Surface and root-zone soil moisture retrieval at basin scale
Active microwave and passive L-band sensors read dielectric contrast to estimate surface soil moisture at basin scale. Exponential filter methods push those estimates to root-zone depth, where irrigation and drought decisions are actually made.
Sensors
- SMAP Radiometer (NASA, L-band 1.41 GHz): Passive brightness temperature at 36 km native resolution; 2–3 day global revisit. Retrieves volumetric surface soil moisture (top ~5 cm) to a target accuracy of 0.04 m³/m³ unbiased RMSE over non-frozen, low-to-moderate vegetation. Dense canopies and frozen ground invalidate the retrieval.
- Sentinel-1 SAR (ESA, C-band 5.4 GHz): Active backscatter in IW mode at 10–20 m spatial resolution; 6-day revisit at the equator, 3–4 days at mid-latitudes with both satellites. VV polarisation is most sensitive to surface roughness and moisture; cross-polarisation adds canopy penetration. Temporal change-detection approaches reduce ambiguity with soil roughness.
- ASCAT on MetOp (EUMETSAT, C-band): Scatterometer delivering relative surface soil moisture (expressed as degree of saturation 0–100%) at 12.5–25 km grid spacing; near-daily global coverage. Long record from 2006 enables climatological anomaly detection. Absolute moisture values require local calibration against soil texture data.
- SMOS MIRAS (ESA, L-band 1.4 GHz): Interferometric radiometer at ~43 km resolution; 3-day revisit. Complements SMAP, particularly over Europe and Africa, with a record beginning in 2010. Radio-frequency interference (RFI) from ground transmitters degrades retrievals in parts of Asia and the Middle East.
What the microwave signal is actually measuring
Soil moisture matters to microwave sensors because liquid water has a dielectric constant around 80, while dry mineral soil sits between 2 and 4. That contrast changes both the emissivity (relevant to passive radiometers) and the backscatter coefficient (relevant to active SAR and scatterometers) in ways that are detectable from orbit. The physics is well established; the difficulty is that soil texture, surface roughness, organic matter content and vegetation water content all modulate the same signal.
L-band (roughly 1–2 GHz) penetrates vegetation canopies and reaches the soil surface more reliably than C-band. SMAP and SMOS operate here. C-band sensors such as Sentinel-1 and ASCAT are more widely available and offer finer spatial resolution, but the signal saturates under dense canopies, particularly closed-canopy forest and tall maize or sugarcane near peak biomass. Buyers should be clear about which land-cover classes dominate their basin before choosing a primary sensor.
From the top 5 cm to the root zone: the exponential filter
Every spaceborne moisture sensor measures the top 2–5 cm of soil. Crop roots and drought stress operate at 20–100 cm depth. The exponential filter, first formalised by Wagner et al. in the late 1990s using ERS scatterometer data, bridges that gap. It treats the soil column as a two-layer system where surface observations drive a recursive update of a soil-water index (SWI). The characteristic time length T, typically between 5 and 40 days depending on soil hydraulic properties, controls how quickly the root-zone estimate responds to surface changes.
The method is computationally lightweight and runs operationally in the Copernicus Global Land Service, which publishes SWI products derived from ASCAT at 12.5 km resolution. The honest caveat: SWI is a relative index, not an absolute volumetric water content. Converting it to plant-available water requires local soil texture and field-capacity data, which may not exist for ungauged basins. For irrigation scheduling at field scale, in-situ calibration remains necessary.
Where the retrievals break down
Clay-rich soils introduce a specific ambiguity. High clay content raises the dielectric constant even when the soil is relatively dry, mimicking a wetter signal. Distinguishing a clay Vertisol from a sandy loam at 60% saturation requires ancillary soil-texture maps, and those maps are often coarse or outdated in the basins where moisture monitoring matters most.
Dense vegetation is the other principal failure mode. The vegetation optical depth (VOD) correction applied in SMAP retrievals works well for grasslands and sparse crops. It degrades over closed-canopy tropical forest, where the canopy itself dominates the emission and the soil contribution is negligible. Sentinel-1 change-detection approaches partially compensate by exploiting temporal backscatter differences rather than absolute values, but they still struggle when canopy water content varies seasonally and mimics soil moisture change.
Frozen ground, permanent snow cover and strong radio-frequency interference each produce additional data gaps. SMOS is particularly affected by RFI across central Asia. Practical basin-scale products should flag these conditions explicitly rather than interpolating through them.
Combining sensors: what multi-source fusion actually buys
No single sensor wins across all conditions. SMAP provides physically calibrated absolute moisture at coarse resolution. Sentinel-1 provides spatial detail at 10–20 m but only relative change. ASCAT provides temporal continuity back to 2006. The standard fusion approach, implemented in products such as the ESA Climate Change Initiative (CCI) combined soil moisture dataset, merges active and passive retrievals using triple collocation to estimate and minimise random errors from each source.
