Study On Soil Moisture Retrieval Method Based On GNSS-R Technology

Soil moisture is an important research object in many scientific fields as an environmental factor in ground dynamic monitoring.It also regulates the exchange of water between the ground and the atmosphere,controlling runoff and erosion caused by rainfall.In climate research,soil moisture is usually analyzed with surface temperature and precipitation as characteristics.In addition,soil moisture is an important indicator of plant growth and crop yield in agriculture and other fields,which has an important impact on human social activities and economy,so it is of great significance to study the theory and method of soil moisture detection.Global Navigation Satellite System reflection(GNSS-R)measurement technology has been widely used in ocean altimetry,sea ice detection and other remote sensing fields due to its advantages of all-weather,high spatio-temporal coverage and low cost.The L-band used by GNSS-R is sensitive to soil moisture changes,showing unprecedented advantages in soil moisture inversion.Based on CYGNSS data and SMAP data,the theory and method of soil moisture retrieval based on satellite-borne GNSS-R technology were studied in this paper.In this paper,based on the application of GNSS-R in soil moisture inversion at home and abroad,the theoretical principle of GNSS-R technology was studied,including the introduction of GNSS-R signal source,geometric structure of reflected signal,polarization characteristics and correlation function analysis.On this basis,the principle of GNSS-R soil moisture inversion is introduced,and two inversion modes of single antenna and double antenna are introduced.Then,the GNSS-R soil moisture inversion model was analyzed,including the calculation and correction process of reflectance.Then,two space-borne GNSS-R soil moisture inversion methods were studied as follows:(1)Satellite-borne GNSS-R soil moisture retrieval method based on phase model modification.Firstly,surface reflectance parameters were extracted from CYGNSS data,and combined with auxiliary information such as vegetation optical thickness,surface roughness and temperature extracted from SMAP data,a theoretical model of soil moisture inversion was initially constructed,and a precise mathematical model of soil moisture inversion was determined by using neural network model.Then,the soil moisture obtained by this model was set at 0.35cm~3/cm~3 as the boundary point,and the proposed stage function model was used to improve the inversion accuracy.The global satellity-borne GNSS-R soil moisture was obtained by using CYGNSS data from October 2018 to May 2019.Finally,the effectiveness of the proposed satellite-borne GNSS-R soil moisture inversion method was evaluated by comparing with the soil moisture data provided by SMAP,and the obtained satellite-borne GNSS-R soil moisture was analyzed in time series.The results showed that the variation trend was consistent.And the mean root mean square error of the statistical test data is 0.07 cm~3/cm~3.(2)A space-borne GNSS-R soil moisture inversion method based on Stacking fusion model.The Stacking model and related machine learning algorithms used are introduced,and the inversion process including data denoising,data matching and training model construction is described in detail.CYGNSS data and SMAP data from October 2019 to September 2020 were used to analyze the characteristic parameters and verify the optimal input feature set on the test data through different combinations.The influence of the partitioning of the training data set on the inversion results is explored to determine the best training data set.After determining the optimal input feature set and the partitioning of the training data set,the inversion model is trained using Stacking models.Finally,the inversion results were evaluated on the test set,and the results showed that CYGNSS soil moisture was consistent with SMAP soil moisture characteristics in the global range.The mean deviation was 0.047 cm~3/cm~3,the mean root mean square error was 0.068 cm~3/cm~3,and the correlation coefficient was 0.8789.