Combining ground- and space-based GNSS observations to mitigate data gaps in numerical weather prediction

Global Navigation Satellite System (GNSS) tomography is a robust technique used to estimate the amount and three-dimensional distribution of water vapour in the troposphere. This information is critical for numerical weather prediction (NWP), as water vapour is a highly variable atmospheric constituent that strongly influences weather processes. The technique relies on observations of GNSS signal delays, which are attenuated and slowed by atmospheric moisture as signals travel from satellites to ground-based receivers. However, the effectiveness of ground-based GNSS tomography is frequently hindered by ill-conditioned or mixed-determined systems, in which model elements become over- or under-determined due to continuously changing satellite geometry. As a result, significant data gaps arise, particularly in regions with sparse ground receiver coverage, such as oceans, deserts, or mountainous areas.

To address these limitations, recent research has focused on integrating space-based GNSS Radio Occultation (RO) observations into tomographic models. The RO technique involves Low Earth Orbit (LEO) satellites receiving GNSS signals that propagate nearly horizontally through the atmosphere, providing high-vertical-resolution profiles of refractivity, temperature, and water vapour. The growing importance of RO data is reflected in international efforts to increase occultation density, with recommendations calling for tens of thousands of daily observations to support NWP applications. In contrast to ground-based observations, which predominantly sample the atmosphere along near-vertical paths, RO measurements supply complementary horizontal information. This complementary geometry improves voxel filling within the tomographic grid and helps resolve the ill-posedness of the inversion problem.

Various tomographic grid parametrisation strategies have been developed to integrate ground- and space-based GNSS observations into a unified tomographic framework. In ground-based GNSS tomography, wet refractivity is estimated by relating it to the lengths of slant wet delay (SWD) ray-path segments within individual voxels. Ray-point coordinates and segment lengths are obtained by reconstructing signal paths using known transmitter and receiver positions through three-dimensional ray-tracing techniques. When combining different types of GNSS observations, the signal reconstruction strategy is observation-type dependent: three-dimensional ray tracing is applied to RO excess phase observations (Level 1a), whereas occultation point coordinates are directly provided for RO wet refractivity profiles (Level 2). Observation-specific uncertainty schemes can further be applied to improve solution robustness.

This study provides a generic assessment of key factors governing tomographic wet refractivity estimation, including ground network density, voxel filling rate, RO event availability, and uncertainty treatment. Results from integrated tomography approaches demonstrate that even a limited number of RO observations can substantially improve wet refractivity estimates, reduce reconstruction errors, and increase the number of filled voxels, particularly for sparse ground networks. Ultimately, the combined ground- and space-based GNSS products are well suited for assimilation into NWP models, enabling a more complete and reliable three-dimensional representation of atmospheric humidity.