Individual tree fall risk assessment along aerial cable corridors
Tall trees adjacent to pole routes, not within them, cause most aerial cable outages in forested terrain. Combining GEDI canopy-height lidar, Sentinel-1 SAR seasonality and wind-exposure modelling from the Copernicus DEM ranks corridor segments by fall-risk probability without a single ground survey.
Sensors
- GEDI (Global Ecosystem Dynamics Investigation): Full-waveform lidar aboard the ISS. Footprint diameter approximately 25 m, along-track spacing approximately 60 m, cross-track beam separation approximately 600 m at the equator. Measures relative height metrics (RH95, RH100) to within roughly 1–3 m RMSE over closed canopy. Coverage limited to latitudes between about 51.6°N and 51.6°S. No cloud sensitivity, but operates only at night.
- Sentinel-1 SAR (C-band, 5.4 GHz): 10 m pixel spacing in Interferometric Wide Swath mode, 6-day revisit at mid-latitudes with both satellites active. C-band backscatter responds to canopy moisture and structure; time-series analysis over a full annual cycle distinguishes deciduous loss of leaves (backscatter drop in winter) from evergreen persistence. Cloud-immune.
- Sentinel-2 MSI: 10 m visible and near-infrared bands, 20 m red-edge and SWIR bands, 5-day revisit at mid-latitudes. Supports species-group discrimination and phenological profiling to complement SAR seasonality analysis. Cloud-affected; requires multi-temporal compositing to achieve consistent cover in persistently overcast regions.
- Copernicus DEM GLO-30: Global digital elevation model at approximately 30 m posting (1 arc-second), derived from TanDEM-X. Used to derive slope, aspect and topographic exposure indices for wind-load modelling. Vertical accuracy typically better than 4 m RMSE over vegetated terrain. Freely available globally.
Why the cleared strip is not the problem
Vegetation-encroachment inspection checks whether branches or stems have grown into the cable corridor itself. That matters, but it is not where most unplanned outages originate. A tree standing five or ten metres outside the cleared right-of-way, tall enough to reach the cable when it falls, is invisible to encroachment surveys yet is the dominant failure mode in forested rural networks. Wind events, saturated root zones and crown asymmetry from adjacent harvesting all cause trees well outside the strip to fall across it.
Quantifying that risk requires knowing three things: how tall individual trees are relative to the cable height, whether those trees are deciduous (and therefore more vulnerable to wind loading in leaf than out), and how exposed each corridor segment is to prevailing winds given local terrain. None of those three questions can be answered efficiently by ground patrol on a network spanning hundreds of kilometres of rural track.
What GEDI waveform lidar actually measures, and where it falls short
GEDI fires full-waveform laser pulses from the International Space Station and records the time-resolved return from the entire vertical canopy column within each roughly 25 m footprint. The RH95 metric, the height below which 95 percent of returned energy originates, is a reliable proxy for dominant canopy height and correlates well with field-measured tree height in closed-canopy forest. Published validation studies report RMSE values of roughly 1 to 3 m depending on forest type and terrain slope.
The limitation is spatial sampling. GEDI footprints along a given orbit track are spaced about 60 m apart, and adjacent tracks are separated by hundreds of metres at mid-latitudes, so the dataset is a sample rather than a wall-to-wall map. For a linear corridor analysis this is workable: GEDI observations within a defined buffer (typically 20 to 80 m) of the pole route are extracted and interpolated to produce a continuous height profile. Gaps are real and should be flagged rather than silently filled. GEDI also has no coverage above roughly 51.6 degrees of latitude, which excludes parts of northern Canada, Scandinavia and Russia. For those regions, commercial airborne lidar or photogrammetric point clouds from very-high-resolution stereo imagery are the practical substitute, at considerably higher cost.
Reading deciduous versus evergreen from radar time series
C-band SAR backscatter from forest canopy is sensitive to the dielectric properties and geometry of scattering elements. Broadleaf deciduous trees lose their leaves in autumn, causing a measurable drop in volume-scattering backscatter of roughly 1 to 3 dB in VV polarisation and a shift in the VV/VH ratio. Evergreen conifers maintain backscatter through winter. A Sentinel-1 time series covering at least one full annual cycle, processed to gamma-naught and filtered with a 3x3 spatial window to reduce speckle, produces a seasonality index that separates deciduous-dominant from evergreen-dominant pixels at 10 m resolution.
This matters for risk ranking because a large deciduous tree in full leaf presents a significantly larger wind-loading surface than the same tree in winter. Conversely, saturated soils in autumn and early winter coincide with peak leaf load, creating a compound risk window. Sentinel-2 phenological composites provide a cross-check on the SAR classification, particularly where mixed pixels complicate the radar signal. Neither sensor alone is definitive; agreement between them raises confidence.
Terrain exposure as a wind-risk multiplier
A tall tree on a sheltered valley floor is a different proposition from the same tree on a ridge crest or a convex slope facing the prevailing westerlies. The Copernicus DEM GLO-30 supports two relevant terrain indices. The topographic wind-exposure index, derived from slope, aspect and the upwind fetch distance across open terrain, identifies locations where accelerated flow is probable. The topographic position index identifies ridge crests and convex breaks of slope where wind shear is concentrated.
