Vegetation encroachment monitoring at operational solar farms
Unmanaged vegetation beneath and around solar arrays shades panels and raises fire risk, yet satellite detection requires careful sensor selection: Sentinel-2's 10 m pixels miss sub-panel patches that Planet's 3–4 m imagery resolves.
Sensors
- Sentinel-2 MSI: 10 m resolution in red, NIR and SWIR bands; 5-day revisit at the equator (2–3 days with both satellites). Sufficient for farm-level NDVI and NDWI trend monitoring but cannot reliably isolate vegetation patches smaller than roughly 20–30 m across.
- Planet SuperDove: 3–4 m resolution across 8 spectral bands including red-edge and NIR; daily revisit over most land areas. Resolves inter-row and sub-panel vegetation patches that fall entirely within a single Sentinel-2 pixel.
- Landsat 8/9 OLI: 30 m resolution; 16-day revisit per satellite, 8-day combined. Too coarse for sub-array detection but provides a free, consistent archive back to 1972 for establishing long-term seasonal baselines and detecting gross encroachment trends.
- MODIS MOD13 (Terra/Aqua): 250 m NDVI and EVI composites at 16-day intervals. Useful only for landscape-scale context around very large farms; individual array vegetation is invisible at this resolution.
Why vegetation beneath a solar array is not a trivial maintenance note
A solar panel shaded by a knee-high thistle patch loses disproportionately more output than the shaded area alone would suggest. Because most utility-scale arrays wire panels in series strings, a single shaded cell forces the string's bypass diode to activate, cutting the contribution of multiple panels simultaneously. The energy penalty can be several times larger than the physical shade footprint.
Fire risk compounds the problem in dry climates. Dry grass and scrub accumulating beneath tracker arrays creates a continuous fuel bed directly beneath electrical infrastructure. Several large farm fires in southern Europe and California have been attributed partly to vegetation contact with wiring or inverters. Neither risk is visible from a site office without systematic ground inspection, which is expensive and infrequent at farms covering hundreds of hectares.
What a floating roof gives away: the spectral logic
Healthy green vegetation has a characteristic spectral signature: low reflectance in the red band, high reflectance in the near-infrared. The Normalised Difference Vegetation Index (NDVI) captures this contrast. NDWI, using green and NIR bands, adds sensitivity to plant water content, which matters when distinguishing actively growing encroachment from dry senescent material. Both indices are computable from Sentinel-2 bands at no cost.
The complication is that solar panels themselves have a distinctive spectral response: low NIR reflectance, moderate SWIR. When vegetation grows beneath or between panels, the mixed pixel seen by the satellite blends panel and plant spectra. At 10 m resolution, a Sentinel-2 pixel covering a typical panel row (roughly 2–3 m wide) and the inter-row gap (1–3 m) is already a mixture. A small weed patch adds a faint NDVI signal that can be lost in that mixture or confused with panel soiling. This is the resolution floor that matters in practice.
The 10 m wall and how Planet steps over it
Sentinel-2's 10 m pixels are well matched to farm-level monitoring. A 50 ha farm will contain roughly 500 pixels, enough to map broad zones of encroachment, track seasonal green-up, and flag areas for ground inspection. What they cannot do reliably is detect a 2 m wide strip of grass growing in a single inter-row gap, or a localised thistle cluster beneath a tracker row. Those features occupy a fraction of one pixel.
Planet SuperDove's 3–4 m pixels change the geometry. An inter-row gap of 2 m now occupies roughly half a pixel rather than a fifth of one, and a 4 m weed patch becomes resolvable as a distinct object. The red-edge band (approximately 700–730 nm) available on SuperDove adds sensitivity to early-stage green-up before NDVI saturates, which matters for catching encroachment before it becomes a shading problem. The trade-off is cost: Planet imagery is licensed commercially, and daily tasking over a large portfolio of farms accumulates quickly.
A practical operational approach combines both. Sentinel-2 time series run continuously at no marginal cost, flagging pixels where NDVI rises anomalously relative to a multi-year seasonal baseline. Those flagged zones then trigger targeted Planet acquisitions for sub-panel resolution mapping. This avoids purchasing Planet coverage everywhere all the time.
Pollinator habitat: the intentional vegetation problem
Many solar farm operators in the UK, Germany and the United States have established managed pollinator habitat beneath and around arrays. These schemes, sometimes required by planning consent, deliberately cultivate wildflower mixes and low-growing grasses. From a satellite sensor's perspective, a thriving pollinator meadow and an unmanaged weed invasion look almost identical in NDVI terms.
Distinguishing the two requires additional information layers. Managed habitat typically follows a documented planting plan with known spatial boundaries, a predictable seasonal phenology tied to the sown species mix, and a controlled height profile. Encroachment tends to appear at array edges first, spread inward, and show a phenology driven by local weed species rather than managed seed mixes. Incorporating farm-specific habitat zone polygons as a mask, and comparing observed phenology curves against expected species-level green-up timing, substantially reduces false positives. Without that contextual layer, any vegetation monitoring system will generate alerts that operations teams learn to ignore, which defeats the purpose.
