Evaluating Trends in Spatial Relationship between NOAA/AVHRR-NDVI and Rainfall as Computed by Geographically Weighted Regression:
A Case Study from a Dry Region in the Middle Kazakhstan

Pavel A. Propastin ^a,b, Martin Kappas ^a and Nadia R. Muratova ^b

^a Department of Geography, Georg-August University, Göttingen, Germany
e-mails: ppavel@gmx.de , mkappas@uni-goettingen.de

^b Laboratory of Remote Sensing and Image Analysis, Kazakh Academy of Science, Almaty, Kazakhstan
e-mail: nmuratova@hotmail.com
 

Abstract

The spatial relationship between vegetation patterns and rainfall as well as its trend over the period 1985-2000 in the shrubland, grassland, and cropland of the Middle Kazakhstan was investigated with Normalized Difference Vegetation Index (NDVI) images (1985-2000) derived from the Advanced Very High Resolution Radiometer (AVHRR), and rainfall data from weather stations. The growing season relationship was examined using a local regression technique known as geographically weighted regression (GWR). Regression models for each pixel and every analysis year (1985-2000) were calculated using this approach. Both the strength of relationship and the regression parameters showed high spatial and temporal non-stationarity. Spatial and temporal drifts of regression parameters and drifts of correlation coefficient were estimated and mapped. There are notable associations between patterns in land cover types and patterns in intercept and slope parameters in the study region. Residuals from the regressions for each pixel and every year were computed. Trend in residual values for each pixel over 16-year period was used to determine a temporal change of conditions of the vegetation cover through the time: pixels with a negative slope are considered to represent ground areas with decreasing amount of vegetation. Four types of trend behavior were determined and analyzed.

Keywords: NDVI, vegetation condition time-trend, geographically weighted regression (GWR) modeling