Enhancing Coastal Hazard Projections in Singapore: Application of Bias Correction Techniques for Monsoon and Storm Surge Modeling

Accurate modelling of monsoon and storm surge heights is crucial for effective coastal and climate resilience management in Singapore. Despite advances in climate modeling and hydrodynamic simulations, systematic biases remain a challenge, often resulting in under- or overestimation of extreme events and coastal hazards. Bias correction is crucial to improve the accuracy of projections. This study explores the application of several bias-adjustment techniques—mean bias correction, variance scaling, and quantile mapping—to improve the accuracy of monsoon and storm surge projections along Singapore’s coastlines. Mean bias correction adjusts the model output to match the observed mean better, whereas variance scaling further refines the distribution by adjusting the model output variance to match the observed variance. Quantile mapping provides a comprehensive approach by modifying the entire distribution of model outputs to match the observed distribution, creating a mapping between the model's Cumulative Distribution Function (CDF) and the observed CDF, which improves the simulation of both median and extreme values. In this study, outputs from the Delft3D FM hydrodynamic model, driven by atmospheric forcings from the Singapore Variable Resolution – Regional Climate Model (SINGV-RCM), which employs six global climate models (GCMs) from the CMIP6 climate projections, were compared with observed data at multiple tide gauge stations in the region. We applied these bias-correction methods individually to historical simulations (1984-2014) and in combination to project future monsoon and storm surge heights (2015-2100). The corrected projections are evaluated through statistical metrics and comparison with historical observations, demonstrating significant improvements in model accuracy and reliability. Our results highlight that quantile mapping provides the most comprehensive bias correction, capturing the full distribution of extreme events, while mean bias correction and variance scaling offer simpler, computationally efficient alternatives.