Geographic Information Systems and Remote Sensing modelling of tree species diversity in the woodlands of Zimbabwe

The continued loss of biological diversity has prompted managers and conservationists to find ways of monitoring ecosystems for the purposes of protecting and preventing further loss of biological diversity. Remote sensing has been suggested as a way of monitoring ecosystems because of its ability to provide spatial data which can be updated relatively faster and at a relatively cheaper than conventional methods. The relationship between tree species diversity and satellite derived indices is well documented. However, not much work has been done in mapping tree species using the obtained relationships, more so in savanna woodlands. Thus, the main objective of this thesis was to investigate whether and to what extent remote sensing can be used to measure tree species diversity in savanna woodlands. In addition the thesis also tested the extent to which linear and spatial regression can be used within a Geographic Information System to predict the spatial distribution of tree species diversity. We based our study on the hypotheses that 1) variance in the reflectance within remotely-sensed images is directly related with tree species diversity in savanna woodlands, and 2) that a remote sensing index of biomass is related to tree species diversity. The former is based on the Spectral Variation Hypothesis, which predicts that species richness can be estimated from habitat heterogeneity while the latter is based on the biomass-diversity or productivity-diversity hypothesis. The biomass-diversity hypothesis predicts that optimum biodiversity is found in ecosystems which have intermediate biomass, while low species diversity is characteristic of ecosystems which have low or high biomass. Specifically, we used regression analysis to test whether and in what way tree species diversity is related to the standard deviation of the Near Infrared (NIR) radiance (a measure of spectral variation) and tree biomass estimated via the Soil Adjusted Vegetation Index (SAVI) in three selected savanna woodlands of Zimbabwe. We also tested the wider applicability of the derived regression models by applying them in a study site away from where the models were developed. Our results showed that tree species diversity has a significant (p< 0.05) hump-shaped response to variations in the standard deviation of NIR radiance and SAVI. Furthermore, results show that the combination of the standard deviation of NIR and SAVI explained between 59% and 73% of the variance in tree species diversity in the study sites. Also, we found that our regression models can be used to spatially predict tree species diversity in landscapes with comparable physical characteristics. Overall, we conclude that remote sensing and GIS can be used to successfully estimate tree species diversity in savanna woodlands