Cotton boll opening and harvest-readiness detection

Sentinel-2 SWIR bands and Landsat OLI track the sharp spectral shift as cotton bolls open and leaves defoliate, giving gin operators and agronomists field-scale readiness signals across large cotton belts. Cloud cover during the narrow harvest window is the method's principal constraint.

Sensors

Sentinel-2 MSI: 10 m visible bands, 20 m SWIR bands (B11 at 1610 nm, B12 at 2190 nm); 5-day revisit at the equator with both satellites. The 20 m SWIR resolution resolves individual commercial cotton fields and captures the reflectance surge that accompanies open boll exposure.
Landsat 8/9 OLI: 30 m multispectral including SWIR1 (1570 nm) and SWIR2 (2110 nm); 16-day single-satellite revisit, 8-day combined. Lower spatial resolution than Sentinel-2 but a consistent archive back to 1984 (Landsat 5 TM) that allows multi-decade phenology baselines to be established for major cotton belts.
Planet SuperDove: 3 m resolution, daily revisit in most latitudes. Eight bands including red-edge but no dedicated SWIR, which limits direct boll-opening detection; most useful for field boundary delineation and catching within-field spatial variability in defoliation progress.
MODIS Terra/Aqua: 250 m (red, NIR) to 500 m (SWIR) resolution; daily revisit. Too coarse for individual field monitoring but useful for regional-scale cotton belt phenology tracking and as a cloud-gap-filling reference when Sentinel-2 and Landsat are both obscured.

What a boll opening looks like from 700 km up

Cotton's harvest-readiness transition is unusually legible from orbit. A closed-boll canopy is green and absorptive in the shortwave infrared. When bolls crack open and expose the white fibre, SWIR reflectance rises sharply. Simultaneously, NDVI collapses as leaves either senesce naturally or are killed by chemical defoliant. The combination of rising SWIR and falling NDVI produces a spectral signature that is distinct from drought stress, disease or simple senescence, all of which affect NDVI without the same SWIR surge.

Sentinel-2 Band 11 (1610 nm) and Band 12 (2190 nm) are the primary detection channels. Published work using these bands shows that the Cotton Boll Opening Index, sometimes formulated as a ratio of SWIR2 to NIR reflectance, can separate open-boll fields from green or partially open ones with reasonable accuracy when cloud-free imagery is available within a few days of the transition. The 20 m resolution of the Sentinel-2 SWIR bands is sufficient to characterise fields of the size typical in Uzbekistan's Fergana Valley, the Texas High Plains or Gujarat.

The defoliation signal and why timing matters to the supply chain

Defoliant application is not cosmetic. Cotton harvested with green leaf material still attached incurs significant quality penalties: leaf trash stains fibre and raises moisture content, both of which reduce gin outturn and depress price. Applied too early, defoliants abort immature bolls. Applied too late, the harvest window closes. In many growing regions that window is two to three weeks before autumn rainfall or frost risk rises sharply.

Satellite monitoring changes the economics of that decision. Rather than relying on field scouts covering a fraction of a large estate or cooperative, an operator can receive a ranked readiness map across every field in a district on each cloud-free Sentinel-2 pass. Fields crossing a defined SWIR threshold get flagged; the gin can begin scheduling picker or stripper sequences days in advance rather than reacting to scout reports. The value is not in the sensor physics alone but in the systematic coverage: a scout misses the field at the far end of the block; the satellite does not.

Building a reliable time series through cloud and revisit constraints

Cloud cover is the method's honest limiting factor. Cotton harvest in the US Mid-South and parts of India coincides with post-monsoon cloud persistence. A 5-day Sentinel-2 revisit cycle can easily produce a 10 to 15-day gap in usable imagery during that period. The standard mitigation is sensor fusion: combining Sentinel-2 SWIR observations with Landsat 8/9 SWIR on its 8-day combined revisit cycle roughly doubles the chance of a cloud-free acquisition in any given fortnight. MODIS daily observations at 500 m can fill regional gaps for trend confirmation, though not for field-level decisions.

A further complication is that boll opening within a single large field is rarely synchronous. Spatial variability in soil texture, irrigation uniformity and plant population means the SWIR signal rises unevenly across a field. A single field-average value can mask the fact that 30 percent of the area is already at peak readiness while the rest is two weeks behind. Analysing the within-field distribution of pixel values, rather than the mean, gives a more actionable picture for variable-rate defoliant application.

Central Asia, the US and India: three contexts, one sensor, different ground truth

The Aral Sea basin cotton belt, principally Uzbekistan and Turkmenistan, operates under state procurement systems where harvest timing is partly administratively set. Satellite phenology monitoring here has a different primary use: independent verification that reported harvest progress matches actual field conditions, which matters for trade finance, international cotton buyers and food security agencies monitoring land-use competition between cotton and food crops.

