Hail damage rapid assessment in standing crops

Hail bruises leaves, snaps stems, and strips canopy in hours. Pre/post red-edge and NIR change detection on sub-10 m imagery can map the damage footprint before recovery or disease obscures what the storm actually did.

Sensors

PlanetScope: 3 m native resolution, daily global revisit from the SuperDove constellation, four to eight spectral bands including red-edge at 700–750 nm. The combination of fine spatial resolution and near-daily cadence makes it the primary sensor for capturing damage within 24 hours of a strike and for tracking recovery day by day.
Sentinel-2 MSI: 10 m in visible and NIR bands, 20 m in red-edge (bands 5, 6, 7) and SWIR. Five-day revisit at the equator, reduced to 2–3 days at mid-latitudes with both satellites. Free and open; the red-edge bands are particularly sensitive to chlorophyll loss from bruising. Cloud cover on the day after a storm is the main operational constraint.
WorldView-3: 31 cm panchromatic, 1.24 m multispectral, 16 bands including SWIR. Tasked on demand; latency depends on orbital geometry and cloud, typically same-day to 2-day. Useful for ground-truth validation of damage class and for high-value crop situations where field-level evidence is legally required.
NEXRAD / OPERA radar composites: Not an imaging sensor for crops, but the essential first step. NEXRAD (US) and EUMETNET OPERA (Europe) provide reflectivity and hail-detection products at roughly 1 km resolution with near-real-time latency. Hail core polygons from these composites define the probable impact zone and drive targeted image ordering, avoiding the cost of tasking unaffected areas.

What a hailstorm does to a canopy, spectrally

Hail damage is not subtle. A strike event can shred leaf tissue, snap stems at the node, strip flag leaves from cereals, and defoliate maize ears in minutes. The physical result is a rapid collapse in canopy greenness confined to the storm track, which may be as narrow as 1 km or as wide as 10–15 km depending on cell width and forward speed. What matters for remote sensing is that the damage is abrupt, spatially bounded, and spectrally distinct from the gradual stress signals produced by drought or disease.

Healthy green vegetation absorbs strongly in the red (around 670 nm) and reflects strongly in the NIR (around 800 nm). Leaf bruising disrupts the spongy mesophyll structure that drives NIR reflectance, while chlorophyll degradation reduces red absorption. Both changes register as a drop in NDVI and, more sensitively, in red-edge indices such as NDRE (normalised difference red-edge index). The red-edge region, roughly 700–740 nm, sits at the inflection point of the vegetation reflectance curve and responds to chlorophyll concentration changes before they are visible to the eye. Sentinel-2 bands 5 and 6 and PlanetScope's red-edge band exploit exactly this physics.

Why timing and resolution are not negotiable

Crops begin to recover within days. Maize can produce new leaf tissue within 3–5 days of a moderate strike if the growing point survives. A Sentinel-2 image acquired five days post-event may already understate the damage extent. This is not a theoretical concern; it is the central operational problem. Same-day or next-day acquisition is the target, which is why PlanetScope's daily revisit is the workhorse here and why Sentinel-2's 2–5 day revisit is a genuine constraint rather than a minor inconvenience.

Spatial resolution matters equally. A hail swath 2 km wide crossing a patchwork of 20-hectare fields will contain sharp internal gradients, with some fields fully in the core and adjacent fields receiving only marginal damage. At 10 m resolution, those gradients are visible. At 30 m (Landsat), many partially damaged fields are averaged into ambiguity. For legal or insurance purposes, the difference between 30 % and 70 % canopy loss in a specific field is not a rounding error. Sub-10 m imagery is the practical threshold for field-level attribution.

Cloud cover is the honest caveat. Convective storms that produce hail also produce anvil cloud that can persist for hours and return the following day. A tasked WorldView-3 or PlanetScope acquisition may be blocked for 24–48 hours. Operators should pre-order imagery for any region where radar indicates a hail core, accepting that some acquisitions will be cloud-contaminated and require the next available pass.

The radar-first workflow

No satellite image analyst should order post-event imagery without first consulting weather radar. NEXRAD in the United States and OPERA in Europe both produce near-real-time hail probability composites, typically at 1 km resolution, derived from reflectivity thresholds and dual-polarisation variables such as differential reflectivity (ZDR) and correlation coefficient. These products delineate the probable hail core to within a few kilometres, which is sufficient to define a tasking polygon for commercial satellites and to identify which Sentinel-2 tiles to prioritise for download.

The radar composite also provides a timestamp for the event, which anchors the pre/post comparison. The pre-event image should be the most recent clear acquisition before the storm, ideally within 7–14 days to minimise phenological drift. Longer gaps introduce noise because the crop itself is changing. In practice, a PlanetScope pre-image from 3–5 days prior and a post-image from the following morning give a clean change signal if cloud cooperates.

Change detection: method and honest limits

The standard approach is pixel-wise differencing of a vegetation index between the pre- and post-event images, after radiometric normalisation to account for differences in solar angle and atmospheric conditions. NDRE is preferred over NDVI for this application because it is more sensitive to moderate chlorophyll loss and less prone to saturation in dense canopies at peak season. The change image is then thresholded to produce damage severity classes, typically light (10–30 % index decline), moderate (30–60 %), and severe (greater than 60 %), though the exact thresholds should be calibrated against any available ground observations.

