Runoff Simulation And Analysis Using SWAT Model And ANN Model In Huotong River Basin

The use of distributed hydrological models to simulate watershed water resources and their changes has been an ongoing research issue in academia and relevant technical departments.The scarcity of measured meteorological data results in the limitation of long-term series simulation of hydrological models.In addition,the development of a hybrid model combining physical hydrological model and machine learning has been proposed,and the improvement of daily runoff prediction accuracy by combining SWAT model and ANN has attracted widespread attention.In this study,the China Meteorological Assimilation Driving Datasets for the SWAT model(CMADS)and its long time Series product(CMADS-L)of the Huotong River Basin were used to analyze several factors of local precipitation and temperature by using Mann-Kendall test and cumulative anomaly method,and to drive the monthly and daily SWAT models.Verify the applicability of CMADS products in SWAT runoff simulation of Huotong River;The annual,flood season,non-flood season,spring,summer,autumn and winter spatial and temporal analysis of CMADS-L driving monthly SWAT flow and water yield were conducted to identify the key areas of runoff process and explore the changes of watershed water quantity under different land use scenarios.The BP neural network of SWAT daily scale results combined with ANN model was used to establish SWAT-ANN model to predict daily runoff.The results show that:(1)The meteorological elements analysis of CMADS-L showed that:(1)In terms of time,annual precipitation in Huotong River Basin was relatively stable from 1979 to2018,the annual precipitation days reduced year by year,annual maximum daily precipitation increased year by year,the plum rains and typhoon season of heavy rain and heavy rain days increased year by year,daily maximum temperature and the annual maximum daily highest temperature warming significantly,daily minimum temperature increased year by year.(2)In terms of spatial distribution,the southern and northern parts of the basin had more precipitation,the northern part had more precipitation days,and the eastern part had higher annual maximum daily precipitation.(2)Applicability study of CMADS products in watershed SWAT model:The results of CMADS driven SWAT model in monthly and daily scale runoff simulation are better than the results of station driven,and CMADS products have good applicability in Huotong River runoff simulation;Based on the CMADS-L driven SWAT model of Huotong River,the R~2and NSE coefficients of the monthly runoff simulation at the two stations were both greater than 0.83 in the rate interval and validation period.In the daily runoff simulation,except for the flood validation period,the R~2and NSE coefficients of flood validation period,Yangzhongban validation period and flood validation period were all greater than 0.69.(3)The spatial and temporal variation of water yield in the sub-basins showed that:(1)In terms of annual water yield and runoff,the spatial variation of water yield was obvious,and the runoff and water yield increased year by year in the northwest of the basin,while the change rate of runoff and water yield was higher in the south of the basin.(2)In flood season/non-flood season,the runoff and water yield in the northwest of the flood season were more,and the runoff and water yield increased year by year,while the change rate of runoff and water yield in the south was higher.In non-flood season,the runoff and water yield decreased year by year in the northwest of the basin,and the decrease rate of runoff and water yield was the highest in the south of the basin.(3)Seasonally,the runoff and water yield decreased year by year in spring,and the change rate of runoff and water yield was the highest in the northwest of the basin.Except for spring,runoff and water yield increased year by year,and the change rate of runoff and water yield was the highest in the southern part of the basin.(4)The change of water yield in different land use change scenarios in key areas:Sub-basins 2,4 and 12 with higher annual water amount were identified as key area A,sub-basins 34,35 and 36 with higher annual water amount change rate from 1979 to 2016were identified as key area B;After the conversion of cropland and grassland into forest land in key area A and key area B,the water volume of the basin decreased the most,and hongkou and Yangzhongban decreased by 0.92%and 0.86%,respectively.(5)Daily runoff simulation results of SWAT-ANN combined model:(1)Daily runoff prediction accuracy of SWAT-ANN model is higher than that of ANN model;Combined with precipitation and flow,SWAT output sub-basin water yield and inflow sub-basin flow are the input layer of SWAT-ANN model,which has the highest prediction accuracy.R~2and NSE coefficients were greater than 0.91 and 0.83 in training period,and greater than 0.83 and 0.69 in verification period.(2)The combined model incorporating PCA method can improve the speed of model learning on the premise of ensuring the prediction accuracy.(3)The prediction effect of the combined model in flood season is better than that of the combined model in the whole year.The R~2and NSE coefficients of daily runoff prediction of the combined model in non-flood season are higher than those of the combined model in the whole year,and the validation period is greater than 0.94and 0.89,respectively.