Grassland degradation and bare-soil encroachment mapping
Multi-year spectral unmixing of Landsat and MODIS time series separates photosynthetic vegetation, dry plant litter and bare soil to track rangeland degradation trajectories and distinguish climate stress from overgrazing.
Sensors
- Landsat 8/9 OLI: 30 m spatial resolution, 16-day repeat per satellite (8-day combined), six reflective bands plus SWIR covering the spectral range needed for linear spectral mixture analysis. Archive from 1984 (TM/ETM+/OLI) enables multi-decadal trajectory analysis.
- MODIS MOD09GA (Terra/Aqua): 500 m daily surface reflectance, seven bands. Coarser than Landsat but daily acquisition reduces cloud-gap probability in tropical semi-arid zones and supports sub-seasonal fractional cover time series at continental scale.
- Sentinel-2 MSI: 10 m (visible/NIR) and 20 m (red-edge, SWIR) resolution, 5-day revisit at mid-latitudes with both satellites. Red-edge bands improve discrimination of sparse green vegetation from dry litter, though the archive begins only in 2015.
- VIIRS (Suomi-NPP / NOAA-20): 375 m and 750 m bands, daily global coverage. Useful for regional-scale bare-soil fraction monitoring and for gap-filling MODIS-era records as Terra and Aqua age.
What the spectrometer actually sees in a degraded pasture
A rangeland pixel is not simply green or brown. It is a mixture: living grass blades, standing dead stems and litter, exposed mineral soil, and sometimes rock. Each component has a distinct reflectance signature. Photosynthetic vegetation absorbs strongly in red wavelengths and reflects in near-infrared. Non-photosynthetic vegetation (dry stems, litter) has high reflectance across visible and SWIR bands with a characteristic cellulose absorption feature near 2.1 µm. Bare soil reflects broadly but with a slope shaped by iron oxides and moisture content. Linear spectral mixture analysis (LSMA) treats each pixel's reflectance as a weighted sum of these pure endmember spectra and solves for the fractional contribution of each component. The fractions must sum to one and remain non-negative, which provides a useful internal consistency check.
The photosynthetic vegetation fraction (PV), non-photosynthetic vegetation fraction (NPV) and bare soil fraction (BS) together constitute the three-endmember model popularised in rangeland science by work using Landsat TM data in Australian and African savannahs. SWIR bands are critical: without them, NPV and BS are easily confused. This is why Landsat OLI bands 6 and 7 (1.61 µm and 2.20 µm) and Sentinel-2 bands 11 and 12 are not optional additions but the load-bearing members of the spectral decomposition.
Turning annual fractions into a degradation trajectory
A single fractional cover image is a snapshot. The diagnostic value comes from stacking annual or seasonal composites across many years and fitting a trend to each pixel's bare soil fraction. A statistically significant upward trend in BS over a decade, coinciding with stable or declining rainfall, is strong evidence of structural degradation rather than a drought response. The Landsat archive, continuous from 1984, is long enough to resolve trends that operate on 10-to-20-year cycles, which is the timescale at which perennial grass communities typically shift to annual-grass or bare-soil dominance under sustained pressure.
Seasonal compositing matters. In semi-arid systems, the green fraction peaks briefly after rainfall. Comparing only dry-season composites across years removes the confounding effect of inter-annual rainfall variability on PV, isolating the structural component of cover change. MODIS MOD09GA, with its daily acquisition, allows construction of consistent phenological windows even where Landsat scenes are cloud-affected for weeks at a time. Combining sensors, using MODIS to identify the appropriate phenological window and Landsat to provide the spatial detail within it, is standard practice in published rangeland monitoring programmes.
Separating what the rain did from what the cattle did
The most contested question in rangeland management is attribution: is this patch bare because it has been dry, or because it has been overstocked? Spatial pattern is the primary discriminant. Grazing-driven degradation concentrates near water points and stock routes, producing radial or linear bare-soil signatures that are geometrically inconsistent with rainfall gradients. Climate-driven decline tracks isohyets and shows coherent regional structure. Overlaying fractional cover trend maps against published livestock census data and gridded rainfall products (CHIRPS at 0.05° resolution, for example) allows a plausibility test even where ground data are sparse.
The method has honest limits here. Livestock density data in many target countries are outdated or spatially coarse. Rainfall products carry their own uncertainty in complex terrain. The satellite can show where bare soil is expanding and at what rate; it cannot by itself prove causation. That attribution requires a second analytical step, combining spatial statistics with ancillary records, and the confidence level should be stated explicitly in any delivered product.
