Tropospheric Delay Modeling Based on Multi-source Data Fusion and Machine Learning Algorithms

摘要

Machine learning algorithms have been widely applied in various fields, including the research of the tropospheric delay of Global Navigation Satellite System (GNSS) signal. In this paper, back-propagation neural network (BPNN), radial basis function (RBF) neural network, and least square support vector machine (LSSVM) algorithm are applied to develop the regional zenith troposphere delay (ZTD) models by merging International GNSS Service ZTD products (GNSS-ZTD) and European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5) data over North America throughout 2020. Among them, two data-fusion strategies are designed with different input parameters of modeling, including the fusion of ERA5 meteorological parameters with GNSS-ZTD and the fusion of ZTD estimations from ERA5 data (ERA5-ZTD) and GNSS-ZTD. With ZTD derived at 77 IGS stations, the accuracy of regional ZTD models is verified by month, as well as the stability and efficiency. The results show that the effect of RBF is best when modeling ZTD with small-size training samples. Among them, the average RMSE of the ZTD models with RBF is 20.8 mm and 20.1 mm for two data-fusion strategies, respectively. The accuracy of RBF is improved by 40.4% and 38.5% over the BPNN in modeling ZTD. The result of the model using the LSSVM is close to the RBF. Moreover, the BPNN has an obvious advantage in modeling ZTD with large-size training samples.