Drought Prediction Using Recurrent Neural Networks and Long Short-Term Memory Model

Abstract

Listen

Translate article

AbstractPredicting drought is the process of forecasting and classifying the dryness of the weather. The availability of long term meteorological data in the form of time-series data allows for various multivariate time-series forecasting techniques to be applied on it. Datasets consist of various meteorological parameters such as ‘Precipitation’ and ‘Surface Pressure’, and a target variable which specifies the degree of dryness. Various machine learning techniques such as recurrent neural networks (RNN) and long short-term memory (LSTM) can be used to forecast the value of score for various lead times. RNN is used to forecast for shorter lead times of 1 to 2 weeks, whereas LSTM is used for forecasting for lead times of 8 to 12 weeks. The results of the prediction can prove to be helpful in many ways. Prior knowledge of weather conditions allows us to minimize the catastrophic effects of extreme weather conditions given appropriate actions are taken for mitigation and prevention. In order to make an action plan for mitigation, it becomes necessary to predict weather conditions as well as the severity of it. Mitigation strategies can include alternate crop recommendations to farmers and special aid to regions which are at most risk.KeywordsRecurrent neural network (RNN)Long short-term memory (LSTM)Time-series data