Biodiversity proxy mapping from forest structural complexity
Three-dimensional canopy structure predicts vertebrate and invertebrate richness better than canopy cover alone. GEDI waveforms and airborne lidar quantify the vertical heterogeneity, gap density and rugosity that field ecologists use as habitat-quality proxies.
Sensors
- GEDI (NASA/ISS): Full-waveform lidar at 25 m footprint diameter, sampling along eight parallel tracks. Provides relative height percentiles (RH10–RH100), plant area index profiles, foliage height diversity and the Plant Area Volume Density (PAVD) profile. Coverage is limited to latitudes between roughly 51.6°N and 51.6°S, and the ISS orbit means spatial coverage is non-contiguous with gaps of tens of kilometres between passes.
- ICESat-2 ATL08: Photon-counting lidar at 100 m along-track segments, 91-day repeat. The ATL08 product gives canopy height and canopy cover fraction. Vertical structure detail is coarser than GEDI waveforms, but polar coverage extends to 88° latitude, filling GEDI's geographic gaps.
- ALOS-2 PALSAR-2: L-band SAR at 1–10 m resolution depending on mode. L-band penetrates canopy to the woody structure layer, giving HV backscatter sensitive to above-ground biomass and, indirectly, to stand structural complexity. Useful for distinguishing structurally simple plantations from multi-layered natural forest where GEDI sampling is sparse.
- Sentinel-2 MSI: 10–20 m multispectral, 5-day revisit at the equator. Provides canopy spectral heterogeneity (coefficient of variation of NDVI or texture metrics) as a two-dimensional proxy for species richness. The Spectral Variation Hypothesis, documented in published literature, links spectral diversity to plant diversity, though the relationship weakens in structurally complex closed-canopy forest where the spectral signal is largely the top-of-canopy layer.
Why structure predicts life
The link between three-dimensional habitat structure and species richness is one of the better-supported generalisations in community ecology. MacArthur and Wilson's foliage height diversity index, proposed in the 1960s, showed that bird species richness in North American forests correlates more tightly with the vertical distribution of foliage than with plant species identity. Subsequent work extended this to bats, beetles, herpetofauna and a range of invertebrate guilds. The physical logic is straightforward: more structural layers create more distinct microhabitats, more niche partitioning, more opportunities for specialised foragers and nesters.
Lidar operationalises this. A full waveform from GEDI encodes the vertical distribution of plant material from ground to canopy top. From that waveform, analysts derive foliage height diversity (Shannon entropy of the PAVD profile), canopy rugosity (standard deviation of canopy height within a neighbourhood), gap fraction at multiple height strata and the ratio of understory to overstory energy return. These metrics map onto the ecological variables that field biologists measure with far more effort and far less spatial coverage.
Which GEDI metrics carry the most predictive weight
Not all GEDI-derived metrics perform equally. Published validation studies, including work using GEDI L2A and L2B products against field inventory plots, consistently find that foliage height diversity and the ratio of mid-canopy to total plant area index are the strongest predictors of vertebrate richness. Canopy height alone (RH100) is a weaker predictor: a tall canopy can be structurally monotonous. Rugosity, computed as the standard deviation of RH98 across a grid of adjacent footprints, adds explanatory power for bird and bat diversity specifically.
Gap density matters most for light-dependent understorey species. GEDI's ground return energy relative to total waveform energy gives a gap fraction estimate. Where this is high, light penetrates, understorey vegetation diversifies and the invertebrate community associated with that layer becomes accessible to insectivorous birds and bats. The practical resolution constraint is the 25 m footprint and the non-contiguous sampling pattern. A single GEDI pass does not give wall-to-wall coverage; it gives a sample. Spatial prediction across unsampled areas requires a fusion model, typically using Sentinel-2 spectral texture or PALSAR-2 backscatter as spatially continuous covariates.
Where the proxy breaks down
Secondary forests are the most important failure case. A twenty-year-old secondary forest in the Brazilian Atlantic Forest or in Borneo can develop canopy heights of 20–25 m, respectable rugosity and a multi-layered PAVD profile. GEDI will score it highly on structural complexity. But its species richness, particularly for habitat-specialist vertebrates and old-growth-dependent invertebrates, may be a fraction of primary forest with similar structural metrics. The structural proxy conflates structural maturity with ecological maturity, and the two diverge on timescales of decades to centuries.
Plantation monocultures with dense planting and some canopy stratification can also produce misleading structural scores. An even-aged Eucalyptus plantation at year ten may have a canopy layer and a sparse understorey return; GEDI will distinguish this from primary forest, but a young mixed-species plantation or an agroforestry system can generate ambiguous waveforms. The honest position is that structural metrics are proxies, not censuses. They are most reliable when used to rank sites within a single forest type and least reliable when comparing structurally similar but floristically different ecosystems. Validation against field data from the target region is not optional.
