Soil moisture retrieval from GNSS reflectometry
Reflected GPS and GLONASS signals carry a measurable imprint of surface dielectric constant, allowing near-surface volumetric soil moisture to be retrieved at basin scale from low Earth orbit, without an active radar transmitter.
Sensors
- NASA CYGNSS constellation (8 microsatellites): Each satellite carries a Delay-Doppler Mapping Instrument (DDMI) that records bistatic reflections of GPS L1 C/A signals from Earth's surface. Effective spatial resolution is approximately 25–36 km (the size of the first Fresnel zone convolved with the glistening zone); median revisit over the tropics and subtropics (38°S–38°N) is roughly 2.8 hours. The constellation has been operational since 2017.
- Spire Global GNSS-R payload (LEMUR-2 satellites): Spire flies GNSS-R receivers alongside its primary GNSS-RO payload. Coverage extends to higher latitudes than CYGNSS. Spatial resolution is comparable (tens of kilometres), and the commercial nature of the constellation allows tasking flexibility not available on NASA research missions.
- ESA SMOS (Soil Moisture and Ocean Salinity): An L-band (1.4 GHz) passive radiometer, not a GNSS-R instrument, but the primary independent validation reference for soil moisture retrieval algorithms. SMOS provides global coverage every 2–3 days at roughly 40–50 km resolution and has operated since 2009, giving a long validation baseline.
- NASA SMAP (Soil Moisture Active Passive): Radiometer-only since its radar failed in 2015. Provides 36 km passive brightness-temperature retrievals of volumetric soil moisture (0–5 cm depth) with 2–3 day global revisit. Widely used as a cross-validation reference for CYGNSS-derived soil moisture products.
- ISMN in-situ sensor networks: The International Soil Moisture Network aggregates ground-based capacitance and TDR sensors from networks including SCAN, COSMOS and OzNet. These provide point-scale volumetric water content (m³/m³) used to validate and bias-correct satellite retrievals; spatial mismatch between point sensors and 25–36 km footprints is itself a known source of validation uncertainty.
Why a navigation signal knows the ground is wet
When a GPS signal strikes a land surface, part of the energy scatters back upward in a pattern governed by the surface's dielectric constant. Water has a dielectric constant near 80; dry mineral soil sits between 2 and 5. Even a modest increase in volumetric water content shifts the dielectric constant substantially, and that shift changes the coherence, power and delay structure of the reflected signal. A receiver in low Earth orbit looking down at the specular reflection point can measure these changes without any active transmitter of its own.
The processing chain begins with the Delay-Doppler Map (DDM): a two-dimensional representation of reflected signal power as a function of code delay (range) and Doppler frequency shift. The shape and peak power of the DDM encode surface roughness, dielectric constant and vegetation attenuation simultaneously. Separating these contributions is the central algorithmic challenge. The Chew et al. algorithm, published and documented through NASA's CYGNSS science team, retrieves soil moisture by normalising DDM observables against a roughness model and applying an empirical vegetation correction derived from ancillary vegetation optical depth data.
What 25 kilometres of footprint actually means in practice
The spatial resolution of a CYGNSS soil moisture retrieval is not a pixel in the photographic sense. It is the area over which the reflected GPS signal is coherently integrated, approximately 25–36 km on a side depending on surface roughness and processing choices. That makes CYGNSS useful at river-basin, agricultural-district and regional drought-monitoring scales. It is not useful for field-level precision agriculture, where you need tens of metres, not tens of kilometres.
Revisit compensates for resolution to some degree. Over the CYGNSS latitude band (roughly 38°S to 38°N, covering most of the world's rain-fed cropland), the eight-satellite constellation achieves a median revisit of around 2.8 hours. That temporal density is genuinely unusual for a passive soil moisture sensor and allows tracking of wetting fronts after rainfall events with a granularity that SMOS or SMAP cannot match.
Latency from observation to publicly archived Level 3 gridded product has historically been one to several days for the NASA science products, though near-real-time processing pipelines have been demonstrated. For operational applications, latency expectations should be confirmed against the specific product version in use.
The three conditions that degrade the retrieval
Vegetation is the most pervasive limitation. Dense canopy attenuates the signal before it reaches the soil and again on the return path. Above a vegetation optical depth of roughly 1.5 (corresponding to closed-canopy tropical forest or dense maize near peak biomass), the soil moisture signal is largely masked. Sparse grassland and early-season crops present manageable attenuation; the algorithm applies a correction, but residual bias grows with canopy density. Published validation studies show retrieval skill degrades noticeably above moderate vegetation optical depth thresholds.
Surface roughness introduces ambiguity because rough surfaces scatter power away from the specular direction in a way that mimics a drier surface. Agricultural tillage, for instance, can produce a roughness signature that biases retrievals low. The Chew et al. approach uses a roughness parameter estimated from ancillary data, but that parameter is not always well-constrained over complex terrain.
