Rainfall and outlier rain prediction with ARIMA and ANN models

Abstract

The precipitation level, a vital agro meteorological factor, holds immense significance in the decision-making process for the promotion of sustainable agriculture, preserving natural resources and improving quality of life. Rainfall prediction is necessary to explore crop environment relationship, water availability, soil erosion, floods and drought disasters. By leveraging Artificial Neural Networks (ANNs) and Autoregressive Integrated Moving Average (ARIMA) techniques, the proposed method utilizes ten input parameters and day-to-day meteorological observations to accurately forecast rainfall events at the Bengaluru station from 2013 to 2017. ANN method is also used to find an outlier during non-monsoon season. The suggested ARIMA model c(2,0,2) forecast daily rainfall 3 days in advance and c(1,0,0) anticipate monthly rainfall 5 months in prior. The model evaluation results are tabulated separately with MSE, RMSE, MAE and R2 values.