Research On Water Vapor Retrieval And Its Application Based On Ground-Based GNSS And Satellite-Borne Remote Sensing Observations

Water vapor in the troposphere is converted from water evaporation in rivers,lakes,and seas.Although water vapor accounts for only 0.25%of the total atmospheric mass,it is the main source of precipitation and the main absorption gas of atmospheric radiation.In geodesy,water vapor content is also the major factor affecting the tropospheric wet delay.Accurate water vapor retrieval is not only important for environmental issues such as global hydrological cycle and climate change,but also can more accurately estimate the tropospheric wet delay in geodesy.Currently,there are two methods commonly utilized for observing water vapor:satellite-borne and ground-based.Satellite-borne observations include observations from satellite-borne remote sensing sensors measuring in various bands,with good spatial continuity,but poor accuracy and temporal resolution.Usually,satellite-borne remote sensing sensors can provide low-frequency and large-scale observation,which is considered as a measurement in the surface domain.Ground-based observations usually refer to ground-based GNSS observation,which has high accuracy and temporal continuity,but cannot provide observations covering a global scale.Therefore,ground-based GNSS observation is usually high-precision,high-frequency,and small-scale,and is considered as a point domain observation method.At present,there is a relatively mature PWV(Precipitable Water Vapor)retrieval algorithm for ground-based GNSS observations.There are various types of algorithms for retrieving PWV from satellite-borne remote sensing observations,which can be mainly divided into traditional physical model and neural network method.Traditional physical model is difficult to accurately deal with the non-linear surface parameters,while neural network model is easy to ignore parameters that can be linearized.Therefore,both methods have their own disadvantages and it is difficult to further improve the retrieval accuracy of a single algorithm.This article is based on satellite-borne and ground-based data to carry out satellite-ground joint PWV retrieval,as well as the study of integrated water vapor retrieval model based on machine learning superimposed on physical models,mainly including cross validation of the accuracy of GNSS PWV and Remote Sensing(RS)PWV,high-precision microwave remote sensing algorithms developed based on ground-based GNSS PWV,and geophysical parameters(such as PWV)responses before and after natural disasters observed by combined satellite-borne and ground-based observations.The main research contents and results are as follows:(1)Using AIRS PWV and ERA5 PWV to evaluate the accuracy of GNSS PWV derived from single C(BeiDou),single G(GPS),and G_C combination systems.Firstly,based on the original observations of 180 GNSS stations on the global scale that can receive BeiDou and GPS signals and the meteorological data provided by ERA5 products,the GNSS PWV for the entire month of May 2021 was calculated.The accuracy was evaluated using ERA5 and AIRS PWV.The evaluation results showed that BDS-3 had a similar retrieval performance to GPS,while the PWV retrieval accuracy of the G_C combination system was the highest,indicating that the addition of BDS-3 signals can improve the PWV retrieval performance of the GNSS system.(2)Deriving the microwave atmospheric radiation transfer model on the land and proposing a new microwave remote sensing PWV retrieval algorithm combining traditional physical model and neural network model.Firstly,based on the traditional microwave atmospheric radiation transfer model,a theoretical linear equation was derived,and the consistency between the theoretical equation and the experimental fitting equation was proved.On this basis,based on the global SuomiNet GNSS stations and microwave sensor AMSR2 data,a new microwave remote sensing integrated algorithm was proposed that combined the advantages of traditional physical models for processing linearized parameters with traditional neural network models for processing nonlinear parameters.Compared with the traditional physical models and traditional neural network models,the accuracy of the integrated model was improved by about 20%-30%,which meant that the performance of the integrated algorithm was significantly better than traditional models.(3)A spatial PWV retrieval algorithm for microwave remote sensing was proposed,which can predict PWV values on the untrained GNSS sites.Firstly,based on the dense SuomiNet GNSS stations in North America and microwave sensor AMSR2,a spatial PWV retrieval model for microwave remote sensing was established using neural network method.Some stations were randomly selected as test set,while the remaining stations were utilized as training set.The accuracy evaluation of the spatial model using GNSS PWV showed that the RMSE of the test set can reach 3.90 mm in ascending orbit and 3.88 mm in descending orbit,the correlation coefficients R were 0.85 and 0.88,respectively,which proved that the spatial model had good prediction performance for untrained stations.Further research on the spatial retrieval accuracy of the model revealed that the spatial model had relatively high accuracy on a single continuous land cover type,but had poor accuracy on mixed land cover types.This was because a single continuous land cover type had stable physical properties,while mixed land cover types had variable physical properties(such as microwave emissivity parameters).The changeable microwave emissivity and other physical properties made the microwave atmospheric radiation transmission process greatly disturbed,which leaded to poor retrieval accuracy in this area.(4)A method for comprehensively characterizing the response of geophysical parameters before and after the eruption of Tonga volcano using low-frequency largescale satellite-borne observations and high-frequency small-scale ground-based GNSS observations was proposed.Based on the multi-source data from the GEO meteorological satellite GOES-R and ground-based GNSS stations around the Tonga volcano,this article comprehensively analyzed the changes of geophysical parameters such as tropospheric PWV,ionospheric TEC(Total Electron Content),sea surface SST(Sea Surface Temperature),and land surface deformation before and after the eruption of Tonga volcano,both at low-frequency large-scale and high-frequency small-scale.The study found that for both the troposphere and ionosphere,there was a process of accumulation of PWV and TEC parameters before the eruption.As TVE(Eruption of the Tonga Volcano)events released a large amount of energy,these parameters rapidly decreased and eventually returned to normal levels.In response to the surface deformation,TVE can cause huge plate compression and movement.During the most intense period of TVE,the AU plate tended to compress the PA plate towards the northeast,the maximum displacement of GNSS station can reach 46 cm.After the calm of TVE,the AU plate and the PA plate tended to separate from each other.