SAR backscatter soil moisture retrieval
C-band SAR backscatter responds to soil dielectric constant, giving a quantitative proxy for near-surface volumetric water content. Sentinel-1's free, 12-day global repeat makes operational soil-moisture mapping feasible, with important caveats on depth and canopy.
Sensors
- Sentinel-1 (C-band, 5.405 GHz): The workhorse for operational retrieval. Interferometric Wide Swath mode delivers 10 m × 10 m pixels across a 250 km swath, with a 6-day repeat at mid-latitudes when both satellites are active. Free and open archive back to 2014. VV polarisation is preferred for bare soil; VV+VH dual-pol enables vegetation correction.
- ALOS-2 PALSAR-2 (L-band, 1.27 GHz): L-band penetrates vegetation canopies and reaches 10–30 cm into the soil, making it complementary to C-band rather than a substitute. Stripmap mode gives 3 m resolution; ScanSAR covers 350 km at 100 m. Revisit is 14 days. Useful for forested or densely cropped areas where C-band saturates.
- SAOCOM-1A/1B (L-band, 1.275 GHz): Argentina's pair of L-band SAR satellites, designed explicitly for soil-moisture applications. Full-polarimetry mode at 10 m resolution, 16-day repeat per satellite. Particularly valuable in South America and for agricultural monitoring where L-band penetration depth matters.
- Capella Space (X-band, ~9.6 GHz): Spotlight mode reaches 0.35 m resolution, useful for field-scale validation and surface roughness characterisation. X-band penetrates soil less than C-band, so moisture sensitivity is shallower still, roughly 1–2 cm. High revisit on demand via commercial tasking. Not suited to operational area-wide moisture mapping but useful for calibration sites.
Why radar cares about water in soil
Soil dielectric constant is the physical property that links radar backscatter to moisture content. Dry mineral soil has a dielectric constant of roughly 3 to 5. Add water, and it climbs steeply: a volumetric water content of 30 percent pushes the dielectric constant above 20. That contrast is large enough for a C-band radar to detect reliably, because the dielectric constant controls how much microwave energy is reflected back toward the sensor versus absorbed or transmitted into the ground.
The relationship is near-linear across the agronomically relevant range of 5 to 40 percent volumetric water content, which is why backscatter intensity, measured in decibels, can be inverted to give a quantitative moisture estimate. The physics is well established: the Dobson mixing model and its successors have described this relationship since the 1980s, and it underpins every operational SAR soil-moisture product in use today.
The depth problem: what the radar actually samples
C-band radar at 5.4 GHz penetrates only the top 2 to 5 centimetres of soil. That is not a sensor limitation to be engineered away; it is a consequence of the electromagnetic skin depth at that frequency and typical soil conductivity. The layer sampled is real and agronomically meaningful, particularly for germination and surface runoff modelling, but it is not the root-zone moisture that drives crop stress or aquifer recharge.
L-band systems such as PALSAR-2 and SAOCOM reach 10 to 30 cm, depending on soil texture and salinity. That is still not the full root zone for deep-rooted crops or trees. Anyone claiming a space-based SAR product gives full-profile soil moisture is overstating the physics. The honest use case for C-band is surface moisture as a boundary condition for hydrological models, or as a rapid indicator of recent rainfall, not a substitute for in-situ tensiometers at depth.
Change detection: how a time series dissolves the roughness problem
Surface roughness scatters radar energy just as moisture does, and the two effects are nearly impossible to separate from a single image. A freshly ploughed field looks wet even when it is dry. The practical solution, operationalised in ESA's own Sentinel-1 soil-moisture products and in the TU Wien change-detection algorithm, is to build a long time series over the same pixel and track relative changes in backscatter rather than inverting absolute values.
The method works by establishing a dry reference (minimum backscatter over the archive) and a wet reference (maximum), then expressing each new observation as a fraction of that range. The result is a relative soil-moisture index, not an absolute volumetric figure, but it is stable, requires no ground calibration, and responds correctly to rainfall events within the 6-to-12-day revisit window. Converting the index to volumetric water content requires local calibration with in-situ data or a soil-texture layer. Without that step, the product is comparative, not absolute.
Vegetation: the signal thief
Green vegetation attenuates the radar signal on the way down and again on the way back up. At C-band, a wheat canopy at full biomass can reduce the soil backscatter contribution by 3 to 6 dB, which corresponds to a spurious apparent drying of 10 to 20 percentage points. The correction requires an estimate of vegetation water content, typically derived from a co-registered optical index such as NDVI or from the cross-polarised VH channel, which responds more strongly to volume scattering from vegetation than to the soil beneath.
