Field-scale nitrogen leaching risk mapping from post-harvest bare-soil indices

Post-harvest bare-soil reflectance composites from Sentinel-2, combined with crop-type history and autumn rainfall, let regulators and agronomists identify which fields carry the highest nitrate leaching risk before winter rains mobilise residual nitrogen.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 20 m in shortwave-infrared (SWIR) bands 11 and 12 at 1610 nm and 2190 nm. Five-day revisit at mid-latitudes with both satellites. SWIR reflectance on bare soils correlates with soil organic matter content and residual crop debris. Cloud contamination in temperate autumn windows can reduce usable observations to fewer than four per month.
Sentinel-1 SAR C-band: C-band (5.405 GHz) backscatter at 10 m resolution, six-day repeat in IW mode. Penetrates cloud and light rain, giving surface roughness and moisture signals when optical data are unavailable. Bare-soil backscatter distinguishes tilled from untilled surfaces and can detect the structural signature of a standing cover crop even under overcast skies, though it cannot directly measure organic content.
ISRIC SoilGrids: Global gridded soil property maps at 250 m resolution, including sand, silt and clay fractions and soil organic carbon to 200 cm depth. Texture class is the primary determinant of hydraulic conductivity and therefore of leaching rate given a nitrogen load. Used as a static covariate layer; uncertainty bands are published per pixel and should be propagated into risk scores.
ERA5 precipitation reanalysis: ECMWF ERA5 provides hourly precipitation estimates at approximately 31 km grid spacing, with a latency of around five days for the preliminary dataset. Autumn cumulative rainfall and storm-event intensity drive the drainage flux that mobilises soil nitrate. ERA5 is not a substitute for dense rain-gauge networks but is consistently available globally and spans back to 1940.

What bare soil after harvest actually tells you

When a nitrogen-intensive crop such as winter wheat, oilseed rape or maize is harvested, the field enters a window of maximum leaching vulnerability. Root uptake stops. Mineralisation of crop residues and soil organic matter continues, releasing ammonium that nitrifies to nitrate. If autumn rainfall arrives before a cover crop establishes, that nitrate moves with drainage water toward groundwater or surface streams.

Sentinel-2 SWIR bands capture this state in two ways. First, the bare-soil composite index, built by selecting cloud-free pixels across a multi-week window and masking vegetated ones using NDVI thresholds below roughly 0.25, reveals the spectral signature of the soil surface itself. Soils with higher organic matter content show a characteristic SWIR response that has been studied extensively in the context of soil organic carbon mapping. Second, and more directly useful here, the complete absence of any green spectral signal through October and November flags fields where no cover crop was sown at all.

Building the composite: the cloud problem in temperate autumn

This is where the method earns its caveats. Northern European autumns routinely produce persistent cloud cover lasting ten to twenty consecutive days. Over a six-week post-harvest window, a field in the UK or northern France may yield only two or three cloud-free Sentinel-2 observations. A single observation is unreliable; compositing over the full window helps, but if cloud breaks are clustered in one fortnight, the composite reflects soil state at that moment, not the average.

Sentinel-1 SAR backscatter fills part of this gap. C-band backscatter from bare agricultural soils is sensitive to surface roughness (tillage state) and volumetric moisture. A freshly tilled, moist bare soil shows a distinct backscatter signature compared with a surface carrying a low green-biomass cover crop. The distinction is not as clean as an optical NDVI signal, but it is available under cloud. Combining the two data streams, using SAR as a gap-filler and confidence flag rather than a primary signal, produces more reliable cover-crop absence maps than either sensor alone.

Persistent snow cover, which can occur in continental climates from November onward, renders both optical and SAR signals ambiguous for this application. Risk mapping should be completed before the first hard frost where possible, or flagged as uncertain for affected pixels.

Crop history and soil texture: the multipliers that turn a bare field into a risk score

Not all bare fields carry the same nitrogen load. A field that grew oilseed rape, which receives up to 200 kg N per hectare in intensive systems and leaves substantial residue nitrogen, presents a higher leaching risk than a field that grew spring barley at lower input rates. Crop-type history from the previous season, derived from multispectral time-series classification (covered separately in the crop-type classification page), is therefore a necessary covariate.

Soil texture from ISRIC SoilGrids or a national soil survey determines how quickly water and dissolved nitrate move through the profile. Sandy loams and gravelly soils with high hydraulic conductivity can transmit nitrate to groundwater within weeks of a rainfall event. Heavy clay soils drain more slowly and may attenuate leaching through denitrification in anaerobic microsites. The risk index weights bare-soil detection by the product of estimated nitrogen loading (from crop type) and drainage class (from texture), then scales by cumulative autumn rainfall from ERA5.

The 250 m resolution of SoilGrids is coarser than the 10 m Sentinel-2 pixels. Within-field soil variability is real and can be large. Where national soil survey data at finer resolution exist, they should replace SoilGrids. Where they do not, the risk score carries a spatial uncertainty that should be communicated explicitly to end users.

