Sensors
- MODIS MOD13/MYD13 (Terra and Aqua): 250 m resolution NDVI and EVI composites at 16-day intervals, with a global archive from 2000. The temporal depth makes MODIS the standard source for multi-year phenology metrics: growing-season length, green-up date, peak NDVI amplitude and inter-annual variability. These phenological signals are the primary covariates for large-range species whose distributions track vegetation productivity cycles.
- GEDI (Global Ecosystem Dynamics Investigation, ISS-mounted LiDAR): Full-waveform LiDAR operating at 1064 nm with a 25 m footprint diameter and approximately 60 m along-track spacing between shots. Provides canopy height, plant area index and vertical foliage profiles between roughly 51.6° N/S (ISS orbital inclination). Canopy height from GEDI is the only globally consistent, direct measurement of forest vertical structure available at this scale. Cover is non-contiguous: each pass samples transects, not a wall-to-wall grid.
- Sentinel-1 SAR (C-band, 5.405 GHz): Interferometric Wide Swath mode delivers 10 m ground range detected imagery with 6-day repeat at mid-latitudes (12-day globally). Repeat-pass coherence over forested areas decays rapidly with canopy disturbance and moisture change, making coherence maps a proxy for forest structural integrity. Backscatter cross-polarisation (VH) correlates with canopy volume. Cloud-penetrating, so it fills gaps in optical time series during monsoon seasons.
- Copernicus DEM (derived from TanDEM-X): 30 m posting globally (GLO-30), with a 10 m version available for most land areas. Used to derive terrain ruggedness index, slope, aspect and topographic wetness index. Ruggedness is a strong predictor for rupiculous (rock-dwelling) reptiles and montane species. The DEM itself is static; its derivatives are stable covariates that do not need refreshing unless used alongside dynamic layers.
- Landsat 8/9 OLI (30 m multispectral): 16-day repeat, 30 m resolution across visible, NIR and SWIR bands. Useful for deriving land surface temperature (via TIRS thermal band at 100 m, resampled to 30 m), surface-water indices (MNDWI) and medium-resolution NDVI time series where MODIS 250 m is too coarse for fragmented habitats. Thermal band data carry a noise-equivalent temperature difference of roughly 0.3 K, adequate for broad thermal-niche characterisation but not fine-scale microhabitat work.
What the model is actually doing
A species distribution model (SDM) is a statistical or machine-learning function that relates known occurrence records to environmental conditions at those locations, then projects that function across a landscape to estimate habitat suitability. The output is a continuous probability or suitability surface, not a binary presence/absence map. MaxEnt, Boosted Regression Trees and Random Forest are the most commonly published approaches; all require the same inputs: georeferenced occurrence records and a stack of environmental covariate layers.
Satellite data supply the covariates. The quality of the model is bounded by the quality of those layers, which means resolution mismatches between covariate and species home-range size are a real source of error. A 250 m MODIS NDVI pixel is adequate for a lion with a home range of hundreds of square kilometres. It is meaningless for a skink whose entire range fits inside a single hectare.
Phenology metrics: what MODIS time series reveal
Vegetation phenology is the calendar of plant growth. From a 23-year MODIS archive, it is possible to extract the start of season, end of season, peak greenness, length of growing season and the coefficient of variation in annual NDVI for every 250 m pixel on Earth. These metrics capture whether a landscape is predictably seasonal or erratically variable, which matters enormously for migratory ungulates, breeding birds and any species whose food supply tracks the green wave.
The standard extraction pipeline uses the MOD13Q1 16-day composite product. Harmonic regression or asymmetric Gaussian fitting smooths the time series before threshold-based or derivative-based phenology extraction. Published studies consistently find that NDVI amplitude and seasonality length outperform single-date NDVI as SDM covariates for savanna mammals. The honest caveat: MODIS phenology metrics are unreliable in persistently cloudy regions where compositing cannot fill gaps, and the 250 m resolution blurs heterogeneous landscapes such as forest-savanna mosaics at their boundary.
Canopy height and forest structure: the GEDI contribution
Many species do not simply require forest. They require forest of a particular vertical structure. Orangutans prefer tall, closed-canopy stands. Certain bat species roost in trees above a minimum height threshold. Ground-dwelling birds in tropical forests often avoid areas where the understorey is too dense. MODIS cannot see any of this. GEDI can.
GEDI's 25 m footprints measure relative height metrics (RH50, RH75, RH98) that describe canopy top and mid-storey density directly. These can be interpolated to continuous rasters or used as point covariates in models that tolerate sparse data. The interpolation introduces uncertainty, particularly in structurally heterogeneous forests where footprint spacing leaves gaps of 500 m or more between shots. Sentinel-1 coherence helps bridge this: coherence at 6-day repeat decays faster over structurally complex or disturbed forest than over intact stands, providing a spatially continuous but indirect proxy for the structure that GEDI measures directly.
Terrain and thermal layers: resolution floors for small taxa
For reptiles, amphibians and small mammals, the relevant environmental gradients operate at scales of tens of metres, not kilometres. Terrain ruggedness at 30 m from the Copernicus DEM resolves the rock outcrops, gullies and aspect differences that determine microhabitat availability for a gecko or a tortoise. At 250 m, those features are averaged away entirely.
Land surface temperature from Landsat TIRS is available at 100 m native resolution (commonly resampled to 30 m). It captures the thermal mosaic of a landscape: south-facing slopes, exposed rock, riparian corridors and shaded forest floors all read differently. For ectotherms, this thermal heterogeneity is a direct predictor of activity time and thermoregulatory opportunity. The limitation is Landsat's 16-day revisit: a single overpass captures one moment in the diurnal and seasonal thermal cycle, so multi-date composites are necessary to represent the thermal regime rather than a snapshot.
