Validating GIS Models 

1. A habitat model - factors
        1. Primary habitat:
            Oak forest on well drained soils

        2. Secondary habitat:
            Oak forest on poorly drained soils, and within 300m from the primary habitat

        3. Non-habitat      
           Oak forest on poorly drained soils and 300m away from the primary habitat

2. Data:
        1. Soil, well/poor drained, source:
        2. vegetation, source:

3. GIS analysis:
        Overlay:
        Oak + well drained, oak + poorly drained

        Proximity:
        300m buffer for secondary habitat and non-habitat

4. Model validation:
       Point data collected in the field: ideally more than 30 observations

       Variables: density of the oak trees, density of the wildlife

        Statistics:
        H1: oak density = normal; K-S test for normality
        Assuming to accept the null hypothesis in this case
        H2: oak in the primary habitat = oak in secondary habitat; t-test 
        accept or reject the null hypothesis
        H3: oak in the secondary habitat = oak in the non-habitat; t-test 
        accept or reject the null hypothesis

        H4: wildlife density = normal; K-S test for normality
        Assuming to reject the null in this case
        H5: wildlife in the primary habitat jay = wildlife in the secondary habitat; Mann-Whitney test
        accept or reject the null
        H6: wildlife in the secondary habitat = wildlife in non-habitat; Mann-Whitney test
        accept or reject the null

        Conclusion: Is the 300m buffer appropriate to define primary, secondary, and non habitats?

5. Error matrix:
                                                                   Mapped category
        True Category                       Primary        Secondary        Total          
         Oak >=50%                            15                5                    20
         Oak< 50                                    3              52                    55
          Total                                       18              57                    75           
Accuracy of the oak forest map: 89%

Overall accuracy  = 160/200 = 80%
Commission error for forest = 10/75 = 13.3%
Omission error for forest = 15/80 = 19.8%

...Back to Ling Bian top page.