A Hybrid BiLSTM-GRU-PINN Deep Learning Framework for Modeling Regional Climate and Extreme Weather Forecasting in the Semi-Arid of Ghana
Year: 2026
Type: article
Abstract: Accurate climate and weather forecasting remains a critical challenge in semi-arid regions like northern Ghana, where climate variability increasingly threatens agriculture, water security, and rural livelihoods.This study proposes a novel hybrid deep learning framework, BiLSTM-GRU-PINN, that combines Bidirectional Long Short-Term Memory (BiLSTM), Gated Recurrent Units (GRU), and Physics-Informed Neural Networks (PINNs) for enhanced regional climate prediction.The framework integrates the temporal modeling capabilities of BiLSTM and GRU with the physical consistency of PINNs, incorporating partial differential equations such as the heat equation directly into the model's loss function.The system is trained on multi-source historical and observational data, including records from GMet, CMIP6, and NASA EarthData, and is evaluated using metrics such as RMSE, MAE, and prediction accuracy.The rresults demonstrate that the BiLSTM-GRU-PINN model significantly outperforms conventional models, including standalone LSTM, GRU, PINN, and statistical or physics-based approaches in both temperature and rainfall prediction.The bestperforming configuration achieved RMSE values of 0.225 for temperature and 0.140 for rainfall, with respective prediction accuracies of 87.78% and 69.76%.Visualization and decomposition analyses show that the model effectively captures long-term trends and seasonal patterns in temperature, while highlighting challenges in predicting the stochastic nature of rainfall.These findings underscore the value of integrating data-driven and physics-based approaches for climate forecasting in data-scarce regions.The proposed framework offers a robust, interpretable, and scalable solution for supporting early warning systems, policy-making, and climate adaptation strategies across semi-arid and similarly vulnerable contexts in Africa..
Author: Toufic Seini
Institution: University for Development Studies
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7225075505
