Root-zone soil moisture estimation for crop water status
Satellite sensors measure the top few centimetres of soil. Getting from that signal to the root-zone moisture that actually governs plant stress requires physics, modelling and honest accounting of what each sensor cannot see.
Sensors
- SMAP L-band Radiometer (NASA): Passive microwave at 1.41 GHz. Penetrates dry soil to roughly 5 cm, less in wet conditions. Native radiometer footprint is 36 km, making field-scale application impossible without downscaling. Global revisit every 2–3 days. Surface volumetric soil moisture retrieved to an unbiased RMSE target of 0.04 m³/m³ over non-frozen, non-dense-vegetation land.
- SMOS MIRAS L-band Radiometer (ESA/CNES): Interferometric passive microwave at 1.4 GHz, ~43 km native resolution. Operational since 2009, giving a 15-plus-year surface moisture archive. Complementary to SMAP for long time-series analysis and data assimilation spin-up, though the coarser effective resolution and higher radio-frequency interference susceptibility in some regions limit direct use.
- Sentinel-1 SAR (ESA, C-band, 5.405 GHz): Active microwave backscatter sensitive to surface dielectric properties and roughness. Interferometric Wide swath mode delivers 10 m pixels with 6-day revisit per satellite (12 days single satellite). Penetration depth at C-band is only 1–2 cm in moist soil, shallower than SMAP, but the spatial resolution makes it the primary tool for downscaling coarse passive-microwave retrievals to field scale via change-detection algorithms.
- Sentinel-2 MSI (ESA, optical): 10–20 m resolution optical imagery used as an ancillary constraint: crop type masks, vegetation indices and bare-soil composites inform the spatial disaggregation of coarse moisture estimates and help flag canopy interception effects. Cannot sense soil moisture directly; cloud cover is a persistent gap that SAR fills.
Why the surface reading misleads you
Soil moisture is not a single number. The top 5 cm can be bone-dry after two days without rain while roots at 30–100 cm are drawing on reserves that accumulated weeks earlier. Satellite microwave sensors measure dielectric permittivity, which responds to liquid water content. At L-band (1.4 GHz), the emission depth in mineral soil is roughly 5 cm under dry conditions and collapses to 1–2 cm when the soil is saturated. C-band SAR, used by Sentinel-1, is shallower still. Neither sensor sees the root zone directly.
This is not a flaw in the sensors. It is physics. The practical consequence is that a surface retrieval showing low moisture after a sunny week tells you the topsoil has dried, not whether the crop is stressed. A surface retrieval showing high moisture after rain tells you the infiltration front has reached the surface layer, not how deep it has gone. Acting on surface readings alone, without a model to propagate the signal downward, risks both over-irrigation and missed stress events.
From surface signal to root-zone estimate: the data assimilation step
The standard approach is to assimilate repeated surface observations into a soil water balance model. The model, typically a two- or three-layer scheme such as those embedded in FAO-56 or the ECMWF land surface model, tracks water movement between layers using precipitation, potential evapotranspiration and soil hydraulic parameters. On its own, the model drifts because forcing data contain errors. Satellite surface retrievals act as corrections: each overpass nudges the modelled surface layer toward the observed value, and the correction propagates downward through the profile according to the model's hydraulic coupling.
The Ensemble Kalman Filter is the most widely published method for this assimilation. Published studies using SMAP and SMOS inputs report root-zone estimates with RMSE values in the range of 0.04–0.06 m³/m³ against in-situ networks such as the International Soil Moisture Network, though performance degrades over dense canopies, frozen ground and organic soils. The quality of the hydraulic parameters, which come from global pedotransfer functions rather than field measurement in most operational settings, is often the dominant source of error.
The resolution problem and what downscaling can and cannot fix
A 36 km SMAP pixel covers roughly 1,300 km². A typical irrigated field in a smallholder landscape is 0.5–5 ha. The mismatch is not cosmetic. Within a single SMAP pixel there may be irrigated fields, rain-fed fields, bare soil, and urban patches, all with different moisture states. The retrieved value is a weighted average that may correspond to none of them accurately.
Downscaling algorithms address this by using the spatial detail in Sentinel-1 backscatter or Sentinel-2 bare-soil reflectance to redistribute the coarse moisture estimate across finer pixels. The DisPATCh algorithm and the Sentinel-1 change-detection approach published by Wagner et al. are among the most referenced methods, producing 100 m to 1 km outputs from SMAP inputs. At 1 km the results are credible for regional irrigation scheduling and drought monitoring. At 100 m the uncertainty is higher and depends heavily on the homogeneity of soil texture and vegetation within the coarse pixel. Honest use means reporting confidence intervals, not just point estimates, and flagging pixels where the within-footprint heterogeneity is large.
What the combined signal actually tells an agricultural buyer
A well-constructed root-zone product, updated every 2–3 days at 1 km and weekly at 100 m, gives an irrigation manager three actionable signals. First, the current root-zone depletion fraction, which maps directly onto FAO-56 stress thresholds for common crops. Second, the trajectory: whether the profile is recharging, stable or depleting, which matters more than any single snapshot. Third, spatial contrast within a district, identifying fields or zones that are systematically drier than neighbours under similar rainfall, which often points to soil texture anomalies, drainage problems or under-irrigation.
The product does not replace field tensiometers or neutron probes. Those remain the ground truth for precise irrigation scheduling on individual fields. What the satellite product adds is spatial coverage at a scale no ground network can match, and the ability to monitor thousands of fields simultaneously without physical access. In food security contexts, that spatial view is often more valuable than sub-field precision.