Downscaling SMAP to Sentinel-1 resolution using statistical or machine-learning disaggregation is an active research area. Published studies report RMSE values of roughly 0.04–0.07 m³/m³ after disaggregation, though performance degrades in heterogeneous landscapes where sub-pixel variability is high. For basin-scale drought monitoring, the coarse-resolution SMAP product is often sufficient. For irrigation-district management, the downscaled product or direct Sentinel-1 inversion is more appropriate.
Operational uses: drought early warning and irrigation scheduling
Drought early warning systems typically use soil moisture anomalies expressed as percentiles relative to a multi-year climatology. ASCAT's record from 2006 and SMAP's from 2015 are long enough to define meaningful climatologies for most climate zones, though 15 years is a thin baseline for characterising decadal variability. Anomalies below the 20th percentile sustained over several weeks are a recognised trigger in famine early warning frameworks.
Irrigation scheduling is more demanding. A water manager needs to know whether root-zone moisture has dropped below a crop-specific threshold, not just whether it is anomalous relative to history. That requires the SWI-to-volumetric-water-content conversion described above, plus crop coefficients and evapotranspiration estimates. Satellize runs this kind of multi-layer soil moisture analysis on open constellations for basin-scale clients; the Tonga crop-estimation programme illustrated how combining moisture and vegetation indices at district scale can sharpen agricultural outlooks without ground sensors. For irrigation scheduling at individual field level, satellite data is best treated as a spatial interpolator between sparse in-situ probes rather than a standalone source.
Alert latency matters operationally. SMAP data is typically available within 24 hours of acquisition through NASA Earthdata. Sentinel-1 IW products appear on the Copernicus Data Space within a few hours. ASCAT near-real-time products from EUMETSAT are available within roughly 3 hours of overpass. None of these latencies are a barrier to operational drought monitoring; they can be a constraint for very fast-response irrigation control systems.
Typical figures
| Surface sensing depth | 2–5 cm (varies with soil moisture, frequency; L-band penetrates slightly deeper than C-band) |
| Spatial resolution (SMAP radiometer) | 36 km native; downscaled products to ~9 km or ~1 km using Sentinel-1 disaggregation |
| Spatial resolution (Sentinel-1 IW SAR) | 10–20 m (range × azimuth); 250 km swath |
| Spatial resolution (ASCAT MetOp) | 12.5 km or 25 km gridded product |
| Revisit frequency | SMAP: 2–3 days global; Sentinel-1: 3–6 days mid-latitudes; ASCAT: ~1 day global |
| Retrieval accuracy (SMAP target) | 0.04 m³/m³ unbiased RMSE over low-to-moderate vegetation on non-frozen soil |
| Frequency bands | L-band 1.41 GHz (SMAP, SMOS); C-band 5.2–5.4 GHz (Sentinel-1, ASCAT) |
| Archive depth | ASCAT from 2006; SMOS from 2010; SMAP from April 2015; Sentinel-1 from 2014 |
| Data latency (operational) | SMAP: ~24 h; Sentinel-1: ~3–6 h on Copernicus Data Space; ASCAT NRT: ~3 h |
| Known retrieval failures | Dense canopy (VOD > ~0.8), frozen ground, snow cover, strong RFI (SMOS in Asia) |
Analytics Satellize can run
| Basin-scale surface soil moisture map | SMAP L3 retrieval or Sentinel-1 change-detection inversion (Ulaby-type backscatter model or machine-learning regression against in-situ reference) | Gridded GeoTIFF or NetCDF, per overpass or composited weekly, with retrieval-quality flags |
| Root-zone soil water index (SWI) | Exponential filter applied to surface moisture time series; T parameter estimated from soil hydraulic conductivity or optimised against in-situ data | Time-series GIS layer per basin sub-unit; CSV export per administrative zone |
| Soil moisture anomaly and drought percentile | Empirical climatology from ASCAT or SMAP archive; percentile ranking per pixel per day-of-year | Weekly anomaly raster and district-level summary report; alert trigger when percentile falls below configurable threshold |
| Downscaled high-resolution moisture field | Statistical or random-forest disaggregation of SMAP using Sentinel-1 backscatter and ancillary land-cover, slope and soil-texture layers | ~100 m to 1 km gridded moisture product; uncertainty band per pixel included |
| Irrigation-district water-stress indicator | SWI converted to plant-available water using FAO soil texture lookup; compared against crop-specific depletion thresholds | Traffic-light alert feed (JSON or email) per irrigation district; updated at each Sentinel-1 overpass |
| Long-term soil moisture trend analysis | Mann-Kendall trend test on merged CCI active/passive soil moisture record; breakpoint detection for land-use or climate attribution | Decadal trend map (PDF and GeoTIFF) with significance levels; narrative interpretation section |
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.