Grassland degradation and bare-soil exposure mapping
Spectral unmixing of Sentinel-2 and Landsat imagery quantifies bare-soil fraction in grasslands, separating chronic overgrazing from seasonal senescence. Soil colour variability across geologies demands regional calibration before results can be trusted.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 20 m in shortwave infrared (SWIR); 5-day revisit at the equator with both satellites. SWIR bands B11 and B12 are essential for the Bare Soil Index and for distinguishing soil brightness from dry vegetation residue.
- Landsat 8/9 OLI: 30 m multispectral resolution; 16-day single-satellite revisit, 8 days combined. Archive extends to 1972 across the full Landsat family, making it the primary source for inter-decadal degradation trends. OLI's coastal-aerosol band aids atmospheric correction over dusty drylands.
- MODIS MOD13 (Terra/Aqua): 250 m NDVI and 500 m surface-reflectance products at daily to 16-day composites. Too coarse to resolve individual paddock-scale patches, but the 20-plus-year archive is indispensable for detecting slow regional declines in vegetation index baselines.
- Planet SuperDove: 3 m resolution, near-daily revisit from the SuperDove constellation. Eight spectral bands including a red-edge channel. Useful for validating sub-field bare-soil patterns identified at Sentinel-2 scale, though archive depth and cost per scene limit routine time-series use.
What bare soil actually signals
Healthy grassland keeps soil covered. Even during dry-season senescence, standing dead biomass and litter maintain a surface layer that moderates temperature, retains moisture and resists wind erosion. When that layer disappears, the cause is almost always mechanical: hooves, tyres, repeated burning at the wrong interval, or some combination of all three.
The distinction matters enormously for management. A paddock that goes brown in August and green in November is behaving normally. One that shows increasing bare-soil fraction each successive dry season, or that fails to recover after rain, is on a degradation trajectory. Satellite time series are the only practical way to document that trajectory across the spatial scales at which rangeland managers actually operate, which can span hundreds of thousands of hectares.
The physics behind the spectral signal
The Bare Soil Index (BSI) exploits the contrasting reflectance behaviour of soil, green vegetation and dry plant material across the shortwave infrared and visible spectrum. The standard formulation combines SWIR1, red, near-infrared and blue reflectances. Soil reflects strongly in SWIR1 and red; green vegetation absorbs red and reflects NIR strongly; dry vegetation sits in between. Sentinel-2's Band 11 (SWIR1, centred at 1610 nm) and Band 4 (red) carry most of the discriminating power.
Spectral unmixing goes further than a single index. A linear mixture model decomposes each pixel into fractional contributions from soil, photosynthetic vegetation and non-photosynthetic vegetation endmembers. The approach, published extensively in the remote sensing literature, allows a 10 m Sentinel-2 pixel to be expressed as, say, 40 percent bare soil, 35 percent dry grass and 25 percent green cover. That fraction map is far more informative than a binary classified map, and it degrades gracefully at the edges of degradation gradients.
The geology problem: why one calibration does not travel
Bare soil is not a single spectral target. Red Kalahari sand, grey vertisol cracking clay, white calcareous crust and black basalt-derived soil each occupy different positions in reflectance space. An index threshold calibrated on one geology will systematically over- or under-estimate bare-soil fraction on another. This is not a minor correction: the error can exceed 30 percentage points in cross-geology comparisons.
Regional calibration requires field reference data collected across the soil-type range of the study area, ideally at the same phenological stage as the imagery. Soil spectral libraries, including those published by USGS, can substitute partially, but ground-truthing remains necessary before any bare-soil product is used to trigger management or policy decisions. We say this plainly because overselling the transferability of index thresholds is a recurring failure mode in applied rangeland remote sensing.
Separating degradation from drought and seasonality
The hardest analytical problem is attribution. A severe drought produces bare-soil signatures that are almost indistinguishable from overgrazing in a single image. The solution is temporal depth. Landsat's archive back to the early 1980s allows analysts to construct a baseline of dry-season bare-soil fraction under known precipitation regimes, using rainfall anomaly data to normalise each year's observation. Persistent bare-soil exposure that tracks stocking density rather than rainfall anomaly is the clearest diagnostic of degradation.
MODIS MOD13 NDVI time series contribute here despite their coarse resolution. The 20-plus-year record of 16-day composites, freely available through NASA Earthdata, allows trend decomposition using methods such as BFAST (Breaks For Additive Seasonal and Trend) to identify structural breaks in vegetation productivity that precede visible bare-soil expansion. Think of it as an early-warning layer that flags where to look at finer resolution.
