Seasonal Arima-Based Forecasting of Monthly Rainfall in Kasese District, Uganda.

Abstract

This study develops and validates a Seasonal Autoregressive Integrated Moving Average (SARIMA) model to forecast monthly rainfall in Kasese District, Uganda, using historical records (1960–2023, number of months, n = 768 months) from the Uganda National Meteorological Authority (UNMA). Exploratory analysis revealed two major rainy seasons (March– May, September–November) and drier months in January–February and June–July. Stationarity tests (Augmented Dickey–Fuller, KPSS) and autocorrelation diagnostics confirmed the suitability of a SARIMA approach. The optimal model, SARIMA (3, 1, 1) (1, 0, 0)[12], selected via Akaike (AIC = 7948.98) and Bayesian Information Criteria (BIC = 7976.55), achieved a mean absolute error (MAE) of 42.60 mm, mean absolute scaled error (MASE) of 0.84, and minimal residual autocorrelation (ACF1 = -0.010). Residual diagnostics, including Ljung– Box and Shapiro–Wilk tests, confirmed white-noise behaviour and normality. Forecasts for 2024–2026 remained within 95% prediction intervals. The results demonstrate that SARIMA models can provide reliable short-term rainfall forecasts, supporting agricultural planning, water resource management, and disaster risk reduction in Kasese District. All analyses were conducted using R statistical software with the forecast package.