Assessing the interrelation between NDVI and climate dependent variables by using granger causality test and vector auto-regressive neural network model

Abstract

Changes in ecosystem structure and function can be revealed by examining the prevailing patterns in vegetation growth and the forces that shape those patterns. The mechanism of ecosystem behaviour may be better understood if the trend of vegetation change and its sensitivity to climatic variation are well understood. The interaction of vegetation and climatic factors (it's driving variables) is non-linear in behaviour and affected by time lag and time accumulation. Jharkhand has a typical plateau in eastern India, having a mixed climate (arid and semi-arid), taken as the study area, and the spatiotemporal distributions of the normalized difference vegetation index (NDVI) were explored with interaction driving factors. This study investigated the time-lag and time-accumulation effects of the NDVI response to climate factors Evapotranspiration (ET), Land Surface Temperature (LST), Potential Evapotranspiration (PET), Precipitation (PREC), and Soil Moisture (SM) and identified the primary controlling factors that affect the vegetation dynamics. The observations indicate that the correlation between NDVI and summer LST (- 0.838) was discovered to be greater than the correlation of NDVI with SM (r = 0.90) and PREC (r = 0.751), showing NDVI as more sensitive to LST when comparing to SM, and PREC, while PET exhibits the significant positive correlation (r = −0.751) with the NDVI in autumn during the studied duration. Higher NDVI values were seen during the monsoon (0.54 ± 0.12), which is correlated with a decrease in the monsoon LST (25.8 ± 0.20), followed by the winter (0.47 ± 0.13), summer (0.33 ± 0.18), and fall (0.37 ± 0.04). Vegetation growth is influenced by both the time-lag and time-accumulation effects of temperature and the time-accumulation impact of precipitation. Regarding the climate-vegetation response mechanism, the application of the Granger Causality (GC) Test and GC-based Vector Auto-Regressive Neural Network (VARNN) Model test reveals that the 0–2 month optimum time lag effect is prevalent in the study area. In addition, the LST and SM have a more prominent stimulating influence on plant growth in the study region than precipitation. The above findings highlight the need to effectively monitor vegetation dynamics under environmental changes by considering the temporal impacts of vegetation response to climate when the current climate models research vegetation-climate interactions.