ANNUAL RAINFALL FORECASTING USING ARIMA MODEL

Abstract

                            Rainfall forecasting is important for food, water management and prevention from flood. Forecast rainfall by using variables such as temperature, wind and humidity. We used the Auto-regressive Integrated Moving Average model (ARIMA) to generate rainfall projections for the selected study area of Tamil Nadu. Rainfall data of Tamil Nadu spanning 115 years from 1901 to 2015 were gathered and statistically evaluated.

                            The Box-Jenkins Autoregressive Integrated Moving Average (ARIMA) approach was used for model identification and diagnostic evaluation for predicting the annual rainfall of the research region. Python programming language was used to identify the appropriate ARIMA (????, ????, ????) ∗ (????, ????, ????) model that fits the rainfall records. The dataset's has been assessed using the Augmented Dickey-Fuller test. The optimal model for forecasting the following decade's rainfall was chosen according to the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). The model's was assessed using root mean square errors (RMSE) and mean squared error (MSE).

                                   Analysis of the rainfall data revealed that ARIMA was the superior model for the annual data with a stationary R-squared value of 0.9978. The optimal ARIMA models were identified, and projections were made for the average annual precipitation for the years 2016 to 2025. The projected rainfall values generated by the ARIMA model were satisfactory and confirmed with normal rainfall data. These findings underscore the reliability of the ARIMA model in climatic studies, particularly in forecasting rainfall patterns. Future research may expand on this model by incorporating additional variables such as temperature, wind and humidity to enhance predictive accuracy.