Enhancing multi-GNSS positioning performances in harsh environments via a refined joint troposphere-multipath hemispherical map

Tropospheric delay and multipath effect are two key errors that are difficult to be accurately corrected in Global Navigation Satellite System (GNSS) Precise Point Positioning Ambiguity Resolution (PPP-AR). The tropospheric residuals due to imperfect modeling can be significant in harsh environments, like in regions with complex terrain and during extreme weather. As the tropospheric delay and multipath effect are coupled in the unmodeled errors, the tropospheric residuals could be misunderstood as multipath. Therefore, accurate correction of the tropospheric delay is crucial for estimating the multipath. However, the coupling effect was not properly considered in previous studies. We propose a refined joint troposphere-multipath hemispherical map (TMM), by constructing a refined troposphere hemispherical map (THM) and an improved multipath hemispherical map (C-TMHM). We use ray-tracing and meteorological data to construct THM correction tables, while tropospheric delays in PPP-AR are corrected by retrieving the corresponding satellite tropospheric delay estimates to obtain "cleaner" multipath C-TMHM model values. Results show that the tropospheric THM model reduces GNSS residuals from about 10-2 mm at low-elevation (7 degrees-30 degrees) compared to the Vienna Mapping Functions 3 (VMF3). Because that the topographic complexity and the rapid variations in atmospheric water vapor are not adequately considered by simply using the elevation-dependent mapping function and horizontal gradients. In particular, part of the tropospheric residuals in the low-elevation are likely to be misinterpreted as multipath. Compared with multi-GNSS PPP-AR positioning performance using traditional model (VMF3 and uncorrected multipath), the proposed TMM (THM and C-TMHM) model improves the positioning accuracy by 32.12% and 36.18% under the cases of complex terrain and extreme weather, respectively, while shortens the convergence time by 33.04% and 30.7%.