Forest canopy height mapping for rural propagation loss estimation

Spaceborne lidar from GEDI retrieves canopy height at 25 m footprints; combined with Sentinel-1 SAR gap-filling, those heights feed ITU-R P.833 vegetation loss models to produce credible propagation budgets for VHF, UHF and low-band LTE in forested rural terrain.

Sensors

GEDI (Global Ecosystem Dynamics Investigation, ISS-mounted lidar): Full-waveform lidar at 1064 nm. Each footprint is nominally 25 m in diameter; eight beam tracks are spaced roughly 600 m apart across a 4.2 km swath. Provides relative height metrics (RH50, RH95, RH100) from which canopy top height is derived. Coverage is limited to latitudes between approximately 51.6° N and 51.6° S, and the ISS precessing orbit means revisit at any given point is irregular, typically accumulating useful density over weeks to months.
Sentinel-1 C-band SAR (ESA): Dual-polarisation (VV+VH) synthetic aperture radar at 5.405 GHz. Ground Range Detected products in Interferometric Wide Swath mode offer 10 m pixel spacing and 250 km swath. VH backscatter saturates at high biomass (roughly above 100–150 t/ha in tropical forest), but cross-polarisation ratio and texture metrics remain useful covariates for canopy density estimation between GEDI tracks. Six-day repeat at mid-latitudes.
Sentinel-2 MSI (ESA): 13-band multispectral imager at 10–20 m resolution. Near-infrared and red-edge bands (bands 5, 6, 7, 8A) support canopy density proxies such as NDVI and LAI. Cloud contamination limits utility in humid tropical forest; temporal compositing over 30–90 days is typically required. Used here as an additional covariate for spatial interpolation of GEDI height, not as a primary height retrieval.
ALOS PALSAR-2 (JAXA, L-band SAR): L-band (1.27 GHz) SAR penetrates deeper into forest canopy than C-band, reducing saturation effects at high biomass. HV backscatter correlates with canopy height in stands up to roughly 40–50 m. Useful in dense tropical forest where Sentinel-1 VH is saturated. Repeat cycle is 14 days; coverage is selective and requires tasking coordination. Archive extends to 2014 for PALSAR-2, with PALSAR data from 2006.

What a lidar waveform tells you that a canopy photograph cannot

A photograph of forest from above gives you a surface. GEDI's full-waveform lidar gives you a vertical profile: the fraction of returned energy at each height above ground, from which the algorithm extracts RH95 (the height below which 95 % of returned energy falls) as a proxy for canopy top height. That distinction matters enormously for propagation. A 30 m closed-canopy stand and a 30 m stand with a broken upper layer and dense understorey at 15 m look similar in an optical image but present very different excess attenuation to a UHF signal.

Published GEDI validation campaigns report RH95 root-mean-square errors in the range of 2–5 m against airborne lidar reference data, depending on terrain slope and canopy closure. On flat terrain with closed canopy, errors sit at the lower end. On steep slopes the waveform broadens due to terrain return, and height retrieval degrades unless a high-quality digital elevation model is used to separate ground from canopy returns. Users should treat GEDI heights in terrain with slopes above roughly 20° with additional caution.

The track-gap problem and how SAR partially fills it

GEDI's eight beam tracks are spaced roughly 600 m apart. In a 10 km radio path through forest, you might intersect four to eight tracks if the geometry is favourable, or none if the path runs parallel to the orbital ground track. Spatial interpolation between tracks is unavoidable, and it introduces uncertainty that compounds along the propagation path.

Sentinel-1 VH backscatter provides a spatially continuous covariate. Regression or machine-learning models trained on collocated GEDI height observations and Sentinel-1 backscatter can produce wall-to-wall canopy height maps at 10–20 m resolution. The honest caveat: C-band backscatter saturates in dense tropical forest, so the model loses sensitivity precisely where canopy is tallest and attenuation is greatest. ALOS PALSAR-2 L-band HV extends the dynamic range, but its irregular coverage means it is a supplement rather than a replacement. Published studies using this three-source fusion (GEDI, Sentinel-1, Sentinel-2) report height map RMSE values of roughly 4–8 m across tropical and boreal test sites, which translates to propagation loss uncertainty of several decibels at UHF frequencies under ITU-R P.833.

From canopy height to decibels: the ITU-R P.833 pathway

ITU-R Recommendation P.833 models excess attenuation through vegetation as a function of frequency, path length through the vegetated medium, and empirical coefficients derived from measurement campaigns. At VHF (30–300 MHz) the dominant mechanism is diffraction over and around the canopy; at UHF (300 MHz–3 GHz) and low-band LTE (700–900 MHz) forward scatter through the canopy becomes significant, and excess attenuation rises steeply with canopy depth and closure.

The practical workflow is as follows. A canopy height raster is derived from the GEDI-SAR fusion described above. Along each candidate propagation path, the depth of forest the Fresnel zone intersects is computed from the height raster and terrain model. That depth feeds the P.833 specific attenuation rate (in dB per metre) at the operating frequency, yielding excess vegetation loss in dB. This figure is added to free-space path loss and terrain diffraction loss to produce a link budget. A 10 m height error in a 500 m forest crossing translates to a Fresnel-zone depth error that, at 900 MHz, can shift the vegetation loss estimate by 3–6 dB. That is the difference between a marginal and a workable link.

