Flash flood susceptibility mapping using LiDAR-Derived DEM and machine learning algorithms: Ljuboviđa case study, Serbia

Flash floods are the result of climatic and hydrological extremes and are manifested by dynamic and complex processes of movement of water and sediment. They represent the most frequent and widespread natural disaster at the global level, with unwanted ecological and economic consequences. The main causes of flash floods are related to numerous meteorological and physical–geographical factors. In the territory of Serbia, flash floods represent the most common natural risk with serious consequences for people’s lives and activities. Flash flood susceptibility mapping plays a crucial role in flood risk assessment and management. The current study prepared a flood inventory using light detection and ranging (LiDAR) derived digital elevation model, and it used integrated tree machine learning models (random forest [RF], classification and regression trees [CART], and support vector machine [SVM]) to predict flood susceptibility in the Ljuboviđa watershed, municipality Ljubovija, western Serbia. First, 12 independent variables were employed as conditioning factors: lithology, rainfall, land use/cover, elevation, slope angle, aspect, plan curvature, topographic wetness index, stream power index, distance from streams, distance from roads, and normalized vegetation index. Using the well-known scikit-learn (train_test_split) Python module, the flood inventory dataset was split into 70 and 30% for training and validation, respectively. The models’ performance was additionally assessed using the area under the curve (AUC). The results of the accuracy assessment demonstrated that the models for predicting flood susceptibility, RF, CART, and SVM, had AUC values of 0.854, 0.802, and 0.831, respectively; it means that RF had 85.4%, CART 80.2%, and SVM 83.1% chance of correctly ranking a random positive example higher than a random negative example, which represents the predictive power of the used models. When it came to predicting flood susceptibility, the RF model outperformed the other models used. This model estimates that 15.49, 16.04, 15.67, 23.10, and 29.70% of the watershed are very low, low, moderate, high, and extremely highly susceptible to floods, respectively. Thus, our study shows that data produced from LiDAR is potentially helpful in managing flood risk, particularly when assessing flood-related issues in the future. Flash-flood susceptibility maps have become a vital tool for risk prevention and management for government and local authorities (particularly national and local civil protection agencies, urban planning and land management departments, Ministries of Water and Environment), emergency response services (police, fire, and medical services), infrastructure and utilities sectors, insurance companies, and others.