You are on page 1of 26

GIS Modelling

GEOG 3731, 8731

Model Parsimony
Broad Definition: The balance between generalisation and precision.

• Over-fitted models may have too many parameters (Betas, knots or nodes) and
while they are precise, they are not generalisable.

• In other words we cannot predict ahead in time or space, nor can we extrapolate.

• Models that are too ‘general’, have no precision.

• For example, a model that says “all-humans live on the earth” is an accurate general
statement but is not precise.
Our model of the real world

Generalisation Balance Precision

Envelope Models
Step 1. Raster data extraction
2. Scatter plot of raster values, for each point location
(the point attributes do not play into this)

Slope Issue:

No presences

3. Bounding envelope created (max and min values from each raster grid)




10 18 Density
4. Raster output grid (the cells satisfying both ranges = 1)
(predictions in other areas, not just where sample points were)

Raster calculator example:

[Slope] >= 30.0 and [Slope] <= 50 and [Density] >= 10 and [Density] <= 18
Envelopes can be multi-dimensional (> =2)
- i.e. any number of raster grids
Bioclim - in R (
> For more information on Bioclim see .ppt:

Classification Trees
Regression Trees
 Parsimony is achieved by looking at the relationship between for
the standard error and the number of nodes in the tree.

 The tree is pruned back until the above relationship reaches and
Generalised Additive Models
GLM - Revision
GLM - Revsion
GAMs – graphically showing non-parametric responses
Penalized (pruned) knots stops over-fitting