Climate Forecasts Using the Hybrid SARIMA-LSTM Onacc Model: Optimizing Predictions for the Bimodal Humid Forest Zone of Cameroon

Abstract

Cameroon is facing increasing climate variability, with direct consequences on agriculture, ecosystems, and rural development. The bimodal humid forest zone-covering the Centre, South, and East regions-is particularly vulnerable due to its specific climatic conditions, which make it both agriculturally productive and highly sensitive to fluctuations in precipitation and temperature. This study aims to calibrate and apply the SARIMA-LSTM ONACC hybrid model, developed under the auspices of the National Observatory on Climate Change (ONACC). Our goal is to optimize climate forecasting for this critical agroecological zone. Historical temperature and precipitation data (1980-2022) from selected meteorological stations were used to train and test the model. The methodology combines the Seasonal AutoRegressive Integrated Moving Average (SARIMA) model to capture linear and seasonal components, with Long Short-Term Memory (LSTM) networks to model nonlinear dependencies and residual dynamics. Preprocessing steps include normalization, seasonal decomposition, and model validation. Model performance was evaluated using RMSE, MAE, and NSE indicators. Results show that the SARIMA-LSTM ONACC model reliably reproduces the interannual variability of the region. These results provide a solid foundation for informed agricultural planning, water resource management, and the development of climate risk mitigation strategies in southern Cameroon.