The method has known weaknesses. First, it cannot distinguish hail damage from other causes of abrupt canopy decline, such as a late frost event or chemical drift, without corroborating evidence from the radar composite. The spatial confinement of the damage to the storm track is the primary discriminator. Second, mixed pixels at field boundaries will produce intermediate values that are genuinely ambiguous. Third, if the post-event image is acquired more than 48–72 hours after the strike, early recovery in younger or more vigorous plants will compress the apparent damage signal. Analysts should report a confidence interval on damage area, not a single number, and flag any fields where the pre/post interval exceeded 48 hours.

WorldView-3's 1.24 m multispectral capability can resolve individual plant rows in maize, which allows stem breakage to be distinguished from leaf bruising by texture analysis. This level of detail is rarely needed for area estimation but is valuable for crop-loss verification in high-value situations.

Connecting the map to a decision

A damage severity map is only useful if it connects to something actionable: an insurance claim, a government disaster declaration, a replanting decision, or a revised yield forecast. The damage area in each severity class, multiplied by a crop-specific yield-loss coefficient, gives a first-order production loss estimate. Published agronomic studies on hail damage in maize and wheat provide these coefficients as a function of growth stage, though they carry substantial uncertainty and should be presented as ranges.

Satellize runs this type of rapid change-detection workflow on open constellations supplemented by commercial tasking on client licence. The Tonga crop-estimation programme demonstrates the organisation's capacity to deliver field-level analytics in geographies where ground-truth is sparse, which is the same operational challenge that arises after a hail event in a remote agricultural district. For a hail assessment engagement, the concrete next step is to share the radar-derived impact polygon and the target crop type, so that image availability and tasking options can be assessed against the event timeline.

Typical figures

Primary spatial resolution	3 m (PlanetScope SuperDove); 10–20 m (Sentinel-2 MSI red-edge bands at 20 m)
Revisit cadence	Daily (PlanetScope); 2–5 days at mid-latitudes (Sentinel-2 twin satellites); on-demand 1–2 day latency (WorldView-3, subject to cloud and orbital geometry)
Key spectral bands	Red-edge 700–740 nm (Sentinel-2 B5/B6, PlanetScope Band 6); NIR 750–900 nm; Red 650–680 nm; SWIR 1550–1650 nm for soil/canopy separation
Weather radar input	NEXRAD Level-III hail products (US, ~1 km); EUMETNET OPERA composite (Europe, ~1–2 km); near-real-time latency typically under 15 minutes
Minimum detectable damage patch	Approximately 0.5–1 ha at 3 m resolution; approximately 4–9 ha at 10 m resolution (single-pixel detection is unreliable; contiguous clusters of affected pixels are the practical threshold)
Optimal post-event acquisition window	Same day to 48 hours; utility degrades significantly beyond 72 hours due to crop recovery
Archive depth for pre-event baseline	PlanetScope: from 2016; Sentinel-2: from 2015; WorldView-3: from 2014 (tasked archive, coverage variable)
Delivery formats	GeoTIFF damage severity raster; vector field-level summary (GeoPackage or Shapefile); PDF damage report with area statistics by severity class
Cloud cover constraint	Optical sensors are blind through cloud; radar composites remain available but do not substitute for post-event imagery. SAR (Sentinel-1) can detect lodging but does not reliably detect leaf bruising at moderate severity.

Analytics Satellize can run

Hail impact zone polygon	Radar reflectivity and hail-detection algorithm output (NEXRAD MESH or OPERA hail product) clipped and vectorised to define the tasking and analysis boundary	GeoJSON polygon delivered within 2 hours of event identification, used to drive image ordering
Pre/post NDRE change map	Pixel-wise differencing of atmospherically corrected NDRE between the most recent clear pre-event acquisition and the first clear post-event acquisition; radiometric normalisation using pseudo-invariant features	GeoTIFF raster classified into light, moderate, and severe decline categories, with per-class area statistics
Field-level damage summary table	Zonal statistics of the change raster against field boundary polygons (client-supplied or derived from Sentinel-2 edge detection); majority-class assignment per field	CSV or GeoPackage with field ID, dominant damage class, mean index change, and area in each severity band
Damage progression time series	Daily PlanetScope NDRE stack over the 7–14 days following the event to track recovery rate by field and severity class	Chart and raster stack showing index recovery trajectory; flags fields where no recovery is observed (indicative of severe stem or root damage)
Production loss estimate	Damage area by severity class multiplied by published crop-stage yield-loss coefficients for the relevant crop (maize or cereal); uncertainty bounds derived from coefficient ranges in the agronomic literature	PDF summary table with low, central, and high production loss estimates in tonnes; caveats on growth stage assumptions stated explicitly
Cloud-gap alert	Automated monitoring of Sentinel-2 and PlanetScope scene metadata within the impact polygon; triggers notification when the first cloud-free acquisition is available	Email or API alert with scene ID, acquisition timestamp, and estimated cloud fraction over the impact zone

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.