Localized Weather Prediction Using Kolmogorov-Arnold Network-Based Models and Deep RNNs

Weather forecasting is crucial for managing risks and economic planning, particularly in tropical Africa, where extreme events severely impact livelihoods. Yet, existing forecasting methods often struggle with the region's complex, non-linear weather patterns. This study benchmarks deep recurrent neural networks such as textttLSTM, GRU, BiLSTM, BiGRU, and Kolmogorov-Arnold-based models (textttKAN and textttTKAN) for daily forecasting of temperature, precipitation, and pressure in two tropical cities: Abidjan, Cote d'Ivoire (Ivory Coast) and Kigali (Rwanda). We further introduce two customized variants of textttTKAN that replace its original textttSiLU activation function with textttGeLU and textttMiSH, respectively. Using station-level meteorological data spanning from 2010 to 2024, we evaluate all the models on standard regression metrics. textttKAN achieves temperature prediction (R2=0.9986 in Abidjan, 0.9998 in Kigali, textttMSE < 0.0014~^circ C 2), while textttTKAN variants minimize absolute errors for precipitation forecasting in low-rainfall regimes. The customized textttTKAN models demonstrate improvements over the standard textttTKAN across both datasets. Classical textttRNNs remain highly competitive for atmospheric pressure (R2 ≈ 0.83-0.86), outperforming textttKAN-based models in this task. These results highlight the potential of spline-based neural architectures for efficient and data-efficient forecasting.