Assessment of Flood Risk with Remote Sensing and Artificial Intelligence Integration: A Case Study of the Saz-Cayirova Stream Basin

Abstract

Floods are one of the most significant natural disasters, leading to substantial material damage Floods are one of the most significant natural disasters, causing both material damages and loss of life. Scientific studies indicate that the intensification of the hydrological cycle due to the effects of climate change can lead to excessive rainfall and, therefore, an increased risk of flooding. In this context, remote sensing, geographic information systems, and artificial intelligence play a crucial role in predicting flood risk and implementing preventive measures by acquiring and analyzing vital data. This integrated approach allows for data-driven, rapid, and accurate results in combating the effects of climate change. In this study, based on the flood disaster that occurred on June 24, 2010, in the Saz-Çayırova Stream Basin, the aim is to generate the current potential flood susceptibility map of the region with the assistance of various artificial intelligence methods. Factors contributing to flooding, such as elevation, slope, aspect, curvature, topographic wetness index, stream power index, soil type, normalized difference vegetation index, land use, distance to the river network, and drainage density, have been considered for flood susceptibility mapping. These parameters have been calculated using Landsat-5 and Landsat-8 satellite images and digital elevation models obtained from the Shuttle Radar Topography Mission (SRTM) data. Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) have been employed for flood susceptibility mapping. When considering the Mean Absolute Error (MAE) values, which is one of the standard accuracy metrics in machine learning, the XGBoost classification model (MAE value of 0.0138) and RF regression model (MAE value of 0.0068) have demonstrated superior performance. According to the machine learning models developed, elevation, stream density, and distance to the river are identified as the most effective factors in flood susceptibility mapping.The results suggest that models produced by integrating artificial intelligence and remote sensing dataset can be utilized in the sustainable management of flood-prone areas.