Spatio-temporal statistics for exploratory NDVI image analysis

Abstract

In this paper, spatio-temporal changes of vegetation in Jilin are explored with MODIS NDVI images time series from 2000 to 2006. MODIS NDVI images time series are organized into a spatio-temporal data cube. We build a linear regression model of NDVI time series at each pixel of the MODIS image. The slope of the linear regression model indicates the change trend of vegetation. Using these change trends data, we further develop three advanced spatio-temporal statistics of regional statistics, the center of gravity, and spatial autocorrelation for exploring spatiotemporal changes of Jilin vegetation. Three method results have commonly showed that Jilin vegetation change is generally stationary and slightly increasing in space. Meanwhile, Jilin vegetation changes are spatially heterogeneous in local regions. In particular, vegetation change is stationary in eastern Jilin, and is increasing in western Jilin. Vegetation change is decreasing in southwestern Jilin, northwestern Jilin, riverside and lakeside regions, and Changchun downtown. Positive spatial autocorrelation of Jilin vegetation changes is significant. Besides, the regions with a similar vegetation change trend are spatially clustered. Three given methods of exploratory NDVI image analysis have emphasized separable spatio-temporal interaction modeling, that is, a twophrase combination of modeling temporal changes firstly and modeling spatial changes secondly.