METEOROLOGICAL DROUGHT FORECASTING USING CLIMATE-INFORMED NON-STATIONARY STANDARDIZED PRECIPITATION INDEX

Abstract

Drought is one of the extreme events, which is not only affecting human society but also the natural environment and economy to a very severe extent. Therefore, drought prediction and forecasting are needed for proper planning and management. Quantification is the first step before forecasting drought characteristics. Current literature forecasted drought through stationary index-based quantification. However, recent studies show clear evidence of significant change in the precipitation. This change can not be captured by the stationary-based indices. Hence this study aims to develop a framework to forecast meteorological drought by preserving the temporal dynamics of drought characteristics through non-stationary SPI. In this study, gridded 0.25x0.25 daily precipitation data from 1951-2018 obtained from IMD is used to forecast the meteorological drought in Upper Bhima Region. A non-stationary standardized precipitation index is developed by using Bayesian Inference Techniques by considering various climate indices such as ENSO, IOD, MEI, GTA, LTA, and DTR as external covariates. These covariates are used to analyze the non-stationary behavior and nonlinear characteristics of precipitation which further helps in predicting extreme drought events. Rainfall variability is explained in a better way by the results of this non-stationary analysis than by stationary analysis and also the same in the case of drought characteristics. Also, the Forecasting of NSPI is done by using the Machine Learning models (i.e.Random Forest regressor and LSTM model). LSTM model shows higher accuracy in predicting NSPI values than Random Forest.