Research On Grid Correlation And Scale Sensitivity Of Precipitation Neighborhood Ensemble Probability Method

Due to the numerical forecast errors and the uncertainty of natural variability of precipitation,quantitative precipitation prediction results are uncertain.The problem of how to reduce the forecast errors of model quantitative precipitation has been widely concerned by meteorologists.The Precipitation Neighborhood Ensemble Probability Method is a new method to deal with the uncertainty of high resolution ensemble forecast.However,in the existing calculation schemes,grids in the neighborhood are calculated with equal weights,and the processing of grids information in the neighborhood may not be objective enough.The probabilistic prediction results may not accurately describe the uncertainty of forecast models.In order to solve the above problems,this paper designs two weight correction schemes which are based on the one-point correlation method,respectively weight correction neighborhood method and binary weight correction neighborhood method.These two methods consider the correlation of neighborhood grids,and in the process of calculating the precipitation neighborhood ensemble probability,weight is assigned to grids according to the correlation coefficient.And neighborhood grid correlation coefficient is established by using the 24-h accumulated precipitation data of GRAPES(Global and Regional Assimilation and Prediction Enhanced System)regional ensemble forecast system from May to September in 2015 and 2016 and in the same period of the24-h accumulated gauge-radar-satellite merged precipitation grid analysis data.On this basis,under different precipitation thresholds and neighborhood radius,the experiments of precipitation weight correction neighborhood ensemble probability method and binary weight correction neighborhood ensemble probability method were carried out from May to July 2017,and the experimental results were compared with ensemble probability forecast and the equal weight neighborhood ensemble probability forecast.Meanwhile,the experiments found that the neighborhood probability methods are sensitive to the scale of the neighborhood.Then the precipitation probability prediction scale sensitivity experiments of ensemble probability forecast,equal weight neighborhood ensemble probability method,weight correction neighborhood ensemble probability method and binary weight correction neighborhood ensemble probability method are compared with neighborhood radius from 1 to 10 grid length,and the results were verified by multiple probabilistic scores.The main conclusions of the research are as follows:(1)Except for the area under the relative operating characteristic curve(AROC)when the neighborhood radius is 15 km,the precipitation probabilistic scores using neighborhood computing schemes are better than the raw ensemble probability forecast methods.(2)The precipitation probabilistic scores of three neighborhood ensemble probability methods have their own advantages and disadvantages,For example,The score of AROC is slightly better for the equal weight neighborhood ensemble probability method.However,the higher reliability of precipitation probability prediction is determined by weight correction neighborhood ensemble probability method and binary weight correction neighborhood ensemble probability method.(3)It can be seen from the probability forecast distribution diagram of individual experiments that the distributions of the three kinds of neighborhood ensemble probability methods can all give more meaningful probability forecast results.The distribution of neighborhood ensemble probability tends to be smooth with the increase of neighborhood radius,and the probability value decreases gradually,but the distribution of weight correction neighborhood ensemble probability method can show more detailed information of precipitation distribution.Moreover,the smooth degree of binary weight correction neighborhood ensemble probability method is between equal weight neighborhood ensemble probability method and weight correction neighborhood ensemble probability method.(4)Equal weight neighborhood probability method,weighted correction neighborhood ensemble probability method and binary weighted correction neighborhood ensemble probability method can provide some uncertain forecast information,but the probability field generated by these methods has a strong correlation with the original forecast precipitation field.The probabilistic predictions of the two weight modification schemes are also related to the selection of the lattice weights in the neighborhood.(5)The experimental results of different neighborhood radii show that different neighborhood methods have slightly diverse optimal neighborhood scales under various precipitation thresholds.In general,the spatial scale range of the optimal statistical score is 5-8 times the horizontal grid scale of the model.More reasonable probability prediction results can be obtained by choosing a proper neighborhood scheme and neighborhood radius.