An integrated statistical and data-driven framework for supporting flood risk analysis under climate change

Abstract

Summary An integrated statistical and data-driven (ISD) framework was proposed for analyzing river flows and flood frequencies in the Duhe River Basin, China, under climate change. The proposed framework involved four major components: (i) a hybrid model based on ASD (Automated regression-based Statistical Downscaling tool) and KNN (K-nearest neighbor) was used for downscaling rainfall and CDEN (Conditional Density Estimate Network) was applied for downscaling minimum temperature and relative humidity from global circulation models (GCMs) to local weather stations; (ii) Bayesian neural network (BNN) was used for simulating monthly river flows based on projected weather information; (iii) KNN was applied for converting monthly flow to daily time series; (iv) Generalized Extreme Value (GEV) distribution was adopted for flood frequency analysis. In this study, the variables from CGCM3 A2 and HadCM3 A2 scenarios were employed as the large-scale predictors. The results indicated that the maximum monthly and annual runoffs would both increase under CGCM3 and HadCM3 A2 emission scenarios at the middle and end of this century. The flood risk in the study area would generally increase with a widening uncertainty range. Compared with traditional approaches, the proposed framework takes the full advantages of a series of statistical and data-driven methods and offers a parsimonious way of projecting flood risks under climatic change conditions.