Grid-connection route planning using satellite-derived land-cover and terrain

Satellite land-cover and elevation data can cut months from grid-connection corridor studies, but classification errors in fragmented farmland and the irreducible need for wayleave negotiation mean satellite outputs are a starting point, not a final answer.

Sensors

Sentinel-2 MSI (ESA WorldCover): 10 m spatial resolution across 13 spectral bands; 5-day revisit at the equator with both satellites. The ESA WorldCover 2021 product delivers an 11-class global land-cover map at 10 m derived from Sentinel-2 and Sentinel-1 composites, with an overall accuracy of around 75 % globally, dropping noticeably in fragmented agricultural landscapes.
Copernicus DEM GLO-30: 30 m posting global digital elevation model derived from TanDEM-X radar interferometry. Vertical accuracy is typically better than 4 m RMSE over flat to moderately hilly terrain; accuracy degrades in dense forest canopy where the radar signal returns from the canopy top rather than bare ground.
SRTM 30 m: NASA Shuttle Radar Topography Mission data at approximately 30 m posting (1 arc-second), covering latitudes 56°S to 60°N. Vertical accuracy is quoted at less than 16 m absolute error at 90 % confidence; Copernicus DEM GLO-30 is generally preferred where available because of improved void-filling and post-processing.
Landsat-derived JRC Global Surface Water: Monthly and annual water occurrence and seasonality layers at 30 m resolution, derived from the full Landsat archive back to 1984. Identifies permanent water bodies, seasonal flooding and historical inundation extents, all of which raise crossing costs and environmental permitting complexity.

What a corridor study actually needs from orbit

Routing a high-voltage transmission line from a renewable energy site to the nearest suitable grid injection point is fundamentally a cost-minimisation problem. The costs that vary across a landscape fall into three broad categories: construction difficulty (slope, ground cover, soil type), environmental and planning sensitivity (protected areas, wetlands, woodland), and land-access complexity (settlement density, parcel fragmentation). Satellite data speaks directly to the first two. It says almost nothing useful about the third.

A least-cost-path algorithm, run on a raster cost surface, needs input layers that assign a relative penalty to each pixel. Slope derived from a 30 m DEM, land-cover class from a 10 m product such as ESA WorldCover, protected-area polygons from WDPA or national equivalents, and water-body masks from the JRC Global Surface Water layer together cover the main deterministic inputs. The algorithm then finds the path of minimum cumulative cost between origin and destination, often using Dijkstra's method or its variants on a graph representation of the raster.

The 10 m land-cover layer is useful and imperfect

ESA WorldCover 2021 is the most practical freely available global product for this work. At 10 m it resolves individual field boundaries in most agricultural regions and distinguishes woodland from scrub with reasonable reliability. The headline overall accuracy figure of roughly 75 % is, however, a global average. Per-class accuracy varies considerably: built-up areas and open water tend to score above 85 %, while heterogeneous cropland and shrubland classes in fragmented landscapes can fall below 65 %. A corridor running through intensively farmed lowlands, where small fields, hedgerows, orchards and peri-urban plots sit adjacent, will accumulate classification errors that misrepresent the true cost surface.

The practical consequence is that satellite-derived routing should be treated as generating candidate corridors for field verification, not as a substitute for it. A 500 m buffer around the preferred centreline is a reasonable working assumption for the zone of uncertainty that ground-truthing must resolve. Where national land-cover products exist at higher accuracy (many European countries maintain cadastral and habitat datasets that exceed WorldCover's per-class performance), those should be fused with the satellite layer rather than replaced by it.

Elevation data: where the DEM misleads the router

Slope and aspect derived from Copernicus DEM GLO-30 are reliable inputs for penalising steep terrain, which raises both construction cost and the structural loading requirements for pylons. The 30 m posting is adequate for identifying ridge crossings, valley floors and major escarpments at corridor-planning scale. It is not adequate for detailed tower-spotting, which requires survey-grade LiDAR or total-station work at individual spans.

Forest cover introduces a systematic bias. TanDEM-X, the source for Copernicus DEM GLO-30, is an X-band radar that partially penetrates vegetation but largely scatters from the upper canopy. In forested areas the DEM surface is elevated relative to bare ground by an amount that varies with canopy height and density, typically 5 to 20 m in closed broadleaf forest. This matters because a routing algorithm using the DEM as a bare-earth surface will underestimate the true ground slope in forested terrain and may underestimate the clearance engineering required. Combining the DEM with a canopy-height model (GEDI or ICESat-2 derived products are publicly available) partially corrects this, though coverage gaps remain.

Protected areas and water crossings: where satellite data is most reliable

The clearest value satellite data adds to corridor planning is in flagging hard constraints early. Ramsar wetlands, national parks, Natura 2000 sites and similar designations are well-documented in polygon form and can be overlaid directly on the cost surface as near-infinite-cost zones, forcing the algorithm to route around them rather than through them. The JRC Global Surface Water product adds a time-dimension that static maps miss: a river crossing that appears straightforward in a dry-season map may sit within a historically flooded floodplain for three months of the year, substantially raising civil works cost and environmental permitting risk.

