<p><span>Bias Correction of Hybrid SARIMA–LSTM Rainfall Forecasts for Climate-Risk Applications in Coastal Cameroon</span></p>

Bias-correction techniques combine observational and modelled climate data over a common reference period to reduce systematic errors and improve the credibility of climate information. In tropical environments, persistent biases in simulated rainfall remain a major limitation for climate-informed planning, early warning, and adaptation. This study assesses the performance of a hybrid SARIMA–LSTM (SARIMA (Seasonal Autoregressive Integrated Moving Average) LSTM (Long Short-Term Memory)) regional climate learning model developed by the National Observatory on Climate Change (ONACC) for monthly rainfall forecasting over Cameroon’s coastal zone. Simulated rainfall is evaluated against in-situ observations from eight rain-gauge stations (Yabassi, Edéa, Nkongsamba, Douala, Kribi, Mamfe, Campo, and Ekona) over the 1950–2022 reference period. To reduce systematic biases, model outputs are corrected using two complementary statistical approaches: (i) the Delta-change (anomaly) method and (ii) Quantile Mapping. Results reveal pronounced station- and season-dependent biases in uncorrected simulations, with the largest absolute deviations occurring during peak rainy months at some locations and relatively larger proportional errors during dry and transition seasons at others. Both bias-correction methods substantially improve agreement with observations, but with distinct characteristics: the Delta method provides stable monthly-scale adjustments, whereas Quantile Mapping better preserves rainfall variability and distributional extremes. Bias-corrected projections indicate sustained strong seasonality and suggest shifts in rainfall regimes across the coastal zone, with changes generally more pronounced under higher radiative forcing. These findings demonstrate the importance of method-aware bias correction when using hybrid statistical–machine learning models for rainfall projections and provide a robust basis for hydrological assessment, climate risk management, and adaptation planning in data-scarce tropical regions.