Study on multiscale-multivariate prediction and risk assessment of urban flood

Abstract

Few studies have explored the impact of drivers on urban flood at multi-time scales, and few studies have assessed the urban flood risk based on accurate description and quantification of the intrinsic correlation between urban flood and its drivers. Therefore, this study develops a multiscale-multivariate prediction model for flood volume (FV) using wavelet transform, and a flood risk assessment model based on the joint distribution of FV and its key drivers using the Copula method. The downtown area of Zhengzhou City, a typical city in North China, is taken as the study area. The results reveal that land use change is the key driver for variation of urban rainstorm flood, and rapid urbanization led to the variation observed in 2005. The impact of land use change on FV primarily manifests at the long cycle scale, and the multiscale-multivariate prediction model demonstrates effective simulation (NSE = 0.957, R2 = 0.958, MAPE = 13.47%) and prediction capabilities (relative errors are all below 20%). Taking the FV exceeding the threshold corresponding to the frequency of 37.5% as an example, the maximum conditional risk probability under nine combinations of key drivers is 88.34%, while the minimum probability is only 6.83%, intuitively indicating the impact of rainstorm and urbanization on this risk. These findings can provide technical references for urban flood forecasting, urban water resources management and urban development planning.