Vegetation and climate interaction patterns in Kyrgyzstan: spatial discretization based on time series analysis

Abstract

Spatio-temporal variations of climate-vegetation interactions in Central Asia have been given a lot of attention recently. However some serious methodological drawbacks of previous studies prevented thorough assessment of such interactions. In order to avoid the limitations and improve the analysis we used spatially explicit time series of NDVI (normalized difference vegetation index), temperature and precipitation which were decomposed to seasonal and trend components on per-pixel basis using STL (seasonal decomposition of time series by loess). Trend and seasonal components of NDVI, precipitation and temperature were assessed pixelwise for temporal correlations with different lags to understand the patterns of their interaction in Kyrgyzstan and adjoining regions. Based on these results a coefficient of determination was calculated to understand the extent to which NDVI is conditioned by precipitation and temperature variations. The images with the lags of time series correlation minima and maxima for each pixel and coefficients of NDVI determination by temperature and precipitation were subjected to cluster analysis to identify interaction patterns over the study area. The approach used in this research differs from previous regional studies by implementation of seasonal decomposition and analyzing the full data without spatial or seasonal averaging within predetermined limits prior to the analysis. NDVI response to temperature and precipitation was assumed to be spatially variable in its sign, strength and lag, thus a separate model was developed for each pixel. The results were assessed with cluster analysis to identify spatial patterns of temporal interactions, decrease dimensionality and facilitate their comprehensiveness. The research resulted in 5 spatial clusters with different patterns of NDVI interaction with temperature and precipitation on intra- and interannual scales. The highest correlation scores between NDVI and temperature at the seasonal scale were found at 0-4 months lag and between NDVI and precipitation at 1-5 months lag. At high elevations of 3000-4000 m above sea level, both precipitation and temperature occurred to be facilitating factors for vegetation development, whereas temperature was rather a limiting factor at lower elevations of 200-1300 m a.s.l. We developed maps of the NDVI coefficient of determination by both temperature and precipitation. Only deserts and glaciers had low coefficients of determination (adjusted R2) on the seasonal scale (0.1-0.3), whereas areas with vegetation were greatly conditioned by temperature and precipitation (0.7-0.95). On the trend scale, dense vegetation and bare areas had low coefficient of determination (0.1-0.3), whereas areas with average vegetation cover were greatly controlled by the climatic factors (0.7-0.9).