Identifying and mapping the spatial distribution of regions prone to snowmelt flood hazards in the arid region of Central Asia: A case study in Xinjiang, China

Abstract

Abstract Snowmelt floods are highly hazardous meteorological disasters that can potentially threaten human lives and property. Hence, snowmelt susceptibility mapping (SSM) plays an important role in flood prevention systems and aids emergency responders and flood risk managers. In this paper, a method of identifying snowmelt flood hazards is proposed, and a large‐scale snowmelt flood hazard zonation scheme based on historical recordings and multisource remote sensing data is established. To assess the quality of our approach, the proposed model was tested in the cold and arid region of Xinjiang, China. Overall, 140 historical snowmelt flood events and 27 explanatory factors were selected to construct a geospatial dataset for SSM of the contemporary period. GridSearchCV was used to comprehensively search the candidate parameters from the grid of given parameters obtained with the random forest (RF) algorithm. Then, the geospatial dataset was divided into two subsets: 70% for training and 30% for testing. Next, SSM results were obtained with the RF algorithm using optimized parameters. The results indicate that our optimized RF classifier performs well for the task of SSM, with a high AUC value (0.975) for the test dataset. The validation and analysis suggest that the proposed method can efficiently identify snowmelt flood hazards in undersampled arid areas at a regional scale.

Keywords

• large scale
• multisource remotely sensed data
• random forest algorithm
• snowmelt susceptibility mapping