Vegetation stress and drought severity for food-security early warning

Satellite vegetation indices detect agricultural drought stress two to six weeks before food insecurity becomes visible on the ground. This page explains the operational methods, honest resolution limits, and the workflows used by FEWS NET and WFP VAM.

Sensors

MODIS Terra/Aqua (MOD13/MYD13): 250 m NDVI and 500 m EVI at 16-day composites; daily overpass gives near-daily cloud-free compositing opportunity. Archive from 2000 onward enables robust 20-plus-year baselines for anomaly calculation. The workhorse of operational food-security monitoring.
VIIRS VNP13 (Suomi-NPP / NOAA-20): 500 m NDVI and EVI at 16-day composites; designed as the MODIS continuity product. Slightly improved radiometric calibration. Daily overpass. Increasingly used by FEWS NET as MODIS instruments age.
Sentinel-2 MSI: 10 m (Red, NIR) to 20 m (Red-Edge) resolution; 5-day revisit at the equator with both satellites. No long archive baseline before 2015, which limits anomaly climatology depth. Cloud cover in humid tropics remains a serious constraint on reliable compositing.
CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data): 0.05-degree (~5 km) gridded rainfall estimates blending TRMM/GPM infrared and station records. Produced by USGS/UCSB. Provides the meteorological drought signal that contextualises vegetation anomalies; archive from 1981.

Three kinds of drought, and why the distinction matters operationally

Drought is not one thing. Meteorological drought is a rainfall deficit relative to a long-term average. Agricultural drought is what happens to crops and pasture when soil moisture falls below what plants need, which may lag rainfall deficits by weeks or lead them if soils were already dry. Hydrological drought shows up in river flows and groundwater, often months later still. Food-security analysts care most about agricultural drought because that is what kills harvests. Satellite vegetation indices measure it directly.

CHIRPS rainfall anomalies flag meteorological drought quickly but tell you nothing about whether the crop in the ground is actually suffering. A late-season rainfall deficit may be irrelevant if the crop has already reached maturity. Conversely, a moderate rainfall deficit on sandy soils can devastate a crop that a clay-soil farmer nearby weathers without damage. Vegetation indices resolve this ambiguity because they measure the plant's actual physiological state, not the weather above it.

What VCI and NDVI anomaly actually measure, and what they miss

The Vegetation Condition Index rescales current NDVI against the historical minimum and maximum for the same pixel and calendar week: VCI = (NDVI_current − NDVI_min) / (NDVI_max − NDVI_min) × 100. A value below 35 is conventionally treated as moderate stress; below 20 as severe. The appeal is that it normalises for land-cover type, so a savanna pixel and a cropland pixel become comparable on the same 0-to-100 scale. FEWS NET and WFP VAM both publish VCI-derived stress maps as operational products.

NDVI anomaly relative to the long-term median (often expressed as a z-score or percentage departure) is simpler and arguably more transparent. Both methods share the same fundamental limit: they measure greenness, not yield. A stressed crop that recovers before grain fill may still produce a reasonable harvest. A crop that looks green because weeds have colonised a failed stand will produce almost nothing. Ground-truthing at the field level remains irreplaceable for translating stress signals into yield estimates.

Phenological timing is a further complication. Comparing this week's NDVI to the multi-year median is only meaningful if the crop is at the same growth stage. Late planting, common when rains are erratic, shifts the phenological calendar and can make a healthy late crop look stressed against a median derived from on-time seasons. Some operational systems now use phenology-adjusted anomalies, but this adds complexity and its own assumptions.

The 250 m to 1 km resolution floor and what it means for smallholder agriculture

MODIS and VIIRS pixels at 250 m to 500 m are large enough to contain multiple fields, fallow strips, and access tracks. In landscapes of smallholder agriculture, which dominate food-insecure regions of sub-Saharan Africa and South Asia, a single pixel routinely mixes stressed crops, healthy kitchen gardens, and bare soil. The resulting NDVI is a weighted average of all of them. Stress in one field can be masked by vigour in an adjacent one.

Sentinel-2 at 10 m resolves individual smallholder plots in principle, but the archive only begins in 2015, giving fewer than ten years of baseline for anomaly calculation. That is thin. Persistent cloud cover during the growing season in humid tropical regions can reduce usable observations to a handful per season, making reliable compositing difficult. The honest position is that Sentinel-2 adds diagnostic resolution for targeted investigation of flagged areas, but MODIS and VIIRS remain the operational backbone for continental-scale early warning precisely because of their depth of archive and reliable compositing.

How FEWS NET and WFP VAM turn pixels into food-security assessments

FEWS NET (the Famine Early Warning Systems Network, funded by USAID) produces monthly food-security outlooks for around 35 countries. Its remote-sensing inputs include MODIS NDVI anomaly, VCI, CHIRPS rainfall, and evapotranspiration estimates from land-surface models. These are combined with market price data, livelihood zone classifications, and field reports to produce Integrated Food Security Phase Classification (IPC) projections. The satellite data provides spatial coverage that field teams cannot match; the field data provides the ground truth that satellite indices alone cannot supply.

WFP's Vulnerability Analysis and Mapping unit (VAM) runs a broadly similar workflow, with particular emphasis on the mVAM mobile survey system to validate remote-sensing signals at the household level. Both systems are explicit that satellite vegetation indices are leading indicators, not conclusions. A VCI below 35 triggers investigation, not a declaration of crisis. That epistemic humility is baked into the operational design, and analysts working with these products should carry it forward.

