Liana infestation and canopy structural disruption mapping
Liana-dominated canopy patches suppress host-tree growth, flatten vertical foliage profiles and shift red-edge reflectance in ways that high-resolution multispectral, hyperspectral and lidar data can resolve. Detection confidence is real but conditional on spatial resolution, phenological timing and forest complexity.
Sensors
- WorldView-3 (Maxar): 0.31 m panchromatic, 1.24 m multispectral across 8 VNIR bands plus 8 SWIR bands. The SWIR bands (1195–2365 nm) add canopy water and lignin sensitivity beyond what most multispectral sensors offer. Revisit roughly 1–4 days depending on latitude and tasking priority. The spatial resolution is the primary reason texture metrics derived from this sensor can resolve individual liana-draped crowns.
- Planet SuperDove (PlanetScope): 3 m multispectral across 8 bands including red-edge (705 nm) and two NIR bands. Near-daily global revisit makes it practical for phenological change detection across large forest areas. At 3 m, individual crown discrimination is marginal in dense canopy; the sensor is more useful for patch-level anomaly mapping and time-series analysis than for sub-crown texture.
- GEDI (NASA, aboard ISS): Full-waveform lidar at 25 m footprint diameter, ~600 m along-track spacing between footprints and ~60 m across-track between adjacent tracks on a given pass. Waveform shape encodes vertical foliage distribution; liana-heavy patches produce characteristically compressed, low-relief waveforms compared with structurally complex host-tree canopy. Coverage is limited to latitudes between approximately 51.6° N and S. Not a wall-to-wall mapping sensor.
- DESIS (DLR/Teledyne, aboard ISS): Hyperspectral imager covering 400–1000 nm in up to 235 bands at roughly 2.55 nm spectral sampling and approximately 30 m spatial resolution. The fine spectral resolution across the red-edge (680–780 nm) and NIR plateau allows retrieval of chlorophyll concentration and canopy water content gradients that broadband sensors miss. The 30 m pixel limits it to patch-level rather than crown-level detection.
What a liana-dominated crown looks like from orbit
Lianas grow upward through host-tree crowns and spread laterally across the canopy surface, replacing the three-dimensional architecture of a healthy tree crown with a relatively flat, horizontally continuous mat. This matters spectrally because liana leaves tend to have higher chlorophyll concentrations at the canopy surface, lower internal leaf water content per unit area than the shaded host-tree foliage they displace, and a fundamentally different leaf-angle distribution. The combined effect is a measurable shift in the red-edge inflection point and a reduction in NIR reflectance relative to structurally intact canopy of the same species composition.
The effect is not subtle in well-chosen imagery, but it is easy to confuse with other stressors. Drought stress, bark-beetle damage and nutrient deficiency all push canopy reflectance in broadly similar directions. Liana infestation is distinguished by its spatial pattern (irregular patches that cross crown boundaries), its persistence across seasons and its structural signature in lidar waveforms. No single spectral index is diagnostic on its own.
Spectral indices that carry signal, and the ones that do not
The red-edge chlorophyll index (CIre), computed as (NIR / Red-edge) minus 1, is among the more sensitive indices to the altered chlorophyll distribution in liana-covered crowns. Studies using airborne hyperspectral data have shown that liana-infested plots score differently on CIre than host-tree canopy at equivalent leaf area index, though the separation depends heavily on the liana species involved and the season of acquisition. The NDVI, by contrast, saturates in closed tropical canopy and offers little discriminating power once green cover exceeds roughly 80 percent.
The red-edge band in Planet SuperDove (centred near 705 nm) and the equivalent band in WorldView-3 give access to CIre at operationally useful spatial scales. DESIS adds the ability to examine the full red-edge slope shape rather than a two-band ratio, which is more informative but also more demanding to process and interpret. SWIR bands on WorldView-3 (particularly the bands centred near 1210 nm and 1650 nm) add sensitivity to canopy equivalent water thickness, which tends to be lower in liana mats than in structurally intact canopy. Honest caveat: cloud cover over tropical forests routinely blocks optical acquisition for weeks at a time, and no current commercial sensor has a SAR equivalent for this type of spectral discrimination.
Texture tells a different story than colour
At WorldView-3 resolution, the surface texture of a liana mat is visually and computationally distinct from a healthy tree crown. Liana mats tend to produce lower grey-level co-occurrence matrix (GLCM) contrast and higher homogeneity scores in the panchromatic band because the flat, interlocking leaf surface lacks the shadow-casting three-dimensional structure of a normal crown. This is a published finding from airborne VHR studies that transfers reasonably well to WorldView-3 imagery, though the exact GLCM parameter thresholds require calibration against field plots for each forest type.
At 3 m Planet SuperDove resolution, crown-level texture is largely lost. The sensor is better used for time-series anomaly detection: a pixel that shifts persistently toward lower NIR and flatter red-edge response across multiple acquisitions is a candidate for follow-up WorldView-3 tasking. This two-stage approach, coarse-resolution screening followed by fine-resolution confirmation, is more cost-effective than blanketing large forest areas with VHR tasking.
GEDI waveforms and the flattened vertical profile
A healthy closed-canopy tropical forest produces a GEDI waveform with a broad, multi-peaked energy return spread across several metres of vertical extent, reflecting foliage distributed from the canopy top down through mid-canopy layers to the ground. A liana-dominated patch compresses this profile. The canopy return is concentrated near the top, the mid-canopy signal is weak, and the ratio of canopy height to waveform extent is lower than for structurally intact forest of comparable height.