Seasonal baselines and change detection in practice
A single NDVI image tells you almost nothing useful. The same pixel value means actively growing encroachment in March and normal senescence in August. Reliable detection rests on comparing each observation against a multi-year seasonal baseline for that specific pixel, typically built from three or more years of Sentinel-2 or Landsat archive data. A z-score or percentile threshold flags pixels where current NDVI exceeds the historical range for that calendar week.
Revisit frequency sets the temporal resolution of the alert system. Sentinel-2's 5-day revisit is frequently interrupted by cloud, particularly in temperate climates. In a typical UK summer, usable cloud-free observations may arrive only once every two to three weeks. In semi-arid climates where fire risk is highest, cloud is less of a constraint and the 5-day cadence is more reliably achieved. Planet's daily tasking helps bridge cloud gaps but does not eliminate them. Operators should expect detection latency of one to four weeks in cloudy seasons rather than days.
Satellize applies this baseline-anomaly framework in its vegetation analytics work, including the crop-estimation programme it runs for the Kingdom of Tonga, where separating seasonal phenology from anomalous change is equally central to the method.
Honest limits of the satellite approach
Satellite monitoring does not replace ground inspection; it directs it. A flagged zone tells a maintenance team where to look, not what species is growing or how tall it has become. Height cannot be inferred from optical NDVI alone. A dense low-growing clover that poses minimal shading risk and a two-metre buddleia that shades an entire string can produce similar NDVI anomalies.
Sub-panel vegetation that grows entirely in the shade of the panels may suppress its own NDVI signal by being light-limited. Ironically, the vegetation most sheltered from satellite view is also the most sheltered from desiccation and therefore the most persistent. Satellite data is strongest at detecting vegetation in inter-row gaps and at array perimeters. For the shaded understorey, periodic drone surveys or ground transects remain necessary. The satellite layer reduces the frequency and cost of those surveys by targeting them rather than eliminating them.
Typical figures
| Best spatial resolution (vegetation detection) | 3–4 m (Planet SuperDove); 10 m (Sentinel-2 MSI) |
| Revisit frequency | Daily (Planet); 2–5 days (Sentinel-2 dual satellite); 8 days (Landsat 8+9 combined) |
| Minimum detectable vegetation patch | ~4–6 m diameter with Planet; ~20–30 m with Sentinel-2 (mixed-pixel constraint) |
| Key spectral bands | Red (665 nm), Red-edge (705–740 nm), NIR (842 nm), SWIR (1610 nm); all available on Sentinel-2 and Planet SuperDove |
| Primary indices | NDVI, NDWI, Red-edge NDVI (for early-stage green-up detection) |
| Detection latency (cloud-dependent) | 1–7 days in arid climates; 1–4 weeks in temperate/cloudy climates |
| Archive depth for baseline construction | Sentinel-2: from 2015; Landsat: from 1972; Planet SuperDove: from approximately 2021 |
| Typical baseline period required | Minimum 3 years of same-sensor data for reliable seasonal anomaly thresholds |
| Coverage | Global land; cloud permitting per acquisition |
| Delivery formats | GeoTIFF anomaly maps, GeoJSON alert polygons, time-series CSV per zone |
Analytics Satellize can run
| Seasonal NDVI baseline and anomaly map | Multi-year percentile baseline from Sentinel-2 archive; z-score change detection per pixel per calendar week | GeoTIFF map updated on each cloud-free acquisition; anomaly zones exported as GeoJSON polygons |
| Sub-panel encroachment patch map | Red-edge NDVI and NDWI from Planet SuperDove at 3–4 m; object-based image analysis to delineate discrete vegetation patches | GeoJSON polygon layer with patch area and NDVI intensity, delivered within 48 hours of cloud-free Planet acquisition |
| Pollinator habitat mask and encroachment separation | Client-supplied habitat zone polygons overlaid on anomaly layer; phenology curve comparison against known species green-up timing | Filtered alert layer excluding managed habitat zones; monthly report on encroachment versus habitat extent |
| Perimeter encroachment risk score | Directional spread analysis from array boundary inward; weighted by prevailing wind direction and adjacent land-cover type for fire-risk context | Risk-scored perimeter segment map; quarterly PDF summary for asset managers |
| Triggered Planet tasking alert | Sentinel-2 anomaly flag used as trigger condition for commercial Planet acquisition order over flagged zones only | Automated tasking request with confirmation; high-resolution image and patch map delivered within 1–3 days of acquisition |
| Multi-site portfolio encroachment dashboard | Standardised NDVI anomaly pipeline applied uniformly across all client farm polygons; ranked by anomaly severity and area | Web-accessible dashboard with per-farm status, trend charts, and downloadable inspection priority list |
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.