Multispectral vegetation index monitoring
Multispectral satellites measure how plant canopies absorb and reflect sunlight, revealing crop stress, biomass change and land-cover shifts weeks before they are visible to the eye. This page explains the physics, the band configurations of the main sensors, and the real limits imposed by cloud cover.
Sensors
- Sentinel-2 MSI (ESA): 13 spectral bands from 443 nm to 2190 nm. Bands 4 (red, 10 m), 8 (NIR broad, 10 m) and 8A (NIR narrow, 20 m) are the workhorses for NDVI and red-edge indices. Band 5 (705 nm) and Band 6 (740 nm) at 20 m resolution give the red-edge chlorophyll index unavailable on older Landsat. Revisit is 5 days at the equator with two satellites combined, though effective cloud-free revisit in humid tropics can stretch to weeks.
- Landsat 8/9 OLI (USGS/NASA): 30 m resolution across 9 bands including red (Band 4, 655 nm) and NIR (Band 5, 865 nm) for NDVI. No dedicated red-edge band. 16-day revisit per satellite; 8-day with both Landsat 8 and 9 operating together. Archive back to 1972 (Landsat 1 MSS) makes it the only sensor capable of multi-decade vegetation trend analysis.
- Planet SuperDove: 8 bands including a red-edge band at 705 nm, 3 m native resolution, daily revisit over most land areas. Radiometric calibration is continuously refined against Sentinel-2 and Landsat. The high revisit is the main argument for SuperDove; cloud-free compositing over a week typically yields usable imagery where Sentinel-2 cannot. Requires a commercial licence.
- MODIS Terra/Aqua (NASA): 250 m resolution for red and NIR bands used in NDVI; 500 m for the additional bands used in EVI. Daily global coverage makes MODIS the standard for large-area phenology tracking and early drought signals. Spatial resolution is too coarse for field-level crop monitoring on smallholder plots below roughly 5 hectares. The MODIS archive runs from 2000.
Why a plant's colour is the least interesting thing about it
Healthy vegetation absorbs red wavelengths (roughly 620 to 700 nm) strongly for photosynthesis and reflects near-infrared (NIR, roughly 700 to 1300 nm) strongly because the spongy mesophyll cell structure in leaves scatters NIR rather than absorbing it. The contrast between these two behaviours is enormous: a healthy canopy might reflect only 5 to 8 per cent of incident red light but 40 to 60 per cent of NIR. Stressed, sparse or senescing vegetation loses that contrast as chlorophyll degrades and cell structure collapses.
The Normalised Difference Vegetation Index (NDVI) formalises this as (NIR minus Red) divided by (NIR plus Red), producing a dimensionless value between minus one and plus one. Dense healthy canopy typically reads 0.6 to 0.9. Bare soil sits around 0.1 to 0.2. Water is negative. The index is simple, sensor-agnostic in principle, and has a 50-year published record, which is why it remains the baseline despite its known saturation problem at high biomass densities.
Where NDVI saturates and why EVI and the red-edge indices exist
NDVI compresses towards its upper limit in dense canopies, meaning a closed-canopy tropical forest and a very productive maize field can return similar values despite very different biomass. The Enhanced Vegetation Index (EVI), developed for MODIS, adds a blue band correction for atmospheric aerosol scattering and a canopy background adjustment, reducing saturation and soil noise. Its formula is 2.5 times (NIR minus Red) divided by (NIR plus 6 times Red minus 7.5 times Blue plus 1). EVI is the standard product for MODIS-based global phenology work.
The red-edge region, from roughly 700 to 740 nm, is where chlorophyll absorption falls off sharply and NIR reflectance begins to rise. The slope and position of this transition are sensitive to chlorophyll concentration before visible yellowing appears. Sentinel-2 Bands 5 and 6 sit directly in this region at 20 m resolution, enabling the Red-Edge Chlorophyll Index (CIre), calculated as (Band 7 divided by Band 5) minus one. Studies published in Remote Sensing of Environment have shown CIre detects nitrogen and water stress several days earlier than NDVI on cereal crops. Planet SuperDove includes a comparable red-edge band at 705 nm, making similar early-stress detection possible at daily revisit.
