Precipitable water vapour mapping from GNSS tropospheric delay
Ground-based GNSS networks convert tropospheric signal delay into precipitable water vapour at sub-hourly intervals, feeding numerical weather prediction with observations that radiosondes and satellites alone cannot match in density or timing.
Sensors
- IGS Global GNSS Network: Roughly 500 continuously operating reference stations worldwide, processing GPS, GLONASS, Galileo and BeiDou signals. Zenith total delay (ZTD) products are available at 5-minute intervals with a formal precision of around 4 mm in ZTD, translating to approximately 0.5–1 mm PWV. Ultra-rapid ZTD estimates are published within 3–9 hours; rapid products within 17 hours.
- EUREF Permanent GNSS Network (EPN): Over 300 stations across Europe and adjacent regions, with station spacing typically 50–200 km. Provides near-real-time ZTD solutions at 1-minute sampling for operational meteorology in several national weather services. Dense enough to resolve mesoscale moisture gradients relevant to convective initiation.
- CYGNSS (NASA, 8-satellite constellation): Launched 2016. Uses GNSS-Reflectometry to measure ocean surface roughness, from which near-surface wind speed is inferred. Wind speed correlates with surface latent heat flux and boundary-layer moisture, giving an indirect constraint on oceanic water vapour transport where no ground stations exist. Revisit is 2.8 hours median over the tropics (38°S–38°N).
- Spire GNSS Radio Occultation Constellation: More than 100 small satellites in low Earth orbit, each receiving GPS/GNSS signals through the atmospheric limb. Occultation geometry yields bending-angle and refractivity profiles; the lowest tropospheric levels constrain moisture in the 0–5 km layer. Spire delivers roughly 10,000–15,000 occultations per day globally. Covered separately in the radio occultation sibling page; noted here because the two techniques are routinely combined in data assimilation.
What a delayed signal actually measures
A GNSS signal travelling from satellite to ground antenna arrives slightly later than it would in a vacuum. That delay has two tropospheric components. The hydrostatic delay, typically 2.0–2.3 m in zenith equivalent path length, is caused by dry gases and is predictable to better than 1 mm from surface pressure alone using the Saastamoinen model. The wet delay, usually 0.01–0.35 m, is caused by water vapour and is far less predictable. It is the wet component that carries meteorological information.
Separating the two requires a surface pressure observation at or near the GNSS antenna. Given that, the zenith wet delay (ZWD) is computed by subtracting the modelled hydrostatic delay from the estimated zenith total delay. A dimensionless conversion factor, Pi (roughly 0.15 for mid-latitudes, varying with atmospheric temperature), then converts ZWD to precipitable water vapour in millimetres of equivalent liquid column. The full chain introduces an uncertainty of roughly 1–2 mm PWV under good conditions, rising toward 3–4 mm when surface pressure is interpolated from distant synoptic stations rather than measured on-site.
Mapping functions and the low-elevation problem
A receiver tracking satellites at high elevation angles sees a short, nearly vertical path through the troposphere. At 5° elevation, the path length is roughly 10 times longer, amplifying both signal and error. Mapping functions such as GMF (Global Mapping Function) and VMF3 (Vienna Mapping Function 3) project slant delays to the zenith, but their accuracy degrades below about 7–10° elevation because the atmosphere is not spherically symmetric and ray bending becomes significant.
In practice, most geodetic processing software applies a cut-off at 7° or 10° elevation. This is a genuine limit, not a software choice. Stations in complex terrain face additional multipath and diffraction problems that inflate ZTD scatter. The consequence for PWV mapping is that horizontal resolution is governed by station spacing, not by the number of satellites tracked. Dense regional networks like EPN can resolve features at 50–100 km scale; the global IGS network leaves gaps of several hundred kilometres over oceans, central Africa, and the polar regions where no permanent stations exist.
Assimilation into numerical weather prediction
The operational value of GNSS PWV became clear during the 1990s GPS-STORM campaign in the United States and was formalised in EUMETNET's E-GVAP programme, which has fed near-real-time ZTD data to European NWP centres since 2004. ECMWF, Météo-France, the UK Met Office and several other centres now assimilate GNSS ZTD directly. Impact studies consistently show improvements in short-range precipitation forecasts, particularly for convective events where rapid moisture build-up precedes initiation by 1–3 hours.
The mechanism is straightforward. Radiosondes launch twice daily at most and are sparsely distributed. Satellite microwave sounders provide good global coverage but are degraded under heavy cloud and have coarser vertical sensitivity in the lowest 2 km. GNSS PWV is insensitive to cloud, operates continuously, and its latency in near-real-time mode is under 15 minutes for well-connected stations. That combination makes it disproportionately useful for the 0–6 hour forecast window where conventional observations are least dense.