In the US, the dominant constraint is gin capacity scheduling. A gin running at full capacity cannot absorb a simultaneous rush from an entire region; satellite readiness mapping distributed across a cooperative allows the gin manager to sequence field pickings and avoid both idle capacity and queue backlogs. In India's Vidarbha and Telangana regions, smallholder field fragmentation (median field sizes often below two hectares) pushes the resolution requirement toward Planet-class imagery for field delineation, with Sentinel-2 SWIR used for the spectral signal itself. The combination is workable but requires careful co-registration.

Honest limits and what ground data cannot be replaced

Satellite detection of boll opening is a proxy, not a direct measurement. The SWIR signal integrates reflectance from open bolls, leaf litter, bare soil and any residual green canopy in the pixel. At 20 m resolution, a pixel over a partially defoliated field contains all of those components mixed. Spectral unmixing can separate them, but the accuracy depends on having good endmember spectra for the specific variety and soil type, which requires local calibration data.

Variety matters more than is often acknowledged. Upland cotton (Gossypium hirsutum) and Pima (G. barbadense) have different boll-to-leaf area ratios and different rates of natural defoliation. A threshold calibrated on one variety in one region may perform poorly on another without retraining. Satellize's crop-estimation work, including the Tonga programme, has reinforced a consistent lesson: satellite analytics built without local agronomic ground truth tend to be precise about the wrong thing. A readiness map is only as good as the boll-count and fibre-maturity samples that anchor it.

From spectral signal to gin schedule

The operational output is not a reflectance map. It is a ranked field list, updated on each cloud-free acquisition, showing which fields have crossed a defined readiness threshold, which are within five days of crossing it based on the rate of SWIR increase over recent passes, and which remain more than ten days out. That list integrates with gin scheduling software or, in simpler deployments, arrives as a CSV or GIS layer that a field manager can open in any standard tool.

Alert latency from satellite overpass to delivered product is typically 12 to 24 hours for Sentinel-2 data processed through the Copernicus Data Space, which is sufficient for defoliant-timing decisions that are made on a daily to weekly basis rather than hourly. The archive depth of Sentinel-2 (from 2015) and Landsat (from 1984 for earlier missions) means that first-year deployments can immediately access multi-season phenology baselines rather than waiting years to accumulate them.

Typical figures

Spatial resolution (SWIR detection)	20 m (Sentinel-2 B11/B12); 30 m (Landsat 8/9 OLI SWIR)
Spatial resolution (field delineation)	3 m (Planet SuperDove); 10 m (Sentinel-2 visible/NIR)
Revisit frequency	5 days (Sentinel-2 dual satellite); 8 days combined (Landsat 8+9); daily (Planet, MODIS)
Key spectral bands	SWIR1 ~1610 nm, SWIR2 ~2190 nm; NIR ~842 nm (Sentinel-2); equivalent OLI bands on Landsat
Minimum detectable field size	Approximately 1 ha at 20 m SWIR resolution; sub-hectare with Planet visible bands for delineation
Alert latency	12–24 hours from satellite overpass to delivered product via Copernicus Data Space
Archive depth	Sentinel-2 from 2015; Landsat OLI from 2013; Landsat TM/ETM+ extends baseline to 1984
Cloud cover constraint	Single-sensor cloud-free gap can reach 10–15 days in post-monsoon regions; sensor fusion reduces this
Delivery formats	GeoTIFF readiness maps, ranked field CSV, GIS vector layer (GeoJSON/Shapefile), dashboard feed
Coverage	Global; Sentinel-2 systematic coverage of all cotton-growing latitudes

Analytics Satellize can run

Field-level boll-opening readiness score	SWIR reflectance ratio (B11/B12 against NIR) thresholded against locally calibrated open-boll endmembers; time-series change detection across Sentinel-2 and Landsat acquisitions	Ranked field list (GeoJSON or CSV) updated on each cloud-free pass, showing readiness stage and days-to-threshold estimate
Defoliation progress map	NDVI decline rate computed from dense Sentinel-2 time series; pixel-level comparison against pre-defoliation baseline NDVI	GeoTIFF per acquisition showing percentage canopy defoliation by field polygon
Within-field variability layer	Pixel-value distribution analysis within field boundaries; coefficient of variation of SWIR signal used to flag spatially heterogeneous fields requiring variable-rate defoliant prescriptions	Per-field heterogeneity score appended to readiness CSV; optional zone map for precision application
Seasonal phenology baseline	Multi-year SWIR and NDVI time-series stack from Landsat archive (2013 onwards); harmonic regression to characterise typical boll-opening date distribution by district	PDF or dashboard report showing current-season timing relative to 5- and 10-year district averages
Cloud-gap-filled readiness composite	Sensor fusion of Sentinel-2 and Landsat SWIR observations; linear temporal interpolation between cloud-free acquisitions to maintain near-daily readiness estimates during cloudy periods	Daily GeoTIFF composite with data-quality flag layer indicating interpolated versus observed pixels
Gin scheduling forecast	Logistic growth model fitted to SWIR time series per field; projected date of threshold crossing with confidence interval based on recent acquisition frequency and cloud probability climatology	Tabular 14-day harvest schedule by field, exportable to gin management software or shared as a structured feed

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.