Rainfall forecast accuracy evaluation method for flood control demand

Using the results of numerical rainfall forecast as the input of the hydrological model can effectively prolong the flood forecast period,and strive more time for flood control.The accuracy of rainfall numerical forecast results used in flood forecast largely determines the accuracy of the hydrological model forecast.Therefore,carrying out evaluation research on the accuracy of rainfall forecast results is of great significance to flood forecasting work.Scholars at home and abroad have carried out researches on the accuracy of rainfall forecasting.However,most of the existing researches focus on the total error of rainfall forecasted by river basins and stations and have not considered the spatial distribution and time history distribution errors of rainfall forecasts. An evaluation method of rainfall forecast accuracy is proposed,which integrates three dimensions of time history,space,and total amount.Because of the different horizontal resolutions of each numerical prediction model,the Kriging interpolation is used to get the grid rainfall value by time,and then the basin area rainfall forecast value is obtained.The measured data from rainfall stations are processed in the same way.Thus,the measured and predicted rainfall of each grid and the whole basin can be obtained.A dimensionless value is selected,and the relative error term is introduced.The relative error of cumulative total rainfall,maximum 1,3,6,12,24 h,and other characteristic periods is comprehensively considered as the order of magnitude evaluation index.To accurately describe the coincidence degree between the predicted rainfall and the measured rainfall spatial distribution,a spatial distribution evaluation method is proposed,which uses the concept of TS score for reference,extends the concept from the field number to the spatial distribution,introduces the concept of tolerance interval,formulates the decision rules of each grid,and obtains the flow domain spatial evaluation index.To evaluate the coincidence degree between the rainfall forecast process and the measured process,the commonly used index "certainty coefficient" of the hydrological forecast is introduced as the time history evaluation index.The comprehensive evaluation index of rainfall is obtained by weighting the above evaluation indexes of magnitude,spatial distribution,and time history distribution. Taking the rainfall data of Yongjiang River basin in Zhejiang Province of China in recent years as an example,the method proposed is used to evaluate a variety of numerical prediction models.The results show that the comprehensive evaluation method proposed can take into account important information such as time history distribution which can not be reflected by TS score and has certain advantages in evaluating the rainfall forecast accuracy of flood control scenarios.At the same time,the rationality of the results is analyzed:under the same level of rainfall,the forecast accuracy of the same numerical forecast model decreases with the extension of the forecast period,and the results conform to the laws of statistics and meteorology.Under the same forecast period,the forecast accuracy of the same numerical forecast model decreases with the increase of rainfall magnitude,and the results conform to the laws of statistics and meteorology.When forecasting light rain and moderate rain,the accuracy of each numerical model has little difference,and the GFS model has a slight advantage in forecasting accuracy.However,with the increase of rainfall magnitude,the rainfall forecasting ability of GFS drops sharply,and its forecasting accuracy is surpassed by GRAPES_MESO and SMSWARMS. The proposed method which integrates three dimensions of magnitude,spatial distribution,and time distribution,can excavate the elements of time distribution that traditional evaluation methods such as TS score do not pay enough attention to.Therefore,this method can better evaluate the accuracy of rainfall forecast based on the demand of flood control scenarios,and can also provide a basis for the fusion application of multiple forecasting models.

Project Supported:

Clc Number: