Vegetation condition indices for agricultural drought monitoring
VCI and TCI normalise live NDVI and land surface temperature against multi-decade baselines to reveal moisture stress weeks before yield loss becomes visible. MODIS, VIIRS and Sentinel-3 SLSTR supply the archive depth and revisit frequency that make the baselines statistically meaningful.
Sensors
- MODIS Terra/Aqua (NASA): 250 m (bands 1-2) to 1 km (thermal) spatial resolution, daily global revisit from two platforms combined. The 2000-to-present archive underpins climatological baselines for VCI and TCI. Band 31/32 brightness temperatures feed LST products (MOD11, MYD11) at 1 km.
- VIIRS SNPP / NOAA-20: 375 m I-band and 750 m M-band imagery, daily global coverage, continuity successor to MODIS. The VNP13 vegetation product and VNP21 LST product extend the MODIS record forward with improved spatial detail and reduced striping artefacts.
- Sentinel-3 SLSTR (ESA/Copernicus): 1 km thermal infrared at 500 m visible/NIR, dual-view geometry, revisit under two days at mid-latitudes with two satellites. SLSTR adds higher-frequency thermal sampling and a free, open data policy suited to national drought monitoring services.
- Landsat 8/9 TIRS: 100 m thermal (resampled to 30 m in products), 16-day revisit per satellite (8-day combined). Spatial detail resolves individual fields below roughly 5 hectares, but the revisit rate is too coarse to catch fast-evolving stress events without MODIS or VIIRS gap-filling.
What the indices actually measure
The Vegetation Condition Index was formalised by Kogan in the early 1990s using AVHRR data. The formula is simple: VCI = (NDVI_current - NDVI_min) / (NDVI_max - NDVI_min), where min and max are drawn from the full multi-year record for the same calendar week and the same pixel. A value near zero means the vegetation is as poor as it has ever been in the archive; a value near one means it is near its historical best. The normalisation strips out the permanent effect of land cover type, so sparse rangeland and dense maize can be compared on the same stress scale.
TCI applies the same logic to land surface temperature, but inverted: high temperature relative to the historical range signals stress, so TCI = (LST_max - LST_current) / (LST_max - LST_min). Combining VCI and TCI into the Vegetation Health Index (VHI = 0.5·VCI + 0.5·TCI, or with locally calibrated weights) captures both the greenness deficit and the thermal loading simultaneously. The physical rationale is coherent: moisture-stressed crops close stomata to limit transpiration, so latent heat flux falls, canopy temperature rises, and photosynthetically active tissue declines, reducing NIR reflectance and therefore NDVI.
Why archive depth is not optional
A baseline built on five years of data will mistake an unusual but non-catastrophic season for extreme stress. The standard recommendation in the peer-reviewed literature is a minimum of ten years; MODIS provides over two decades, AVHRR extends the record to the early 1980s for coarser-resolution continuity. VIIRS is now accumulating its own archive from 2012 onward. The practical consequence for a new monitoring programme is that buying or commissioning fresh commercial imagery cannot substitute for these open archives: no commercial constellation yet has the dwell time to generate statistically stable percentile baselines at field scale.
Archive consistency matters too. MODIS Collection 6.1 applies retrospective reprocessing so that older and newer data share the same atmospheric correction coefficients. Mixing collections, or mixing MODIS with VIIRS without cross-calibration, introduces spurious trends that can look like drought signals. This is a real operational hazard, not a theoretical one.
Honest limits of the thermal signal
Cloud cover is the most immediate constraint. Thermal sensors cannot see through cloud, and drought monitoring is most urgent during the growing season, which in many regions coincides with convective cloud activity. Compositing over 8- or 16-day windows suppresses cloud contamination but delays the signal. A single cloud-free observation in a 16-day window may not represent the mean canopy temperature for that period.
Spatial resolution creates a second limit. At 1 km, a MODIS or SLSTR pixel in a fragmented smallholder landscape contains a mixture of crops, soil, and possibly fallow land. The retrieved LST is an area-weighted blend, not a pure crop canopy temperature. Landsat TIRS at 100 m resolves individual fields in most commercial farming systems, but its 16-day revisit means it misses the peak temperature events that a daily sensor would catch. There is no current operational sensor that offers both sub-100 m thermal resolution and daily revisit; this is a genuine gap in the constellation.
VCI and TCI are relative indices. They indicate how bad conditions are compared with history, not whether absolute thresholds for crop failure have been crossed. A region with consistently poor vegetation will show moderate VCI values even in a catastrophic year, simply because its historical minimum was already low.
Turning indices into operational drought products
National meteorological and agricultural agencies typically publish weekly or dekadal (10-day) VHI maps during the growing season. The FAO's Global Information and Early Warning System and USAID's FEWS NET both use VCI-family products as one layer in multi-indicator drought assessments. The indices are most reliable when interpreted alongside rainfall anomaly data, soil moisture estimates, and crop calendar information: a VCI of 25 in the vegetative stage carries different implications than the same value at grain filling.
Anomaly persistence matters for impact. A single week of low VCI may reflect a transient dry spell from which a crop recovers. Three consecutive weeks below VCI 35 during a critical growth window is a different situation. Operational systems therefore track the duration and spatial extent of stress as well as its intensity, producing area-under-stress statistics that feed into national food security assessments.
Satellize runs VCI, TCI and VHI derivation pipelines on open MODIS, VIIRS and Sentinel-3 archives, producing dekadal anomaly layers and alert triggers calibrated to client-defined thresholds. The Tonga crop-estimation programme demonstrated how index products derived from open constellations can be operationalised for small island states where ground-truth data are sparse.
Calibration and the ground-truth problem
VCI and TCI are dimensionless and relative. Translating them into yield impact or area-of-loss estimates requires calibration against historical yield statistics or field survey data. Where national agricultural statistics are reliable and spatially disaggregated, regression models linking VHI to yield anomaly can be trained and validated. Where statistics are sparse or politically sensitive, the indices remain useful for ranking relative severity across regions and years, but quantitative impact estimates carry wide uncertainty bands.
Phenological correction is a related requirement. Crops at different growth stages respond differently to the same moisture deficit. A pixel-level VCI calculated without reference to the local crop calendar will conflate stress in early-season establishment with stress at flowering, which have very different yield consequences. Integrating crop calendar data, whether from published FAO sources or from satellite-derived planting date estimates, substantially improves the operational relevance of the output.
Typical figures
| Typical spatial resolution (VCI/TCI) | 250 m to 1 km (MODIS/VIIRS); 1 km (Sentinel-3 SLSTR thermal); 100 m (Landsat TIRS) |
| Revisit frequency | Daily (MODIS Terra+Aqua combined, VIIRS SNPP+NOAA-20 combined); 1-2 days (Sentinel-3 dual satellite); 8-day combined (Landsat 8+9) |
| Compositing period (standard) | 8-day or 10-day (dekadal) to suppress cloud contamination |
| Key spectral bands | Red (~620-670 nm) and NIR (~841-876 nm) for NDVI; thermal infrared (~10.5-12.5 µm) for LST |
| Archive depth for baseline | MODIS: 2000-present; AVHRR continuity: 1981-present (coarser); VIIRS: 2012-present |
| Minimum meaningful baseline period | 10 years recommended; shorter archives increase percentile uncertainty |
| Latency (near-real-time products) | MODIS and VIIRS NRT products available within 3-6 hours of overpass via NASA FIRMS/LANCE |
| Cloud limitation | Thermal and optical bands fully obscured by cloud; compositing required; persistent cloud cover can leave gaps of 2-4 weeks in tropical regions |
| Delivery formats | GeoTIFF anomaly layers, NetCDF time series, dekadal PDF bulletins, GIS-ready vector alert zones |
Analytics Satellize can run
| Weekly VCI / TCI / VHI anomaly maps | Kogan (1990s) percentile normalisation applied to MODIS MOD13 / VNP13 NDVI and MOD11 / VNP21 LST products against full archive baseline | GeoTIFF raster layers clipped to client area of interest, dekadal cadence |
| Stress persistence and area-under-stress statistics | Consecutive-week threshold exceedance counting (e.g. VHI < 35 for 3+ dekads) spatially aggregated by administrative unit or agro-ecological zone | Tabular CSV and choropleth map report, updated dekadally during growing season |
| Early warning alert triggers | Client-defined VHI threshold and spatial extent rules; automated flag when conditions met during crop-calendar critical windows | Email or API alert with supporting map and anomaly magnitude |
| Crop-calendar-adjusted stress assessment | VHI values weighted by growth stage (vegetative, flowering, grain fill) using published FAO or satellite-derived planting date layers | GIS layer and narrative bulletin distinguishing recoverable early stress from critical-window stress |
| Multi-year drought frequency climatology | Pixel-level frequency of VHI below threshold across full MODIS/VIIRS archive, expressed as return period | Static GeoTIFF climatology layer for baseline risk mapping and insurance product design |
| Sentinel-3 SLSTR thermal anomaly monitoring | LST anomaly derivation from Copernicus SLSTR Level-2 products; comparison against MODIS baseline with cross-calibration adjustment | High-frequency (near-daily) thermal anomaly feed for rapid-onset heat stress events |
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.