Satellite Navigation Signals Turned Into a Storm Early-Warning System for Rain

Every satellite navigation receiver on Earth is quietly doing meteorology. As GPS, BeiDou, GLONASS and Galileo signals travel from orbit to the ground, they pass through the atmosphere, and the water vapor in that atmosphere delays them by tiny but measurable amounts. A new study published in Frontiers in Astronomy and Space Sciences shows that these delays, harnessed through an integrated modeling framework, can be sharpened into a genuine severe-weather tool — one that was tested against one of the most devastating rainstorms in modern Chinese history, the Zhengzhou ‘7·20’ deluge of July 2021, which dropped a record 374.3 millimeters of rain in a single day.

The research, led by Ying Liu of the Beijing Institute of Tracking and Telecommunications Technology together with Yuze Yang of Shanghai Astronomical Observatory and colleagues, tackles three stubborn weaknesses in the way scientists currently retrieve precipitable water vapor, or PWV, from satellite navigation data. PWV measures the total mass of water vapor stacked in a vertical column of air above a given point, and it is one of the most important variables in short-term forecasting of heavy rainfall. The trouble is that the standard retrieval technique, known as Precise Point Positioning or PPP, treats the atmosphere as if it were perfectly uniform above each station — and it only describes the sky directly over that one station, leaving forecasters blind to the regional picture.

That first assumption breaks down exactly when it matters most. During severe convective weather, water vapor is distributed highly unevenly across the sky, and the resulting horizontal gradients in tropospheric delay are far too large to ignore. The team therefore built an enhanced multi-GNSS PPP model that explicitly estimates tropospheric gradient parameters in both the north–south and east–west directions, updated as a piecewise linear function every two hours. When validated against roughly 120 stations from the International GNSS Service MGEX network, the gradient-enhanced solution reduced the mean bias of the retrieved zenith wet delay by 0.8 to 1.1 millimeters — an improvement of about 40 percent — and cut the root-mean-square error by 0.4 to 0.8 millimeters, up to 8.8 percent for the multi-constellation solution.

Combining four satellite constellations provided its own gains. Multi-GNSS processing adds many more observations to the Kalman filter that underpins PPP, stabilizing the wet-delay estimates, although the study found that BeiDou solutions were still about two millimeters less consistent with the IGS reference products than GPS, largely because the orbit modeling of BeiDou’s geostationary and inclined geosynchronous satellites remains less mature. Independent confirmation came from Beijing Miyun Station, where colocated GNSS equipment and a water vapor radiometer produced PWV time series correlating at up to 98.59 percent — remarkable agreement between two completely different measurement technologies.

But a single station, however accurate, sees only its own patch of sky. To extend the picture across an entire region, the researchers turned to a regional modeling approach called the Modified Optimal Fitting Coefficients, or MOFC, model. Unlike simple polynomial fits, MOFC includes an exponential height-dependent term that captures the physical reality that water vapor thins rapidly with altitude. Using zenith wet delays from a network of reference stations, the model’s coefficients are estimated on a server side, encoded in the RTCM broadcast standard, and sent to users, who can then compute their local wet delay from the coefficients alone. In the Zhengzhou test, the regional model achieved a zenith wet delay accuracy of about 1.83 centimeters, corresponding to PWV accuracy of roughly 2.97 millimeters — sufficient to track the moisture swings that drive heavy rain.

The regional model turned out to have a bonus benefit that has nothing to do with weather. When the modeled wet delay was fed back into PPP processing as a tight constraint, the convergence time for ambiguity resolution — the step that unlocks centimeter-level positioning — dropped from 0.53 minutes to 0.33 minutes on average, with the east and up components improving most. In other words, the same atmospheric correction that improves rain monitoring also makes precise positioning faster, a two-for-one result that will interest surveyors, autonomous vehicles and anyone who depends on rapid high-precision satellite positioning.

The third and arguably most consequential innovation addresses a long-standing frustration: knowing how much water vapor is in the air does not tell you when it will actually fall. PWV typically climbs before rainfall and drops afterward, but moisture alone is not sufficient. Precipitation requires not just abundant vapor but saturation — air cooled to the point where it can no longer hold its water. The study therefore proposed a joint judgment criterion combining PWV with air temperature and dew point temperature, the latter being the temperature to which air must be cooled at constant pressure for saturation to occur.

The framework faced its ultimate test during the Zhengzhou rainstorm of July 2021. Analysis of eight continuous GNSS reference stations showed the wet delay and PWV climbing steadily from July 15 onward, peaking exactly on July 20 when rainfall hit its maximum, and then declining through July 25. Precipitation occurred only when PWV exceeded about 60 millimeters — but that threshold alone was not the whole story. On July 16, the air temperature approached the dew point, signaling near-saturation with a temperature–dew point depression below one to two degrees Celsius, yet no rain fell because the moisture reservoir was still too shallow. Only when both conditions coincided — PWV above 60 millimeters and near-saturation — did the catastrophic rainfall begin. The joint criterion correctly captured the onset while avoiding false alarms.

The authors are careful to note that the 60-millimeter threshold is specific to the Zhengzhou event and to its mid-latitude summer climate; other regions and seasons will demand different values. The real contribution is the physical logic of the combined criterion: PWV supplies the moisture background, while the convergence of temperature and dew point indicates that the air is ready to release it. The researchers also acknowledge that their regional model validation relied on a single test station in a relatively flat area, and that performance in mountainous terrain or different climate regimes remains to be confirmed.

Even with those caveats, the significance of the work is hard to overstate. Ground-based GNSS stations already blanket much of the planet, delivering all-weather, around-the-clock measurements at a fraction of the cost of radiosonde balloons, which launch only twice daily, or water vapor radiometers, which fail precisely when it rains because liquid water scrambles their microwave signals. By folding horizontal gradients into the retrieval, stitching station measurements into a regional field, and pairing moisture with saturation conditions to time the rain, this framework converts the planet’s navigation infrastructure into a dense, continuous, real-time atmospheric observatory. For a world facing increasingly extreme rainfall, the satellites that guide us may also become our earliest warning that the sky is about to open.

Subject of Research: GNSS-based precipitable water vapor retrieval and precipitation monitoring using tropospheric gradient estimation and regional tropospheric modeling

Article Title: Enhanced precipitable water vapor inversion and rainstorm monitoring based on GNSS PPP regional tropospheric modeling

Article References: Enhanced precipitable water vapor inversion and rainstorm monitoring based on GNSS PPP regional tropospheric modeling. (n.d.). https://doi.org/10.3389/fspas.2026.1954361

Image Credits: AI Generated

DOI: 10.3389/fspas.2026.1954361

Keywords: GNSS, precipitable water vapor, precise point positioning, tropospheric gradients, MOFC model, zenith wet delay, rainstorm monitoring, Zhengzhou 7·20 rainstorm, dew point temperature, PPP ambiguity resolution, severe weather forecasting, multi-GNSS