Simulation of Meteorological Drought of Bankura District, West Bengal: Comparative Study between Exponential Smoothing and Machine Learning Procedures

Abstract

Simulation of drought is needed for proper planning and management of water resources. This study has been developed using the following five key points: a) primarily from rainfall Standard Precipitation Index (SPI), Percentage to Normal (PN), Decile based drought index (DI), Rainfall Anomaly Index (RAI), China Z Index (CZI), and Z-score are estimated on yearly basis (1901-2017), those indices are added and a new index standardized total drought (Sd) has been established. b) Considering Sd as the input parameter a comparative assessment has been made between 4 individual models (3 models from exponential smoothing, 1 model from machine learning) in simulation and prediction of drought status of next 18 time steps (years) in Bankura District and Winexpo model outperforms the other models as it obtains minimized Standard Error (SE), Random Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) and highest Correlation coefficient (R2) value. c) The cumulative drought proneness of the region is also assessed and it is found that the whole district will be drought-prone within the year 2100. This region is historically a drought prone region and agricultural shock is the common issue. In such a circumstance simulation of drought is a good attempt. Though a lot of models already developed in case of simulation of drought but still a perfect, continuous long term prediction is a big issue to solve. This study provides a comparative study between exponential smoothing and machine-learning procedures and also introduces a new combined index standardized total drought.