Assessing Meteorological Drought and Detecting LULC Dynamics at a Regional Scale Using SPI, NDVI, and Random Forest Methods

The current research explores the relationship between meteorological drought and land-cover changes locally via the normalized difference vegetation index (NDVI) and standard precipitation index (SPI). The historical time-series dataset of rainfall from 1981 to 2018 was used to compute the SPI, whereas the NDVI and land use/land cover (LULC) for the year 2016–2018 were calculated from the Sentinel-2 dataset of the studied region. The positive ( r = 0.82, r = 0.79, and p = 0.001) and negative ( r = 0.51 and p = 0.087) correlations were observed between NDVI and SPI data during 2016–2018. The 1 month scale of the SPI was positively correlated with NDVI. It was noticed that the maximum and minimum correlations occurred during the starting and end of the growing period, respectively. The multiple regression models were developed based on the correlation coefficients to predict the NDVI and investigate the relationship between the NDVI and SPI. The models have predicted accuracy ( R 2 ) of 0.68, 0.63, and 0.26 for the normal (2016), moderate (2017), and severe drought (2018) years, respectively. The drastic changes in an LULC were noticed during the regular and severe drought years.