Detecting spatial and temporal patterns in NDVI time series using histograms

Abstract

The aim of this study was to analyse bimodal histogram patterns of monthly National Oceanic and Atmospheric Administration (NOAA) advanced very high resolution radiometer (AVHRR) normalized difference vegetation index (NDVI) global area coverage (GAC) data and their relation to vegetation dynamics and climatic conditions for the period 1982-1991 in Argentina. The proposed method was to split up bimodal histograms by the median criterion and to study each mode as a separate unimodal frequency distribution. Modes were analysed based on their histogram shape and statistical parameters, geographical distribution and dynamics, and climatic significance. For the latter, a multinomial statistical analysis was used. The split-up criterion yielded coherent results. Histogram shapes and statistical parameters changed according to season. For geographical dynamics, 84% of pixels remained in the same mode through the seasons, and 16% shifted temporarily to the other mode. Changes from low-NDVI mode to high-NDVI mode were caused by an improvement in water supply, rainfall or irrigation, and higher temperatures. Changes in the opposite direction were due to a reduction in vegetation cover produced by drought, harvest, or autumn effects. The low-NDVI mode was strongly related to the arid zone with 74.6% probability (α = 0.05), and the high-NDVI mode was related to humid (58.8%) and semiarid zones (38.4%). This contribution helps explain the dynamics of vegetation cover along the latitudinal range from 22° to 55°S, for nine growing cycles, with a simple methodology. Improving the knowledge of multimodal histograms may allow a better understanding of difficult classification results.