An InSAR Atmospheric Correction Method Based on GNSS and Machine Learning

 Objectives:   The InSAR technique has been widely applied to monitoring land deformation with very high spatiotemporal resolution. The elimination of atmospheric delay in the interferometric phase is critical to achieving higher accuracy of land deformation monitoring with the technique.   Methods:   Hence this study proposes a GNSS atmospheric delay modeling method based on the combination of the machine learning approach and the K-means algorithm. In the experiment on the performance evaluation of the proposed, two sentinel-1 short-time baseline interferograms in southern California for summer 2019 and winter 2020 are used, respectively.   Results:   The results after the correction of tropospheric delays show that: (1) The root mean square error (RMSE) of the interferometric phase was reduced by an average of 78%, which is significantly better than the conventional methods, including thepower law function (highest 73%), GACOS estimation (61%), and weather models (58%); (2) The average correlation between the phase and elevation greatly decreased from 0.56 to 0.23.   Conclusions:   The experiment results indicate that the proposed method can effectively reduce the stratified tropospheric delay, and the modeled tropospheric delays are robust to surface deformation. Besides, the proposed method is potentially used in areas with low spatial density GNSS stations through simulation experiments. This study can provide a reference for InSAR tropospheric delay spatiotemporal mapping based on GNSS atmospheric data.