Performance Improvement of a Multiple Linear Regression-Based Storm Surge Height Prediction Model Using Data Resampling Techniques

Abstract

Storm surges present a major hazard to coastal areas worldwide, a risk that is further amplified by ongoing sea-level rise associated with climate warming. The purpose of this study is to enhance the prediction performance of a storm surge height model by incorporating data resampling techniques into a multiple linear regression framework. Typhoon-related predictors, such as location and intensity-related parameters, were used to estimate observed storm surge heights at eleven tide gauge stations in southeastern Korea. To address the data imbalance inherent in storm surge height distributions, we applied combinations of over- and under-sampling methods across various threshold levels and evaluated them using four statistical metrics: root mean square error (RMSE), mean absolute error (MAE), mean squared error (MSE), and the coefficient of determination (R²). The results demonstrate that both threshold selection and sampling configuration significantly influence model accuracy. In particular, station-specific sampling strategies improved R² values by up to 0.46, even without modifying the regression model itself, underscoring the effectiveness of data-level balancing. These findings highlight that adaptive resampling strategies—tailored to local surge characteristics and data distribution—can serve as a powerful tool for improving regression-based coastal hazard prediction models.