Pollinator habitat quality mapping from satellite-derived floral-resource phenology
Gaps in bloom succession starve pollinator populations before anyone notices. Dense Sentinel-2 and PlanetScope time series reconstruct sub-seasonal flowering phenology across landscapes, exposing the temporal bottlenecks that patch-area maps miss entirely.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; red-edge bands B5, B6, B7 at 20 m resolve canopy chlorophyll transitions that precede and follow peak bloom. Five-day revisit at the equator (two-satellite constellation) supports sub-seasonal phenology curve fitting, though cloud cover frequently reduces effective sampling to 10–20 clear observations per growing season in temperate climates.
- PlanetScope: 3 m resolution, near-daily revisit from the SuperDove constellation. Captures fine-scale spatial heterogeneity within habitat patches and fills Sentinel-2 cloud gaps through temporal density. Lacks dedicated red-edge bands on most archive imagery, limiting chlorophyll-transition sensitivity compared with Sentinel-2.
- Landsat 8/9 OLI: 30 m resolution, 16-day revisit. Its primary value here is historical depth: the Landsat archive extends to 1972, allowing multi-decade phenology baselines against which current bloom-period shifts can be measured. Red-edge coverage is absent; NDVI and EVI remain the primary indices.
- MODIS MOD13 vegetation indices: 250–500 m resolution, near-daily revisit. Too coarse to resolve individual habitat patches in fragmented landscapes, but the 20-year MOD13 time series provides landscape-scale phenological context and helps calibrate the start-of-season and end-of-season parameters used in finer-resolution models.
Why timing matters more than area
Conservation planning has long treated habitat quality as a function of patch size and connectivity. For pollinators, that framing is incomplete. A bumblebee colony foraging radius of roughly 500 m to 2 km means that what happens within that radius across a twelve-week flight season determines colony success or failure. A landscape can contain abundant floral resources in aggregate and still collapse a colony if those resources peak simultaneously in week four and leave nothing for weeks eight through twelve.
Phenological continuity, the unbroken succession of bloom from early-season species through to late-summer foragers, is the structural property that satellite time series can actually measure. Patch area is visible in a single image. Phenological gaps are only visible across a dense stack of images taken through the season. That distinction is the entire basis for this approach.
What the red-edge bands are actually detecting
Sentinel-2's red-edge bands (B5 at 705 nm, B6 at 740 nm, B7 at 783 nm) are sensitive to chlorophyll concentration in the upper canopy. Flowering in many plant species is preceded by a measurable shift in canopy reflectance: green leaf area changes, senescent tissue appears, and in open grassland the spectral mixture shifts as flower heads emerge above the sward. The red-edge chlorophyll index (CIre = B7/B5 minus 1) has been shown in published grassland studies to track these transitions at sub-seasonal timescales.
This is a proxy, not a direct flower count. The satellite is measuring canopy optical properties; it is inferring flowering phenology from those properties. The inference is most reliable in open habitats where flower heads constitute a meaningful fraction of the canopy signal: calcareous grassland, heathland, hay meadows. In closed-canopy woodland or dense scrub, the flower signal is buried under tree canopy and the method loses most of its discriminating power. That is a hard physical limit, not an engineering problem.
Building the phenology curve and finding the gaps
The workflow begins with a dense time series: ideally 30 or more cloud-free or cloud-masked observations between March and October for a temperate Northern Hemisphere site. Sentinel-2's Sen2Cor atmospheric correction and the associated scene-classification layer are used to mask cloud and cloud shadow before any index calculation. PlanetScope observations are fused into the stack to fill gaps, accepting the trade-off that PlanetScope NDVI is not directly interchangeable with Sentinel-2 CIre without cross-calibration.
A smoothing function, commonly a Savitzky-Golay filter or a double-logistic model fitted to the NDVI or CIre time series for each pixel, produces a continuous phenology curve. From that curve, standard phenometrics are extracted: start of season, peak greenness, rate of green-up, and the date and duration of the post-peak decline. The spatial analysis then asks a landscape-level question: across the set of habitat patches within a defined foraging radius, is there a period when all patches simultaneously show low index values? That temporal trough is the phenological gap.
Gap severity can be quantified as the number of consecutive days across which the area-weighted mean index across all patches falls below a threshold, typically set relative to the within-season maximum. The threshold is empirical and site-specific; there is no universal value. Published work on UK calcareous grasslands and Dutch heathlands has used thresholds of 40–60% of peak CIre, but these should be treated as starting points rather than constants.
Honest limits: what this method cannot tell you
Species composition is invisible to this approach. A pixel with high CIre in July could represent a diverse mix of late-flowering forb species or a monoculture of a single grass species entering seed-head stage. Ground-truth survey data from the site, or at minimum a habitat classification layer, is needed to distinguish these cases. Without it, the phenology curve is a structural signal of uncertain ecological meaning.
