Crop disease and pest stress detection using hyperspectral sensing
Pathogen infection and pest feeding alter leaf pigments, water content, and cell structure days before visible symptoms appear. Narrow-band hyperspectral sensing can read those chemical shifts; broadband multispectral sensors mostly cannot.
Sensors
- ASI PRISMA: Italian Space Agency hyperspectral imager: 239 contiguous bands from 400 to 2500 nm at approximately 12 nm spectral resolution, 30 m spatial resolution, 30 km swath, revisit roughly 29 days at nadir but programmable for off-nadir tasking. The spectral sampling is fine enough to resolve the red-edge inflection point, carotenoid absorption near 510 nm, and the 970 nm and 1200 nm water absorption features.
- DESIS on ISS: DLR/Teledyne hyperspectral sensor: 235 bands from 400 to 1000 nm at approximately 2.55 nm sampling, 30 m spatial resolution, 30 km swath. Covers the visible and near-infrared range critical for pigment indices. ISS orbital inclination (51.6°) limits coverage to latitudes below about 52°N and revisit is irregular, which constrains time-series applications.
- EMIT on ISS: NASA Earth Surface Mineral Dust Source Investigation: 285 bands from 380 to 2500 nm at approximately 7.4 nm spectral resolution, 60 m spatial resolution. Primarily a mineralogy mission but its shortwave-infrared coverage resolves lignin and cellulose absorption features useful for advanced canopy stress work. Freely available via NASA Earthdata.
- Sentinel-2 MSI red-edge bands: Three red-edge bands at 705 nm, 740 nm, and 783 nm, 20 m spatial resolution, 5-day revisit at mid-latitudes with two satellites. Not hyperspectral, but the red-edge bands support chlorophyll indices (e.g. CIre, NDRE) that detect moderate-to-severe stress. Useful as a dense time-series complement when hyperspectral tasking is infrequent.
What a sick leaf looks like in 200 bands
Healthy leaves absorb strongly in the blue and red, reflect in the green, and show a steep reflectance rise across the red edge (roughly 680 to 750 nm) driven by chlorophyll concentration and mesophyll cell structure. When a fungal pathogen or feeding insect disrupts that structure, three things happen in sequence: chlorophyll degrades, carotenoid pigments become proportionally dominant, and cell-wall integrity falls, reducing the scattering that produces near-infrared reflectance. Water stress from pest feeding also deepens absorption at 970 nm and 1200 nm.
A broadband sensor such as Landsat OLI or even Sentinel-2 MSI (outside its red-edge bands) integrates reflectance across 30 to 100 nm windows, averaging over the fine spectral features that encode these changes. A hyperspectral sensor sampling every 5 to 12 nm can resolve the red-edge inflection point precisely, calculate the ratio of carotenoid to chlorophyll absorption, and detect the water-band deepening. The difference between 'plant is stressed' and 'plant is specifically chlorotic from rust versus water-deficit' often lives in spectral features narrower than 20 nm.
The indices that matter and what they actually measure
The red-edge chlorophyll index (CIre) uses the ratio of near-infrared reflectance near 800 nm to red-edge reflectance near 720 nm, minus one. Published work on wheat yellow rust and grapevine downy mildew has shown CIre declining measurably 7 to 14 days before lesions become visible to the naked eye, though the exact lead time depends on pathogen load and canopy architecture. The carotenoid-to-chlorophyll ratio (sometimes expressed as the SIPI index, using bands near 800, 680, and 445 nm) rises as chlorophyll degrades faster than carotenoids under pathogen pressure. The Photochemical Reflectance Index (PRI), which uses bands at 531 and 570 nm, tracks xanthophyll de-epoxidation and responds to both water stress and light-use efficiency changes associated with disease.
For insect pest damage specifically, feeding that removes mesophyll tissue (thrips, spider mites) reduces near-infrared scattering without immediately altering pigment ratios, producing a distinctive signature in the 750 to 900 nm plateau. Piercing-sucking insects that inject saliva alter cell turgor and water content, deepening the 970 nm absorption band. These are detectable in hyperspectral data but require careful separation from drought stress, which produces similar water-band signals. Without ancillary weather records or ground truth, the spectral ambiguity is real and should not be understated.
Where spaceborne hyperspectral falls short
Thirty metres is the honest resolution floor for current spaceborne hyperspectral systems. At that scale, a single pixel over a mixed cereal field integrates reflectance from healthy plants, stressed plants, soil gaps, and sometimes standing water. Early disease foci are often smaller than a single pixel. Airborne hyperspectral systems routinely operate at 1 to 5 m resolution, which resolves individual plant rows and early-stage infection patches of a few square metres. For field-scale precision agriculture decisions, airborne data is frequently the more useful product.
Cloud cover is the other hard constraint. PRISMA and DESIS acquire optical data and are fully blocked by cloud. In humid tropical and subtropical regions where many high-value disease-prone crops grow (cocoa, banana, rice), cloud-free acquisitions during the growing season can be rare. Revisit rates of 29 days for PRISMA mean that a cloud-free window may not coincide with the optimal detection window for fast-moving pathogens. Sentinel-2's 5-day revisit partially compensates, but only for the coarser pigment indices its red-edge bands support.
