Ecological corridor planning using satellite land-cover time series
Least-cost path analysis across satellite-derived resistance surfaces can identify viable wildlife corridors, but only if the land-cover record is long enough to separate stable natural cover from temporarily green degraded ground.
Sensors
- Landsat 8/9 OLI: 30-metre multispectral resolution, 16-day revisit per satellite (8-day combined), archive back to 1972 via earlier Landsat missions. The temporal depth is the primary asset here: multi-decadal composites reveal whether a candidate corridor has been stable forest for 30 years or was cleared and partially regrown in the last decade.
- Sentinel-2 MSI: 10-metre resolution in visible and near-infrared bands, 5-day revisit at mid-latitudes with both satellites. Adds spatial detail to land-cover classification and allows detection of narrow riparian strips or hedgerows down to roughly 10 metres wide, though sub-canopy structure remains invisible.
- VIIRS VNP46 (night lights): Daily 500-metre night-light composites. Artificial light at night is a well-documented barrier to nocturnal wildlife movement; VNP46 monthly composites quantify road and settlement illumination intensity as a resistance-surface input, distinguishing lit highways from unlit tracks.
- Copernicus DEM (TanDEM-X derived): Global 30-metre and 10-metre digital elevation model. Slope and terrain ruggedness are standard inputs to species-specific resistance surfaces; steep ridgelines or deep valleys constrain movement for many taxa independently of land cover.
Why temporal depth matters more than spatial resolution
A single land-cover map taken this year can show a patch of scrub between two forest blocks and flag it as a plausible corridor. What it cannot show is whether that scrub has been there for twenty years or appeared after a clearance event three years ago and is still declining. A recently disturbed patch may carry an NDVI signature indistinguishable from recovering secondary forest, yet offer none of the structural complexity that movement-sensitive species require.
The Landsat archive, continuous in useful form from the early 1980s, is the only freely available source with enough temporal depth to answer that question at landscape scale. By building annual or biennial land-cover composites from Landsat surface-reflectance products and tracking per-pixel trajectories, analysts can classify each candidate corridor cell as persistently natural, recently degraded, recovering, or chronically disturbed. Least-cost path models built on stable-cover layers produce substantially different corridor routes than those built on a single-date classification, and the difference is operationally significant for any reintroduction or land-acquisition decision.
Building the resistance surface: what goes in and why
A resistance surface assigns a movement cost to every pixel in a landscape. Low cost follows natural land cover, gentle terrain, and darkness at night. High cost follows roads, settlements, intensive agriculture, and artificial light. The least-cost path algorithm then finds the route between two habitat patches that minimises accumulated cost, analogous to finding the path of least effort across a friction map.
Sentinel-2 MSI at 10 metres contributes the land-cover classification layer, using spectral indices such as NDVI, the normalised difference built-up index, and shortwave-infrared bands to separate forest, shrubland, grassland, cropland, bare ground, and impervious surface. Road density is typically derived from OpenStreetMap or national mapping agency data rather than imagery, because asphalt roads and dirt tracks are often below the reliable detection threshold of 10-metre imagery unless they are wide enough to occupy a full pixel. VIIRS VNP46 night-light intensity adds a barrier layer that is particularly relevant for nocturnal species. The Copernicus DEM contributes slope as a terrain-resistance term. Each layer is resampled to a common grid and weighted according to species-specific or guild-specific movement parameters drawn from published literature.
What 10-metre imagery honestly cannot resolve
The 10-metre floor of Sentinel-2 is a genuine constraint. Hedgerows narrower than one pixel, culverts under roads, badger setts, dry-stone walls used as movement corridors by small mammals, and sub-canopy root networks are all invisible. A corridor that appears blocked on a satellite-derived map may be functionally permeable at ground level through features the sensor cannot see. Conversely, a corridor that appears green and connected may be structurally simplified plantation with low permeability for forest-interior species.
Cloud cover compounds the problem in humid tropical regions. A 16-day Landsat revisit and a 5-day Sentinel-2 revisit sound adequate, but persistent cloud in equatorial zones can reduce usable observations to a handful per year in some pixels. Annual compositing using the best available clear observations mitigates this, at the cost of temporal precision: the composite may blend observations from different seasons. Analysts should report the per-pixel observation count used in any composite and flag areas with fewer than, say, four clear observations per year as lower-confidence classifications.
Corridor stability scoring across decades
The most defensible corridor prioritisation combines a current least-cost path with a stability score derived from the time series. One published approach, applied in various forms across tropical forest monitoring studies, assigns each corridor pixel a persistence score equal to the fraction of years in the archive in which it was classified as natural land cover. A pixel that has been forest in 28 of the past 30 annual composites scores differently from one that has been forest for only 4 years following agricultural abandonment.
