling noisy data
Is not sensitive to data quality issues common in ecology
Best when mean > variance
Identi es regime shifts known a priori
/ 27
Next Steps
Numerical identi cation of change point in v
Compare to distance-based metrics (feedback?)
Compare to ordination techniques
Predictive capacity
/ 27
Related Software
R packages (dev versions):
trashbirdecology/distanceTravelled (calculate s and v)
trashbirdecology/regimeDetectionMeasures (calculate multiple regime

detection metrics)
trashbirdecology/bbsRDM (application to spatial data)
natbprice/tvdiff (regularized differentiation)
/ 27
/ 27

Awm April2019 4april PDF

Загружено:

Сведения о документе

Исходное описание:

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

Awm April2019 4april PDF

Загружено:

Авторское право:

Доступные форматы

4/5/2019 Velocity of Ecological System Trajectory

Velocity of Ecological System Trajectory

Ecosystems are Complex

Ecosystems are Complex

- difficult to model mathematically

Ecosystems are Complex

The Holy Grail of Ecology

forecasting change in time to prevent or

The Holy Grail of Ecology

Ecological Regime Shifts

how? loss of negative feedback(s) maintaining the

goal? predict in time to prevent

Methods for Detecting Regime Shifts

few explicitly handle noisy data

Regime Shifts Detection Methods

Principal Coordinates Analysis

Regime Shifts Detection Methods

Principal Coordinates Analysis

Some Multivariable Regime Shifts Detection

variance index (max eigenvalue of covariance matrix)

Principal Coordinates Analysis

x normal(μ = 100, σ = 50) for t ≥ 50

the linear speed of a system's trajectory in phase space

Step 1: Calculate s , 'distance travelled'

Step 2: Calculate v , t he linear speed of s

Velocity, Toy System

v increases with increasing e ect size

Smoothing Noisy Data Before Calculating v

[1] Price NP, JL Burnett. R Package for regularized differentiation.

v Identi es Potential Periodicities in

Empirical System: Paleodiatoms

Data published in T. Spanbauer et al. 2014. Prolonged Instability Prior to a

Empirical System: Paleodiatoms

Distance travelled s & Velocity v identify

Conclusions: v is Simple, Intuitive

Capable of handling noisy data

Is not sensitive to data quality issues common in ecology

Best when mean > variance

Identi es regime shifts known a priori

Compare to distance-based metrics (feedback?)

Compare to ordination techniques

trashbirdecology/regimeDetectionMeasures (calculate multiple regime

trashbirdecology/bbsRDM (application to spatial data)

natbprice/tvdiff (regularized differentiation)

Вам также может понравиться