Research On Data Set Constructing And Rainstorm Forecasting Of Mountain Rainstorm Of Southern Shaanxi Based On TRMM Data

Rainstorms cause floods,especially in mountain areas.In addition to floods,rainstorms cause landslides,avalanches,and mudslides,these disaster can easily lead to people 's loss of life and property.Therefore,research on rainstorms and forecasting of rainstorms are very meaningful.Based on the TRMM satellite rainfall data,this paper uses the Qinba mountain area in southern Shaanxi as the research,TRMM data is spatially downscaled.Based on the TRMM data after downscaling,the rainstorm data set is constructed,and apply the rainstorm data to rainstorm forecasting research,provide a new way for rainstorm simulation and forecasting in areas with little or no surface data.Due to the lack of low resolution of the original TRMM precipitation data,the TRMM data needs to be downscaled,and there are few optional intermediate parameters in the study area.Therefore,two methods are used to regression downscale the TRMM precipitation data:(1)One site corresponds to one model,uses univariate non-linear downscaling(UNR)with remote sensing surface temperature as the intermediate parameter.This method can obtain the daily precipitation sequence.(2)Per year corresponds to one model,use multiple linear regression(MLR)downscaling model with remote sensing surface temperature,elevation,and latitude and longitude as intermediate parameters.This method obtains high-resolution annual precipitation data for the entire study area.This paper analyzes the impact of the downscaling process on the accuracy of TRMM data compared with the measured data,and conducts an accuracy test on the entire study area and each station.After that,the annual maximum precipitation was screened according to 1d,3d,and 5d scales to construct storm data sets of downscaled TRMM data and measured data,and the accuracy of the storm data sets was analyzed.The rainstorm data of Hongchun Station and Zhulinguan Station were selected for fitting and forecasting in the storm forecasting research.Due to the short rainstorm sequence,the gray-period extension model and BP neural network model were selected to study the rainstorm fitting and forecasting,and the fitting and forecasting effects of two methods were compared.The main research conclusions of this paper are as follows:(1)On the whole,the UNR downscaling and the MLR downscaling have little effect on the accuracy of TRMM data.this conclusion indicating that downscaling improve the resolution of TRMM data,but not affect the accuracy of TRMM data.(2)In the accuracy test of comparing the scale data with the measured data,from the overall view of the study area,both types of downscale precipitation data show that when the precipitation is low,the downscale precipitation is greater than the measured precipitation,and the precipitation is large When the scaled down precipitation is less than the measured precipitation.In terms of different time scales at different stations,the downscale data BIAS is significantly smaller in summer months than in winter months,and RMSE is the opposite.From the perspective of different data types,the accuracy of MLR data is slightly better than that of UNR data.(3)During the construction of the rainstorm data sets,the accuracy of the rainstorm data gradually improved with the rainstorm time scale from 1d to 3d to 5d.In the research of rainstorm fitting and forecasting,the fitting and forecasting effect of BP neural network model is better than the gray-periodic extension model;the fitting and forecasting effect of MLR storm data is better than the UNR storm data;the fitting and forecasting effect of 3d and 5d scale storm data is better than 1d scale;The fitting effect of building the model period is better than that the forecasting effect of forecasting validation period.