Band configurations: what each sensor can and cannot do
Sentinel-2 MSI is the most capable freely available sensor for vegetation index work. Its 10 m red and NIR bands support field-level NDVI mapping, and its 20 m red-edge bands support CIre and the MERIS Terrestrial Chlorophyll Index (MTCI). The shortwave infrared bands (Band 11 at 1610 nm, Band 12 at 2190 nm) add sensitivity to canopy water content and allow separation of green vegetation from dry biomass. The main constraint is the 5-day revisit, which is a statistical revisit under clear skies; in practice, cloud-free observations over a given field in the UK or equatorial Africa may arrive every 2 to 6 weeks during wet seasons.
Landsat OLI at 30 m lacks red-edge bands entirely, limiting it to NDVI and EVI. Its value is the archive. Trend analysis across decades, such as detecting gradual rangeland degradation or the expansion of irrigated agriculture, is only possible with Landsat. MODIS at 250 to 500 m is unsuitable for individual field monitoring but indispensable for national or continental phenology tracking, where its daily revisit outweighs the coarse resolution. Planet SuperDove fills the temporal gap between Sentinel-2 observations but introduces a cost and a calibration consideration: inter-satellite consistency across a constellation of hundreds of small satellites requires careful normalisation before multi-temporal analysis.
Cloud contamination: the number that buyers rarely see in demos
Optical vegetation indices require cloud-free observations. This sounds obvious but the operational consequences are frequently understated. The European Space Agency's own Sentinel-2 cloud statistics show that tropical regions can have cloud cover exceeding 70 per cent of observations year-round. Temperate maritime climates, including the UK and much of Western Europe, commonly experience 50 to 60 per cent cloud cover in winter months. A nominal 5-day revisit can therefore translate to a cloud-free observation once every 3 to 8 weeks during the growing season in these regions.
Mitigation strategies include cloud-shadow masking with the Sen2Cor or s2cloudless algorithms, temporal compositing (selecting the least-cloudy pixel from a rolling window of observations), and fusion with SAR backscatter data to maintain a signal through cloud. SAR-optical fusion can sustain a weekly biomass proxy even when no optical image is usable, though the relationship between SAR backscatter and vegetation indices is crop-type dependent and requires local calibration. For buyers in persistently cloudy regions, the honest advice is to plan data acquisition strategies around compositing windows rather than individual overpass dates.
From index values to operational decisions
Raw index values become useful when they are compared against a baseline: the same field in the same phenological stage in prior years, or a regional mean for the crop type. Anomaly detection, expressed as the z-score or percentile departure from a multi-year mean, is the standard method for identifying stressed areas within a season. The USDA Foreign Agricultural Service and FAO both publish NDVI anomaly products derived from MODIS for food security monitoring, providing a well-documented methodological reference.
At field scale, zonal statistics extracted from Sentinel-2 or SuperDove imagery, aggregated to cadastral parcel boundaries, produce per-field index time series. These feed yield-estimation models, insurance trigger assessments and irrigation scheduling tools. Satellize runs this type of analysis operationally: the Kingdom of Tonga crop-estimation programme uses Sentinel-2 derived vegetation indices aggregated to smallholder plot boundaries to produce seasonal yield forecasts for government planning. The accuracy of any yield model depends on ground-truth calibration data collected in the field; satellite indices alone are proxies, not measurements of yield.
Honest limits before you commission an analysis
Several constraints are worth stating plainly. NDVI and EVI saturate in high-biomass canopies above roughly 0.8 to 0.9 index values, making them poor discriminators of productivity differences in dense forest or high-yield crops at peak greenness. Sub-field variability on plots smaller than two to three pixels (60 to 90 m for Sentinel-2 at 20 m bands, 30 m for Landsat) is not reliably resolved. Atmospheric correction quality varies with aerosol loading and sensor viewing geometry; uncorrected surface reflectance data should not be used for multi-date comparisons.
