Research On Regionalization Of Hydrological Model Parameters Based On CART Algorithm

Against the background of global climate change,torrential flood disasters occur frequently and are extremely destructive,causing serious harm to the safety of our lives and property.Hydrological models can be an effective tool for predicting these events and timely release of forecast and warning information.However,the application of hydrological models requires accurate and reliable model parameter values to support them.In previous studies,the determination of hydrological model parameters was based on a large number of measured rainfall Runoff data.At present,most of the mountainous and hilly areas in China lack or do not have hydrological monitoring stations.In addition,with the impact of human activities and the intensification of global climate change,the underlying surface conditions of the river basin have undergone tremendous changes,making early hydrology Observation data is no longer applicable,and the area with monitoring sites has also become an area lacking data.Based on the above background,the problem of hydrological forecasting in data-free river basins is a technical problem urgently to be solved in the current hydrological circles.At present,there are two main ways to solve this problem: one is to develop a distributed hydrological model with a clear physical mechanism to reduce the uncertainty of the hydrological model;the other is to seek an effective and feasible regionalization scheme for the data model of the hydrological model of the non-database watershed.The thesis is based on the above two points,and the main research contents are as follows:1)A total of 13 small and medium-sized watersheds in Liaoning Province and Jilin Province were selected as the research object,and a total of 93 floods were selected.Based on the spatiotemporal variable source mixed runoff model,the SCE-UA global search algorithm is used to rate and verify the model parameters.The average Nash coefficients during the verification period and rate are 0.86 and 0.82,and the relative errors of the peaks are 5.8% and 6.5%,Indicating that the spatio-temporal variable source mixed runoff model has good applicability in the study area.Based on the Sobol sensitivity analysis method,the sensitivity analysis of the model parameters is conducted to study the difference in the contribution of each model parameter to the model output.2)The model parameters of all watersheds are mutually transplanted to form training samples,and the effect of the parameter set transfer between watersheds is evaluated.A set of watershed feature attribute description sets is proposed to provide prerequisites for identifying similar watersheds in the target watershed below.Principal component analysis(PCA)was used to perform principal component analysis on watershed attribute factors to extract uncorrelated factors.3)The classification and regression tree(CART)split rules are used to find suitable donor watersheds for the dataless watersheds.The rules for transplanting model parameters between basins are formulated,and the results of model parameter transplantation are evaluated.The research results show that the CART algorithm can be effective The success rate of model parameter transplantation is improved because the rules generated by the CART algorithm optimally consider the spatial proximity and physical similarity between watersheds.This paper summarizes the problems and shortcomings of the research content,and puts forward the prospects and Suggestions for the future solutions to the problem of hydrology prediction in the basin without data.