Archive depth is also limited. PRISMA launched in March 2019; DESIS began operations in mid-2018. Neither has the 50-year archive of Landsat or the decade-plus record of Sentinel-2. Establishing historical baselines for disease pressure mapping requires either supplementing with multispectral archives or accepting shorter reference periods.
From spectral signal to field alert: the analytic chain
The practical workflow starts with atmospheric correction, which is non-trivial for hyperspectral data. Small errors in correcting path radiance propagate directly into the narrow-band indices that carry the disease signal. Published methods such as ATCOR and FLAASH are standard; the quality of the correction is often the dominant source of uncertainty in the final stress map, not the sensor itself.
After correction, pixel-level index calculation produces a stress-severity raster. Thresholding against a healthy-canopy baseline, ideally derived from earlier-season imagery of the same field, identifies anomalous pixels. Spatial clustering then separates isolated noisy pixels from genuine infection foci. The output is a disease-pressure map showing probable stress location and relative severity, not a confirmed pathogen identification. Ground verification remains essential before any spray or intervention decision.
Satellize runs this analytic chain on PRISMA and Sentinel-2 red-edge data, with outputs calibrated against agronomic ground-truth where clients provide it. The Tonga crop-estimation programme demonstrated that even in small-island contexts, integrating spectral stress signals with field records improves estimation accuracy meaningfully.
Honest limits and how to work within them
No spaceborne hyperspectral sensor today reliably detects disease at sub-hectare scale under operational conditions. The technology is genuinely useful for regional surveillance, prioritising which fields warrant ground inspection, and tracking the spread of established outbreaks across a landscape. It is not a substitute for scouting.
Spectral ambiguity between disease, drought, nutrient deficiency, and senescence is the central interpretive challenge. Nitrogen deficiency and chlorophyll loss from rust produce overlapping red-edge signatures. Separating them requires either multi-temporal analysis (drought stress builds gradually; rust can appear suddenly) or ancillary data such as soil maps, rainfall records, or variety-specific susceptibility information. A stress map delivered without that context can mislead as easily as it informs.
The field is moving. Several commercial hyperspectral constellations are in development or early deployment, promising sub-10 m resolution at higher revisit. When those systems mature, the resolution constraint that currently limits spaceborne detection to field-scale rather than plant-scale analysis will diminish substantially. The spectral physics will remain the same; the spatial precision will improve.
Typical figures
| Spatial resolution (spaceborne hyperspectral) | 30 m (PRISMA, DESIS, EMIT) |
| Spectral resolution | ~2.5–12 nm depending on sensor; DESIS finest in VNIR, PRISMA covers VNIR+SWIR |
| Spectral range | 400–2500 nm (PRISMA, EMIT); 400–1000 nm (DESIS) |
| Revisit (PRISMA) | ~29 days nadir; off-nadir tasking can shorten this, cloud permitting |
| Revisit (Sentinel-2 red-edge supplement) | 5 days at mid-latitudes (two-satellite constellation) |
| Minimum detectable stress patch (spaceborne) | Practically ~1 ha or larger; sub-hectare foci typically below detection threshold |
| Archive depth | PRISMA from March 2019; DESIS from mid-2018; EMIT from 2022 |
| Cloud sensitivity | Full blockage; no signal through cloud or heavy aerosol |
| Latency (tasked acquisition to processed product) | Typically 3–10 days depending on tasking queue and processing pipeline |
| Delivery formats | GeoTIFF stress-index rasters, vector anomaly polygons, CSV field-summary reports |
Analytics Satellize can run
| Red-edge chlorophyll stress map | CIre and NDRE index calculation from PRISMA or Sentinel-2 red-edge bands, change-detected against seasonal baseline | GeoTIFF raster with per-pixel stress severity class; field-boundary summary report |
| Carotenoid-to-chlorophyll anomaly layer | SIPI index computed from PRISMA VNIR bands; anomaly flagged where ratio exceeds healthy-canopy threshold | GIS polygon layer of probable early-stage chlorotic stress foci |
| Canopy water-content change detection | Reflectance difference at 970 nm and 1200 nm water absorption features across two PRISMA acquisitions | Change raster indicating areas of declining water content; alert shapefile for priority scouting |
| Multi-date stress progression report | Time-series of Sentinel-2 NDRE and CIre over the season, with anomaly onset date estimated per field polygon | PDF report with per-field charts and map; CSV of onset dates by field ID |
| Spectral stress classification (disease vs. drought separation attempt) | Multi-index combination (PRI, SIPI, NDWI, NDRE) with decision-tree classification; requires ancillary rainfall input for disambiguation | Classified raster with confidence score; honest uncertainty flag where ambiguity is unresolvable from spectral data alone |
| Priority scouting zone delineation | Spatial clustering of stress anomaly pixels above threshold; ranked by cluster size and severity index | Ordered list of field zones with GPS centroids for ground inspection; exportable to mobile scouting apps |
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.