Phenological stage matters: an index value means something different at emergence than at canopy closure. Any analysis that does not account for crop calendar or land-cover type will produce misleading anomalies. Finally, the red-edge indices require sensors that have those bands. Analysts who promise red-edge outputs from Landsat data are either fusing data from a different sensor or are mistaken.
Typical figures
| Best available spatial resolution (optical) | 3 m (Planet SuperDove); 10 m (Sentinel-2 red/NIR); 20 m (Sentinel-2 red-edge/SWIR); 30 m (Landsat OLI); 250 m (MODIS red/NIR) |
| Revisit (cloud-free, practical range) | Daily (MODIS, Planet SuperDove constellation); 5 days nominal / 2–8 weeks effective (Sentinel-2, cloud-dependent); 8–16 days (Landsat 8+9) |
| Data latency (open sensors) | Sentinel-2 Level-2A: typically 3–6 hours after overpass via Copernicus Data Space. Landsat Collection 2: 24–48 hours via USGS EarthExplorer. MODIS daily products: same-day via NASA FIRMS/Earthdata. |
| Key spectral bands for vegetation indices | Red (~655–670 nm), NIR (~842–865 nm), Red-edge (~705–740 nm, Sentinel-2 and SuperDove only), Blue (~450–490 nm, for EVI), SWIR (~1610 nm and ~2190 nm for canopy water content) |
| Index range and saturation threshold | NDVI: –1 to +1; saturation above ~0.8 in dense canopy. EVI: –1 to +1; less prone to saturation. CIre: typically 0.5–5 for vegetated surfaces. |
| Minimum detectable plot size (practical) | ~0.1 ha at 3 m (SuperDove); ~0.5 ha at 10 m (Sentinel-2); ~2–3 ha at 30 m (Landsat) |
| Archive depth | Landsat: 1972 (MSS) to present. MODIS: 2000 to present. Sentinel-2: 2015 to present. Planet SuperDove: ~2017 to present (commercial). |
| Cloud contamination (indicative) | Tropical regions: >70% of Sentinel-2 observations affected. Temperate maritime: 50–60% in winter. Arid/semi-arid: <20% year-round. |
| Standard delivery formats | GeoTIFF (index rasters), Cloud-Optimised GeoTIFF (COG), NetCDF (time series stacks), CSV/GeoJSON (zonal statistics per parcel) |
Analytics Satellize can run
| NDVI / EVI time series per field parcel | Zonal statistics from atmospherically corrected Sentinel-2 or Landsat surface reflectance, aggregated to cadastral or user-defined polygon boundaries | GeoJSON or CSV time-series feed, updated each cloud-free overpass |
| Red-edge chlorophyll index (CIre) stress map | Sentinel-2 Band 7 / Band 5 ratio, compared against phenology-adjusted baseline from prior seasons | Weekly GeoTIFF raster with per-parcel anomaly scores; alert flag when CIre drops below user-defined threshold |
| Seasonal vegetation anomaly report | Z-score departure from 5-year MODIS or Sentinel-2 mean NDVI for the same day-of-year, following FAO/USDA crop monitoring methodology | PDF report with maps and tabular summary, delivered at user-specified intervals during the growing season |
| Cloud-gap-filled vegetation index composite | Temporal compositing (minimum cloud, maximum NDVI selection) over rolling 16- or 32-day windows; optional SAR-optical fusion for persistent cloud regions | Monthly COG raster stack suitable for ingestion into GIS or crop modelling pipelines |
| Crop yield proxy estimate | Accumulated growing-season NDVI or EVI correlated against historical yield statistics; methodology class follows published USDA NASS and FAO remote sensing yield models | End-of-season yield index per administrative unit or per parcel, with confidence interval based on cloud-free observation count |
| Land-cover and crop-type classification | Random forest or support vector machine classifier trained on multi-date Sentinel-2 spectral and index features, validated against ground-truth samples | Annual classified raster (GeoTIFF) with per-class accuracy statistics in accompanying report |
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.