Resolution floors and the problem of sparse cover
LSMA performs well when the landscape is genuinely sub-pixel mixed, which is the normal condition in semi-arid rangelands at 30 m resolution. It struggles when patches of bare soil are smaller than a pixel, as can occur in early-stage degradation where trampling creates small gaps between grass tussocks. At 30 m, a pixel needs roughly 10–15% bare soil before the BS fraction reliably exceeds noise. Sentinel-2 at 10–20 m lowers this threshold but introduces a shorter archive and a more demanding atmospheric correction requirement.
Cloud cover is a persistent constraint in sub-Sahelian and monsoon-influenced rangelands. A single Landsat scene may be cloud-free fewer than four times per year at some latitudes, which limits the temporal density of the time series. MODIS partially compensates through volume, but at 500 m it cannot resolve field-scale management units. There is no sensor combination that eliminates this trade-off entirely; the honest answer is to report data gaps alongside trend estimates and widen confidence intervals accordingly.
From trend map to management decision
The practical output of this analysis is a degradation severity map: pixels classified by the rate of BS fraction increase over the analysis period, typically binned into stable, slow degradation, rapid degradation and recovery classes. Overlaid on a cadastral or communal boundary layer, it becomes a prioritisation tool for rangeland restoration investment or destocking interventions. Time-series charts for individual management units give range managers a record they can present to herders or to financing bodies.
Satellize runs this class of fractional cover analysis on open Landsat and MODIS archives, with Sentinel-2 added where finer spatial detail is required. The Tonga crop-estimation programme demonstrated the organisation's capacity to build multi-year phenological baselines from open constellations in data-sparse environments, a directly transferable methodology. For rangeland clients, the standard deliverable is an annual GIS layer set with accompanying trend statistics, updated each dry season. Governments using this for national reporting to the UNCCD (United Nations Convention to Combat Desertification) can align outputs with the Land Degradation Neutrality indicator framework, which explicitly uses fractional cover as a tier-one variable.
Typical figures
| Spatial resolution (primary) | 30 m (Landsat 8/9 OLI); 10–20 m available with Sentinel-2 |
| Spatial resolution (regional synoptic) | 500 m (MODIS MOD09GA) |
| Revisit (Landsat combined) | 8 days (Landsat 8 + 9 combined at equator) |
| Revisit (MODIS) | Daily (Terra + Aqua) |
| Key spectral bands | Red (~0.65 µm), NIR (~0.86 µm), SWIR1 (~1.61 µm), SWIR2 (~2.20 µm); red-edge optional via Sentinel-2 |
| Archive depth | Landsat: 1984–present (TM/ETM+/OLI); MODIS: 2000–present; Sentinel-2: 2015–present |
| Minimum detectable BS fraction change | ~10–15 percentage points at 30 m; lower (~5–8 pp) at 10 m with Sentinel-2 |
| Trend detection latency | Annual update after dry-season composite; near-real-time monitoring not applicable at this temporal scale |
| Coverage | Global; cloud-affected regions require multi-year compositing to fill gaps |
| Delivery formats | GeoTIFF fractional cover layers, trend rasters, CSV trend statistics per management unit, PDF summary report |
Analytics Satellize can run
| Annual fractional cover map (PV / NPV / BS) | Linear spectral mixture analysis (LSMA) on dry-season Landsat or Sentinel-2 composites | GeoTIFF triplet (three fraction layers) per year, one file per season |
| Multi-year bare-soil trend surface | Pixel-wise linear regression of annual BS fraction over the full archive; Mann-Kendall significance test applied | Trend slope raster (percentage points per year) with p-value mask; GIS layer |
| Degradation severity classification | Thresholded trend slope binned into stable / slow / rapid degradation / recovery classes, calibrated against published semi-arid rangeland benchmarks | Classified GIS polygon layer with area statistics per management unit or administrative boundary |
| Attribution overlay (climate vs. grazing pressure) | Spatial correlation of BS trend with CHIRPS rainfall anomalies and livestock density grids; radial pattern detection around water points | Annotated map with attribution confidence ratings; PDF narrative report |
| UNCCD Land Degradation Neutrality indicator layer | Fractional cover change aligned to LDN tier-one methodology using Landsat-derived PV fraction as proxy for SDG 15.3.1 sub-indicator | Indicator-compliant GeoTIFF and summary table formatted for national UNCCD reporting |
| Management-unit time-series dashboard | Zonal statistics of annual fractional cover composites aggregated to cadastral or communal boundaries | Interactive chart set (CSV + web embed) showing PV, NPV, BS fractions by season for each unit |
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.