Honest limits: what the index cannot resolve

The method produces a relative risk ranking, not a nitrate concentration measurement. It cannot tell you how much nitrate will reach a watercourse; that requires a hydrological model, field drainage maps, and ideally in-situ water quality data. Regulators using this output for enforcement should treat it as a prioritisation tool, directing field inspections toward high-scoring parcels rather than as standalone evidence.

SWIR-based organic matter proxies are sensitive to soil moisture as well as organic content. A wet bare soil and a dry organic-rich soil can produce similar SWIR signals. Normalisation approaches such as the Bare Soil Index or the SWIR-based Soil Composition Index help, but residual moisture confounding remains, particularly in the wet autumn window when this analysis is most needed. Published studies using Sentinel-2 bare-soil composites for organic carbon mapping report R-squared values typically in the 0.5 to 0.7 range against field measurements, which gives a sense of the signal-to-noise ratio one is working with.

Field boundaries are essential. Without accurate parcel polygons, the bare-soil composite bleeds across field edges, mixing signals from adjacent fields with different management histories. National agricultural parcel registries, where available, should anchor the analysis. In their absence, automated field boundary delineation introduces its own error.

Integration with national regulatory frameworks

Nitrate Vulnerable Zones under the EU Nitrates Directive (91/676/EEC) and equivalent national designations require member states and successor regulators to monitor and enforce nitrogen management rules on designated agricultural land. Satellite-derived cover-crop absence maps offer a scalable audit layer that ground inspection alone cannot match at national scale. Several European environment agencies have piloted remote-sensing approaches to Nitrates Directive compliance monitoring, and the method described here is consistent with those published frameworks.

Satellize can configure this workflow on open Sentinel data for any national jurisdiction, with outputs registered to the client's parcel registry. The Tonga crop-estimation programme demonstrated the organisation's ability to adapt open-constellation analytics to non-standard agricultural contexts; the nitrogen leaching workflow applies the same compositing and classification infrastructure to a temperate regulatory problem.

Delivery formats for regulatory clients typically include a ranked parcel-level GIS layer updated at the end of the post-harvest window, a summary report quantifying the area at high, medium and low risk by administrative unit, and an audit trail of the cloud-fraction and observation-count metadata for each parcel, so that assessors can identify parcels where data quality was insufficient to support a confident classification.

Typical figures

Primary optical resolution	10 m (VIS/NIR bands), 20 m (SWIR bands 11 and 12) — Sentinel-2 MSI
SAR resolution	10 m ground range, IW mode — Sentinel-1
Optical revisit	5 days at mid-latitudes (Sentinel-2A + 2B combined)
SAR revisit	6 days per track, IW mode — Sentinel-1
Soil texture covariate resolution	250 m — ISRIC SoilGrids; finer where national surveys are available
Rainfall forcing resolution	~31 km grid, hourly — ERA5 reanalysis; ~5-day latency for preliminary product
Minimum detectable cover-crop signal	NDVI > ~0.15 reliably distinguishes sparse cover from bare soil at 10 m; very sparse or recently sown cover may be missed
Sentinel-2 archive depth	From June 2015 (Sentinel-2A launch); systematic global coverage from 2017
Typical analysis window	6 to 10 weeks post-harvest (approximately September to November in northern temperate zones)
Output formats	GeoTIFF risk raster, parcel-level GeoPackage or Shapefile, PDF summary report

Analytics Satellize can run

Post-harvest bare-soil composite	Cloud-masked median or percentile compositing of Sentinel-2 SWIR and red-edge bands over the post-harvest window; NDVI threshold masking to isolate bare-soil pixels	GeoTIFF raster at 10–20 m resolution, with per-pixel observation-count and cloud-fraction metadata layer
Cover-crop absence map	Binary classification of parcel pixels as cover-cropped or bare, using NDVI time-series thresholding on Sentinel-2 and SAR backscatter anomaly as a cloud-period gap-filler	Parcel-level GeoPackage with confidence score and SAR-versus-optical source flag per parcel
Field-level nitrogen leaching risk index	Multiplicative risk score combining cover-crop absence, prior-season crop nitrogen loading class (from crop-type history), soil drainage class (from SoilGrids texture), and cumulative ERA5 autumn rainfall	Ranked parcel table (high/medium/low risk tiers) in GeoPackage and CSV, suitable for direct import into regulatory GIS
Temporal cover-crop establishment tracking	Dense Sentinel-2 NDVI time series per parcel from harvest date onward, flagging the date at which NDVI first exceeds the bare-soil threshold, or confirming absence through the window	Per-parcel establishment-date raster and summary table; exportable as a feed updated on each cloud-free overpass
Data-quality audit layer	Per-parcel count of cloud-free Sentinel-2 observations and SAR acquisitions within the analysis window; parcels below a minimum observation threshold flagged as data-insufficient	Audit GeoPackage appended to the risk output, enabling regulators to exclude or downweight uncertain parcels
Multi-year risk trend analysis	Repeat application of the compositing and risk-scoring workflow across the Sentinel-2 archive (2017 onward) to identify fields with persistent non-compliance or structural cover-crop absence	Multi-year parcel-level trend report in PDF and GeoPackage, with year-on-year risk score time series

Sources

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.