Honest limits: cloud, canopy, frozen ground and organic soils
Passive microwave retrievals are not affected by cloud, which is one of their genuine advantages over optical sensors. But dense vegetation attenuates the soil emission signal significantly. SMAP applies a vegetation opacity correction using ancillary vegetation water content estimates, but over canopies with water content above roughly 5 kg/m², the soil signal is largely masked and retrievals are flagged as low quality. Maize at peak biomass, dense rice paddies and closed-canopy orchards all fall into this category for much of the growing season.
Frozen soil is another hard limit. Microwave emission from ice-bearing soil is fundamentally different from liquid water, and SMAP masks frozen pixels. In temperate and boreal regions this creates winter data gaps of several months. Organic soils, including peatlands and heavily amended agricultural soils, have dielectric properties that differ from the mineral-soil assumptions built into retrieval algorithms, introducing systematic biases that are not fully corrected in standard products.
Satellize runs root-zone assimilation pipelines over open SMAP and Sentinel-1 archives, with Sentinel-2 ancillary layers for crop masking, as part of the same analytics infrastructure used in the Kingdom of Tonga crop-estimation programme. Outputs are delivered as GIS-ready rasters with per-pixel quality flags rather than as clean maps that hide the uncertainty.
Getting from a moisture map to an irrigation or stress decision
Root-zone volumetric water content needs to be translated into a management-relevant quantity before it is useful to a non-specialist buyer. The most direct translation is the fraction of total available water depleted from the root zone, expressed relative to crop-specific thresholds. FAO Irrigation and Drainage Paper 56 tabulates depletion fractions for major crops above which yield-reducing stress begins. Wheat, for instance, begins to experience stress at roughly 55% depletion of total available water; maize is closer to 50%.
Combining a satellite-derived root-zone depletion estimate with these published thresholds produces a stress probability map that an irrigation district or national food security agency can act on without needing to understand the underlying retrieval physics. The key discipline is propagating the moisture uncertainty through to the stress probability rather than presenting a binary stressed or not-stressed classification. A pixel with a depletion estimate of 48% and an uncertainty of ±8% is genuinely ambiguous; a pixel at 70% depletion with the same uncertainty is not.
Typical figures
| Native passive microwave resolution (SMAP radiometer) | 36 km footprint; downscaled products available at 9 km (SMAP-Sentinel fusion), 1 km and experimentally 100 m |
| Native passive microwave resolution (SMOS MIRAS) | ~43 km effective resolution; variable across swath |
| SAR resolution for downscaling (Sentinel-1 IW) | 10 m ground range; 6-day revisit (two-satellite constellation) |
| Sensing depth | L-band passive: ~5 cm (dry) to ~1 cm (saturated); C-band SAR: ~1–2 cm; root-zone (30–100 cm) via model assimilation only |
| SMAP global revisit | 2–3 days at mid-latitudes; daily near poles |
| Surface retrieval accuracy (SMAP design target) | Unbiased RMSE ≤ 0.04 m³/m³ over non-frozen, non-dense-vegetation land |
| Root-zone assimilation accuracy (published range) | RMSE 0.04–0.06 m³/m³ against in-situ networks; degrades over organic soils and dense canopy |
| Archive depth | SMAP: April 2015 to present; SMOS: November 2009 to present; Sentinel-1: April 2014 to present |
| Vegetation masking limit | Passive microwave retrievals flagged unreliable above ~5 kg/m² vegetation water content |
| Delivery formats | GeoTIFF raster with per-pixel quality flags; NetCDF time-series stacks; optional WMS/WMTS tile service |
Analytics Satellize can run
| Root-zone depletion fraction map | Ensemble Kalman Filter assimilation of SMAP surface retrievals into a two-layer soil water balance model (FAO-56 or equivalent); downscaled using Sentinel-1 change-detection | Weekly GeoTIFF at 1 km, with per-pixel uncertainty band and crop-specific stress threshold overlay |
| Irrigation stress alert | Threshold exceedance on root-zone depletion fraction relative to FAO-56 published crop coefficients; alert triggered when depletion probability exceeds configurable confidence level | District-level alert feed (JSON or email digest) updated on each SMAP overpass cycle |
| Seasonal soil moisture trajectory report | Time-series analysis of assimilated root-zone estimates against historical percentiles from SMOS/SMAP archive; anomaly expressed as percentile rank | Monthly PDF report with district-level percentile maps and trend charts, suitable for food security briefings |
| Downscaled 100 m soil moisture layer | DisPATCh-class spatial disaggregation using SMAP coarse retrievals and Sentinel-1 backscatter spatial patterns; Sentinel-2 bare-soil composites used to constrain soil texture proxy | Per-overpass GeoTIFF at 100 m with heterogeneity quality flag; flagged pixels where within-footprint variance is high |
| Irrigation performance assessment | Spatial contrast analysis comparing root-zone depletion across fields within a district under similar rainfall forcing; outlier detection identifies systematically under- or over-irrigated zones | GIS polygon layer with per-field depletion statistics and anomaly classification; annual summary table |
| Data assimilation spin-up archive | Retrospective assimilation run using SMOS archive (2009 onwards) to initialise soil moisture state for current-season modelling; reduces cold-start bias in seasonal forecasts | NetCDF stack of historical root-zone estimates for client's area of interest, delivered once at project start |
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.