Frequency matters more than most planning tools acknowledge

VHF signals (e.g. 150 MHz) diffract over canopy relatively efficiently; the canopy acts more as an elevated obstacle than a lossy medium. UHF and low-band LTE signals interact with the canopy volume itself. At 700 MHz, measured excess attenuation through tropical forest can reach 0.2–0.4 dB/m of path length through the canopy, meaning a 200 m forest block adds 40–80 dB of excess loss. That figure is from published measurement campaigns in tropical environments; temperate broadleaf and boreal conifer stands show lower values, typically 0.05–0.2 dB/m at similar frequencies, partly because canopy closure and leaf area index are lower.

Seasonal variation adds another layer of complexity in deciduous forests. Sentinel-2 NDVI time series can flag deciduous versus evergreen canopy, allowing the propagation model to apply leaf-on and leaf-off attenuation coefficients by season. This is not a refinement for its own sake: a rural LTE site designed on leaf-on attenuation will show several decibels of unexpected gain in winter, potentially causing interference to adjacent cells.

What the data cannot do, and where ground truth is irreplaceable

Canopy height is not canopy attenuation. P.833 coefficients are empirically derived and carry their own uncertainty, particularly in mixed-species stands with complex vertical structure. GEDI height metrics describe the top of the canopy; they say little about stem density, leaf area distribution, or the presence of a dense understorey that may intercept signals at antenna height. A 25 m canopy height in a sparse savannah woodland and a 25 m height in a closed-canopy rainforest are not equivalent propagation environments.

For planning purposes, the satellite-derived canopy height map is best understood as a first-pass screening tool that identifies which path segments carry material vegetation loss risk. Segments flagged as high-risk warrant drive-test or field measurement campaigns before final site design. Satellize's analytics pipeline, which runs on open constellations including GEDI and Sentinel, can deliver the canopy height raster and per-path vegetation loss profiles as GIS layers ready for import into standard propagation planning software. The Tonga crop-estimation programme demonstrated the same GEDI-Sentinel fusion approach in a Pacific island context where ground survey access is limited, which is a common condition in rural telecoms planning.

Archive depth and the case for historical canopy baselines

GEDI has been acquiring data from the ISS since April 2019. Sentinel-1 archive runs from 2014; ALOS PALSAR-2 from 2014, with predecessor PALSAR data from 2006. That depth allows multi-year canopy height trend analysis, which matters for network planning in regions experiencing active deforestation or reforestation. A site designed against 2019 canopy conditions may face materially different propagation in 2027 if the forest has been cleared or has regrown. Annual canopy height updates derived from the GEDI-SAR fusion can flag paths where vegetation loss assumptions in the original link budget have drifted beyond the fade margin.

Typical figures

GEDI footprint diameter	Nominally 25 m; footprint spacing along-track approximately 60 m
GEDI cross-track beam spacing	Approximately 600 m between adjacent beam tracks
GEDI canopy height RMSE (flat terrain, closed canopy)	2–5 m (RH95 vs airborne lidar reference; higher on slopes above 20°)
Sentinel-1 spatial resolution (IW GRD)	10 m pixel spacing; 250 km swath; 6-day repeat at mid-latitudes
Fused canopy height map resolution	10–20 m; RMSE approximately 4–8 m in published tropical and boreal studies
Frequency range for ITU-R P.833 application	VHF (30–300 MHz), UHF (300 MHz–3 GHz), low-band LTE (700–900 MHz)
Typical excess attenuation range (tropical forest, UHF)	0.2–0.4 dB/m of path length through canopy
GEDI latitudinal coverage limit	Approximately 51.6° N to 51.6° S
Sentinel-1 archive depth	From 2014 (Sentinel-1A launch); Sentinel-1B data 2016–2021
GEDI archive depth	April 2019 to present (ISS-dependent; mission extensions subject to NASA review)

Analytics Satellize can run

Wall-to-wall canopy height raster	GEDI RH95 metric extraction fused with Sentinel-1 VH backscatter and Sentinel-2 red-edge indices via regression or gradient-boosted model; ALOS PALSAR-2 HV added where available for high-biomass stands	GeoTIFF raster at 10–20 m resolution, delivered per planning area with per-pixel height uncertainty estimate
Per-path vegetation loss profile	Fresnel-zone intersection with canopy height raster along defined propagation paths; excess attenuation computed via ITU-R P.833 specific attenuation rates at specified operating frequency	CSV or shapefile of path segments with vegetation loss in dB, flagging segments exceeding a client-specified threshold
Canopy type classification (evergreen vs deciduous)	Sentinel-2 NDVI phenology analysis over 12-month time series; seasonal amplitude thresholding to separate evergreen, deciduous and mixed canopy classes	Classified raster layer for seasonal P.833 coefficient assignment in propagation planning software
Annual canopy height change detection	Year-on-year comparison of GEDI-SAR fused height maps; change pixels flagged where height delta exceeds estimated retrieval uncertainty	Annual update GeoTIFF with change magnitude layer; summary report of paths where vegetation loss budget has shifted beyond specified fade margin
High-risk path segment ranking	Ranked list of candidate propagation paths by total vegetation excess loss, combining canopy height, path geometry and operating frequency; integrates terrain diffraction loss from a digital elevation model	Prioritised site-survey list in PDF and GIS format, identifying segments warranting field measurement before final link budget sign-off

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.