Evaluation And Optimization Of Microwave Soil Moisture Products For Fengyun-3 Satellite

Fengyun satellite microwave remote sensing soil moisture products play an important role in agricultural applications,especially in crop monitoring and disaster warning,and the evaluation of the reliability of its products is very important.This study is based on the soil moisture data of the China Automatic Soil Moisture Observation Station of the National Meteorological Administration of China,and systematically analyzed the quality of temporal and spatial distribution of the level 2 soil moisture products retrieved by FY-3B and FY-3C satellites in Shandong in 2018.Contrast with SMAP and SMOS satellite retrieving level 3 soil moisture products.The results show that in Shandong,FY-3B,FY-3C and SMAP have good time consistency with the automatic station observation data,and the root mean square error(RMSE)is 0.09 m³·m^-3,and the correlation coefficient(R)is greater than 0.3.The unbiased root mean square error(ub RMSE)of SMAP is 0.05 m³·m^-3,indicating that it has higher application value after removing the systematic error,while SMOS has poor applicability in Shandong.After adding the normalized difference vegetation index(NDVI)of the MODIS satellite as a reference,the correlation and error of FY-3B,FY-3C and automatic stations have obvious seasonal changes,and FY-3B and FY-3C often overestimate the soil moisture in May,August and September corresponding to the maturity period of winter wheat and summer maize,but underestimate in the rest of the time.This is because the X-band detection depth is shallow,which is used by Fengyun Satellite,and the results are greatly affected by surface vegetation,and the L-band is used by SMAP and SMOS,and the detection depth of which is deeper,and the results are less affected by surface vegetation.In order to improve the integrity of the Fengyun-3 soil moisture product and optimize the results,the Fengyun-3 passive microwave soil moisture complement model and optimization model based on the random forest algorithm are developed.The brightness temperature under ten channels of the FY-3B microwave radiometer,NDVI,latitude,longitude,date and FY-3B soil moisture are taken as input,and the automatic station observation data are used as the"true value"for learning and training,The optimized soil moisture product has a high accuracy.After the decision tree is greater than 100,R2 stabilizes at 0.83,and the root mean square error is within 0.03.For the missing values of FY-3B soil moisture,the random forest algorithm is also used to construct a prediction model to complete it,and the model accuracy is 0.92.Evaluating the developed FY-3B completed products and optimized products,we find that the continuity and relevance of FY-3B completed products are improved,but the statistics parameters are not much different from the original product.The FY-3B optimized product has a very high consistency with the automatic station observation data.The average difference,root mean square error and unbiased root mean square error are almost 0,and the correlation coefficients are all approaching 1,indicating that FY-3B soil moisture data are optimized well,and it can be further promoted to China and the world to generate larger-scale soil moisture products.At the same time,it shows that the random forest algorithm model has a very high accuracy and use value for soil moisture prediction.