| The spatial and temporal distribution and changes of atmospheric water vapor are an important piece of information in meteorological applications.On the one hand,conventional water vapor detection technology cannot perform all-weather observations and the sparse distribution of detection sites results in low water vapor observation accuracy and spatial and temporal resolution;It is expensive and cannot be deployed on a large scale.Due to the above-mentioned defects of conventional precipitation water vapor detection technology,it is difficult to obtain high-precision continuous precipitation water vapor observation values using conventional water vapor detection technology.With the rapid development of Global Navigation Satellite Systems(GNSS),countries around the world have successively established high-density ground-based GNSS observation networks.GNSS remote sensing water vapor technology as GNSS expansion application has also been greatly developed.The GNSS reference station network deployed by various countries to remotely sense the precipitation water vapor content in the troposphere has the advantages of all-weather,high spatial and temporal resolution and low cost,these advantages make GNSS remote sensing water vapor technology one of the important means in the field of water vapor detection.GNSS remote sensing water vapor can be divided into precipitation inversion and water vapor tomography technology.At present,the main problems of water vapor inversion and GNSS water vapor tomography are: how to choose the best solution strategy for atmospheric precipitable water under the influence of multiple variable factors;The traditional BP neural network predicts the atmospheric precipitation are easy to make the prediction model iteratively unstable and local minimum;the commonly used equal layering in the height direction when discretization of water vapor tomography is inconsistent with the actual distribution of water vapor in the vertical direction,reducing the accuracy of water vapor tomography results.In order to solve the above-mentioned problems of GNSS water vapor detection technology,this paper has done corresponding research on the analysis of the best solution strategy of water vapor inversion,water vapor prediction and non-uniform stratified water vapor tomography.Therefore,the main research results of this article are:(1)In view of the fact that there was little research on the effect of mapping function on the inversion of atmospheric precipitation and the influence of various variables on the inversion of water vapor accuracy of the mapping function was not considered,three mapping functions of NMF,VMF1,and GMF at different altitude angles were studied.The accuracy of the model was calculated by seasons and the seasonal variation characteristics of the three types of mapping functions under different constraints were analyzed.The test results show that the accuracy of inversion of atmospheric precipitation in different seasons was quite different,and the accuracy of inversion in autumn and summer was low,the inversion accuracy of different stations varies greatly.Taking into account the baseline calculation error and the invertible precipitation accuracy,the GMF mapping function model has the highest accuracy in the GNSS water vapor inversion test in Hong Kong,and the 10° altitude angle was selected as the satellite cutoff altitude angle.(2)Aiming at the shortcomings of the traditional neural network to predict the precipitable water use of random initialization parameters tends to make the prediction model computationally large and iteratively unstable.An improved BP neural network algorithm was proposed to predict atmospheric precipitable water,this algorithm is based on the historical data of water vapor Iteratively obtain the initialization parameters with better effects,and substitute the initialization parameters with better iteration effects into the neural network for modeling and prediction.Verification through experiments: The improved neural network algorithm has higher prediction accuracy than the traditional neural network algorithm in the sounding station water vapor prediction test and GNSS water vapor prediction test.(3)In-depth research on the theory and method of water vapor tomography technology,introduced the slope path wet delay and horizontal gradient function model calculation method.Discretize the tomographic test area and establish the tomographic observation equation.Taking into account the uneven distribution of the wet refractive index and the vertical distribution of water vapor parameters in the discretized grid,different types of constraints were added to the tomography equation to form the tomography equation group,and the method of solving the tomography equation system was introduced.Since the conventional tomography modeling uses uniform stratification along the vertical direction of the ground which does not meet the law of water vapor parameters decreasing exponentially in the vertical direction,an improved stratification method was proposed which is to perform non-uniform stratification according to the vertical distribution characteristics of water vapor Layered equally.The experimental results show that: compared with the results of uniform layered tomography,the wet refractive index parameters obtained by non-uniform layering were closer to the change curve of the measured values,and the correlation coefficient and tomographic accuracy were higher than that of uniform layering,which effectively improves Accuracy of water vapor tomography results.And according to the plane distribution map of water vapor parameters obtained by tomography,the characteristics of water vapor space-time changes were analyzed. |
PDF Full Text Request