Optimizing Weather Forecasting Accuracy via Radial Basis Function Networks, Convolutional Neural Networks and Convolutional Neural Networks

Weather forecasting is crucial for various sectors, including agriculture, disaster management, and infrastructure planning. However, traditional prediction models such as Numerical Weather Prediction (NWP) and ARIMA often struggle with long-term accuracy due to the nonlinear and chaotic nature of atmospheric phenomena. The primary objective of this study was to develop and compare the performance of Radial Basis Function Networks (RBFNs), Convolutional Neural Networks (CNNs), and Long Short-Term Memory Networks (LSTMs) in predicting key weather variables such as temperature, humidity, sea-level pressure, windspeed, and rainfall. Historical meteorological data from the Kenya Meteorological Department, spanning a decade, was used to train and evaluate the models. The methodology employed included data preprocessing, model training with a 70-15-15 split for training, validation, and testing, and performance evaluation using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and training time as key metrics. The results revealed that RBFNs consistently outperformed CNNs and LSTMs, particularly for stable variables like temperature and sea-level pressure, with lower RMSE and faster training times. CNNs and LSTMs, while better at capturing complex temporal patterns, struggled with chaotic variables such as rainfall and windspeed, exhibiting higher error rates and longer training times. In conclusion, RBFNs were found to be the most efficient and accurate model for real-time weather forecasting in resource-constrained environments, whereas CNNs and LSTMs may require additional tuning or hybrid approaches to improve their forecasting of more dynamic weather variables.