Surface-water permanence, derived from multi-year Landsat or Sentinel-2 water-frequency products, provides a covariate that is distinct from both vegetation and terrain. Permanent water attracts very different species assemblages than seasonal pans. This layer is covered in detail on a sibling page; here it enters the covariate stack as a static or annually updated input.
Putting the covariate stack together: choices and honest limits
A typical covariate stack for a large-mammal SDM might include: MODIS phenology metrics at 250 m, GEDI-derived canopy height interpolated to 100 m, Sentinel-1 mean VH backscatter and 6-day coherence at 10 m (aggregated to 100 m to reduce noise), Copernicus DEM terrain derivatives at 30 m, and a surface-water frequency layer at 30 m. All layers are resampled to a common grid, usually the coarsest resolution in the stack, before model fitting.
For small taxa, the same workflow shifts: MODIS drops out, Landsat thermal and Copernicus DEM terrain derivatives at 30 m become primary, and GEDI footprints may be too sparse to interpolate reliably in the study area. This is not a failure of the sensors. It is an honest statement about what each instrument was designed to measure.
Model transferability is the other persistent limit. An SDM fitted on occurrence records from one region and projected onto another assumes that the covariate relationships are stable across space. They often are not. Satellize's approach to this, informed by the same rigorous covariate thinking applied in the Tonga crop-estimation programme, is to validate out-of-sample before any projection is used operationally.
From suitability surface to conservation decision
A suitability raster is not a conservation plan. It becomes useful when it is intersected with land-tenure data, threat layers and management boundaries. The immediate analytic products are: a continuous suitability map with uncertainty bands, a binary threshold map at a user-specified sensitivity/specificity trade-off, a ranked list of unprotected high-suitability patches, and a change surface comparing current suitability against a baseline year.
The update cadence matters. MODIS phenology metrics can be refreshed annually. Sentinel-1 coherence can be updated every 12 days. GEDI canopy height is a slower-moving input that changes meaningfully only after disturbance events. A well-designed monitoring workflow distinguishes which covariates are dynamic and schedules updates accordingly, rather than reprocessing the entire stack on every cycle.
Typical figures
| Phenology covariate resolution (MODIS MOD13Q1) | 250 m, 16-day composite, global archive from 2000 |
| Canopy height measurement footprint (GEDI) | 25 m diameter LiDAR footprint; ~60 m along-track spacing; coverage 51.6° N/S |
| Forest-structure proxy resolution (Sentinel-1 IW) | 10 m ground range detected; 6-day repeat at mid-latitudes, 12-day globally |
| Terrain covariate resolution (Copernicus DEM GLO-30) | 30 m posting globally; 10 m available for most land areas |
| Land surface temperature resolution (Landsat TIRS) | 100 m native (resampled to 30 m); NEDT ~0.3 K; 16-day revisit |
| Thermal band noise floor | ~0.3 K noise-equivalent temperature difference; adequate for broad thermal-niche characterisation |
| MODIS archive depth | February 2000 to present (Terra); July 2002 to present (Aqua) |
| GEDI operational period | April 2019 to present (with gaps for ISS manoeuvres); non-contiguous transect sampling |
| Covariate stack common grid | Typically resampled to coarsest layer in stack; 250 m for large-mammal models, 30 m for small-taxa models |
| Deliverable format | Cloud-optimised GeoTIFF, GeoPackage or NetCDF; uncertainty rasters included |
Analytics Satellize can run
| Multi-year MODIS phenology metrics stack | Harmonic regression or asymmetric Gaussian fitting on MOD13Q1 16-day NDVI composites; standard published phenology-extraction algorithms | GeoTIFF layers: start of season, end of season, peak NDVI, season length, inter-annual NDVI coefficient of variation |
| GEDI canopy height interpolated raster | Spatial interpolation (kriging or random forest regression) of GEDI RH98 footprint values against Sentinel-1 backscatter and DEM covariates | 30 m or 100 m continuous canopy height GeoTIFF with interpolation uncertainty layer |
| Sentinel-1 coherence and backscatter forest-structure proxy | 6- or 12-day repeat-pass interferometric coherence computation (IW mode, VV/VH); multi-temporal mean backscatter stack | 10 m coherence and backscatter GeoTIFF time series; annual summary statistics layer |
| Terrain ruggedness and topographic wetness index | Standard DEM derivatives (Vector Ruggedness Measure, TWI) computed from Copernicus DEM GLO-30 or GLO-10 | 30 m terrain covariate GeoTIFF stack (slope, aspect, VRM, TWI) |
| Species habitat suitability surface | MaxEnt, Random Forest or Boosted Regression Tree SDM fitted on occurrence records and assembled covariate stack; cross-validated and threshold-mapped | Continuous suitability raster, binary threshold map, uncertainty band raster, summary PDF report |
| Unprotected high-suitability patch ranking | Intersection of suitability surface with WDPA protected-area boundaries; patch-level zonal statistics | GeoPackage of ranked unprotected patches with area, mean suitability score and proximity to existing protected areas |
| Annual suitability change surface | Differencing of suitability rasters between baseline and current year, driven by updated dynamic covariates (MODIS phenology, Sentinel-1) | Change GeoTIFF with gain/loss classification; tabular summary by administrative unit or management zone |
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.