Crop residue cover and tillage practice detection

Post-harvest fields carry a spectral fingerprint that distinguishes no-till from ploughed ground. Short-wave infrared absorption by cellulose and lignin makes tillage practice legible from orbit, with important caveats about resolution and cloud.

Sensors

ASI PRISMA (PRecursore IperSpettrale della Missione Applicativa): Hyperspectral imager covering 400–2500 nm in 239 contiguous bands at roughly 30 m spatial resolution and ~12 nm spectral sampling. Directly resolves the cellulose absorption feature near 2100 nm. Revisit is opportunistic, typically 29 days at nadir, limiting time-critical post-harvest windows.
DLR/GFZ EnMAP (Environmental Mapping and Analysis Programme): Hyperspectral sensor covering 420–2450 nm at 30 m ground sampling distance and ~10 nm spectral resolution. Designed explicitly for surface mineralogy and vegetation biochemistry. Revisit approximately 27 days at nadir; swath 30 km, which constrains regional coverage per pass.
USGS/NASA Landsat 8 and 9 OLI: Two SWIR bands: Band 6 (1566–1651 nm) and Band 7 (2107–2294 nm) at 30 m. Band 7 straddles the cellulose absorption region but integrates across it rather than resolving the feature. Useful for the Cellulose Absorption Index (CAI) approximation and SWIR ratio indices. 16-day revisit per satellite; combined 8-day revisit with both.
ESA Sentinel-2 MSI: SWIR bands B11 (1565–1655 nm) and B12 (2100–2280 nm) at 20 m. Band placement is comparable to Landsat OLI SWIR. The 10-day revisit per satellite (5-day with both Sentinel-2A and 2B) is useful for catching narrow post-harvest windows before secondary tillage or re-wetting. No band centred shortward of 2100 nm limits CAI precision.
Commercial hyperspectral smallsats (e.g. Planet Tanager, Wyvern): Emerging commercial hyperspectral constellations aim for VSWIR coverage at 8–30 m with improved revisit. Spectral sampling varies by vendor and is not yet fully characterised in peer-reviewed literature at operational scale. Treat published specifications as aspirational until validated.

What dry straw absorbs that bare soil does not

Cellulose and lignin, the structural polymers that persist in crop residue long after harvest, produce a characteristic absorption trough centred near 2100 nm. A second, shallower feature sits around 2300 nm. Together they form the physical basis of the Cellulose Absorption Index, defined as CAI = 0.5 × (R2000 + R2200) − R2100, where R denotes surface reflectance at the subscripted wavelength in nanometres. Positive CAI values indicate residue; bare mineral soils and green vegetation return near-zero or negative values.

The physics is straightforward. The difficulty is spectral resolution. Resolving the trough properly requires bands narrower than roughly 20–30 nm centred on 2100 nm. Landsat OLI Band 7 and Sentinel-2 Band 12 bracket the feature but do not sit inside it with enough precision to compute a true CAI. Broadband SWIR ratios (such as B7/B6 on Landsat) can still separate high-residue from bare fields under favourable conditions, but they conflate residue type, soil brightness and moisture in ways that hyperspectral data do not.

Conventional tillage, conservation tillage, no-till: what the thresholds mean

Agronomic convention defines three broad practice classes by residue cover remaining after primary tillage. No-till leaves more than 30 percent of the soil surface covered. Conservation tillage retains 15–30 percent. Conventional tillage, which inverts the topsoil, typically leaves less than 15 percent. These thresholds matter because they correspond to measurable differences in erosion risk, soil moisture retention and, over time, soil organic carbon accumulation.

Satellite-derived residue cover estimates are calibrated against these thresholds, but the mapping is not clean. Residue cover changes rapidly after harvest: rain mats it down, wind moves it, secondary tillage buries it, and decomposition reduces it. The post-harvest window in which spectral contrast is highest can be as short as two to four weeks in humid climates. A 27-day revisit on a hyperspectral sensor may simply miss it. Multispectral sensors with shorter revisit offer a partial substitute, at the cost of spectral precision.

