Spatio-temporal dependency of vegetation dynamics on climatic variables during 1982–2015 over India

The abnormally changing climate has a direct or indirect impact on the vegetation dynamics. The main objective here is to understand how the vegetation dynamics respond to climate change during the historical period, which could act as a clarion call for the future assessment as well. To accomplish the same, role of potential climatic variables viz., solar radiation, precipitation, maximum and minimum temperatures on changing normalized difference vegetation index (NDVI) is analysed. Vegetation dynamics largely depend on geographical conditions. This study is applied to whole India, comprising 1139 grid points over the period 1982–2015. This study made use of satellite derived longest available data record from Global Inventory Monitoring and Modelling System (GIMMS) and Climate data records (CRU) from climate research unit to study the spatio-temporal changes and assess the driving forces for change in vegetation dynamics. The trends of NDVI and all the climatic variables are analysed using modified Mann–Kendall approaches. Significant climatic trends along with NDVI trends are evidenced over India during period of analysis. Linear association between NDVI and each climatic variable are analysed based on Pearson correlation coefficient at monthly and seasonal scales. In this study, we have proposed spatio-temporal dependency modelling of vegetation dynamics (i.e., NDVI) on climatic variables. The stepwise linear regression analysis is performed for dependency modelling at grid level covering whole of India. Concurrent and Lag-1 climatic variables are used to consider temporal dependency, and for spatial dependency both average- and max-pooling-based climatic features are extracted. It is evident from the present study that consideration of spatio-temporal dependency could improve the modelling performance (of NDVI-climatic variable) to a great extent in more number of grid points. Spatio-temporal dependency modelling could achieve higher coefficient of determination (R2) values (≥0.5) in 314 grids, which was possible only in 157 grids in case of simple dependency modelling.