Accuracy Evaluation And Detailed Simulation Of GPM IMERG Satellite Daily Precipitation Products In Fujian, Zhejiang And Jiangx

Precipitation has an important impact on human society,so there is a need to refine the spatial distribution of precipitation.At present,there is a common problem in the general precipitation estimation models,in which the precipitation accuracy of the estimation models is overestimated or underestimated when the daily precipitation is low or high.Based on this,this paper takes the mountainous area of Fujian-Zhejiang-Jiangxi as the study area,and firstly classifies the precipitation from meteorology and the actual situation of the study area,and then uses the principal component analysis to select suitable precipitation influence factors for light rain,medium-large rain and heavy rain,and brings the precipitation influence factors of light rain,medium-large rain and heavy rain into the respective models to obtain the regional spatial precipitation of light rain,medium-large rain and heavy rain,respectively;among them,in the light rain and medium-large rain models The model coefficients and residuals are obtained,and the estimated model precipitation is calculated,and then the residuals are superimposed with the estimated model precipitation to obtain the model estimated precipitation spatial distribution map with a spatial resolution of 1 km×1 km.Finally,the Tyson polygon and natural watershed combination method is used to fuse the light rainfall,medium-large rainfall and heavy rainfall models to obtain regional refined daily spatial precipitation estimates.The main conclusions of the paper are as follows.(1)From the evaluation of downscaling models with different rainfall classifications,the correlation coefficients of all three rainfall classifications are above 0.83,which is more than0.58 improvement over the original GPM IMERG precipitation product,and the correlation coefficient R of light rain is the highest at 0.944,with the lowest error index and the best fitting effect,indicating that the downscaling models based on rainfall classifications can improve the models to some extent and effectively effectively improve the spatial resolution and accuracy of GPM IMERG precipitation products in the study area.(2)From the spatial distribution of daily precipitation,the graded fusion downscaling model is consistent with the spatial distribution of precipitation from meteorological stations on January 17,April 7 and October 27;from the range of daily precipitation,the extreme range of precipitation from the graded fusion downscaling model is consistent with that from meteorological stations,so the graded fusion downscaling model can improve the accuracy of the original GPM IMERG precipitation products to some extent and reduce the accuracy of the original GPM IMERG precipitation products.GPM IMERG precipitation products.(3)The long and short term memory network model is improved by selecting the model parameters: the hidden layer is 4,the number of neurons N is 300,and the number of iterations is 150,and the accuracy index of the model is optimal,neither overfitting nor underfitting.(4)Compared with the vegetation NDVI downscaling model and MOD05 downscaling model,the graded fusion downscaling model takes into account different rainfall levels and has better simulation ability than the other two downscaling models,and the root mean square error RMSE is 7.130 mm lower than the original GPM IMERG data,the correlation number R is improved to 0.872,and the relative error MRE is reduced by 6.624 percentage points.6.624 percentage points.In precipitation fusion,the MOD05 downscaling model has higher simulation ability than the vegetation NDVI downscaling model,and the precipitation maximum range is consistent with the precipitation from meteorological stations.