Study On The Real-time Correction Method Of Flood Forecast In The Basin Above Gagdache Of Gan River

Compared with the distributed hydrological model,the aggregate hydrological model has lower requirements in terms of the number of parameters and the accuracy of information required for modeling,and is an important tool for flood forecasting in watersheds,especially in areas where information is relatively scarce.However,due to the uncertainty of forecast models and the inexperience of forecasters,real-time correction is needed to improve the accuracy and timeliness of forecasts.Real-time flood forecast correction can be combined with real-time correction technology based on real-time hydro-meteorological data to update the parameters of the original model or forecast results by adding observation information time by time during the forecast,thus improving the forecast accuracy.Therefore,the key to improving the accuracy of real-time flood forecasting is not only to obtain accurate and reliable real-time observation information,but also to rely on models that can accurately respond to the hydrological processes in the study basin and effective real-time correction methods.In this paper,we will construct a real-time correction model for flood forecasting based on the ensemble Kalman filter method(EnKF)in the context of the ensemble model.The uncertainties of the model under the existing observation conditions are also analyzed based on numerical experiments and practical arithmetic examples.The prediction effects of different combinations of runoff,surface soil water and water surface evaporation on hydrological status and flow are compared and analyzed based on actual calculations in the Gagdache control basin.It provides a reference for the future development of real-time correction techniques for data assimilation by making full use of multivariate/source observation data.The main research and findings of this paper are as follows:(1)This paper summarizes the current research status and problems of real-time flood forecast correction,couples the Xin'an River model with the EnKF algorithm to construct an EnKF-based real-time flood forecast correction model,and analyzes the forecast correction effect of the model,which can improve the flood forecast accuracy by more than 20% and the Nash coefficient to more than 0.9.(2)In this paper,we analyze the uncertainties of the model in three aspects: the estimation effect of soil water content in different soil layers,initial conditions and input perturbation errors when adding flow observations,considering the existing observation conditions.The analysis shows that the upper soil water content is estimated better than the lower and deep layers when flow observations are added.Under the same initial soil water content conditions,the real-time calibration model can improve the flood error by more than 40% compared with the Xinanjiang model,which can effectively reduce the dependence on the initial conditions.The input perturbation error has a great influence on the model forecast accuracy,especially in the case of large rainfall,it is easier to overestimate the flow,and the setting of input perturbation error should be paid attention to in the model application.(3)In this paper,three satellite remote sensing products,namely,evaporation from ground observation and evaporation and surface soil water content data from GLEAM product,and surface soil water content data from GLDAS product,were selected,and the three remote sensing products were evaluated and calibrated for their availability.The corrected water surface evaporation and surface soil water content data were combined with three observations of flow,and nine sets of arithmetic examples were set up to compare and analyze the effects of flow and soil water content estimation in different soil layers when adding different sources and types of observations.The results show that the flow estimation results basically match when different sources of observations are used,indicating that the EnKF-based real-time flood forecast correction model has strong inclusiveness and robustness;the forecast accuracy meets the requirements when a combination of multiple observations is used,and can be improved by up to 10% compared with flow observations only.