Study On The Three-dimensional Water Vapor Tomography Method Combining GNSS And RS Data And Its Applications

In recent years,extreme weather events have occurred globally.China has also faced enormous challenges from extreme weather,such as typhoons and heavy rainfall.In 2021,there were precipitation extremes in the Henan region and autumn floods in the Yangtze river basin,which resulted in meteorological disasters,e.g.,severe urban flooding,flash floods,and geological landslides.These disasters have had a serious impact on the ecological environment,the economic society,and people’s properties and livelihoods.Therefore,it is urgent to develop a high-precision,omnibearing,and refined atmospheric water vapor monitoring technique,which can provide accurate and reliable data support for forecasting studies of extreme weather,having important application value and scientific significance in enhancing the prevention capability of dealing with extreme meteorological events.Due to the significant potential of atmospheric water vapor detection in high space-time,the Global Navigation Satellite Systems(GNSS)water vapor tomography technique has made a booming development.However,the existing water vapor tomography models still face the critical problem of spatial mismatch between the inverted cone-shaped GNSS signals and box-shaped three-dimensional(3D)tomography regions,which limits the retrieval accuracy of 3D tomographic atmospheric water vapor fields.Moreover,no studies have been conducted to develop rainfall forecasting tools based on tomography water vapor density(WVD)fields,which makes it difficult for 3D WVD products to be widely used in meteorological studies such as rainfall prediction.This dissertation focuses on the key problem of geometrical distribution defects in the GNSS water vapor tomography model,and systematically constructs the theory and method of GNSS and Remote Sensing(RS)water vapor tomography using the new generation GNSS and RS technologies,proposing a new GNSS-RS water vapor tomography technique.Besides,the positive cone-shaped RS-like virtual signals with high time resolution are innovatively constructed,and their application potential in the retrieval of high-quality and continuous 3D water vapor products has been deeply explored in the dissertation.Furthermore,the optimal solution algorithm of tomographic observation equations and the meteorological application of 3D water vapor tomographic products are investigated and developed.Based on deep learning technique,the dissertation initially establishes the rainfall forecasting models using 3D WVD fields,which builds a “technical bridge” between the water vapor tomography model and rainfall forecasting,and is further expected to promote the application potential of GNSS/RS technologies in meteorological monitoring and early warning.The dissertation performs research in several aspects of model optimization,algorithm improvement,and product application.The detailed work and research contents are outlined as follows:(1)With the breakthrough point of integrating GNSS and RS water vapor products,the dissertation first proposes an improved GNSS water vapor tomography algorithm constrained with RS precipitable water vapor(PWV)data.The algorithm detailedly explores the feasibility of incorporating RS PWV observations into GNSS water vapor tomography models,and then gives ideas for parameterizing RS PWV signals.GNSS data and RS water vapor products in the Xuzhou area are used to validate the proposed algorithm.Compared with the radiosonde profiles and the ERA5-derived 3D water vapor field,the improved algorithm can properly improve the accuracy of water vapor tomography results.However,due to the disadvantage of the RS PWV as constraints in the voxel-based tomographic model,the improvements in tomographic results are not significant.(2)To address the above problem that the RS PWV constraints cannot fully show its observational advantages,the dissertation introduces a new idea of retrieving RS slant water vapor(SWV)observations for the first time,which forms the approximately positive cone-shaped RS water vapor signals.The water vapor tomography method combining GNSS/RS SWV joint observations is then developed.Referring to the estimation strategy of GNSS SWV,this method first retrieves the atmospheric horizontal gradient terms of RS pixels,and then reconstructs the high-resolution RS SWV water vapor observations.The experimental results show that the proposed method can make full use of the spatial observation advantage of the GNSS/RS joint signals,with the average of observations and the mean number of punctured voxels of the 3D tomographic model increased by 32.50% and 33.78%,respectively.Compared to the traditional tomography method and the above optimal method with RS PWV constraints,the retrieval accuracy of the new method is improved by 32.11% and26.48%,respectively,which reveals that the new method has a good reconstruction capability for the 3D water vapor fields.(3)Taking into account the characteristics of high horizontal resolution RS water vapor products,the dissertation presents the node parameterization process of RS PWV/SWV observations based on the modeling advantages of the node-based tomographic model,and further develops the node-based GNSS/RS joint water vapor tomography algorithm.Both the node-based algorithm with GNSS as well as RS PWV and the node-based algorithm using GNSS as well as RS SWV can retrieve more accurate 3D water vapor tomographic products than the node-based algorithm only with GNSS observations.Furthermore,the performance of the tomographic solutions obtained by the two node-based algorithms is slightly better than that of the voxel-based tomography method using GNSS/RS SWV observations.(4)There are obvious time resolution differences between RS water vapor products and GNSS data,which restricts the meteorological application prospect of the optimal3 D water vapor tomography products obtained from the GNSS-RS water vapor tomography technique.To do this,the dissertation proposes an optimized water vapor tomography method combining GNSS and RS-like virtual signals.According to the retrieval idea of approximately positive cone-shaped RS SWV signals,the optimized method innovatively introduces the positive cone-shaped RS-like virtual signals with the same time resolution as GNSS signals.The effectiveness of the improved method in different types of GNSS station networks is fully validated using the dense GNSS network in the Hong Kong area and the sparse GNSS network in the Xuzhou region,respectively.Compared to the conventional GNSS tomography method,the optimized method substantially improves the crossing rate of tomographic voxels in both regions,and performs a significant improvement of the water vapor tomographic results,with an accuracy improvement rate of 18.18% and 38.28%,respectively.The proposed optimization method shows a higher contribution in improving the quality of water vapor tomographic results than the above new method using GNSS and RS water vapor observations.(5)Given the spatial distribution characteristics of atmospheric water vapor,an adaptive algebraic reconstruction algorithm(ART)for 3D water vapor tomography based on the dynamic error allocation principle and the weight matrix model of elevation angles is proposed.The adaptive algorithm constructs a new error allocation principle considering the variation of WVD values,which helps the corrections of the iterative process to follow the vertical distribution characteristics of the atmosphere.The experimental results in the Xuzhou area highlight that the adaptive ART algorithms have a significant advantage over the common ART algorithms in reconstructing highprecision 3D water vapor fields.In addition,a quantitative comparison shows that the solution accuracy of the proposed adaptive algorithms is improved by about 9% to 18%when adding the weight matrix model of elevation angles to the adaptive algorithms.(6)According to the high spatiotemporal detection of 3D tomographic water vapor products,a short-term rainfall prediction method based on the 3D tomographic WVD fields and deep learning is proposed.The dissertation preliminarily investigates the correlation between the variability characteristics of the vertical WVD observations as well as the 3D WVD fields and the rainfall evolution process.Besides,the ANN-driven short-term rainfall prediction model based on vertical water vapor information and the CNN-driven short-term rainfall prediction model based on water vapor images are established,respectively.The rainfall prediction results illustrate that the proposed two models can properly forecast rainfall events,with the probability of detection around60%,which basically demonstrates the significant potential of the 3D tomographic water vapor products in short-term rainfall forecasting.This dissertation has 91 figures,27 tables,and 234 references.