Winter crop area estimation and freeze-damage detection
SAR backscatter and optical NDVI together track winter wheat and barley from sowing through vernalisation, then expose canopy damage after hard freezes before spring regrowth closes the window for accurate loss assessment.
Sensors
- Sentinel-1 C-band SAR (ESA): 10 m ground range resolution in IW mode, 6-day repeat at mid-latitudes with both satellites active. C-band (5.405 GHz) backscatter responds to canopy dielectric properties; cell rupture from hard freezes reduces canopy water content, producing a measurable drop in VV and VH backscatter within one to two acquisition cycles. Operates through cloud and darkness, critical during northern-hemisphere winters.
- Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands, 20 m in red-edge and SWIR. Five-day revisit at mid-latitudes (both satellites). NDVI and red-edge chlorophyll index track canopy greenness; post-freeze imagery shows abrupt NDVI decline in damaged zones. Cloud cover is the primary constraint during winter months, making SAR the fallback.
- MODIS Terra and Aqua (NASA): 250 m resolution in bands 1–2, daily revisit. Too coarse for field-level mapping but invaluable for regional phenology baselines and climatological NDVI composites against which anomalies are measured. MODIS 16-day NDVI composites (MOD13Q1/MYD13Q1) provide the long archive, dating to 2000, needed to establish what a normal February looks like for a given region.
- Landsat 8 and 9 OLI (USGS/NASA): 30 m resolution, 8-day combined revisit. Supports cross-calibration of Sentinel-2 NDVI time series and extends the optical archive to 1984 via earlier Landsat missions. Useful for validating crop-area boundaries and providing an independent greenness estimate when Sentinel-2 is cloud-obscured.
What the backscatter drop is actually measuring
When air temperatures fall below roughly minus six to minus eight degrees Celsius for several hours, ice crystals form inside leaf mesophyll cells. The cell walls rupture. Free water migrates out of the tissue. This is not a slow process: it happens overnight. The agronomic consequence is well known. The remote-sensing consequence is less widely appreciated.
Sentinel-1 C-band backscatter is sensitive to the dielectric constant of vegetation, which is dominated by liquid water content. A hard freeze that kills a significant fraction of the canopy reduces that dielectric constant sharply. In VH polarisation, which is more sensitive to volume scattering from the canopy than to the underlying soil, the drop can reach several decibels relative to the pre-freeze acquisition. That signal is detectable against the normal seasonal trajectory of backscatter, which rises gradually through autumn tillering and then plateaus. An abrupt mid-winter dip, coinciding with a recorded frost event, is the diagnostic pattern.
The honest caveat: partial damage is harder to read. A field that loses thirty per cent of its tillers will show a smaller backscatter change, potentially within the noise of day-to-day soil moisture variation. Soil freeze-thaw cycles produce their own backscatter transients. Separating canopy damage from soil effects requires comparing VH and VV together, and anchoring the analysis to meteorological records of the frost event.
Mapping the crop area before damage can be assessed
You cannot measure damage to a crop you have not yet located. Winter wheat and barley are sown between September and November across most of Europe and Central Asia. By December, a dense time series of Sentinel-1 acquisitions already separates emerged cereals from bare soil, oilseed rape, and grassland. Cereals at this growth stage show a characteristic VH backscatter rise as the canopy develops, distinct from the flat or declining signal of bare or lightly vegetated soil.
Sentinel-2 NDVI adds a complementary classifier once cloud conditions permit. Red-edge bands (705 nm and 740 nm) are particularly useful for separating early-season cereals from other green covers because of their sensitivity to chlorophyll concentration at low leaf area index. The combination of SAR phenological trajectory and optical spectral signature, used together in a supervised classifier trained on ground-truth parcels, consistently achieves overall accuracies above ninety per cent in published studies on European winter cereal mapping, though performance degrades in fragmented smallholder landscapes where fields are smaller than a few Sentinel-1 pixels.
Field boundaries matter here. Without them, mixed pixels at parcel edges dilute both the area estimate and the damage signal. Where cadastral data are unavailable, boundary delineation from Sentinel-2 texture is a prerequisite step, covered separately in the sibling page on smallholder field boundary delineation.
The climatological baseline and why it matters
A single post-freeze image is not enough. NDVI in January varies enormously between a mild maritime winter and a continental one, even in undamaged crops. The analytical anchor is a multi-year climatological baseline: the expected NDVI or backscatter value for that pixel, in that week of the year, given the current season's trajectory up to the frost event.
MODIS provides the archive depth for this. MOD13Q1 sixteen-day composites run from 2000 onwards, giving more than two decades of winter phenology for any location. The baseline is typically constructed as the median or interquartile range of the same calendar period across reference years. Anomaly detection then flags pixels where the post-freeze value falls more than one or two standard deviations below the baseline expectation, after accounting for the current season's pre-freeze trajectory.
This approach was used in the European Commission's MARS Crop Monitoring service during the February 2012 and January 2017 frost events across Eastern Europe, when satellite-derived damage maps were compared against field survey data to calibrate national production revisions. The method is not proprietary; it is published practice. What varies between implementations is the quality of the baseline, the timeliness of the post-event acquisition, and the rigour of the meteorological co-registration.
The timing problem: a closing window
Spring regrowth is the enemy of post-freeze damage assessment. Winter cereals are remarkably resilient. Surviving tillers compensate. By March or April across much of Europe, a field that lost half its stand in January may look nearly normal in NDVI terms, because the survivors have filled the gaps. The satellite signal of damage is strongest in the two to four weeks immediately following the freeze event.
