Agricultural Drought Index Selection using Probability Distribution: Statistical and Linear Regression Approach

Understanding drought and its trend analysis is becoming more challenging as intensified changes in climate lead to negative impacts on agriculture. A key challenge is to formulate a drought index for developing a drought forecasting model to identify drought severity. This paper focuses on assessing the efficiency of remote sensing and GIS techniques for monitoring the spatiotemporal extent of agricultural drought using the Standardized Precipitation Index (SPI). SPI is the major index to assess the agriculture drought and it is further challenging to select the best scale of SPI calculation on various time scales of 1-, 3-, 6-, 9-, and 12- months scale. This research aims to attempt to identify the best SPI scale among 1-, 3-, 6-, 9-, and 12-month scales. A database of SPI is generated (period 1991–2022) from daily gridded precipitation data from the India Meteorological Department Pune for the Jaisalmer region in Rajasthan state of India. Statistical method- Kolmogorov-Smirnov (KS) and Anderson-Darling (AD) tests are used to investigate three fitting distribution models (Gamma, Weibull, and Normal Distribution Function) for cumulative precipitation data 1-, 3-, 6-, 9-, 12- months scale. In order to pick the optimal time scale, the gamma distribution function is evaluated using the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) as goodness-of-fit criteria with AIC 1928.68 and BIC 1934.76. In contrast, the machine learning approach of a linear regression model is applied to get the best SPI scale out of 1-, 3-, 6-, 9-, and 12-month scales. In both the statistical method (KS & AD test ) and machine learning approach of linear regression, SPI at a 3-month time scale is selected with a significant mentioned result. The SPI -3 scale validation is done using linear correlation with the daily gridded precipitation dataset, in which SPI at a 3-month time scale is selected as a stronger correlation (0.65) with precipitation i.e. good enough for dry spell Rajasthan state.