Where the ocean gap bites, and what GNSS-R offers
Over the open ocean, the absence of ground stations is a hard constraint. No amount of processing sophistication recovers a ZTD estimate without a receiver on the surface. GNSS-Reflectometry missions such as CYGNSS partially address this by inferring ocean surface wind speed, which constrains boundary-layer moisture fluxes and provides an indirect input to NWP moisture analysis. The relationship is physical but indirect: wind speed drives evaporation, and evaporation drives column moisture, but the connection involves assumptions about sea-surface temperature and atmospheric stability that introduce their own uncertainties.
Ship-borne and buoy-mounted GNSS receivers offer a more direct solution in principle, and several research campaigns have demonstrated PWV retrieval from vessels with accuracy comparable to land stations. Operational deployment remains limited. The practical result is that GNSS PWV mapping is genuinely excellent over continental Europe, Japan, North America and parts of East Asia, patchy over the tropics and southern hemisphere, and essentially absent over most of the Southern Ocean and Arctic Ocean. Users planning applications in those regions should treat GNSS PWV as a complement to microwave sounder retrievals rather than a replacement.
From raw ZTD to an operational PWV product
A production pipeline has four stages: GNSS data collection and quality screening, precise point positioning or network adjustment to estimate ZTD, meteorological ancillary data ingestion for the hydrostatic correction, and spatial interpolation or kriging to produce gridded PWV fields. Each stage has latency implications. Ultra-rapid IGS orbits and clocks are available within 3–9 hours; if a user needs sub-15-minute latency, they must rely on predicted orbits with somewhat larger errors in ZTD of around 6–8 mm versus the 4 mm achievable with final products.
Gridded PWV fields at 0.25° or finer resolution are routinely produced over Europe using EPN data and over Japan using the GEONET network of roughly 1,300 stations, the densest national GNSS network in the world. GEONET's density (average station spacing around 25 km) allows it to resolve PWV gradients associated with the Baiu front and typhoon moisture inflow at scales that no other observing system can match over land. Satellize can ingest third-party ZTD streams or processed PWV grids and integrate them into client weather-risk or infrastructure-monitoring workflows, as it does with open satellite data streams in programmes such as the Tonga crop-estimation work.
Archive depth for IGS ZTD products extends to 1994, giving a 30-year climatological baseline. This is long enough to compute PWV anomalies relative to a stable mean and to detect multi-decadal moistening trends consistent with a warming atmosphere, a signal that has been documented in several regional studies using EPN and GEONET data.
Typical figures
| ZTD formal precision (IGS final products) | ~4 mm, equivalent to ~0.5–1 mm PWV |
| Temporal sampling | 1–5 minutes (station-dependent); near-real-time products typically 5 min |
| Latency (near-real-time mode) | <15 minutes for well-connected EPN/IGS stations |
| Latency (IGS ultra-rapid products) | 3–9 hours |
| Spatial resolution (gridded product) | Governed by station spacing: ~25 km (GEONET), ~50–200 km (EPN), several hundred km (global IGS) |
| Minimum detectable PWV change | ~1–2 mm under optimal conditions; ~3–4 mm with interpolated surface pressure |
| Elevation angle cut-off (processing) | Typically 7–10°; mapping function errors increase sharply below this |
| Archive depth (IGS ZTD) | 1994 to present (~30 years) |
| GNSS signals used | GPS L1/L2, GLONASS, Galileo E1/E5, BeiDou B1/B2 (multi-constellation improves ZTD by ~10–15%) |
| Coverage gaps | Open ocean, central Africa, polar regions: no ground stations; GNSS-R provides indirect constraint only |
Analytics Satellize can run
| Near-real-time PWV time series per station | Precise point positioning (PPP) ZTD estimation with VMF3 mapping function and on-site pressure correction | JSON or CSV feed, sub-15-minute latency, per-station uncertainty flag included |
| Gridded PWV analysis field | Ordinary kriging or variogram-based spatial interpolation of station PWV values, with orographic correction using digital elevation model | NetCDF or GeoTIFF grid at client-specified resolution, hourly or 3-hourly cadence |
| PWV anomaly map relative to 30-year climatology | Comparison of current PWV to IGS archive climatological mean and standard deviation for the same calendar week | Standardised anomaly GIS layer with threshold alerts at ±1.5 and ±2 sigma |
| Convective moisture index for severe-weather early warning | Rate-of-change of column PWV over 30–60 minute windows; threshold exceedance correlated with convective initiation studies | Automated alert with station location, PWV tendency (mm/hr), and confidence tier |
| Multi-year PWV trend analysis | Linear regression on deseasonalised monthly PWV anomalies from IGS archive; Mann-Kendall significance test | PDF report with trend maps, station-level trend coefficients and p-values |
| NWP assimilation-ready ZTD dataset | SINEX or BUFR-formatted ZTD with full covariance, formatted to EUMETNET E-GVAP standard | Operational data feed compatible with ECMWF and WMO-standard NWP ingest pipelines |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.