Study On Key Techniques Of Measuring Soil Moisture Change Based On L-band GNSS-R

Soil moisture refers to the amount of water in the soil.It plays an irreplaceable role in the global water cycle or in the entire ecosystem.It affects the weather system on a large scale and affects the weather system on a small scale.Above,it determines the growth of crops,so accurate measurement and prediction of soil moisture is very necessary in terms of ecology,climate and agricultural development.In recent years,the remote sensing technology based on the reflected signal of the L-band global navigation satellite system has provided a new observation method for the measurement and prediction of soil moisture.Based on the theories and methods of remote sensing of soil moisture by GNSS-R technology at home and abroad,this paper uses GNSS-R technology to carry out a series of studies on soil moisture in China.The research content is as follows:(1)The 2018-2019 two-year database of key parameters of soil moisture retrieval by spaceborne GNSS-R remote sensing has been established,mainly for CYGNSS(Global Cyclone Navigation Satellite System)surface reflectivity,incident angle,and simulated scattering power DDM(extended Time Doppler diagram)and SMAP(Soil Moisture Active and Passive Task)data distribution and characteristics of soil moisture,vegetation transmission coefficient,and surface roughness are analyzed.(2)The data quality control method and data flow of the inversion of soil moisture by spaceborne GNSS-R have been established.The main process of inverting soil moisture includes:integrating and extracting CYGNSS and SMAP data in the Chinese region,data screening,solving the time correspondence and resolution problems of CYGNSS data and SMAP data,data smoothing,using multi-linear regression or random forest regression Methods Train the model to retrieve and test the soil moisture.(3)On the basis of comparative research,a soil moisture retrieval model based on spaceborne GNSS-R data is constructed.Two models of random forest regression and linear regression are trained and tested.The fit of random forest training data is better than that of linear regression training data.Its R value can reach 0.9912,which is close to 100%,RMSE=0.0134cm~3/cm~3,The R value of the linear regression training data is only 0.677,RMSE=0.0766cm~3/cm~3,the R of the linear regression test data is lower than the R of the random forest regression data,and the RMSE value is higher than the RMSE of the random forest regression data,but compared to the training data The R value increased by 0.223.The random forest test data is just the opposite.The R value of the test data is lower than the training data by about 0.11.(4)Using SMAP SM data,the CYGNSS linear regression SM and CYGNSS random forest regression SM were spatially tested for the average,quarterly,and daily accuracy of 2019.The average SM of CYGNSS linear regression in 2019 was significantly higher in high altitude areas.2019 SMAP average SM,CYGNSS random forest 2019 average SM in these regions is relatively close to the SMAP 2019 average SM,but in the coastal areas of Taiwan,Hainan,southern Yunnan and other regions,the difference between the SM values is relatively large;SM in each season for the spatial distribution of low and high values,the three types of SM are relatively close,especially CYGNSS random forest regression SM and SMAP SM.CYGNSS linear regression SM is worse than the two.In 2019,the R value of the daily CYGNSS linear regression SM can be as high as 0.8364,and the minimum is 0.5002.The average value of R throughout the year is 0.7225,with a large variation.The RMSE values are all around 0.08 cm~3/cm~3.The ratio is much more stable,and the correlation is high.The R value can be as high as 0.9359,and the minimum is 0.8025.The average value of R for the whole year is 0.8826,and the value of RMSE is around 0.05 cm3/cm3.(5)The ground station SM is used to analyze the time continuity of the two CYGNSS SMs.The ground stations that are within 18km from the CYGNSS SM point and whose correlation R(Pearson correlation coefficient)with CYGNSS SM is greater than 0.4 are selected.CYGNSS linear After the regression SM is compared with the nearest station SM,there are 55 ground observation sites that meet the requirements.After the CYGNSS random forest regression SM is compared with the nearest site SM,there are 243 ground observation sites that meet the requirements.Then,the R and RMSE of the two CYGNSS SMs compared with the same ground observation site SM were counted.A total of 48 observation sites participated in the comparison.The time span is 2018-2019.There are 48 random forest regressions between SM and SMAP SM.The R values of the points are all higher than the R values of the linear regression SM and SMAP SM,and the RMSE values of the 48 points between the random forest regression SM and SMAP SM are all lower than the RMSE values of the linear regression SM and SMAP SM.