Drought prediction based on SPI and SPEI with varying timescales using LSTM recurrent neural network

Over years, natural calamities like drought have taken a huge toll on human life and resources. As the prediction methods increase, the effects of natural calamities can be reduced to an extent by preplanning and providing warnings to the people. Metrological drought indices like standardized precipitation index and standardized precipitation evapotranspiration index are used to identify drought and its severity level. By forecasting these indices, the occurrences of drought are predicted using the prediction models which help the society to take preventive measures due to the effect of drought. Many research works on prediction majorly focused on statistical methods such as Holt-Winters and ARIMA, but these methods lack accuracy to provide long-term forecasts. However, with advances in the area of machine learning especially artificial neural networks and deep neural networks, there seems to be a method to predict drought in the long term with a good accuracy. Long short-term memory is used in recurrent neural network to predict the drought indices which handle the real-time nonlinear data well and good that can help authorities better prepare and mitigate natural disasters. In this paper, we compare the 1-, 6- and 12-month prediction of the ARIMA statistical model with LSTM using multivariate input in hopes of bettering said performance.