Farm-level crop loss verification for index and indemnity insurance
High-resolution optical and SAR imagery can verify crop losses at individual field level, but resolution floors, cloud cover, and evidentiary standards vary sharply by sensor and peril type.
Sensors
- Sentinel-1 SAR (C-band, ESA): 10 m ground range resolution in Interferometric Wide swath mode, 6-day repeat at the equator with both satellites active. All-weather, day-night imaging. Backscatter drops sharply over open water and flattened crop canopies, making flood inundation and lodging detectable regardless of cloud cover, which is critical in monsoon-affected insurance markets.
- Planet SuperDove: 3 m multispectral imagery (eight bands including red-edge and NIR) with near-daily revisit globally. Sufficient to isolate plots as small as 0.1 ha in open terrain. Cloud dependence remains a hard constraint; optical data gaps of 10–20 consecutive days are common during wet seasons in South and Southeast Asia.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, with short-wave infrared bands. At this resolution individual plant rows are visible, and hail-strike bruising on leaf canopies produces detectable SWIR reflectance changes. Tasked on demand; cost and cloud risk make it unsuitable as the primary monitoring layer but valuable for spot adjudication of contested claims.
- Sentinel-2 MSI (ESA): 10 m visible and NIR, 20 m red-edge and SWIR, 5-day revisit with both satellites. The free archive back to 2017 enables pre-season baseline construction. At 10 m, many smallholder plots in fragmented landscapes fall below two pixels across, creating boundary-mixing errors that inflate apparent damage or mask it entirely.
What the physics actually detects
Crop damage changes how a canopy interacts with electromagnetic radiation. Flood inundation replaces a rough, vegetated surface with open water, causing C-band SAR backscatter to fall by 3–8 dB relative to a pre-event baseline. Drought-induced senescence reduces chlorophyll, collapsing the red-edge reflectance peak that healthy plants show between 700 and 740 nm. Hail damage ruptures cell structure, raising SWIR reflectance as internal water content drops. Each of these signals is physically grounded and repeatable, which is precisely what makes satellite evidence admissible in an insurance context.
The critical word is 'relative'. A single post-event image proves nothing without a pre-event baseline from the same sensor and season, ideally from multiple prior years to account for natural phenological variation. An insurer presenting satellite evidence without a documented baseline methodology is presenting an anecdote, not a measurement.
Index products versus parcel-specific loss: a meaningful distinction
Area-yield index insurance pays out when a regional index, typically derived from NDVI or rainfall, falls below a threshold. It does not require individual field verification. The basis risk, meaning the gap between what the index says and what a specific farmer actually lost, is the product's known weakness, not a satellite problem.
Parcel-specific indemnity insurance is different. It requires evidence that the loss occurred on the named insured parcel, not the neighbouring field. This demands spatial resolution fine enough to isolate the plot boundary, which in fragmented smallholder landscapes in sub-Saharan Africa or South Asia can mean plots of 0.2 to 0.5 ha. At Sentinel-2's 10 m resolution, a 0.3 ha square plot is roughly 17 pixels, and boundary pixels are contaminated by adjacent land cover. Planet SuperDove at 3 m gives roughly 330 pixels over the same area, with far cleaner boundaries. The resolution choice is therefore not a technical preference; it is a legal one.
Cloud cover is not a nuisance. In the tropics it is a structural problem.
The perils that generate the most insurance claims, floods and cyclone-related inundation, occur during seasons of persistent cloud. A study of Sentinel-2 availability over Bangladesh during the 2017 monsoon found usable optical scenes were available on fewer than 15 days between June and September. If the loss event falls inside that optical gap, an optical-only workflow cannot produce evidence. This is not a solvable problem with better scheduling; it is a consequence of cloud physics.
Sentinel-1 SAR penetrates cloud and rain (with some attenuation in very heavy precipitation) and maintains its 6-day revisit regardless of season. The practical workflow for tropical indemnity insurance is therefore SAR-primary for flood and lodging detection, with optical imagery used when available to confirm canopy condition and distinguish crop type. Attempting to run an optical-only tropical insurance programme is a design error, not a data gap.
Evidentiary standards: what an insurer's legal team will actually accept
Satellite evidence enters the claims process at two points: automated triage, where imagery flags parcels for follow-up, and adjudication, where it supports or contests a specific claim. The evidentiary bar is higher for adjudication. Most reinsurers and national agricultural insurance programmes require that the methodology be documented, reproducible, and independently auditable. That means a written specification of the pre-event baseline period, the change-detection algorithm (change vector analysis, thresholded difference index, or supervised classification), the geometric accuracy of parcel boundaries, and the uncertainty estimate on the loss fraction.
