Satellite vegetation index triggers for drought index insurance

Parametric drought insurance can pay out without loss adjustment, but only if the trigger index actually tracks crop stress. Satellite vegetation indices offer spatially continuous drought signals where rain gauges are absent, at the cost of residual basis risk when canopy condition diverges from root-zone moisture.

Sensors

MODIS Terra/Aqua: 250 m resolution in red and NIR bands (MOD09GQ/MYD09GQ), daily revisit per satellite, 16-day composite products (MOD13Q1) reduce cloud contamination. Archive from 2000 onward gives the 20-plus-year climatology needed to compute VCI anomalies. Thermal bands at 1 km support land-surface temperature stress indices.
VIIRS VNP09 (Suomi-NPP / NOAA-20): 375 m I-band daily surface reflectance, continuing the MODIS record with improved radiometric calibration. The VNP13 vegetation index product provides NDVI and EVI composites at 500 m and 1 km. Operationally useful for near-real-time trigger monitoring where daily cloud-free observations are needed.
Sentinel-2 MSI: 10 m resolution in red and NIR (bands 4 and 8), 5-day revisit at the equator with both Sentinel-2A and 2B operating. Enables field-level NDVI mapping, resolving individual smallholder plots down to roughly 0.1 ha. Cloud cover in humid tropics limits effective revisit; temporal compositing over 10 to 20 days is typically required.
Planet SuperDove: 3 m resolution, daily revisit over most land areas, 8 spectral bands including red-edge. Useful for validation of coarser index products and for resolving heterogeneous smallholder landscapes where a 250 m MODIS pixel mixes stressed and healthy crops. Requires a commercial licence; archive depth is shorter than MODIS.

Why rain gauges cannot anchor a payout

Parametric drought insurance replaces subjective loss adjustment with an objective index: if the index crosses a threshold, the policy pays. The appeal is speed and low administrative cost. The problem is that most of the world's agricultural land sits far from a functioning rain gauge. Sub-Saharan Africa, Central Asia and the Pacific islands all have gauge networks too sparse to represent the spatial variability of a drought event at the scale of an individual farm or even a district.

Satellite vegetation indices sidestep the gauge problem entirely. They measure what the plant is actually doing rather than inferring it from rainfall. The Vegetation Condition Index, first formalised by Kogan in the early 1990s and subsequently operationalised through NOAA's drought monitoring work, expresses current NDVI as a percentile of the historical range for that pixel and that calendar week. A VCI of 20 means the vegetation is greener than only 20 percent of the historical record for that location and time of year. That is a drought signal the insurer can write into a contract.

Building the index: climatology, smoothing and thresholds

A credible VCI product needs at minimum 10 years of consistent satellite data to estimate the local NDVI range reliably; 20 years is better. MODIS provides a continuous archive from February 2000, making it the workhorse for historical calibration. The raw daily reflectance is noisy: cloud shadow, aerosol loading and bidirectional reflectance effects all depress NDVI transiently. Standard practice is to use maximum-value compositing over 16-day windows, as in the MOD13Q1 product, which selects the least-cloud-contaminated observation in the period.

Temporal smoothing matters enormously for insurance design. A single bad 16-day composite caused by a dust storm should not trigger a payout. Most operational drought index products apply a further smoothing step, such as a Savitzky-Golay filter or a running three-period weighted mean, before computing the VCI percentile. The insurance contract must specify the exact compositing and smoothing algorithm, the reference archive period and the spatial aggregation unit (pixel, administrative zone or insured parcel). Ambiguity in any of these parameters creates legal exposure.

Threshold calibration requires historical loss data. Without ground-truth yield records or indemnity loss ratios from prior seasons, a VCI threshold of 35 is as defensible as one of 25. Where such data exist, logistic regression or simple correlation analysis can identify the VCI level that best separates loss years from non-loss years at a given location. Where they do not, the insurer is essentially setting a threshold by agronomic judgement, which should be disclosed to the policyholder.

The basis risk problem: what the canopy hides

Basis risk is the gap between what the index measures and what the farmer actually loses. It is the central weakness of any parametric product, and satellite vegetation indices do not eliminate it; they redistribute it.

The most common failure mode is the canopy-soil moisture divergence. In the early stages of a drought, plants draw on root-zone soil moisture reserves that are invisible to optical sensors. NDVI can remain near-normal for two to four weeks after rainfall has ceased, depending on soil type and rooting depth. A policy triggered on VCI alone would not pay during this lag, even though the agronomic damage is accumulating. Conversely, a late-season rain event can green up a canopy that has already experienced irreversible reproductive failure, producing a false-negative: the index recovers but the yield does not.

A second failure mode is spatial mismatch. At 250 m, a single MODIS pixel covers 6.25 ha. In a smallholder landscape with plot sizes of 0.5 to 2 ha, the pixel aggregates stressed and unstressed fields. The VCI for that pixel may sit above the trigger threshold even when a majority of the constituent farms are in drought. Sentinel-2 at 10 m reduces this problem substantially but introduces its own trade-off: higher cloud sensitivity and a shorter effective revisit in tropical climates.