Corridors that are both low-cost on the current resistance surface and high-scoring on the persistence index are the strongest candidates for protection or restoration investment. Corridors that are currently low-cost but recently converted from higher-scoring land cover are candidates for urgent intervention before the window closes. This two-dimensional prioritisation is a practical output from the time-series archive and is not achievable from any single-date analysis regardless of resolution.
Night lights as a barrier layer: the physics behind the proxy
VIIRS VNP46 detects radiance in the day-night band, a panchromatic channel sensitive to low-light emission in the 500 to 900 nanometre range. Monthly composites, available from 2012, distinguish persistent artificial illumination from ephemeral sources such as fires. Road lighting and settlement glow are consistent across months; their intensity correlates with traffic volume and urbanisation density, both of which are documented barriers to movement for mammals, amphibians, and invertebrates in the ecological literature.
The 500-metre pixel of VNP46 is coarser than the Sentinel-2 land-cover layer, so night-light values must be spatially disaggregated or used as a zone-level modifier rather than a per-pixel resistance value. An unlit rural track and an unlit forest path will both appear dark in VNP46, so the layer distinguishes lit from unlit infrastructure but cannot differentiate among types of unlit surface. That distinction still falls to the land-cover classification.
From analysis to a decision a land manager can use
The deliverable from a corridor analysis is not a single optimal path. It is a ranked set of candidate corridors with associated uncertainty, stability scores, and the specific land-cover transitions that currently constrain each route. A land manager or conservation finance team needs to know which 2-kilometre gap in an otherwise intact corridor is the binding constraint, what land-cover class occupies that gap, and how long that class has been present. That is an answerable question from the satellite archive.
Satellize runs this class of analysis on open Landsat and Sentinel archives, with optional commercial tasking for higher-frequency monitoring of priority gap zones. The Tonga crop-estimation programme demonstrated the same underlying workflow of multi-temporal compositing and per-pixel trajectory classification applied to agricultural land cover; the corridor context adds the resistance-surface and path-finding steps. For a new corridor study, the practical starting point is agreeing the target species or guild, the spatial extent, and the archive period, then reviewing the cloud-cover statistics for the region before committing to a temporal resolution the data can actually support.
Typical figures
| Spatial resolution (land cover) | 10 m (Sentinel-2 MSI); 30 m (Landsat 8/9 OLI) |
| Revisit frequency | 5 days at mid-latitudes (Sentinel-2 two-satellite pair); 8 days combined (Landsat 8+9) |
| Archive depth | Landsat continuous from 1984 in useful multispectral form; Sentinel-2 from 2015 |
| Night-light layer resolution | 500 m (VIIRS VNP46 monthly composite); available from 2012 |
| Elevation model resolution | 30 m or 10 m (Copernicus DEM, TanDEM-X derived) |
| Minimum detectable linear feature | Approximately 10 m width at Sentinel-2 resolution; sub-canopy features undetectable |
| Key spectral bands | Visible (Blue, Green, Red), NIR, SWIR-1, SWIR-2 for land-cover discrimination; Day-Night Band for night lights |
| Typical cloud-free observation count | 4 to 20+ clear observations per year per pixel depending on region and cloud climatology |
| Deliverable formats | GeoTIFF resistance surfaces, GeoPackage or Shapefile corridor routes, PDF prioritisation report, optional GIS-ready stability-score rasters |
Analytics Satellize can run
| Multi-decadal land-cover time series | Annual best-available-pixel compositing of Landsat surface reflectance; supervised classification using spectral indices (NDVI, NDBI, SWIR ratios) | Per-pixel annual land-cover raster stack with class labels and observation-count layer, GeoTIFF |
| Corridor stability score layer | Per-pixel persistence fraction calculated across annual composites; flags persistent natural cover versus recently converted or recovering pixels | Continuous 0-to-1 stability raster, GeoTIFF, with accompanying histogram report by candidate corridor zone |
| Resistance surface | Weighted combination of land-cover class costs, slope from Copernicus DEM, road density, and VIIRS VNP46 night-light intensity; weights configurable by species guild | Single-band resistance raster at 10 m or 30 m, GeoTIFF, with documented weighting table |
| Least-cost corridor routes | Circuit theory or least-cost path algorithms (e.g. Circuitscape-class methods) applied to resistance surface between defined source and destination habitat patches | Ranked corridor polylines and corridor-width polygons, GeoPackage or Shapefile |
| Pinch-point and gap analysis | Current-flow mapping to identify high-resistance bottlenecks within candidate corridors; cross-referenced against land-cover class and stability score | Annotated map of binding constraints per corridor, with land-cover class and years-since-change for each pinch-point, PDF and GIS layer |
| Night-light barrier assessment | Monthly VIIRS VNP46 composites aggregated to annual median; thresholded to identify persistently illuminated road and settlement zones intersecting corridor routes | Night-light barrier layer overlaid on corridor routes, GeoTIFF and summary table of illuminated corridor-crossing lengths |
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.