Crop type also matters. Wheat and maize residues have different lignin-to-cellulose ratios and different geometric arrangements on the surface, producing different CAI magnitudes at equivalent fractional cover. A classifier trained on maize stubble in the US Midwest will not transfer directly to rice straw in Southeast Asia without retraining.

Soil carbon accounting needs tillage maps, and tillage maps need honesty about error

Carbon credit schemes for agricultural soils, including those operating under the Verified Carbon Standard and the Gold Standard, require evidence of practice change. A farmer claiming credits for switching from conventional to no-till needs to demonstrate that change spatially and temporally. Satellite-derived tillage maps are the only scalable verification tool available at field level across large areas.

The published literature, much of it using Landsat SWIR data, reports overall classification accuracies in the range of 70–85 percent for two-class (tilled versus untilled) discrimination under good conditions. Three-class discrimination, separating no-till from conservation tillage, is harder and typically less accurate. Errors concentrate in fields with mixed residue from partial operations, in areas with bright soils that mimic residue spectrally, and wherever cloud cover forces the analyst to use imagery from outside the optimal post-harvest window.

For carbon accounting, this matters. A 15 percent misclassification rate applied across a large enrolled area can produce systematic over- or under-crediting. Honest programme design buffers against this with conservative crediting factors and ground-truth sampling, not by assuming the satellite is always right.

Erosion risk: the agronomic case for mapping residue before the rains

Crop residue cover is one of the primary inputs to the Revised Universal Soil Loss Equation (RUSLE), specifically the C-factor, which represents the ratio of soil loss from a cropped field to that from a continuously tilled bare plot. A field with 30 percent residue cover can have a C-factor an order of magnitude lower than a ploughed field. Mapping residue cover before the first significant autumn or spring rainfall event therefore gives land managers a lead time to intervene, whether by no-till contracts, cover cropping incentives or targeted extension advice.

Sentinel-2's 5-day revisit (with both satellites) is the most practical tool for catching this window at continental scale, despite its spectral limitations. Paired with a digital elevation model to weight slope-aspect combinations, a SWIR-derived residue index can generate an erosion risk surface that is actionable within days of harvest. The output is an estimate, not a measurement, and should be presented with uncertainty bounds rather than false precision.

Where hyperspectral data genuinely earns its cost

For national-scale annual reporting, Sentinel-2 and Landsat are likely sufficient if the methodology is appropriately conservative. For high-value applications, including carbon project verification, precision agronomy consultancy, or policy compliance monitoring across contested land, PRISMA or EnMAP data justify their additional complexity.

PRISMA, operated by the Italian Space Agency, has been used in published studies to map CAI across agricultural regions with classification accuracies approaching 90 percent for two-class problems when imagery falls within two weeks of harvest. EnMAP, launched in April 2022, offers comparable spectral capability with a slightly wider swath. Neither provides the revisit frequency needed to guarantee cloud-free coverage of a specific field within the optimal window, which is the central operational constraint for both sensors.

Satellize runs residue and tillage analytics on open Sentinel and Landsat archives and can commission tasking on PRISMA or EnMAP under client licence. The methodology is consistent with the published CAI literature and is calibrated against field-survey data where available, as in the Tonga crop-estimation programme where ground-truth collection informed index calibration. Outputs are delivered as georeferenced GIS layers with per-field confidence scores, not as clean maps that imply certainty that the physics does not support.

The limits worth stating before the contract is signed

Cloud cover is the bluntest constraint. In humid tropical and temperate maritime climates, cloud-free imagery within a two-week post-harvest window may simply not exist in a given year. Synthetic aperture radar does not solve this: SAR backscatter responds to surface roughness and moisture, not to cellulose chemistry, and cannot compute a CAI equivalent. It can distinguish rough ploughed fields from smooth no-till surfaces under some conditions, but the two methods are not interchangeable.