Verification and contract audit

Insurers and reinsurers increasingly require independent verification that the trigger index was computed correctly and consistently with the contract specification. This is not a trivial audit. The VCI value for a given pixel on a given date depends on the surface reflectance product version, the compositing window, the smoothing algorithm and the reference archive. A change in the upstream MODIS collection version, for instance the shift from Collection 6 to Collection 6.1, can alter historical NDVI values and shift percentile rankings.

A verification workflow typically involves three steps. First, reproduce the index from raw Level-2 reflectance data using the contract-specified algorithm, independently of the data provider's operational system. Second, compare the reproduced index against the trigger threshold for each spatial unit and each monitoring period covered by the policy term. Third, document any data gaps caused by persistent cloud cover and specify how the contract handles them, whether by interpolation, by extending the monitoring window or by defaulting to a secondary index such as land-surface temperature anomaly.

Satellize's analytics work on the Kingdom of Tonga crop-estimation programme involved exactly this kind of independent index computation from open constellation data, which is directly transferable to trigger verification for parametric products in island and smallholder contexts.

Honest limits and what to do about them

No satellite vegetation index is a perfect proxy for crop loss. Cloud cover in humid tropical regions can produce data gaps lasting several weeks during the critical flowering and grain-fill stages, precisely when drought stress is most damaging. MODIS and VIIRS daily imagery mitigates this through temporal compositing, but a persistent monsoon break can still leave a monitoring gap. Sentinel-2's 5-day revisit is not 5 cloud-free days.

Persistent cloud is the most tractable problem. Synthetic aperture radar, which penetrates cloud, can provide a complementary soil moisture signal, though SAR-derived soil moisture at field scale remains an active research area rather than an operational product. A hybrid trigger that combines VCI from optical data with a microwave soil moisture anomaly from Sentinel-1 or SMAP reduces the probability of both false positives and false negatives, at the cost of a more complex contract specification.

The deeper limit is agronomic: no remote sensing system currently measures grain yield directly. NDVI and VCI are proxies for green biomass, not for harvestable output. For crops where the economic value is in the grain rather than the canopy, the relationship between VCI and loss is statistical and location-specific. Buyers of parametric products based on these indices should ask their insurer for the historical correlation between the trigger index and actual loss ratios in their region, and for the confidence interval around that correlation. If the insurer cannot provide it, the basis risk is unquantified.

Typical figures

Spatial resolution (MODIS VCI)	250 m (red/NIR bands); 500 m for multi-band composites
Spatial resolution (Sentinel-2 NDVI)	10 m (bands 4 and 8)
Revisit frequency	Daily (MODIS/VIIRS); 5 days at equator (Sentinel-2A+B combined)
Standard compositing window	16-day maximum-value composite (MODIS MOD13Q1); 10 to 20 days typical for Sentinel-2 in cloud-prone areas
Archive depth for climatology	MODIS from February 2000 (24-plus years); VIIRS from 2012; Sentinel-2 from 2015
Key spectral bands	Red (~665 nm) and NIR (~865 nm) for NDVI; thermal infrared at 10.8 µm for land-surface temperature stress index
Minimum spatial aggregation unit	Single pixel (~250 m MODIS) to administrative polygon; field-level with Sentinel-2 or Planet
Index latency (operational products)	MODIS 16-day composites available within 2 to 4 days of period end; near-real-time daily VIIRS within 24 hours
Delivery formats	GeoTIFF raster, polygon-aggregated CSV, API time-series feed, PDF trigger verification report
Cloud-cover limitation	Optical indices unavailable under persistent cloud; gap tolerance depends on compositing window length

Analytics Satellize can run

VCI time series per insured zone	NDVI percentile ranking against pixel-level MODIS/VIIRS climatology (Kogan VCI method)	Season-long weekly VCI raster and zone-aggregated CSV with trigger-threshold flags
NDVI anomaly map at field scale	Sentinel-2 NDVI deviation from multi-year seasonal mean, Savitzky-Golay smoothed	10 m GeoTIFF anomaly layer per monitoring period, clipped to policy boundary
Trigger verification report	Independent recomputation of contract-specified index from Level-2 reflectance; comparison against threshold for each spatial unit and monitoring window	Auditable PDF report with index values, threshold crossings, data-gap log and methodology appendix
Basis risk assessment	Correlation analysis between historical VCI and available yield or loss-ratio records; spatial mismatch quantification by plot-size distribution	Statistical summary report with correlation coefficients, confidence intervals and recommended threshold ranges
Cloud-gap detection and interpolation log	Per-pixel cloud-mask analysis across monitoring period; linear or harmonic interpolation where gap duration is below contract-specified tolerance	Gap-flagged raster stack with interpolation confidence layer and gap-duration statistics per zone
Historical trigger back-test	Application of contract algorithm to full MODIS archive to reconstruct hypothetical payout history; comparison against historical drought event records	Annual trigger/no-trigger table with VCI values, payout flags and overlay against publicly documented drought years

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.