Combining sensors to extend coverage
GEDI's sampling gaps are a real operational constraint. The ISS precesses continuously, so repeat coverage of any specific location is irregular and the inter-track spacing at the equator can exceed 600 m. For national-scale biodiversity proxy mapping, the standard approach is to use GEDI footprints as training data for a spatially continuous model driven by Sentinel-2 texture metrics and PALSAR-2 HV backscatter. Sentinel-2's 10 m bands allow computation of canopy spectral heterogeneity (the coefficient of variation of reflectance across a moving window), which correlates with plant species diversity in open and semi-open canopies. PALSAR-2's L-band penetrates canopy to the woody layer, adding sensitivity to stand structural complexity independent of illumination conditions.
ICESat-2 ATL08 fills the latitude gap above 51.6°N, covering boreal and temperate forests that GEDI cannot reach. Its photon-counting approach gives canopy height and cover at 100 m segments, sufficient for landscape-scale stratification. The fusion of ICESat-2 heights with Sentinel-2 spectral texture and, where available, national airborne lidar surveys produces the most spatially complete structural complexity maps. Several European countries now have open airborne lidar archives at 1–4 point returns per square metre, which allow rugosity and gap metrics at 10–25 m resolution without any spaceborne lidar at all.
From structural map to conservation decision
A structural complexity map is most useful when it informs a prioritisation decision: where to place a new protected area, where to target restoration, which forest concessions carry the highest biodiversity risk. For that purpose, the output needs to be a ranked surface with honest uncertainty bounds, not a single predicted richness value. Bootstrapped ensemble models that propagate GEDI waveform noise, atmospheric correction uncertainty in Sentinel-2 and spatial prediction error give confidence intervals that planners can actually use.
Satellize runs this class of fusion analysis on open constellations, applying the same structural complexity workflow that underpins its Tonga crop-estimation programme to forest monitoring contexts. The deliverable is a GIS layer of structural complexity indices with per-pixel uncertainty, accompanied by a site-ranking report that flags areas where secondary-forest ambiguity or plantation interference requires ground-truth verification. The ranking report is the product; the map is the evidence behind it.
Typical figures
| GEDI footprint diameter | 25 m; non-contiguous sampling, inter-track spacing up to ~600 m at equator |
| GEDI latitude coverage | 51.6°N to 51.6°S (ISS orbital inclination limit) |
| ICESat-2 ATL08 segment length | 100 m along-track; 91-day exact repeat; coverage to 88° latitude |
| Sentinel-2 spatial resolution | 10 m (visible/NIR), 20 m (red-edge, SWIR); 5-day revisit at equator |
| PALSAR-2 resolution (stripmap mode) | 3–10 m; L-band 1.27 GHz; ~14-day repeat |
| Key GEDI structural metrics | RH10–RH100, PAVD profile, foliage height diversity, cover fraction, ground return ratio |
| Minimum detectable structural difference | GEDI vertical accuracy ~10 cm RMSE on flat terrain; degrades on steep slopes (>30°) |
| GEDI archive depth | April 2019 to present (with mission gaps); data publicly available via NASA Earthdata |
| Fusion model output resolution | Typically 25–100 m depending on dominant sensor and training density |
| Delivery formats | GeoTIFF (structural index rasters), GeoPackage (site rankings), PDF summary report |
Analytics Satellize can run
| Foliage height diversity surface | Shannon entropy of GEDI PAVD profile per footprint, spatially interpolated with Sentinel-2 texture covariates using random forest regression | GeoTIFF raster at 25–100 m resolution with per-pixel 90% confidence interval |
| Canopy rugosity map | Standard deviation of RH98 across 3×3 GEDI footprint neighbourhoods; gap-filled using PALSAR-2 HV backscatter | GeoTIFF raster; landscape summary statistics by forest management unit |
| Gap density index | GEDI ground-return energy fraction per footprint aggregated to 100 m grid; validated against Sentinel-2 shadow fraction | GeoTIFF raster; tabular summary by protected area or concession polygon |
| Structural complexity site ranking | Composite index from foliage height diversity, rugosity and gap density; bootstrapped uncertainty propagation; secondary-forest flagging via stand-age proxy from Sentinel-2 time series | PDF prioritisation report with ranked polygon table and uncertainty flags |
| Spectral heterogeneity layer | Coefficient of variation of Sentinel-2 band reflectance in 50 m moving window; Spectral Variation Hypothesis applied as two-dimensional plant diversity proxy | GeoTIFF; used as spatial covariate in fusion model or standalone for open-canopy areas |
| Secondary-forest structural ambiguity flag | Comparison of GEDI structural scores against Landsat-derived disturbance history (time since last clearance); sites with high structure but short recovery time flagged for ground verification | Polygon layer with ambiguity classification; appended to site-ranking report |
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.