Advanced weighing of GIS layers

Issue: Modeling habitat of Red Squirrel in Mt. Graham.

Factors:
        a. Topography:
            elevation
            slope
            aspect (E-W)
            aspect (N-S)

        b. Vegetation:
            land cover 
            canopy closure 
            food productivity 
            tree dbh
            distance to openness (road network and canopy closure)
 
Data: DEM, vegetation cover, roads
           200 presence cells (observed)
           200 absence cells (randomly located)

Logistic Regression:
           The dependent variable is dichotomous
           The independent variables can be numerical or categorical

           Dependent variable (binary): presence or absence
           Independent variables(14):
                1.elevation 2.slope 3.aspect N-S 4.aspect E-W
                5.distance to opening
                6-8.food productivity, 3 categories, presence or absence for each variable
                9-11.canopy closure, 3 categories, presence or absence for each variable
                12-14. dbh, 3 categories, presence or absence for each variable

Statistical Testing:
            t-test for continuous independent variables:
                mean of the indep var at presence sites = mean of absence sites
            C²-test for categorical independent variables:
                distribution of the indep var at presence sites = absence
                aspect N-S is not significant

Data Partition:
            Data partition for model development vs. validation

            75% of sample are used to develop model, 150 presence sites, 150 absence sites
            25% of sample are used for model validation, 50 presence sites, 50 absence sites

The Logistic Model:
                Y = 0.002ele - 0.228slope + 0.685canopy1 + 0.443canopy2
                      + 0.481canopy3 + 0.009aspectE-W
                                      1
                P(Y) = --------------, the probability of hawks presence
                               1 + exp^(-Y)

Accuracy Assessment:
        decide a cut-off value for P, convention: 0.5
        convert logistic regression output (P) into suitable/unsuitable
        cell value < cut-off = unsuitable, cell value >= cut-off = suitable

        Error Matrix for 150 presence sites and 150 absence sites
        82% correct for presence sites, 76% correct for absence sites

Model Validation:
        Error Matrix for the 50 presence and 50 absence sites
        74% correct for presence sites, 68% for absence sites

GIS Overlay:
       
 
 
 
 

...Back to Ling Bian top page.