Research on the Application of BP Neural Network Algorithm in FY-4A-SSI Product Correction
In order to compensate for the shortcomings of ground observation radiation data and obtain high-precision solar radiation spatial distribution data,a monthly FY-4A SSI product correction model based on BP neural network is established based on daily solar radiation data from seven ground radiation stations in Guizhou from March 1,2018 to April 30,2020.The model is used to correct and analyze the total radiation of FY-4A,and the correction results are compared with those obtained using an unary linear regression model.The results show that:(1)The total radiation value of FY-4A retrieval has a biased distribution characteristic of overestimating low-value radiation and underestimating high-value radiation,and overall,the total radiation value of FY-4A is relatively high.(2)The correlation coefficient R between the total radiation value of FY-4A and the station value is between 0.767 and 0.926,the mean absolute error MAE is between 5.04 and 6.98 MJ·m-2,the mean error ME is between 4.43 and 6.68 MJ·m-2,and the root mean square error RMSE is between 5.70 and 7.44 MJ·m-2.(3)After the linear correction,the R of the two remains unchanged,with MAE between 1.45 and 3.09 MJ·m-2,ME between-0.25 and 0.42 MJ·m-2,and RMSE between 2.12 and 4.70 MJ·m-2.(4)After corrected by the BP neural network method,the R of the two is between 0.856 and 0.962 MJ·m-2,MAE between 1.02 and 2.43 MJ·m-2,ME between-0.48 and 0.23 MJ·m-2,and RMSE between 1.36 and 3.77 MJ·m-2.Therefore,the error between the total radiation value of FY-4A correction based on the BP neural network model and the station value is small.The correction accuracy of FY-4A correction based on the BP neural network model is higher than that by the linear correction,and its stability is good.Therefore,the FY-4A correction based on the BP neural network model can be applied to the correction of FY-4A surface solar irradiance products in mountainous areas of Guizhou Plateau.
Solar radiationFY-4A-SSIBP neural networkunary linear regressioncorrection
NETL
NSTL
万方数据
