Pasture productivity loss estimation for livestock drought insurance

Vegetation index time series from MODIS, Sentinel-2 and Landsat quantify pasture biomass deficits against long-term baselines, providing objective trigger data for index-based livestock insurance products and post-drought indemnity calculations.

Sensors

MODIS MOD13 / MYD13 (Terra and Aqua): 250 m NDVI and 500 m EVI at 16-day composites; the workhorse for index-based livestock insurance triggers because its archive runs to 2000, giving 25-plus years of baseline climatology. Cloud contamination is reduced by maximum-value compositing, though persistent cloud in humid rangelands still degrades quality.
VIIRS VNP13 (Suomi-NPP / NOAA-20): 500 m vegetation index product with daily overpass and 8-day composites; effectively extends and cross-calibrates the MODIS record as MODIS instruments age. Useful for near-real-time anomaly monitoring during an active drought event.
Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with both satellites. Allows NDVI and SAVI computation at paddock scale, resolving spatial heterogeneity invisible to MODIS. The archive begins in 2015, too short for standalone climatology but useful for downscaling MODIS anomalies.
Landsat 8 / 9 OLI (USGS / NASA): 30 m resolution, 16-day revisit per satellite (8-day combined). The Landsat archive from 1972 is the longest continuous land-surface record available; Collection 2 surface-reflectance products are well-suited to multi-decadal baseline construction. Revisit is too infrequent for operational in-season triggers but valuable for historical percentile calibration.

Why an insurer cannot simply send an adjuster

Livestock drought insurance covers millions of hectares of rangeland across Mongolia, the Sahel, the Horn of Africa and the pastoral belts of South America. Ground-based loss adjustment at that scale is not slow, it is impossible. A herder in Dundgovi province cannot wait three months for an adjuster who may never arrive. Index-based products solve the logistics problem by replacing field inspection with a satellite-derived trigger: when the vegetation index in a defined zone falls below a threshold percentile of its historical distribution, a payout is made automatically.

The trigger mechanism works because photosynthetically active biomass is the proximate driver of livestock body condition and mortality risk. Grasses reflect strongly in the near-infrared and absorb in the red; the ratio, expressed as NDVI, tracks green biomass with reasonable fidelity across most rangeland types. A sustained negative anomaly in NDVI relative to the long-run mean is a defensible proxy for forage deficit, even if it is not a direct measurement of tonnes of dry matter per hectare.

Building a baseline that a regulator will accept

The statistical credibility of any index product depends entirely on the quality of the historical baseline. MODIS MOD13 provides 16-day composite NDVI from 2000 onward at 250 m, giving roughly 25 growing seasons for most rangeland zones. From that record, analysts compute pixel-level percentile distributions for each compositing period, typically expressed as the 25th, 10th and 5th percentile thresholds. A current-season observation falling below the 25th percentile might trigger a partial payout; below the 10th percentile, a full payout. The exact thresholds are set during product design and written into the policy.

Landsat's 30 m archive extends back to the 1970s but requires careful inter-sensor calibration across five instrument generations. It is most useful for validating whether the MODIS baseline period is representative or happens to include an unusually wet or dry cluster of years. VIIRS VNP13 is now the preferred operational continuity instrument as the Terra and Aqua satellites approach end of life, and cross-calibration studies published in the remote sensing literature show good consistency with MODIS for rangeland NDVI.

Soil adjustment matters more than it sounds

In sparse rangelands, bare soil contributes substantially to the sensor signal, particularly early in the season or during severe drought when canopy cover drops below 30 percent. NDVI overestimates greenness in these conditions because bright soil in the red band suppresses the denominator. The Soil-Adjusted Vegetation Index (SAVI), which introduces a soil-brightness correction factor L, reduces this bias. Published values of L around 0.5 are commonly used for semi-arid rangelands, though the optimal value varies by soil type.

For insurance applications, the choice of index is not merely technical. If the index systematically overestimates biomass during drought because bare soil is brightening the signal, triggers will fire less often than they should, and herders will be undercompensated. Insurers and reinsurers reviewing index product design should ask specifically whether SAVI or EVI (which incorporates a canopy background adjustment) has been tested against the NDVI baseline, and whether the trigger thresholds were calibrated on the same index that will be used operationally.

Basis risk: the honest problem at the centre of the product

Index insurance does not pay when the cow dies. It pays when the index says the cow probably should have died. The gap between those two things is called basis risk, and it is the most serious limitation of the approach. A herder whose pasture sits on a productive valley floor may experience good forage even when the surrounding 250 m MODIS pixel averages a severe deficit. Conversely, a herder on degraded land may lose animals even when the index shows only a moderate anomaly.

Sentinel-2 at 10 m reduces but does not eliminate this problem. Paddock-level heterogeneity within a 10 m pixel is still real, and Sentinel-2's archive is too short to serve as a standalone trigger baseline. The practical approach is to use Sentinel-2 to identify within-pixel spatial variance and flag zones where basis risk is structurally high, then price those zones differently or exclude them from the product. Mongolia's Index-Based Livestock Insurance programme, supported by the World Bank, has grappled with basis risk since its inception in the early 2000s and remains the most thoroughly documented case study of both the method's promise and its limits.