These indices do not replace meteorological modelling, and they carry the resolution limitations of a 30 m DEM: fine-scale terrain features narrower than roughly 60 to 90 m are smoothed out. What they provide is a consistent, automated, wall-to-wall exposure score that can be multiplied against the tree-height and species-type layers to produce a composite risk score per corridor segment. Segments scoring high on all three inputs, tall trees, deciduous or mixed canopy, exposed terrain, are the priority for targeted ground inspection or pre-emptive clearance.
Honest limits of the composite model
The output is a probability ranking, not a guarantee. Individual tree condition, root health, proximity to recent harvesting and species-specific failure mechanics are not visible from orbit. A diseased tree of modest height can fall further than a healthy tall one. The model is best understood as a triage tool: it identifies the corridor segments where the probability of a damaging fall is materially higher than average, allowing ground survey effort to be concentrated rather than spread uniformly.
Cloud cover limits Sentinel-2 optical contributions in persistently overcast climates; a full annual SAR time series is more reliable in those regions. GEDI's sampling gaps mean that isolated tall trees between footprints can be missed. Pole-route geometry must be digitised or provided as a GIS input; the method does not locate poles from imagery. Accuracy improves substantially when the client can supply even a coarse existing asset register, since buffer widths and cable heights can then be set per span rather than assumed from a network-wide default.
Satellize has applied similar multi-sensor canopy analysis in agricultural contexts, including the Kingdom of Tonga crop-estimation programme, and the same phenological separation logic transfers directly to forested corridor risk work.
From risk layer to actionable maintenance schedule
The practical output is a GIS polygon layer segmenting the pole route into risk bands, typically three to five classes, with supporting attributes: estimated maximum adjacent tree height from GEDI, canopy type classification from Sentinel-1 seasonality, terrain exposure score from the DEM, and a composite risk index. Each high-risk segment carries a recommended buffer distance for ground inspection.
Annual refresh is sufficient for most networks, timed to coincide with post-leaf-drop Sentinel-1 acquisition so that the deciduous classification is at its clearest. Networks in regions with a history of ice-storm or wet-snow loading events may benefit from a secondary update after significant precipitation events, though that requires near-real-time SAR tasking rather than open-archive processing. The layer integrates directly into standard network GIS platforms and can be overlaid against outage history to calibrate the risk thresholds against observed failure rates as data accumulates.
Typical figures
| Canopy height spatial resolution | GEDI footprints ~25 m diameter, ~60 m along-track spacing; interpolated to corridor profile |
| SAR classification pixel spacing | 10 m (Sentinel-1 IW mode, ground range) |
| DEM posting | ~30 m (Copernicus GLO-30, TanDEM-X derived) |
| Sentinel-1 revisit | 6 days at mid-latitudes (two-satellite constellation); 12 days with one satellite |
| GEDI latitude coverage | Approximately 51.6°S to 51.6°N only |
| Minimum detectable canopy height (GEDI) | Reliable above ~5 m; accuracy degrades on steep slopes (>30°) |
| SAR time series required for deciduous classification | Minimum 12 months of acquisitions; 2 years preferred for phenological stability |
| DEM vertical accuracy | Typically <4 m RMSE over vegetated terrain (Copernicus GLO-30 specification) |
| Archive depth | Sentinel-1 from 2014; GEDI from April 2019; GLO-30 static product |
| Delivery format | GeoPackage or Shapefile risk-band polygons with attribute table; GeoTIFF rasters on request |
Analytics Satellize can run
| Adjacent canopy height profile | GEDI RH95 extraction within user-defined buffer (20–80 m) of pole-route centreline, gap-flagged interpolation | GIS line layer with per-segment maximum and 90th-percentile adjacent tree height attributes |
| Canopy type classification (deciduous / evergreen / mixed) | Sentinel-1 VV and VH annual backscatter seasonality index; cross-validated against Sentinel-2 NDVI phenological composites | 10 m raster and corridor segment attribute; confidence score per pixel |
| Terrain wind-exposure index | Topographic wind-exposure and topographic position index derived from Copernicus GLO-30 slope, aspect and upwind fetch | 30 m raster clipped to corridor buffer; segment-mean score in attribute table |
| Composite fall-risk score | Weighted combination of normalised height, species-type and exposure indices; weights adjustable against client outage history | Corridor segmented into 3–5 risk bands; GeoPackage layer ready for network GIS import |
| Priority inspection segment list | Threshold filter on composite score; ranked by estimated consequence (span length, network criticality if supplied) | CSV and PDF report listing segment coordinates, risk drivers and recommended ground-survey buffer width |
| Annual risk refresh | Repeat SAR seasonality classification on new 12-month Sentinel-1 stack; GEDI update where new orbits intersect corridor | Updated GIS layer with change flags highlighting segments where risk class has increased since prior cycle |
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.