Regionalization of Calibrated Curve Number Using Multivariate Regression for Flood Simulation in the Moluccas Islands
Abstract
Accurate estimation of calibrated Curve Number (CN) is essential for reliable flood simulation, particularly in island watersheds characterized by short hydrological response times, high rainfall variability, and heterogeneous land cover. This study regionalizes calibrated CN values in the Moluccas Islands, Indonesia, using multivariate regression for flood simulation. Seven watersheds in the Southern Halmahera and Ambon–Seram River Basin Regions were analyzed using hourly flood events during 2022–2025. Calibrated CN values derived from accepted HEC-HMS event-based simulations were related to woodland, grassland, cropland, unused land, construction land, rainfall amount, and rainfall duration. The linear and non-linear multivariate regression forms were compared at various watersheds, including subregional and regional scales, using Nash-Sutcliffe Efficiency (NSE), Pearson's correlation coefficient ( ), and Root Mean Square Error (RMSE). The results indicate that rainfall and rainfall duration were the most consistent predictors across nearly all models, confirming the dominant role of event-scale rainfall in controlling calibrated CN behavior. At the watershed scale, linear regression yielded NSE values of 0.776–0.996, values of 0.835–0.951, and RMSE values of 3.369–11.249, whereas non-linear regression yielded NSE values of 0.756–0.996, values of 0.844–0.964, and RMSE values of 2.930–12.574. At the subregional scale, non-linear regression slightly outperformed linear regression in Southern Halmahera, whereas linear regression remained superior in Ambon–Seram. At the regional scale, both forms exhibited nearly identical performance, with RMSE of 4.320 for the linear model and 4.332 for the non-linear model. These findings demonstrate that calibrated CN regionalization is feasible for data-limited island watersheds. At the same time, the relative suitability of linear and non-linear regression depends on hydrological heterogeneity and spatial scale.