Comparative Analysis Of Arima, Sarima And Ets Models For Forecasting Monthly Precipitation In Manipur, India Using High-Resolution Satellite Data

Abstract

This research study examines the Autoregressive Integrated Moving Average (ARIMA), Seasonal ARIMA (SARIMA), and Exponential Smoothing State Space Model (ETS) models for the monthly precipitation forecast over Manipur, India with the high resolution satellite data, PERSIANN-PDIR-Now data set of the time span between 2001 and 2023. Spatially consistent preprocessing was carried out for the data by using ArcGIS. Analysis of R for the data sets to analyze time series. Additive Decomposition, which is of type Decomposition, is conducted as seasonality changed without correlation to the trend magnitude. Three models have been developed and validated with different metrics such as RMSE, MAE, and MASE. The best model is an ARIMA(5,0,1) that captures short-term time dependencies, and hence, indicates moderate accuracy but less ability to deal with seasonal effects. The SARIMA(1,0,0)(1,1,0)[12] model is more reliable because it captures both seasonal and non-seasonal components with minimal bias. ETS (A,N,A) shows good positive additive seasonality but with mild residual autocorrelation, which indicates that it lacks proper freedom to deal with irregular fluctuations. There is a steady increase of monsoonal intensity over the years, according to the trend analysis. Seasonal components present predictable peaks in the case of monsoon. The random component presents anomalies, such as extreme rainfall in 2018, which presents global climatic events like El Niño. SARIMA achieves a good balance in terms of being accurate yet uncomplicated for places that exhibit the pronounced seasonality. The implications of this paper are climate resilience and support better water resource management, agricultural planning, and disaster preparedness in the state of Manipur.