Soil moisture is the second major confound. Wet residue spectrally resembles bare wet soil more than dry residue. Residue that has been rained on and dried repeatedly loses spectral contrast as it decomposes and mixes with soil particles. The practical consequence is that imagery acquired more than three to four weeks after harvest in wet climates is often too ambiguous to classify reliably, regardless of sensor quality.

At 30 m resolution, fields smaller than roughly one to two hectares are subject to significant mixed-pixel contamination from field margins, hedgerows and access tracks. Sub-field variability in residue distribution, common where harvest equipment leaves uneven windrows, is invisible at this scale. Anyone using these maps for within-field management decisions needs finer spatial resolution than current hyperspectral satellites provide.

Typical figures

Spatial resolution (hyperspectral)	30 m (PRISMA, EnMAP); limits field detection to approximately 1 ha minimum clean pixel
Spatial resolution (multispectral SWIR)	20 m Sentinel-2 MSI; 30 m Landsat 8/9 OLI
Spectral sampling (CAI-capable)	~10–12 nm (EnMAP, PRISMA); broadband SWIR ratios only for Sentinel-2 and Landsat
Revisit (hyperspectral)	~27–29 days nadir (PRISMA, EnMAP); off-nadir tasking can improve to ~4–7 days at reduced resolution
Revisit (multispectral)	5 days (Sentinel-2A+2B combined); 8 days (Landsat 8+9 combined)
Key spectral bands for residue detection	2000 nm, 2100 nm, 2200 nm for CAI; 1600 nm and 2200 nm SWIR ratio for broadband approximation
Minimum detectable residue cover	~15–20% fractional cover under dry conditions with hyperspectral data; higher threshold (~25–30%) with broadband SWIR
Optimal acquisition window	Within 2–4 weeks of harvest, before secondary tillage, rain-matting or decomposition reduces spectral contrast
Archive depth	Landsat: 1972 to present (SWIR from Landsat 5 TM, 1984); Sentinel-2: 2015 to present; PRISMA: 2019 to present; EnMAP: April 2022 to present
Typical classification accuracy (published range)	70–85% overall for two-class (tilled/untilled) with SWIR multispectral; up to ~90% with hyperspectral CAI in optimal conditions

Analytics Satellize can run

Post-harvest residue cover fraction map	Cellulose Absorption Index (CAI) from hyperspectral data, or SWIR band ratio (B7/B6 Landsat; B12/B11 Sentinel-2) with empirical calibration	GeoTIFF layer with per-pixel fractional residue cover estimate and confidence band, delivered within 5 days of cloud-free acquisition
Tillage practice classification (three-class)	Threshold-based classification of residue cover fraction into no-till (>30%), conservation tillage (15–30%), conventional tillage (<15%) per agronomic convention	Field-boundary-clipped GIS polygon layer with practice class and per-field classification confidence score
Annual tillage practice change detection	Year-on-year comparison of classified post-harvest imagery using consistent acquisition-window matching; change flagged where class shifts by one or more categories	Change map GIS layer with field-level transition matrix (e.g. conventional to no-till), suitable for carbon programme baseline and monitoring reports
Pre-rain erosion risk surface	RUSLE C-factor estimation from SWIR-derived residue cover index, weighted by slope derived from a digital elevation model	Raster erosion risk surface (low/medium/high classes) delivered before first significant rainfall event; updated on each cloud-free overpass during the post-harvest window
Hyperspectral tasking and CAI time series for verification plots	Commissioned PRISMA or EnMAP acquisitions over enrolled carbon or agri-environment scheme parcels; CAI computed per parcel with uncertainty propagation	Per-parcel CAI time series report with acquisition metadata, uncertainty bounds and comparison to scheme thresholds; formatted for auditor review
Cloud-gap-filled seasonal residue composite	Multi-date Sentinel-2 SWIR compositing using maximum SWIR2 value within a defined post-harvest window to reduce cloud and moisture confounds	Seasonal best-observation composite raster with provenance layer showing acquisition date per pixel; delivered as analysis-ready GeoTIFF

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.