Cloud cover is a genuine constraint in humid savannas and highland grasslands. Sentinel-2 optical imagery is unusable under cloud, and revisit effectively drops to whatever clear-sky frequency the climate allows. In persistently cloudy regions, seasonal compositing over three to six months may be the only way to obtain a clean image, which sacrifices temporal precision. Synthetic aperture radar can detect canopy structure changes but does not directly measure bare-soil fraction with the same fidelity as optical SWIR bands.
Resolution floors and what they hide
At 10 m, Sentinel-2 can resolve bare patches roughly 20 m across with reasonable confidence. Smaller patches, the kind that form first around watering points and along fence lines, require Planet SuperDove's 3 m imagery to detect reliably. The practical implication is that early-stage degradation, when intervention is cheapest, is often sub-pixel at Sentinel-2 scale. By the time a 10 m pixel shows majority bare soil, the degradation is already advanced.
MODIS is essentially useless for paddock-scale work. Its value is continental trend monitoring, not site management. Analysts should match sensor choice to the spatial scale of the management question before committing to a processing approach.
From index to decision
A bare-soil fraction map becomes actionable when it is layered against stocking records, paddock boundaries and rainfall history. The combination allows a land manager to see not just where degradation is occurring but which paddocks are degrading faster than their neighbours under similar rainfall, pointing directly at management differences rather than climate. That comparison is the analytical step most remote sensing products omit.
Satellize applies spectral unmixing and BSI time-series analysis on open Sentinel-2 and Landsat archives, with regional soil endmember calibration as a standard part of project scoping. The Tonga crop-estimation programme demonstrated the value of phenological time-series methods in small-island contexts; the same temporal decomposition logic transfers directly to rangeland degradation monitoring. Outputs are delivered as annual or seasonal GIS layers with change-magnitude rasters, suitable for integration into existing rangeland management information systems.
Typical figures
| Spatial resolution (primary) | 10 m (Sentinel-2 visible/NIR), 20 m (Sentinel-2 SWIR), 30 m (Landsat OLI) |
| Spatial resolution (validation) | 3 m (Planet SuperDove) |
| Revisit frequency | 5 days (Sentinel-2 combined); 8 days (Landsat 8+9 combined); near-daily (MODIS composites) |
| Key spectral bands | SWIR1 (~1610 nm), SWIR2 (~2190 nm), Red (~665 nm), NIR (~842 nm), Blue (~490 nm) |
| Minimum detectable bare patch | ~20 m diameter at Sentinel-2 scale; ~6 m at Planet SuperDove scale |
| Archive depth | Landsat: 1972 to present; Sentinel-2: 2015 to present; MODIS: 2000 to present |
| Cloud-cover constraint | Optical sensors fully obscured under cloud; seasonal compositing required in humid zones (3–6 month windows typical) |
| Bare-soil fraction accuracy | ±5–15 percentage points typical after regional calibration; ±30 pp or worse without geology-specific endmembers |
| Delivery formats | GeoTIFF rasters, GeoPackage vector summaries, time-series CSV per management unit |
| Processing latency | Near-real-time composites: 3–7 days after scene acquisition; annual trend products: commissioned delivery |
Analytics Satellize can run
| Seasonal bare-soil fraction map | Spectral unmixing (linear mixture model) using soil, photosynthetic vegetation and non-photosynthetic vegetation endmembers from Sentinel-2 SWIR and visible bands | GeoTIFF raster layer per season, with per-pixel fraction values 0–1 |
| Bare Soil Index time series per paddock or management unit | BSI computed from Sentinel-2 B11, B8, B4 and B2 at each available clear-sky acquisition, aggregated to user-defined polygons | CSV time series and annual summary table; trend direction and magnitude per unit |
| Inter-annual degradation trend map | BFAST or linear regression on annual dry-season bare-soil composites from Landsat archive, rainfall-normalised using published gridded precipitation products | GeoTIFF showing trend slope (fraction per year) and statistical significance; PDF summary report |
| Degradation hotspot alert | Threshold exceedance on rolling bare-soil fraction anomaly relative to site-specific historical baseline | Polygon GIS layer flagging paddocks exceeding user-defined degradation threshold, updated each clear-sky pass |
| Soil endmember calibration report | Field spectroscopy or USGS spectral library matching against dominant soil types in the study area, used to parameterise regional unmixing model | Calibration report with endmember spectra, validation accuracy statistics and recommended index thresholds |
| Long-term vegetation productivity baseline | MODIS MOD13 NDVI 16-day composites decomposed for trend and seasonality over 20-year archive | Regional trend map and per-pixel productivity anomaly raster, suitable for prioritising finer-resolution analysis |
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.