Seasonal inundation data from JRC, which draws on the Landsat archive back to 1984, is particularly useful for identifying crossings that look viable on a single-date image but carry a history of flooding. That 38-year record is difficult to replicate from any other freely available source.

What satellite data cannot do, and what replaces it

Subsurface geology is invisible to optical and radar sensors at the wavelengths used by Sentinel-2 or TanDEM-X. Rock type, soil bearing capacity, groundwater depth and karst risk all affect foundation design and construction cost per tower, but none of them appear in any satellite-derived product with the specificity a civil engineer needs. Geotechnical survey remains irreplaceable for final route selection.

Wayleave and easement negotiation, the process of securing legal access rights from landowners along the corridor, is entirely absent from satellite analysis. Land parcel fragmentation visible in a 10 m land-cover product gives a rough proxy for negotiation complexity (many small parcels mean many separate agreements), but the actual negotiation timelines and costs depend on landowner disposition, local land law and political context. In practice, wayleave is frequently the dominant source of delay and cost overrun in grid-connection projects. Satellite routing can help identify corridors with fewer parcels or lower sensitivity, but it cannot shorten the legal process once a corridor is chosen.

Satellize's analytics work on open constellations including Sentinel-2 and Landsat, and can fuse those with commercial tasking where finer resolution or more recent imagery is needed for a specific study area. The Tonga crop-estimation programme demonstrated that open-satellite classification pipelines can be adapted to fragmented, small-plot landscapes, which is directly relevant to the accuracy problem in agricultural corridor routing.

Delivering a corridor study: from raster to ranked options

A practical satellite-derived corridor study produces not a single recommended route but a ranked set of candidate corridors, typically three to five, each with an associated cost-surface score and a breakdown of the penalty contributions (slope, land-cover class, protected-area proximity, water crossings). That breakdown tells the project team which constraints are driving the cost and which are amenable to engineering mitigation.

Outputs are most useful as GIS layers that can be loaded directly into the planning team's own environment, overlaid with local data the satellite products do not capture. A PDF summary of ranked corridors with per-corridor statistics is useful for stakeholder briefings, but the underlying raster cost surface and vector corridor centrelines are the deliverables that actually save time in the next phase of work. Latency from data acquisition to delivered corridors depends on study-area size; for a 50 km by 50 km area with existing WorldCover and Copernicus DEM coverage, processing time is measured in days, not weeks.

Typical figures

Land-cover spatial resolution	10 m (ESA WorldCover / Sentinel-2 MSI)
Elevation model posting	30 m (Copernicus DEM GLO-30; SRTM 30 m as fallback)
Copernicus DEM vertical accuracy	Typically < 4 m RMSE over flat to moderate terrain; degrades under dense forest canopy
WorldCover overall classification accuracy	~75 % globally; per-class accuracy in fragmented agricultural areas can fall below 65 %
Sentinel-2 revisit	5 days at equator (both satellites); cloud cover reduces usable acquisition frequency
JRC Global Surface Water archive depth	1984 to present, derived from Landsat 5, 7, 8 and 9
Protected-area overlay source	WDPA polygons (UNEP-WCMC); updated monthly
Typical study-area processing latency	Days for a 50 × 50 km area using pre-processed WorldCover and Copernicus DEM tiles
Delivery formats	GeoTIFF cost-surface rasters, GeoJSON or Shapefile corridor centrelines, PDF summary report
Minimum corridor buffer for field verification	500 m either side of centreline recommended given classification uncertainty

Analytics Satellize can run

Raster cost surface	Weighted overlay of slope (DEM-derived), land-cover penalty, protected-area exclusion zones, and water-body crossing flags; weights calibrated to client cost assumptions	GeoTIFF raster at 10–30 m resolution covering the study corridor
Ranked candidate corridors	Least-cost-path analysis (Dijkstra or similar graph algorithm) run on the cost surface between defined origin and destination nodes; three to five corridors generated with distinct spatial separation	Vector centrelines (GeoJSON / Shapefile) with per-corridor cumulative cost scores and penalty-class breakdowns
Land-cover classification accuracy assessment	Stratified random sampling of WorldCover classes within the study area, cross-checked against available high-resolution imagery; confusion matrix and per-class user/producer accuracy reported	Accuracy report flagging classes and sub-areas where field verification is highest priority
Historical flood-risk overlay	JRC Global Surface Water occurrence and seasonality layers clipped to corridor buffers; crossings ranked by maximum historical inundation frequency	GIS layer and tabular summary of water-crossing risk by corridor
Protected-area and sensitivity map	Intersection of WDPA polygons, Ramsar boundaries, and national-level designations with corridor buffers; sensitivity tiered by designation type	Vector overlay and PDF table of sensitive intersections per corridor option
Canopy-height correction for DEM	Fusion of Copernicus DEM GLO-30 with GEDI or ICESat-2 derived canopy-height estimates to produce a corrected bare-earth surface in forested segments	Corrected DEM GeoTIFF for forested corridor segments, with uncertainty band where GEDI coverage is sparse

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.