Simulation And Prediction Of Lake Water Level Based On Vine Copula Function And Bayesian Model Averaging

Accurate simulation and prediction of lake water level changes is of great significance for artificial lake control,treatment and restoration,and giving full play to the functions of lakes.Lake water level prediction is the result of many nature-society binary factors,with very complex nonlinear relationships and uncertainties.How to combine these factors to establish a prediction model that accurately reflects the change of lake water level has become one of the urgent problems to be solved in current hydrological research.Under the influence of global climate change and high-intensity human activities,Erhai and Qilu Lake in Yunnan Province suffered continuous drought disasters around 2010,resulting in a decline in lake water level and serious threats to the water ecological security of the river basin.The establishment of early warning indicators of water level drought is of great significance to drought prevention in river basins.In this study,we constructed a long-term lake level prediction model based on the vine copula function and bayesian model average that combined lake water level and the main hydro-meteorological factors.The model was analyzed from two perspectives,namely,single model and combination model,and predicted the water level of Erhai and Qilu Lake respectively.We also introduced the fuzzy membership function to divide the water level of drought limt in Erhai,and scheduling measured and predicted water levels.The results of this study will be used to quantify the relationships between hydrological and meteorological variables on the lake water level,support the strategies to plan and allocate the water resources,and could guide the lake scheduling,and protection and governance in the basin,and drought resistance decision-making.Firstly,we constructed a model based on the vine copula function that combined water level data and data for the main meteorological influences on the water level,namely evaporation(E),temperature(T),precipitation(P),and runoff flowrate(F),and incorporated rolling decisions and real-time correction of prediction results.The correlations between the different variables were determined.The model was then applied to predict the long-and short-term water levels in Erhai Lake.The results showed that(1)The predicted daily water levels(with ME=0.02～0.09,RMSE=0.02～0.024,NSE=0.99,and IA=0.99)were more accurate than the predicted monthly water levels(with the ME=0.039～0.444,RMSE=0.194～0.279,NSE=0.913～0.958,and IA=0.977～0.989),and the accuracy of the predictions improved as the number of variables increased.(2)The prediction accuracy of the vine copula model was lower for small sample sizes and when there was a lack of runoff data.By improving the analysis of the model’s errors,the percentages of the relative errors of the prediction accuracy less than 5%,10%,15%,and 20% increased to 70%,83%,95%,and98%,respectively.(3)The vine copula model outperformed the back-propagation neural network and support vector regression models,and,of the three model types,gave the best estimate of the nonlinear relationships between the predicted water level and climatic factors,especially in the wet season(May to October).The prediction accuracy of the three models was higher for middle water level,but lower for low and high water level.In order to improve the prediction effect,based on the above research,the optimal variable combination of water level and hydrometeorological factors was selected to construct Vine Copula,BP and SVR models to obtain water level prediction values,and the model prediction values were used as input to construct BMA(VC,BP,SVR)and BMA(BP,SVR)combination models using the BMA method,which were applied to the monthly water level of Qilu Lake for prediction and uncertainty analysis,and was compared with the(VC,BP,SVR)combinatorial models constructed by the Vine Copula and Copula methods,they had the same input as the BMA method.The results show that:(1)The BMA method improved the prediction accuracy.The prediction accuracy of BMA(VC,BP,SVR)and BMA(BP,SVR)combined models was higher than that of a single model,and the correlation coefficients reached 0.99 and 0.98,respectively.(2)The BMA95% confidence interval indicates that the BMA(VC,BP,SVR)combined model had a range coverage of 90%,with less uncertainty and greater uncertainty in wet season water level prediction.(3)In the(VC,BP,SVR)combination model of three different methods,the Vine Copula method improved the prediction accuracy higher than the BMA method,and the Copula method did not improve the prediction accuracy.The three methods had better effect on medium water level prediction,and high water level was greatly affected by precipitation and runoff,resulting in high water level prediction accuracy was lower than low water level prediction.Finally,the fuzzy membership function was introduced to divide the drought limit water level in Erhai,and the water supply reduction coefficient was determined by comprehensively considering the priority and guarantee level of life,industry,agriculture,drainage ecological base flow and power generation demand in different typical hydrological years,and the measured and predicted water level scheduling of lake water and water use could provide scientific basis and technical support for drought defense decision-making.