Frozen ground is a hard limit. Ice has a dielectric constant close to that of dry soil, so a frozen surface and a dry surface are nearly indistinguishable to the algorithm. Retrievals over frozen or snow-covered ground are flagged and typically masked in operational products. This matters most for high-latitude agricultural regions in spring, precisely when soil moisture monitoring is agronomically important.
Validation: what the published record actually shows
Cross-validation against SMAP radiometer products over vegetated land surfaces has shown unbiased RMSE values in the range of 0.04–0.06 m³/m³ for CYGNSS retrievals in low-to-moderate vegetation conditions, which is within the 0.04 m³/m³ target that the soil moisture community treats as the threshold for useful agricultural and hydrological applications. Performance is better over bare or sparsely vegetated land and worse over dense vegetation, as noted above.
Validation against ISMN in-situ networks introduces a representativeness problem: a point sensor and a 25 km footprint are not measuring the same thing, especially in heterogeneous landscapes. Published studies acknowledge this explicitly. The practical implication is that CYGNSS soil moisture is best treated as a basin-average or regional-average product rather than a ground-truth substitute at field scale.
SMOS provides a longer archive (since 2009 versus CYGNSS's 2017 start) and is often used to extend climatological baselines. The two datasets are complementary rather than redundant, and a combined analysis can separate long-term trends from short-term anomalies.
Where the method earns its place in an operational system
The strongest operational case for GNSS-R soil moisture is in regions where ground sensor networks are sparse and cloud cover limits optical and thermal remote sensing. Tropical and sub-tropical agricultural zones, including much of sub-Saharan Africa, South and Southeast Asia, and the Sahel, sit within the CYGNSS latitude band and are chronically underserved by in-situ networks. For a government or development agency trying to monitor drought onset or estimate crop water stress across a large area, a 25–36 km product with sub-daily revisit is genuinely useful even with its resolution limitations.
Flood inundation detection is a secondary application that the same DDM observables support. Open water produces a very strong coherent reflection, and the transition from soil to standing water is detectable even under vegetation. This is a distinct product from soil moisture retrieval but uses the same raw data stream.
Satellize runs soil moisture analytics on open constellation data for sovereign clients requiring basin-scale agricultural monitoring. The Tonga crop-estimation programme demonstrated how coarse-resolution soil moisture inputs can constrain yield models when combined with optical vegetation indices. Clients interested in applying similar methods to larger continental domains can discuss a scoping call with the Satellize analytics team.
Typical figures
| Spatial resolution (CYGNSS Level 3 gridded) | 25–36 km effective footprint (first Fresnel zone, GPS L1 C/A) |
| Latitude coverage | 38°S to 38°N (CYGNSS); higher latitudes accessible via Spire GNSS-R |
| Revisit (CYGNSS, 8 satellites) | Median ~2.8 hours over tropical/subtropical land |
| Signal frequency | GPS L1 C/A at 1575.42 MHz (CYGNSS); multi-constellation receivers also use GLONASS L1 |
| Retrieval depth | Near-surface (0–5 cm); L-band does not penetrate significantly deeper |
| Retrieval accuracy (published, low-to-moderate vegetation) | Unbiased RMSE ~0.04–0.06 m³/m³ vs SMAP; degrades above vegetation optical depth ~1.5 |
| Archive depth (CYGNSS) | March 2017 to present; SMOS validation baseline from November 2009 |
| Product latency (NASA science products) | 1–several days for standard Level 3; near-real-time pipelines demonstrated but not universally operational |
| Known masking conditions | Frozen ground, snow cover, dense closed-canopy forest, open ocean (separate product) |
| Delivery formats (Satellize analytics output) | GeoTIFF, NetCDF, GIS-ready shapefiles, tabular basin summaries |
Analytics Satellize can run
| Basin-scale volumetric soil moisture anomaly map | CYGNSS DDM-based retrieval (Chew et al. algorithm) normalised against multi-year climatology | Weekly GeoTIFF layer with anomaly expressed as standard deviation from mean, flagged for frozen/dense-vegetation pixels |
| Drought onset and intensification alert | Time-series change detection on CYGNSS Level 3 soil moisture, thresholded against user-defined percentile triggers | Automated alert report with affected basin polygons, onset date and severity tier |
| Crop water stress index input layer | Soil moisture retrievals combined with MODIS or Sentinel-2 vegetation optical depth to estimate plant-available water | GIS layer ingested into yield or irrigation-scheduling models; compatible with FAO AquaCrop inputs |
| Flood inundation extent (complementary DDM product) | Coherent reflection power anomaly detection in CYGNSS DDMs; distinct from soil moisture algorithm but same raw data | Polygon shapefile of probable inundation extent, updated at each overpass, with confidence flag |
| Multi-sensor soil moisture fusion product | Optimal interpolation or data assimilation merging CYGNSS, SMOS and SMAP retrievals to reduce retrieval gaps and noise | Gridded NetCDF at 0.25° resolution with per-pixel uncertainty estimate |
| Seasonal soil moisture trend report | Mann-Kendall trend analysis on multi-year CYGNSS archive, stratified by land-cover class | PDF technical report with time-series charts, trend significance maps and narrative interpretation for non-specialist decision-makers |
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.