Flower density below canopy is also undetectable. A sparse sward with ten flowers per square metre and a dense sward with two hundred flowers per square metre can produce similar canopy reflectance if leaf-area index is similar. The method identifies when and where the canopy is in a state consistent with flowering, not how much forage is actually available per unit area.
Cloud is the operational constraint that limits everything. In maritime climates, a growing season may yield fewer than fifteen usable Sentinel-2 observations. Phenology curve fitting with that few points produces wide confidence intervals around the derived metrics. PlanetScope's daily revisit mitigates this but does not eliminate it; persistent cloud systems affect both sensors simultaneously.
From landscape map to conservation decision
The primary output is a map of phenological gap severity, expressed in days, across the landscape at 10–20 m resolution. Patches that consistently show early-season or late-season deficits are candidates for targeted agri-environment scheme prescriptions: late-cut hay management, overwintering flower strips, or introduction of early-flowering species mixes. The satellite analysis identifies where the problem is and when it occurs; it does not prescribe the agronomic solution.
Multi-year stacking adds a second dimension. Running the same analysis across five or more growing seasons reveals whether phenological gaps are structurally stable features of the landscape or artefacts of a single anomalous year. Stable gaps are management priorities. Variable gaps may reflect drought years or management changes and warrant different interventions.
Satellize has applied analogous dense time-series phenology methods in the Kingdom of Tonga crop-estimation programme, where sub-seasonal NDVI trajectories are used to track crop development across small, fragmented parcels. The same curve-fitting and gap-detection logic transfers directly to pollinator habitat contexts, though the ecological interpretation differs. For a site-specific phenological gap analysis covering a defined landscape, the practical starting point is a boundary file and a target growing season.
Typical figures
| Primary spatial resolution | 10 m (Sentinel-2 visible/NIR); 20 m (Sentinel-2 red-edge); 3 m (PlanetScope); 30 m (Landsat 8/9) |
| Effective revisit (cloud-dependent) | 5 days (Sentinel-2 two-satellite); near-daily (PlanetScope); 16 days (Landsat); 250 m near-daily (MODIS MOD13) |
| Key spectral bands | Sentinel-2 B5 705 nm, B6 740 nm, B7 783 nm (red-edge); B4 665 nm, B8 842 nm (NDVI); PlanetScope red and NIR |
| Minimum habitat patch detectable | Approximately 0.1 ha at 10 m resolution (Sentinel-2); approximately 0.01 ha at 3 m (PlanetScope); smaller patches are subpixel and unresolved |
| Phenology curve observations required | Minimum 15–20 cloud-free scenes per growing season for reliable double-logistic fitting; 30+ preferred |
| Archive depth | Sentinel-2: from 2015; Landsat: from 1972 (30 m); MODIS MOD13: from 2000; PlanetScope: from approximately 2016 |
| Latency (operational monitoring) | Sentinel-2 Level-2A available within 3–5 days of acquisition via Copernicus Data Space; PlanetScope varies by licence |
| Delivery formats | GeoTIFF phenometric layers, GeoPackage patch-statistics tables, CSV phenology curve data per patch, PDF seasonal summary report |
| Cloud masking | Sen2Cor scene-classification layer (Sentinel-2); USGS CFmask (Landsat); manual QA flag filtering recommended for high-accuracy phenology work |
Analytics Satellize can run
| Sub-seasonal phenology curves per habitat patch | Savitzky-Golay or double-logistic smoothing of dense NDVI/CIre time series; standard in published remote-sensing phenology literature | GeoTIFF stack of smoothed index values at 10-day intervals; per-patch CSV of phenometrics (start of season, peak date, peak value, end of season) |
| Phenological gap map | Temporal trough detection: days per season where area-weighted patch index falls below site-calibrated threshold within a defined foraging-radius buffer | GeoTIFF layer of gap duration in days per landscape cell; foraging-radius polygons with gap-severity scores in GeoPackage |
| Multi-year phenological stability assessment | Year-on-year comparison of gap metrics across Sentinel-2 and Landsat archive; coefficient of variation across seasons per patch | PDF report with ranked patch list by gap severity and inter-annual stability; GIS layer of stable versus variable gap locations |
| Bloom-succession continuity score | Spatial overlap analysis of patch-level peak-bloom windows within foraging radius; adapted from published landscape phenology connectivity frameworks | Per-landscape-unit continuity score (days of overlapping bloom coverage) exported as attribute table and choropleth GeoTIFF |
| Agri-environment scheme targeting layer | Intersection of gap map with land-parcel cadastral boundaries; priority scoring based on gap duration and parcel accessibility | GeoPackage of priority parcels with gap metrics attached; ready for import into farm advisory GIS platforms |
| Historical phenology baseline (Landsat) | NDVI time-series reconstruction from Landsat 8/9 and available Landsat 5/7 archive; comparison with current Sentinel-2 phenology | Decadal shift report: change in peak-bloom date and gap duration relative to 10- or 20-year baseline, per habitat class |
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.