New publication
"High-resolution tropospheric refractivity fields by combining machine learning and collocation methods to correct earth observation data" by Shehaj et al. (2022)
Modeling tropospheric effects in GNSS signals has been addressed for decades through empirical models (low-cost receivers) or parameter estimation (for high accuracy applications). Previously, we have successfully applied machine learning to predict time series of GNSS tropospheric delays at known geodetic locations; the difficulty arises when we predict the delays at new locations. At our institute we have utilized least-squares collocation to interpolate in space tropospheric parameters using simplified physical models.
In this work, we propose a combination where we exploit their complementary characteristics. We use neural networks to predict time series of tropospheric delays and collocation to produce high resolution fields of tropospheric delays and/or refractivity. For the entire territory of Switzerland, we report and accuracy of 1-2 cm.
The paper can be found: external page https://doi.org/10.1016/j.actaastro.2022.10.007.
