Regional Nonstationary Extreme Flood Frequency Analysis Research In The Xiangjiang River Basin

Accurate estimation of design floods is an important part of the design and operation of hydraulic engineering.The traditional flood frequency analysis method calculate the design flood values of different frequencies,which is the basis of the flood control scheme.However,with the changes of basic physical conditions of extreme flood such as climate and land use,the stationary assumption of traditional hydrological frequency calculation may not be satisfied,and the reliability of traditional hydrological frequency analysis results was seriously questioned;While a single station non-stationary model will raise the uncertainty of design flood estimation,so the regional non-stationary model that combine single station nonstationary extreme flood and inter-station flood correlations was built to improve design flood calculation accuracy.In this research,we take the annual maximum flood peak series(AMFPS)of the Xiangjiang River Basin——the largest basin in Hunan Province,as the research data to explore the model of a single-station and regional extreme non-stationary flood frequency calculation and quantitatively estimates the extreme flood risk changes caused by the climatedriven.The main research contents and results are as follows:(1)Choosing the climate driving factors(climate indices,sea surface anomaly,sea level pressure anomaly)that have significant correlation with AMFPS of 24 hydrological stations,using the Spearman nonparametric rank correlation test for preliminary choose,then using the generalized linear model and leave-one-out cross-validation(LOOCV)to dermine the optimal combination of climate drivers for AMFPS.Results show that the most significant climatic driving factors related to the AMFPS of Xiangtan Station——the hydrological control station of the Xiangjiang River Basin,are JJA AO(average of the Arctic Oscillation Index in June,July and August)and the first principal component PC1 of the JJA SLPa2(sea level pressure anomaly in the Northwest Pacific region).The climate index with the most significant correlation with the AMFPS of most hydrological stations in the Xiangjiang River Basin is JJA AO.(2)Building the the non-stationary flood frequency calculate model of single station.Considering the time-varying moment model under three conditions:traditional stationary,time trend,and climate driving factors as covariates,the non-stationary frequency is calculated for the AMFPS in the Xiangtan Station.The models are compared through the Akaike Information Criterion(AIC)and the Deviation Information Criterion(DIC),and the optimal model is obtained as whose mean is linear function of JJA AO and JJA SLPa2-PC1 but variance is linear function of JJA SLPa2-PC1.The single-station non-stationary frequancy calculate model can fully simulate and describe the trend of changing extreme flood risk caused by climate driving factors in the Xiangjiang River Basin.Based on the Bayesian model,the model can obatain climate-informed flood quantiles uncertainty interval easyly,which provides a scientific basis for flood control measures in the basin.(3)Building the regional non-stationary frequency analysis model of the Xiangjiang River Basin based on Hierarchical Bayes(HBM-RNFA).Based on whether the correlation between JJA AO and the extreme floods is significant,24 hydrological stations divide into two clusters.In both the whole basin case and the clusters case,the regional non-stationary flood frequencies based on hierarchical Bayesian simulation are constructed respectively.In both HBM-RNFA model(full watershed)and HBM-RNFA-Cluster model(significant clusters),this model is divided into 3 types according to the degree of comprehensive stations information:no pooling,partial pooling and full pooling.The AIC criterion and the DIC criterion are used to compare the models.The results show that the regional non-stationary partial pooling model that not only reflects the characteristics of regional homogeneity but also considers site differences is the best model.The best model averages the spatial distribution response of extreme floods to climate driving factors,reduce the uncertainty significantly,and improve the calculation accuracy.The return period based on HBM-RNFA-Cluster partial pooling model is significantly smaller than the traditional,which shows that the probability of extreme flood events increases in most stations of the Xiangjiang River Basin,due to the influence of the west Pacific sea level pressure anomaly.The analysis and application of regional non-stationsry extreme flood frequency further enriches the method system of non-stationary hydrological analysis and calculation research,and provides a scientific reference for the designated countermeasures for flood control and disaster reduction and flood risk management in the Xiangjiang River Basin.