Geometric accuracy deserves particular attention. If the cadastral parcel boundary used to extract the satellite signal is displaced by 15 m relative to the true field edge, and the field is 50 m wide, a significant fraction of the extracted pixels belong to the wrong field. GPS-verified boundary data, or at minimum a documented co-registration step, is not optional for indemnity products. It is the difference between evidence and noise.
Where the method reaches its honest limits
Partial damage within a field is harder than total loss. A hail event that kills 40% of a maize stand in a patchy pattern may not produce a statistically separable signal at 10 m resolution if the damaged and undamaged areas are interleaved at sub-pixel scale. Sub-metre imagery from WorldView-3 can resolve this, but tasking latency and cost make it impractical as a routine product for low-premium smallholder policies.
Drought-induced senescence and natural crop maturity produce similar spectral signatures. A field that is brown because the farmer harvested early looks identical from orbit to one that is brown because drought killed the crop. Distinguishing the two requires either a time series dense enough to identify the onset date (was browning gradual or abrupt?) or ground-truth data from the field. Neither is always available. Honest loss verification programmes document this ambiguity explicitly and define the conditions under which satellite evidence is treated as indicative rather than conclusive.
Satellize's crop-estimation work for the Kingdom of Tonga involved constructing seasonal baselines from open-constellation data, a methodology directly transferable to insurance loss verification where pre-event normals are the foundation of any defensible claim assessment.
Building a workflow that survives audit
A credible farm-level loss verification workflow has four documented components. First, a pre-season baseline built from at least two prior years of the same sensor and phenological window. Second, a change-detection step with a defined threshold and uncertainty band, not just a visual interpretation. Third, geometric validation of parcel boundaries against a known reference. Fourth, a confidence tier that distinguishes high-confidence total loss from ambiguous partial damage, so that adjusters know which cases need ground inspection.
The output is not a single damage map. It is a structured dataset with per-parcel loss estimates, confidence scores, and the dates of the pre- and post-event imagery used. That structure is what allows a reinsurer, a regulator, or a court to audit the finding. Programmes that skip the documentation to accelerate claims processing tend to find that speed creates liability rather than reducing it.
Typical figures
| Spatial resolution (SAR, Sentinel-1 IW mode) | 10 m range × 10 m azimuth (after multi-looking) |
| Spatial resolution (optical, Planet SuperDove) | 3 m multispectral |
| Spatial resolution (optical, Maxar WorldView-3) | 31 cm panchromatic, 1.24 m multispectral, 3.7 m SWIR |
| Revisit (Sentinel-1, both satellites) | 6 days at equator; shorter at higher latitudes |
| Revisit (Planet SuperDove) | Near-daily globally, subject to cloud cover |
| Minimum detectable parcel (10 m SAR) | Approximately 0.5 ha for reliable boundary isolation; smaller parcels subject to boundary-mixing error |
| Minimum detectable parcel (3 m optical) | Approximately 0.1 ha in open terrain with clean cadastral boundaries |
| SAR frequency and polarisation | C-band (5.405 GHz); VV and VH dual-polarisation in IW mode |
| Archive depth (Sentinel-1 and Sentinel-2) | From 2014 (Sentinel-1A launch) and 2015 (Sentinel-2A launch) respectively |
| Typical post-event data latency | Sentinel-1 and Sentinel-2 products available within 1–3 hours of downlink via Copernicus Data Space; commercial tasking 24–48 hours depending on cloud and orbit |
Analytics Satellize can run
| Pre-event canopy baseline | Multi-year median compositing of NDVI and SAR backscatter over the same phenological window, using Sentinel-1 and Sentinel-2 archives | Per-parcel baseline raster and tabular statistics, GIS layer |
| Flood inundation flag per parcel | Thresholded SAR backscatter change detection (VV polarisation, pre/post difference) with water-body masking to suppress false positives | Binary flood-affected / not-affected parcel dataset with confidence score, GIS layer and CSV |
| Drought senescence severity score | Red-edge chlorophyll index (Sentinel-2 bands B5/B4) change from seasonal baseline, with onset-date estimation from dense time series | Per-parcel loss-fraction estimate (0–100%) with onset date and confidence tier, tabular report |
| Hail damage detection | SWIR reflectance anomaly mapping (Sentinel-2 B11/B12 or WorldView-3 SWIR) combined with rapid post-event tasking within 48 hours of reported event | Damage probability map at parcel level, flagged parcels for ground-truth adjuster routing |
| Ambiguity classification (harvest vs. damage) | Time-series shape analysis distinguishing abrupt spectral change (damage) from gradual senescence (maturity/harvest) using Planet SuperDove dense stack | Per-parcel classification with supporting time-series chart, PDF adjudication report |
| Audit-ready evidence package | Structured documentation of baseline period, algorithm specification, geometric co-registration log, and uncertainty estimates per parcel | Signed PDF methodology report plus versioned GIS dataset suitable for reinsurer or regulatory submission |
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.