GEDI's 25 m footprint means it cannot resolve individual crowns, and the sparse sampling pattern (tracks separated by roughly 600 m on a single overpass) means wall-to-wall structural mapping requires aggregating many passes over time, which introduces phenological and seasonal inconsistency. The sensor is most useful for validating patch-level structural anomalies identified in optical data, not for discovering them independently. GEDI coverage also excludes higher latitudes, which is not a constraint for tropical liana mapping but is worth noting for any temperate application.
Where detection confidence actually breaks down
Below 30 m spatial resolution, liana patches smaller than roughly a quarter of a hectare are effectively invisible as distinct features. They may still contribute to a mixed-pixel spectral signal, but separating their contribution from co-occurring stressors becomes speculative rather than analytical. This is the hard floor for operational mapping.
Phenological complexity compounds the problem in mixed forests. In seasonally dry tropical forests, host trees may shed leaves while lianas retain them, making liana patches conspicuous. In aseasonal humid forests, the contrast is smaller and more variable. Acquisition timing relative to local phenology is not a minor methodological detail; it is often the deciding factor in whether a detection programme produces actionable results or noise. Cloud persistence over humid tropical forests means that achieving a cloud-free acquisition at the right phenological moment may require waiting months or accepting partial coverage. Satellize's analytics work for the Kingdom of Tonga crop-estimation programme involved similar cloud-management trade-offs in a tropical island context, and the same planning discipline applies here.
Finally, liana species vary enormously in their spectral and structural signatures. A programme calibrated against one dominant liana genus in one forest type will not transfer without retraining to a different system. Ground-truth plot data are not optional; they are the difference between a map and a guess.
From anomaly map to forest management decision
The practical output of a liana infestation mapping programme is a probability surface: a raster layer assigning each analysis unit (crown, or patch at coarser resolution) a likelihood of heavy liana infestation, derived from a combination of spectral indices, texture metrics and, where GEDI sampling permits, waveform shape features. Thresholding this surface at a chosen confidence level produces a binary map suitable for prioritising ground intervention or further survey.
Change detection between two or more epochs adds a temporal dimension, identifying areas where infestation is spreading, stable or retreating following management. Annual or biannual Planet SuperDove time series provide the temporal backbone; WorldView-3 tasking provides spatial precision where the probability surface exceeds a trigger threshold. The two sensors are complementary rather than interchangeable, and a programme that uses only one of them will sacrifice either coverage or resolution.
Typical figures
| Finest spatial resolution (optical) | 0.31 m panchromatic / 1.24 m multispectral (WorldView-3) |
| Operational detection floor | Approximately 30 m resolution for patch-level anomalies; sub-crown texture requires ≤1.5 m |
| Revisit (Planet SuperDove) | Near-daily global; cloud-free acquisition over humid tropics may still require weeks of waiting |
| Revisit (WorldView-3) | 1–4 days depending on latitude and tasking priority |
| GEDI footprint and sampling | 25 m footprint diameter; ~600 m along-track spacing; coverage 51.6° N–S latitude |
| DESIS spectral range and resolution | 400–1000 nm; up to 235 bands at ~2.55 nm sampling; ~30 m spatial resolution |
| Key spectral bands for liana detection | Red-edge (~705 nm), NIR (~865 nm), SWIR (~1210 nm, ~1650 nm); full red-edge slope via DESIS |
| Archive depth | WorldView-3 from 2014; Planet daily archive from approximately 2016; GEDI from 2019; DESIS from 2018 |
| Minimum mappable patch size (practical) | ~0.25 ha at WorldView-3 resolution; ~1 ha at Planet SuperDove; larger at DESIS/GEDI |
| Deliverable formats | GeoTIFF probability raster, vector polygon layer (GeoPackage / Shapefile), change-detection report |
Analytics Satellize can run
| Liana infestation probability map | Random forest or gradient-boosted classifier trained on CIre, GLCM texture and SWIR water-content indices from WorldView-3 and/or Planet SuperDove, with ground-truth calibration plots | GeoTIFF probability raster per analysis epoch, with confidence intervals per pixel class |
| Crown-level texture anomaly layer | GLCM contrast and homogeneity computed on WorldView-3 panchromatic band at 0.31 m; anomaly scoring relative to forest-type baseline | Vector polygon layer of flagged crowns with texture score attributes, delivered as GeoPackage |
| Red-edge phenological time series | CIre time series derived from Planet SuperDove 8-band archive; change-point detection to identify pixels with persistent red-edge suppression across seasons | Per-pixel time-series chart and spatial layer of persistent anomaly candidates, exported as GeoTIFF and CSV |
| GEDI waveform structural anomaly flags | Waveform metrics (relative height percentiles, plant area index, waveform extent) extracted from GEDI Level 2A/2B products; comparison against forest-type structural envelope to flag compressed profiles | Point layer of flagged GEDI footprints with waveform metric attributes, delivered as GeoPackage |
| Infestation spread change detection | Bi-temporal or multi-temporal differencing of liana probability rasters across Planet SuperDove epochs; net change area statistics by forest management unit | Change polygon layer and summary report with area statistics (ha gained, lost, stable) |
| VHR tasking trigger alerts | Automated threshold on Planet SuperDove anomaly score to flag areas exceeding infestation probability threshold, triggering WorldView-3 tasking request | Alert feed with geographic coordinates and recommended tasking footprint, delivered as GeoJSON |
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.