This creates an operational constraint that shapes the entire workflow. Post-freeze imagery must be acquired, processed, and delivered to national forecasting agencies before the recovery signal closes. Sentinel-1's six-day repeat (three days with both satellites active over Europe) is well-suited to this. Sentinel-2's five-day repeat is adequate if cloud cover cooperates, which in a January frost event over Ukraine or Kazakhstan it frequently does not. SAR is not optional; it is the primary sensor for timely damage detection.
Latency from acquisition to analysis matters too. ESA's Copernicus Data Space Ecosystem typically makes Sentinel-1 IW SLC and GRD products available within an hour of downlink. Processing to analysis-ready backscatter and change maps adds further time depending on infrastructure. A well-configured pipeline can deliver a damage layer to a ministry within twelve to twenty-four hours of the post-freeze overpass.
From damage map to production forecast
A freeze-damage layer is not itself a production forecast. It is an input. The translation requires knowing the pre-freeze crop area, the severity of the damage by zone, the probability of compensatory tillering (which depends on growth stage at the time of the freeze), and the subsequent weather trajectory. These are agronomic and meteorological judgements that sit outside the remote-sensing workflow.
What satellite data can provide is the spatial disaggregation that national statistics often lack. A ministry may know that a frost event occurred but have no regional breakdown of where damage was concentrated. A damage-probability map at ten to thirty metre resolution, aggregated to administrative units, gives forecasters a spatial prior that ground surveys can then validate selectively rather than exhaustively. This is where the satellite product earns its place in the forecasting chain: not by replacing agronomic expertise, but by directing it.
Satellize has built crop-estimation pipelines for sovereign clients, including the Kingdom of Tonga programme, and applies the same baseline-anomaly architecture to winter cereal monitoring on request.
Honest limits of the method
Partial canopy damage below roughly twenty to thirty per cent stand loss is difficult to detect reliably from Sentinel-1 alone at ten metre resolution. The backscatter change is small relative to soil moisture noise. Optical NDVI is more sensitive to partial chlorophyll loss but requires cloud-free conditions that a winter frost event rarely provides promptly.
C-band SAR cannot distinguish between freeze-killed tissue and canopy flattened by wet snow load, which produces a similar backscatter reduction. Meteorological records are essential for attribution. In regions with persistent snow cover, the snow itself dominates the backscatter signal and masks the canopy response entirely until melt.
Finally, the method is calibrated primarily for temperate continental and maritime climates where winter wheat and barley phenology is well documented. Applying it to higher-latitude spring-sown crops, or to winter cereals in subtropical climates with different vernalisation patterns, requires re-validation of the backscatter trajectory models and the climatological baselines.
Typical figures
| Primary SAR resolution | 10 m (Sentinel-1 IW GRD mode) |
| Primary optical resolution | 10 m visible/NIR, 20 m red-edge/SWIR (Sentinel-2 MSI) |
| SAR revisit (mid-latitudes) | 6 days single satellite, ~3 days with Sentinel-1A and 1C combined |
| Optical revisit | 5 days (Sentinel-2 pair); 8 days (Landsat 8+9 combined) |
| Climatological baseline archive | MODIS from 2000; Landsat from 1984; Sentinel-2 from 2015 |
| SAR frequency | C-band, 5.405 GHz; VV and VH polarisations |
| Minimum detectable damage | ~30% stand loss at field scale (10 m); lighter damage requires optical confirmation |
| Post-event delivery latency | 12–24 hours from acquisition to damage layer, subject to downlink and processing pipeline |
| Coverage | Global; Sentinel-1 IW swath 250 km; systematic coverage of European and Central Asian cereal belts |
| Delivery formats | GeoTIFF damage-probability raster, vector damage zones by administrative unit, tabular area estimates (CSV/XLSX) |
Analytics Satellize can run
| Winter cereal area map | Supervised classification on Sentinel-1 VH backscatter time series plus Sentinel-2 NDVI and red-edge indices; random forest or similar ensemble classifier trained on reference parcels | GeoTIFF crop-area raster and area summary by administrative unit, issued at sowing completion and updated monthly through winter |
| Post-freeze backscatter anomaly layer | Change detection: post-event VH backscatter compared against pre-event seasonal trajectory and multi-year climatological median; z-score or threshold-based flagging | GeoTIFF anomaly map with damage-probability classes, delivered within 24 hours of post-freeze Sentinel-1 overpass |
| NDVI departure from baseline | Pixel-wise comparison of post-freeze Sentinel-2 NDVI against MODIS-derived climatological envelope for the same calendar week; anomaly expressed as standard deviations below median | GeoTIFF anomaly raster and regional summary report; conditional on cloud-free acquisition |
| Damage-area statistics by administrative unit | Zonal aggregation of damage-probability raster over crop-area mask; area-weighted severity classes (light, moderate, severe) assigned per district or oblast | Tabular report (CSV/XLSX) with hectares by damage class per administrative unit, formatted for direct input to national production forecast models |
| Seasonal phenology deviation index | Comparison of current-season Sentinel-1 and Sentinel-2 phenological trajectory against historical range derived from MODIS archive; flags fields developing anomalously slowly after a frost event | Time-series chart per region and GIS layer, updated at each satellite overpass through spring green-up |
| Frost-event attribution report | Co-registration of satellite damage signal with ERA5 or national meteorological station records of minimum temperature, duration below threshold, and snow cover presence; separates freeze damage from snow-load or soil-moisture artefacts | PDF technical report with methodology, confidence assessment, and spatial damage summary; suitable for submission to government forecasting agencies or insurers |
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.