Tower-site access route mapping in remote terrain
Satellite-derived slope, land-cover and seasonal-wetness data can screen candidate tower-site access routes before a single vehicle is dispatched, cutting wasted mobilisation in mountainous or forested terrain. Outputs are least-cost path rasters, not turn-by-turn navigation, and ground truth always has the final word.
Sensors
- ALOS World 3D 30 m (AW3D30): Global digital surface model at 30 m horizontal posting, vertical accuracy approximately 5 m RMSE over open terrain. Used to derive slope grids, curvature and ridge-line geometry that determine whether a tracked or wheeled vehicle can physically reach a candidate site.
- Sentinel-1 SAR (C-band, 5.405 GHz): 6-day repeat at mid-latitudes in Interferometric Wide Swath mode, 10 m ground range resolution. Backscatter time-series distinguishes seasonally waterlogged ground from dry trafficable surface; coherence change detection flags newly inundated floodplain segments that would not appear on any static road map.
- Sentinel-2 MSI: 10 m visible and near-infrared bands, 5-day revisit with two satellites. NDVI and NDWI layers identify dense canopy requiring clearance and surface-water bodies that interrupt candidate routes. Cloud cover over tropical terrain can block optical acquisition for weeks at a time, which is why SAR is the primary seasonal-wetness source.
- Planet SkySat: Tasked sub-metre optical imagery (0.5 m native resolution) used for spot-checking specific route segments where AW3D30 slope ambiguity is high, for example narrow switchbacks or bridged stream crossings. Tasking adds cost and latency of one to three days; it is not part of the baseline analysis.
Why a road that exists on a map may not exist in practice
National road databases in mountainous and forested regions are frequently years or decades out of date. A track surveyed during a dry season may be impassable for four months of every year once a stream crossing floods, a slope softens or a landslide deposits material across the carriageway. Sending a survey crew to find this out costs a day's mobilisation at minimum, and in genuinely remote terrain the bill can reach tens of thousands of dollars before anyone has looked at a candidate tower site.
The satellite approach does not replace that ground visit. It sequences it. The goal is to eliminate the obvious failures before anyone packs equipment into a vehicle, and to give field teams a ranked shortlist of routes worth investigating rather than an unfiltered list of everything that appears passable on a topographic sheet.
What slope grids and SAR backscatter actually tell you
The AW3D30 digital surface model, released by JAXA and freely available, provides 30 m slope grids that can be reclassified against published vehicle-capability thresholds. A standard four-wheel-drive utility vehicle struggles above roughly 20 to 25 degrees sustained gradient; tracked plant equipment extends that to perhaps 35 degrees depending on surface condition. These thresholds are not precise, and the 30 m posting means narrow switchbacks or short steep ramps may be averaged away. That is a known limit of the method, not a defect that can be engineered out at this resolution.
Sentinel-1 C-band backscatter behaves differently over wet and dry soil. Saturated ground returns a characteristically lower backscatter in VV polarisation than dry compacted surface, and a multi-temporal stack across a full seasonal cycle can map which route segments are reliably wet during the rainy season. The approach has been documented in published literature for agricultural soil-moisture mapping and transfers directly to trafficability assessment, though it conflates surface roughness with moisture in some conditions, particularly over rocky terrain. Coherence between repeat passes drops sharply over flooded surfaces, providing a secondary wetness indicator that is less sensitive to roughness ambiguity.
Least-cost path analysis: what the output is, and what it is not
The analytic product is a least-cost path raster computed across a friction surface built from slope, land-cover class and seasonal-wetness probability. Each cell in the raster is assigned a traversal cost derived from gradient penalty functions and wetness flags; Dijkstra or similar graph-search algorithms then identify the lowest-cost corridor between a road network entry point and a candidate tower location. The result is a ranked set of candidate corridors with associated cost scores, not a routable path with turn instructions.
This distinction matters. The output tells a planner which general corridor is least likely to produce an impassable surprise, and at what point along that corridor the highest-risk segments sit. It does not account for bridge load ratings, locked gates, land tenure or the specific ground condition on the day of the survey. Those factors require a person on the ground. The satellite analysis is a prior probability, not a guarantee.
Optical imagery as a cross-check, not a primary source
Sentinel-2 at 10 m resolution contributes land-cover classification: forest requiring clearance, open scrub, agricultural land and surface water are distinguishable with reasonable confidence in most climatic zones. The NDWI index (Green minus NIR, normalised) is a reliable surface-water indicator at this resolution. Where a candidate route crosses a river or seasonal stream, Sentinel-2 time-series can show whether that crossing is inundated during the wet season and for roughly how many months.
Cloud cover is the limiting factor in tropical and montane regions. In parts of Southeast Asia, Central Africa and the Andean highlands, Sentinel-2 may return fewer than four cloud-free acquisitions per year over a given point. This is precisely why SAR is the primary seasonal-wetness sensor: C-band penetrates cloud and light vegetation. Planet SkySat tasking can resolve ambiguous segments at sub-metre resolution but introduces cost and scheduling lead time. It is most useful for confirming a specific bridge crossing or a short section of track that the 30 m DEM cannot resolve.