The Water Cloud Model is the standard correction approach: it parameterises the canopy as a layer of water droplets with known attenuation coefficients, then subtracts its contribution from total backscatter. The parameters are crop-type dependent and need calibration. For bare or sparsely vegetated soils, no correction is needed and C-band retrieval is most reliable. For dense maize or soybean canopies at peak growth, the retrieval becomes unreliable regardless of correction, and L-band is the better choice. Knowing which regime you are in before commissioning an analysis saves time and money.
Operational products and their honest limits
Several global soil-moisture products run operationally on Sentinel-1 data. ESA's H SAF programme and the Copernicus Global Land Service both publish near-real-time surface soil-moisture layers at 1 km resolution, with latency of roughly 3 hours after acquisition. These are validated against global in-situ networks and report unbiased RMSE values of around 0.05 m³/m³ over bare and lightly vegetated surfaces, degrading to 0.08 to 0.12 m³/m³ over cropland. Those figures are from published validation reports and should be treated as typical, not guaranteed, for any specific site.
Field-scale retrievals at 10 to 100 m resolution are possible but require more careful processing: speckle filtering, local incidence angle correction, and often multi-temporal averaging that trades temporal resolution for spatial precision. A single Sentinel-1 pass over an agricultural field gives a noisy pixel-level estimate; a three-image stack reduces speckle enough to be useful for field-boundary level mapping. Satellize applies this workflow for crop-monitoring programmes, including the Tonga crop-estimation engagement, where surface moisture is one input to a broader yield model. The fundamental constraint remains the 2 to 5 cm sensing depth: any application requiring root-zone moisture must couple the SAR surface observation with a land-surface model that propagates the signal downward.
Choosing the right sensor for the application
For most agricultural and hydrological applications, Sentinel-1 is the starting point: free data, dense archive, consistent calibration, and a well-documented processing chain. SAOCOM adds L-band depth and full polarimetry for situations where canopy cover or required penetration depth rule out C-band. PALSAR-2 covers the same L-band niche with longer heritage and global coverage, though access is less open than Sentinel-1.
Capella and other commercial X-band systems are best reserved for site characterisation: mapping surface roughness at very high resolution to parameterise a retrieval model, or validating a coarser operational product at a specific field. Using commercial tasking for routine area-wide moisture monitoring is expensive relative to the marginal gain over free Sentinel-1 data. The decision tree is straightforward: bare or sparse vegetation with no budget constraint on data, use Sentinel-1; dense canopy or need for deeper sensing, add L-band; need sub-metre surface characterisation, task commercial X-band for a specific date and location.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 10 m × 10 m (ground range detected, multi-look) |
| Swath width (Sentinel-1 IW mode) | 250 km |
| Revisit (Sentinel-1, both satellites active) | 6 days at mid-latitudes; 12 days with one satellite |
| Radar frequency (C-band) | 5.405 GHz (wavelength ~5.6 cm) |
| Sensing depth (C-band) | 2 to 5 cm, depending on soil texture and moisture state |
| Sensing depth (L-band, PALSAR-2 / SAOCOM) | 10 to 30 cm in typical mineral soils |
| Typical retrieval accuracy (bare soil, validated) | RMSE ~0.05 m³/m³ volumetric water content |
| Operational product latency (Copernicus Global Land) | ~3 hours after satellite overpass |
| Archive depth (Sentinel-1) | From April 2014 (Sentinel-1A launch) |
| Delivery formats | GeoTIFF, NetCDF, COG; ingested via Copernicus Data Space |
Analytics Satellize can run
| Relative surface soil-moisture index | TU Wien change-detection algorithm applied to Sentinel-1 VV time series; dry/wet reference normalisation | GeoTIFF layer per acquisition, clipped to area of interest, with per-pixel confidence flag |
| Vegetation-corrected volumetric water content | Water Cloud Model canopy correction using co-registered Sentinel-2 NDVI; calibrated against soil-texture class | Field-boundary GIS layer with absolute VWC estimate and uncertainty range |
| Rainfall event detection and spatial extent | Anomaly detection on Sentinel-1 backscatter time series; threshold exceedance against rolling 30-day baseline | Alert report with mapped wet-area extent and estimated onset date, delivered within 24 hours of overpass |
| Seasonal moisture climatology | Multi-year Sentinel-1 archive aggregation; percentile stack to characterise normal dry and wet envelopes | Annual summary raster set showing median, 10th and 90th percentile moisture by calendar week |
| L-band vs C-band depth comparison | Paired SAOCOM or PALSAR-2 and Sentinel-1 acquisition analysis; difference map highlights canopy attenuation and deeper moisture signal | Technical memorandum with co-registered difference maps and site-specific sensor recommendation |
| Soil-moisture input layer for crop yield model | Surface VWC time series ingested as boundary condition into FAO AquaCrop or similar land-surface model | Processed input dataset formatted for model ingestion, with provenance metadata and gap-fill log |
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.