Development Of Rapid Flood Models And Machine Learning Methods For Multi-scale Flood And Waterlogging Simulation And Management

The frequency and intensity of extreme flooding events have been increasing in the context of climate and land changes.However,there are significant differences between flooding and waterlogging in term of definitions and causes,and the traditional flood simulation models are not able to satisfy the needs of rapid flood management due to complex model structure and time-consuming modeling process.Therefore,how to construct a rapid flood modeling framework for different flood types that can fully balance the model performance and running time,to quantify the flooding effect of each flood-inducing indicator and to propose adaptive management strategies have become current research hotspotsThe flooding is extraordinary serious under the dual influence of extreme weather and tropical cyclone in the coastal regions of southeastern China,where are experiencing unprecedented human activities and economic development.The Jiulong River Watershed(JRW),a coastal watershed in southeast China,where is witnessing fastestgrowing in economy and experiencing severity of climate-related disasters in the past several decades,was chosen as the study area.Machine learning method was integrated into a rapid flood modeling process to spatially identify the inundation areas for different flood types and quantify the flooding effect of each flood-causing factor.The major findings of this study are as follows:1)A rapid flood modeling and evaluation framework for different flood types was constructed,including HAND model for riverine flooding,random forest model for flood prediction and evaluation,and Blue Spot model for urban waterlogging.Based on the proposed framework,the flood inundation areas and depth were spatially identified at different recurrence periods,the effect of each flood-causing indicator on flooding was quantified,and the potential waterlogging areas and waterlogging-induced factors were identified.2)According to historical flooding records and simulation results,different areas of the Jiulong River Basin suffered different levels of flood disasters.However,the frequency and density of historical flood disaster events in Xinluo district,Longyan city were significantly higher than other areas.Bare land and buil-up areas were the two major land use types which were identified to be sensitive to flooding in Jiulong River Watershed.The flooding effect of geographical indicator,hydro-morphological indicator,land cover indicator and meteorological indicator factors were 48%,34%,12%,and 5%,respectively.3)Zhangzhou city was dominated by continuous precipitation that was normally dual-core and dynamic,and a significant spatial correlation found between hazard factors and exposure indicators.Longyan city was dominated by short-duration rainstorm that was single-core and static,and only 2 hazard factors(the flood-causing precipitation threshold and the number of waterlogged areas)was found to be spatially correlated with exposure indicators.This research can be used to predict the key features of flood occurrence,assess the systemic risks of disaster-bearers,evaluate disaster mitigation measures,screen flood disaster adaptation strategies,and provide support for integrated adaptive watershed management.