Empirical And Statistical Validation Of Rainfall Asymmetry And Phase Shifts In Ghana’s Western Region

Abstract

Rainfall in Ghana’s Western Region remains bimodal, yet recent seasons display distortions in timing and intensity that complicate agricultural and water-resource planning. Monthly data from 12 stations (2017–2023) were analysed via a Gamma-linked generalized additive model (GAM), seasonal-trend decomposition (STL), and PELT change-point detection, with independent validation from corrected empirical sinusoidal and accumulation-based amplitude metrics. The GAM revealed pronounced nonlinearity in the seasonal cycle (edf ≈ 4.6; p < 0.001) but no monotonic trend once seasonality was accounted for (p ≈ 0.61). Predictive diagnostics (pseudo-R² = 0.50; deviance explained = 25.5%; RMSE ≈ 48.5 mm; MAE ≈ 38.1 mm) indicated a satisfactory model fit with residuals consistent with the assumed distribution. The empirical sinusoid, incorporating a phase-correction term, successfully reproduced the asymmetric bimodality (major peak May–June; minor peak October–November) and approximated the observed mean (≈ 310 mm). Change-point analysis revealed that the episodic mean and variance realignments—most notably from 2019–2021—were consistent with short-lived regime shifts rather than persistent drift. These findings indicate that intra-annual phase realignment and amplitude modulation, rather than long-term wetting or drying, dominate rainfall variability from 2017–2023. The incorporation of phase-adaptive empirical diagnostics into agro-climate early warning systems is recommended to enhance onset and cessation guidance. This study uniquely integrates statistical modelling with an empirically corrected sinusoidal framework, providing the first phase-adaptive validation of Ghana’s bimodal rainfall regime using combined GAM, SHAP, and change-point techniques. The integration of empirical and advanced statistical diagnostics offers a replicable blueprint for regional rainfall assessment and early-warning system enhancement.