Honest limits of the method
Three limitations deserve explicit statement. First, AW3D30 is a surface model, not a bare-earth terrain model, in forested areas. Canopy height is included in the elevation values, which means slope calculations over dense forest carry additional uncertainty. Where GEDI lidar canopy-height data is available (coverage between approximately 51.6 degrees north and south latitude), it can be used to subtract canopy from the surface model, improving slope accuracy beneath the forest. Outside GEDI coverage, the uncertainty remains.
Second, the seasonal-wetness classification from Sentinel-1 backscatter is probabilistic. A route flagged as high-wetness-risk may be perfectly passable in an unusually dry year, and a route flagged as low-risk may be cut by a localised storm event the day before the survey. The analysis characterises climatological risk, not the specific conditions on a given date.
Third, the method says nothing about road surface quality, potholes, erosion gullies or the condition of any structure on the route. Satellize's access-route analysis, like all remote-sensing screening tools, is a filter applied before ground truth, not a substitute for it. The Kingdom of Tonga crop-estimation programme, Satellize's published analytics engagement, operates under the same principle: satellite data narrows the question, field confirmation answers it.
From raster output to a decision a planner can act on
The deliverable from an access-route screening analysis is typically a GIS layer set: a slope-classified raster, a seasonal-wetness-risk overlay, and a ranked least-cost path corridor per candidate site, packaged with a concise written assessment of the highest-risk segments on each route. Planners can load these directly into QGIS, ArcGIS or equivalent tools.
The practical value is in sequencing field resources. A portfolio of thirty candidate tower sites in a mountainous region might yield eight sites with clearly accessible routes, twelve with moderate-risk wet-season constraints and ten with slope or wetness conditions that make access genuinely doubtful. Prioritising the eight for immediate survey, scheduling the twelve for dry-season visits and commissioning targeted SkySat tasking on the ten before any mobilisation is a straightforward planning decision. The satellite analysis does not make that decision; it gives the planner the information to make it without sending a vehicle into the mountains first.
Typical figures
| Slope grid resolution | 30 m (AW3D30); vertical accuracy ~5 m RMSE over open terrain, higher uncertainty under dense canopy |
| SAR backscatter resolution | 10 m ground range (Sentinel-1 IW mode); 6-day repeat at mid-latitudes |
| Optical land-cover resolution | 10 m (Sentinel-2 MSI); 5-day revisit with two-satellite constellation |
| Spot-check optical resolution | 0.5 m native (Planet SkySat); tasking latency 1 to 3 days, coverage on request |
| Seasonal-wetness archive depth | Sentinel-1 data available from 2014; AW3D30 derived from ALOS PRISM acquisitions 2006 to 2011 |
| Least-cost path output format | GeoTIFF raster (EPSG:4326 or client CRS); optional vector corridor polygons in GeoPackage or Shapefile |
| Slope threshold for standard 4WD | Approximately 20 to 25 degrees sustained gradient; tracked plant equipment to ~35 degrees (terrain and surface dependent) |
| Cloud-cover constraint | SAR unaffected; Sentinel-2 optical may yield fewer than 4 cloud-free acquisitions per year in tropical montane zones |
| Latency from order to delivery | 5 to 10 working days for standard analysis using open-archive data; longer if SkySat tasking is included |
Analytics Satellize can run
| Slope-classified access corridor raster | AW3D30 DEM slope derivation with vehicle-capability threshold reclassification | GeoTIFF layer with per-cell gradient class (passable, marginal, impassable) per candidate site portfolio |
| Seasonal wetness risk overlay | Sentinel-1 VV backscatter time-series analysis and coherence change detection across full annual cycle | GeoTIFF probability layer showing fraction of wet-season acquisitions flagged as saturated or inundated, per route segment |
| Land-cover and surface-water classification | Sentinel-2 MSI supervised classification with NDVI and NDWI indices; multi-date compositing to reduce cloud gaps | GeoPackage land-cover layer with forest, open, agricultural and water classes; seasonal inundation extent polygons |
| Least-cost path corridors | Friction-surface construction from slope, wetness and land-cover inputs; Dijkstra graph-search from road network entry points to candidate sites | Ranked vector corridors per candidate site with associated cost score and highest-risk segment annotations, in GeoPackage format |
| Spot-check high-resolution segment review | Planet SkySat tasked sub-metre imagery over flagged ambiguous segments; manual interpretation of track condition, bridge presence and stream crossings | Annotated image chips with written assessment notes, delivered as PDF report with embedded georeferenced imagery |
| Canopy-corrected slope layer (GEDI-assisted) | GEDI Level 2 canopy-height product subtracted from AW3D30 surface model to approximate bare-earth terrain in forested cells, within GEDI coverage latitude limits | Revised slope GeoTIFF with canopy-correction applied where GEDI data is available; uncertainty flag where correction is not possible |
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.