Detail Abstract

Flood susceptibility mapping and prediction for extreme flood in Vu Gia Thu Bon River Basin

Abstract

Flooding in humid regions have been increasing recently and can have devastating effects. Therefore, predicting flood-prone areas is essential for proactive disaster management. However, with limited data due to the scarcity of flood observation stations and the lack of monitoring systems, such predictions through physical hydrological models are challenging to accurately obtain. This study aims to evaluate two new boosting machine learning models, namely, LightGBM and CatBoost, for the first time in the prediction of flooding susceptibility (FS) in Humid Region (VGTB River basin in Vietnam). Then, the performance of these models is compared with that of the common random forest (RF) method, and also compared with the rainfll-runoff modeing. Fourteen independent factors that influence FFS in the study area, including elevation, slope, plan curvature, aspect, vertical and horizontal distance from main streams, hillshade, flow accumulation, the topographic wetness index (TWI), rainfall, land use (LU), lithology, the normalized difference vegetation index (NDVI), and the sediment transport index (STI), were assessed. Approximately 445 flash flood sites were identified through field visits and records of historical flood events. Accordingly, the dataset was divided into two groups for training (70%) and testing (30%) through a random selection scheme. The results show that the area under the curve (AUC) values of the receiver operating curve (ROC) were above 97% for all tested models, which indicates excellent accuracy. The FS mapping (FSM) results showed that downstream areas that are highly populated are prone to flooding and are characterized by high and very high levels of susceptibility. The results also compared with the 2D Rainfall Runoff Model showing acceptable agreement in terms of flood extend. The study verified that the newly employed algorithms (LightGBM and CatBoost) can be used for FFS prediction in other similar arid environments with reasonable accuracy. The outcomes of this study can be used by planners and officials for flash flood risk reduction.