Latency, archive depth, and the practical delivery chain

MODIS 16-day composites are typically available within two to three days of the composite period closing, through NASA's EARTHDATA platform. VIIRS VNP13 follows a similar schedule. CHIRPS dekadal (10-day) products are released roughly five days after the period ends. Sentinel-2 Level-2A surface reflectance is available within hours of acquisition through the Copernicus Data Space. The operational latency that matters is the time from a stress event to a decision-maker seeing an alert. With automated processing, that can be under a week for MODIS-based products.

Archive depth is a genuine competitive advantage of the MODIS record. Twenty-plus years of consistent observations from the same sensor family means that anomalies can be calculated against a climatology that spans multiple El Niño and La Niña cycles, Sahelian wet and dry phases, and the slow greening trend attributed to elevated CO₂. Shorter records produce noisier anomalies and higher false-alarm rates. This is one reason why Satellize's crop-estimation work in Tonga, where the archive is shorter and the island geography creates its own compositing challenges, required careful baseline construction before anomaly products became reliable.

A note on what this page does not cover: flood extent, burn scar mapping, and cyclone damage assessment are handled in sibling pages in this library. Vegetation stress from waterlogging looks superficially similar to drought stress in NDVI space and requires ancillary data to distinguish.

Honest limits, and where additional data closes the gaps

Cloud cover is the single biggest operational constraint in humid and sub-humid regions. A 16-day MODIS composite will include cloud-contaminated pixels that degrade the NDVI signal even after quality filtering. In the West African Sahel, the constraint is less severe because the growing season coincides with a relatively predictable monsoon that still allows clear-sky compositing. In Ethiopia's highlands or the Great Lakes region, cloud cover can be persistent enough to produce large data gaps at critical crop-growth stages.

Aerosol loading from dust and biomass burning depresses NDVI independently of vegetation health. The Sahara-to-Sahel dust transport season overlaps with early growing-season monitoring windows. Quality flags in the MODIS and VIIRS products identify aerosol-contaminated observations, but flagging removes data rather than correcting it. Analysts should inspect quality-layer statistics before interpreting anomaly maps in dust-prone regions.

Finally, VCI and NDVI anomaly measure stress at the time of observation. They do not predict whether stress will persist or recover. A single anomalous dekad followed by good rains may leave no lasting yield impact. Sustained anomalies across three or more consecutive composites are a far stronger signal. Operational systems weight persistence explicitly, and single-dekad alerts should be treated with proportionate scepticism.

Typical figures

Spatial resolution (operational)	250 m (MODIS NDVI band), 500 m (MODIS EVI, VIIRS VNP13); 10–20 m (Sentinel-2)
Temporal compositing period	16-day (MODIS MOD13/MYD13, VIIRS VNP13); 10-day dekadal products derived operationally by FEWS NET
Revisit (raw overpass)	1–2 days (MODIS Terra + Aqua combined); 1 day (VIIRS); 5 days (Sentinel-2 constellation at equator)
Archive depth	MODIS: 2000–present (~24 years); VIIRS: 2012–present; Sentinel-2: 2015–present; CHIRPS: 1981–present
Key spectral bands	Red (~620–670 nm) and NIR (~841–876 nm) for NDVI; Red-Edge (705 nm, 740 nm) available on Sentinel-2 MSI
Rainfall input resolution	CHIRPS: ~5 km (0.05°); dekadal and monthly products
Minimum detectable stress area	Approximately 1–4 km² for reliable anomaly detection with MODIS; ~0.1 ha in principle with Sentinel-2 (cloud permitting)
Product latency	MODIS/VIIRS 16-day composites: 2–3 days post-period; CHIRPS dekadal: ~5 days post-period; Sentinel-2 L2A: hours post-acquisition
Cloud cover impact	Major constraint in humid tropics; 16-day compositing mitigates but does not eliminate; quality flags identify contaminated pixels
Delivery formats	GeoTIFF raster (VCI, NDVI anomaly z-score); vector polygons (stress zones by IPC-aligned severity class); CSV time-series per admin unit

Analytics Satellize can run

VCI stress map	Pixel-wise VCI computed from MODIS MOD13 or VIIRS VNP13 against full archive min/max; severity classes at <20, 20–35, 35–50	GeoTIFF raster + admin-unit summary table, updated each 16-day composite period
NDVI anomaly (z-score)	Current NDVI minus long-term median divided by standard deviation, calculated per pixel per calendar dekad against the full archive baseline	GeoTIFF raster with z-score values; alert polygon layer where z-score < −1.5 for two or more consecutive composites
Rainfall deficit contextualisation	CHIRPS dekadal anomaly overlaid with VCI stress zones to distinguish meteorological from agricultural drought signal	Combined PDF bulletin with dual-layer map and narrative interpretation per livelihood zone
Phenology-adjusted stress index	Growing-season start detected from NDVI time-series inflection; anomaly calculated relative to same phenological stage in prior years rather than fixed calendar week	GeoTIFF raster; flagged where phenological shift exceeds two weeks relative to median onset
Persistent stress alert	Temporal persistence filter: pixel flagged only if VCI < 35 or NDVI z-score < −1.5 for three or more consecutive 16-day composites	Vector polygon layer of persistent-stress zones, updated fortnightly; suitable for direct input to IPC analysis
Admin-unit time-series dashboard	Area-weighted mean VCI and NDVI anomaly aggregated to district or livelihood-zone boundaries; trend line against prior five seasons	Interactive web dashboard or static CSV/JSON feed; configurable alert thresholds per country context

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.