Application of Geographically Weighted Regression Analysis to Assess Human-Induced Land Degradation in a Dry Region of Kazakhstan

SUMMARY The primary objective of this study was to assess a human-induced dryland degradation in the cachment basin of the Balkhash Lake in the Middle Kazakhstan based on time series of rainfall data and normalized difference vegetation index (NDVI) for the period 1985-2000. We developed a method to remove the climatic signal from the change in vegetation activity over the study period. By applying a local regression technique known as geographically weighted regression (GWR), relationship between spatial patterns of the growing season NDVI and the growing season rainfall were estimated for every pixel and every year. In geographically weighted regression, the regression parameters are estimated using an approach in which the contribution of an observational site to the analysis is weighted in accordance to its spatial proximity to the specific location under consideration. The weighting is a function of location and it declines the further the observation is from the location for which predictions and parameter estimates are required. The regression models identified a strong dependence of spatial patterns of NDVI on that of precipitation parameter. The relationship between NDVI and the explanatory variable was found to vary spatially and temporally. At local scales, the regression models indicate that over 90% of spatial variations in NDVI is accounted for by the climatic predictor. Deviations in NDVI from this relationship, expressed in regression residuals, were calculated for each year of the study period 1985-2000. Residuals, laying out of the " Standard Error of the Estimate " are regarded as outliers and interpreted as human-induced. The results of the modelling were validated by comparison of the remote sensing data of high spatial resolution (Landsat TM and ETM) and the data from field trips to degrading areas.