East African programmes, including those operating in Kenya and Ethiopia under the IBLI framework developed at the International Livestock Research Institute, face additional complications from cloud cover during the long rains and from the spatial heterogeneity of Sahelian vegetation. Persistent cloud in a compositing window can cause a pixel to be filled with lower-quality observations, introducing noise into the trigger calculation. Maximum-value compositing mitigates this but does not eliminate it.

From anomaly map to indemnity figure

The analytic workflow runs in four stages. First, current-season composites are compared against the historical percentile surface to produce a deficit anomaly map, expressed in percentile units or as a z-score. Second, the anomaly is spatially averaged within the defined insurance zones, which may be administrative units, agro-ecological zones or custom polygons agreed with the regulator. Third, the zone-level index value is compared against the trigger and exit thresholds written into the policy. Fourth, the payout fraction is calculated according to the policy's linear or step function and reported to the insurer.

Satellize structures this workflow as a scheduled analytic feed, delivering zone-level index summaries at each compositing period during the policy season. The Tonga crop-estimation programme uses a comparable vegetation-index pipeline on Sentinel-2 and Landsat data, which informs the methodology applied here. Delivery formats are typically a structured data feed alongside a GIS layer showing the spatial anomaly pattern, so that both the insurer's actuarial team and any reinsurance counterparty can audit the underlying spatial data rather than relying on a single reported number.

What the method cannot do

Vegetation indices measure greenness, not nutritional quality. A pasture recovering from a previous drought may show near-normal NDVI while carrying grasses of poor digestibility or high fibre content. Livestock condition can deteriorate even when the index looks acceptable. This is a known and unresolved limitation of the approach.

The 250 m MODIS pixel is also simply too coarse for small-holder pastoral systems with paddocks under 10 hectares. Sentinel-2 at 10 m is the practical floor for paddock-level work, but its short archive requires careful statistical treatment. And no satellite index replaces a functioning livestock mortality reporting system: the index tells you the pasture was stressed; it cannot tell you whether the herder moved animals to better ground, purchased supplementary feed or suffered total loss. Honest product design acknowledges these gaps and sets trigger thresholds conservatively enough that basis risk does not systematically disadvantage the insured.

Typical figures

Spatial resolution (MODIS MOD13)	250 m (NDVI), 500 m (EVI)
Spatial resolution (Sentinel-2 MSI)	10 m (red, NIR bands used for NDVI/SAVI)
Spatial resolution (Landsat 8/9 OLI)	30 m
Revisit / compositing period	MODIS: 16-day composite (daily overpass); VIIRS: 8-day composite; Sentinel-2: 5-day (two satellites); Landsat: 8-day (combined 8 and 9)
Archive depth for baseline	MODIS: 2000 to present (~25 years); Landsat: 1972 to present; Sentinel-2: 2015 to present; VIIRS: 2012 to present
Key spectral bands	Red (~665 nm) and near-infrared (~865 nm) for NDVI/SAVI; blue (~490 nm) added for EVI atmospheric correction
Minimum resolvable insurance zone	Practical minimum ~1 km² for MODIS-based triggers; ~1 ha feasible with Sentinel-2 (subject to archive-length caveats)
Trigger latency after compositing period closes	Typically 3–10 days for official MODIS/VIIRS product release; Sentinel-2 L2A available within 1–3 days via Copernicus Data Space
Cloud contamination risk	Reduced by maximum-value compositing; persistent cloud in humid rangelands can still degrade 16-day composites
Delivery formats	Zone-level index summary (CSV/JSON feed), GeoTIFF anomaly raster, GIS polygon layer with percentile attribution

Analytics Satellize can run

Historical percentile baseline surface	Pixel-level percentile distribution (5th, 10th, 25th) computed from MODIS MOD13 archive; cross-validated against Landsat Collection 2 for representativeness	GeoTIFF raster stack of percentile thresholds per compositing period, delivered once at policy inception
In-season NDVI / SAVI anomaly map	Current composite compared against baseline percentile surface; SAVI computed with L=0.5 for sparse canopy correction	GeoTIFF anomaly layer and zone-level summary table, updated each compositing period
Zone-level trigger assessment	Spatial averaging of anomaly within defined insurance polygons; comparison against policy trigger and exit thresholds	Structured JSON/CSV feed per zone per period, flagging trigger status and payout fraction
Basis risk variance map	Within-zone coefficient of variation of Sentinel-2 NDVI used to quantify spatial heterogeneity invisible to MODIS	GIS polygon layer annotating each zone with a basis-risk rating; input to actuarial pricing review
End-of-season deficit accumulation index	Cumulative sum of below-threshold anomaly days across the growing season, analogous to published MODIS-based vegetation condition index methods	Single-value zone summary report for post-drought indemnity calculation and reinsurance settlement
Cloud-contamination quality flag	Per-pixel quality assurance band from MOD13 and Sentinel-2 scene classification layer used to flag composites with insufficient clear observations	Quality flag column appended to zone summary feed; alerts insurer when a trigger decision rests on degraded data

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.