Comparison of Time Series Forecast Models for Rainfall and Drought Prediction

Abstract

Forecasting of rainfall is important to be prepared for future weather-related disasters. Rainfall data can be categorized as time series data because rainfall data can be recorded in chronological order. Time series forecast is used in fields like economics, environmental, and engineering predictions as a decision support factor. Due to the importance, many models and methodologies have been developed for time series forecasts according to the types of inputs, expected outcomes, and easy applicability. This research was conducted to identify the most appropriate time series forecast model for rainfall prediction. A regression type model and a neural network model were selected to identify which type of forecast model is more suitable for rainfall prediction. ARIMA model and Recurrent Neural Network model of Non-linear Auto-Regressive Moving Average were selected as the candidate prediction models for time series forecast and the models were developed for rainfall forecast. From the developed models, it was observed that the RNN models are suitable for long-term prediction of rainfall and drought with the availability of a higher number of past rainfall data while the ARIMA model is more suitable for prediction of rainfall where there is less past recorded rainfall data for a short-term forecast period.