Multi-Frequency And Multi-Track GNSS Remote Sensing Soil Moisture Fusion Inversion

Soil moisture regulates the interaction between the surface and atmosphere,thereby affecting climate and weather.Quickly and accurately obtaining soil moisture content in a certain area plays a very important role in fields such as agriculture and forestry.In recent years,the Global Navigation Satellite System Reflectometry has rapidly developed and become a research hotspot in remote sensing for earth observation.At present,the ground-based GNSS-R soil moisture inversion method has problems such as weak model adaptability and insufficient utilization of multi-system GNSS reflection signal data.This thesis mainly studies from three aspects based on the observation data of the Medium Earth Orbit,Geostationary Orbit,and Inclined Geosynchronous Orbit satellites in multi-system,One aspect is the adaptability of the model: combining optimization algorithms and deep learning algorithms to establish a soil moisture inversion model,reducing the impact of nonlinear factors on different underlying environments while improving the accuracy and adaptability of the model inversion;The second is to weaken the vegetation interference: on the basis of the correlation between the characteristic parameters and the vegetation biomass,vegetation growth state,the reflection signal data of MEO,IGSO and GEO are integrated into the vegetation information to further enhance the robustness of the inversion model;The third aspect is data fusion: comprehensively utilizing data from multiple GNSS systems,multiple orbits,and multiple frequency bands,using data quality analysis,trajectory data screening,and clustering methods,introducing multiple weighting methods,and establishing a multi-system,multi frequency,and multi orbit GNSS data fusion inversion model to solve the problem of low utilization rate of multisource GNSS data caused by inconsistent satellite revisit cycles.The specific research content is as follows: The study of the thesis includes:(1)The method of establishing soil moisture inversion model based on adaptive deep learning is studied.Aiming at the problems of low modeling accuracy and weak environmental adaptability of traditional soil moisture inversion model,the sensitivity of GNSS reflected signal characteristic parameters to soil moisture changes is analyzed,and appropriate reflected signal characteristic parameters are selected.The Convolutional Neural Network’s advantages of strong self-learning ability and high modeling accuracy are fully utilized to establish a soil moisture inversion model,which alleviates the influence of the underlying surface changes to a certain extent.In addition,by combining with Particle Swarm Optimization,which has the characteristics of fast parameter optimization speed and high stability,the adaptability of the deep learning model is further improved.The model is verified by three test sites of the Plate Boundary Observatory in the United States.The experimental results indicate that the characteristic metrics of reflected signal can effectively reflect the change of soil moisture content,and the soil moisture inversion model integrated by PSO and CNN can weaken the coupling effects of multiple factors and effectively improve the accuracy of soil moisture inversion.At the same time,the inversion accuracy of soil moisture under different vegetation coverage and relief sites is effectively improved by the PSO-CNN inversion model,which verifies the environmental adaptability of the model.(2)The method of MEO/IGSO/GEO fusion inversion model considering vegetation information is studied.A satellite with a revisit period of 1 day is selected to increase the types of satellite data of different orbital altitudes.The soil moisture retrieval capability of different orbital satellites is analyzed,and the inversion model of MEO/IGSO/GEO data fusion is constructed considering that the amplitude could reflect the vegetation change state.The IGSO and GEO reflected signal data incorporated into the method in this thesis improved the reliability and stability of the model.The experiment is carried out with 31 days of measured site data.The experimental results indicate that the characteristic parameters of the reflected signals of satellites with different orbital altitudes can reflect the variation of soil water content.The inversion accuracy of the multiple linear regression model based on MEO/IGSO/GEO fusion is improved compared with the traditional multiple linear regression model,and the inversion accuracy is also improved when the amplitude data is added into the input set of the deep learning model.(3)The multi-system MEO combined soil moisture estimation method based on track clustering is studied.In view of the ground soil moisture retrieval requirements of sites without in-situ moisture modeling data,clustering is carried out on the trajectory of satellite reflection points with a long return visit period,and the phase information of reflection signals is calculated.All estimated values within a day are equal-weighted,and the inversion capabilities of soil moisture in different frequency bands of different systems are compared and analyzed.Finally,the objective weighting method is used to determine the system weight to realize the soil moisture estimation of the multinavigation satellite system combination.The method in this thesis makes full use of the reflected signal data of MEO satellites of different systems to improve the utilization rate of GNSS data.GNSS sites provided by IGS are selected to carry out experiments.The experimental results indicate that for the soil moisture retrieval performance of a single system,the accuracy of BDS and GALILEO inversion is equivalent to but higher than that of GPS and GLONASS inversion,and the weight determination method using Bayesian triangulated hat has the best effect,while the combination scheme of GPS,BDS and GALILEO is the best.