Flood Feature Analysis And Classified Flood Prediction Based On Watershed Heterogeneity

Flood forecasting is the most effective measure for flood control and disaster reduction.Among them,classified flood prediction can significantly improve the prediction accuracy,and it has become a hot spot in the field of flood prediction.Most of the studies are usually based on meteorological and hydrological data to classify and forecast different floods for a specific river basin,which fails to fully consider the important factor affecting flood generation-the spatial heterogeneity of the watershed.In order to improve the accuracy of flood prediction and explore the relationship between basin heterogeneity and flood characteristics,this paper takes 354 basins in China as the research object,uses hierarchical clustering method to classify flood events in each basin,and uses partial correlation analysis and factor analysis to filter out the basin heterogeneity factors that affect flood characteristics of each type of basins,and then establishes a flood feature prediction model and a flood process discharge prediction model.The main conclusions of this study are as follows:(1)In this paper,eight flood characteristics indicators are selected as clustering factors,and the flood events in 354 watersheds in China from June 19,2020 to June 19,2021 are classified into three categories by principal component analysis and hierarchical clustering algorithm.The contour factor is 0.68,and the clustering effect is good.Type 1contains 165 watersheds,which are mainly distributed in the middle and lower reaches of the Yangtze River,the lower reaches of the Huai River,the middle and upper reaches of the Yellow River,the Liao River basin and the Pearl River basin.Type 1 belongs to the small floods with severe changes because its skewness and variation coefficient are large and the total volume of flood and peak discharge are small.Type 2 contains eight watersheds,which are mainly distributed in the middle and lower reaches of the Yangtze River.Type 2 belongs to the long-duration large flood because it has a long flood duration,a large total volume of flood and peak discharge.Type 3 contains 181 basins,which are mainly distributed in the Northwest River Basin,the Southwest River Basin,the main stream of the Yellow River,the Songhua River Basin,and the upper reaches of the Huaihe River,etc.Type 3 belongs to the conventional small flood because its flood volume and flood change are small.(2)Through partial correlation analysis and factor analysis,this paper selects the main influencing factors of flood characteristics of each type of watersheds and all watersheds.There are 10 factors affecting the flood characteristics of the first type of watershed,including previous moisture conditions,total precipitation in the previous 2days,total precipitation in the previous 15 days,NDVI,average elevation,terrain moisture index,clay content,sand content,CONTAG and the percentage of impervious area.There are 5 factors affecting the flood characteristics of the second type of watershed,including total precipitation in the previous 15 days,NDVI,terrain moisture index,SHDI and percentage of impervious area.There are 12 factors affecting the flood characteristics of the third type of watershed,including previous moisture conditions,total precipitation in the previous 11 days,total precipitation in the previous 12 days,NDVI,vegetation coverage,forest coverage,perimeter,slope,clay content,sand content,SHEI and percentage of impervious area.There are 13 factors affecting the flood characteristics of all watersheds,including total precipitation in the first 3 days,total precipitation in the first 6 days,previous moisture conditions,NDVI,forest coverage,area,average elevation,elevation difference,slope,terrain moisture index,clay content,sand content,and NP.(3)Based on the selected basin heterogeneity factors,the flood characteristic prediction model and flood process discharge prediction model are established by using random forest and xgboost respectively,and the flood process discharge prediction model is explained by using the SHAP.For the flood characteristic prediction model,the R~2 of test set of each type and all watersheds are 0.88,0.84,0.89 and 0.86 respectively,which are more than 0.8.The simulation effect is good.Except for the type 2 model with too few samples,the simulation effect of type 1 and type 3 flood characteristic prediction models is better than all watershed flood characteristic prediction model.For the flood process discharge prediction model,the R~2 of test set of each type and all watersheds are0.84,0.86,0.89 and 0.83 respectively.The result shows that classified flood prediction can improve the accuracy of flood prediction.For the first type of watershed,previous moisture conditions,NDVI,and sand content have a greater impact on the flood process discharge.For the second type of watershed,SHDI,the total precipitation in the first 15days,and NDVI have a greater impact on the flood process discharge.For the third type of watershed,perimeter,slope,and percentage of impervious area have a greater impact on the flood process discharge.For all watersheds,previous moisture conditions,NP and mean elevation had a greater impact on flood process discharge.