Solar farm surface heat-island effect and microclimate characterisation

Large photovoltaic arrays suppress latent heat flux, raising land-surface temperature by measurable margins detectable in Landsat TIRS and ECOSTRESS imagery. The thermal footprint matters for environmental permitting and for module-efficiency accounting.

Sensors

Landsat 8/9 TIRS: Two thermal infrared bands centred near 10.9 µm and 12.0 µm, 100 m native resolution resampled to 30 m in the data product, 16-day repeat at the equator. The split-window pair enables atmospheric correction and land-surface temperature (LST) retrieval to roughly ±1–2 K accuracy under clear skies. Archive extends to 2013 for Landsat 8, giving a decade of comparable thermal scenes.
ECOSTRESS (ISS-mounted): Five-band thermal radiometer covering 8–12.5 µm at approximately 70 m spatial resolution. Irregular revisit driven by the ISS orbit precession, typically 1–5 days at mid-latitudes, with overpass times varying across the day, which is useful for capturing diurnal LST cycles that a sun-synchronous sensor misses. Published LST accuracy is around ±1.5 K.
Sentinel-2 MSI: No thermal capability, but the 10 m shortwave and near-infrared bands (B4, B8, B11, B12) support retrieval of the Normalised Difference Vegetation Index and shortwave infrared reflectance for adjacent land cover, enabling comparison of vegetative vigour and soil moisture proxies in the downwind zone. 5-day revisit with two satellites.
MODIS Terra/Aqua LST: 1 km resolution, twice-daily overpass per satellite, providing the longest continuous LST archive (2000 to present). Too coarse to resolve individual panel rows, but useful for establishing the regional background temperature climatology against which farm-scale anomalies are measured, and for capturing seasonal and interannual LST variability.

Why a solar farm is warmer than the land it replaced

A vegetated or bare-soil surface partitions incoming solar energy between sensible heat (warming the air), latent heat (evaporation and transpiration), and ground heat flux. Photovoltaic panels intercept much of the incoming radiation, convert roughly 15–22% to electricity depending on technology and temperature, and re-emit the remainder as longwave radiation and sensible heat. The critical difference is that panels do not transpire. Latent heat flux, which is the primary cooling mechanism over vegetated land, is largely suppressed beneath and between arrays.

The result is a surface heat island analogous in mechanism, though not in scale, to the urban heat island. Published field studies at sites in the Mojave Desert and in China's Qinghai province have found daytime LST differentials of 3–8 K between panel arrays and adjacent undisturbed land, with the magnitude depending on panel density, local climate, and the nature of the replaced surface. Night-time differentials can reverse or diminish as panels lose stored heat quickly. Satellite thermal imagery captures this contrast directly, provided the scene is cloud-free and the analyst applies appropriate atmospheric and emissivity corrections.

What the retrieval actually measures, and where it can mislead you

Landsat TIRS and ECOSTRESS measure top-of-atmosphere radiance in the thermal infrared window. Converting that to land-surface temperature requires atmospheric correction (removing water vapour absorption and emission) and an emissivity estimate for each surface type. Panel emissivity in the 10–12 µm window is typically around 0.94–0.96, close to vegetation and soil, so emissivity error is modest. Atmospheric correction using MODTRAN-based approaches or the split-window algorithm reduces residual error to roughly 1–2 K for Landsat TIRS under clear skies.

The honest limits are significant. Cloud cover renders thermal retrieval impossible: a single opaque cloud pixel contaminates the scene, and semi-arid regions that host most utility-scale solar farms can still accumulate 30–60 cloudy days per year. At 100 m native resolution, Landsat TIRS cannot resolve individual panel rows (typically 3–6 m wide) or inter-row gaps; it measures a blended signal from the array footprint. ECOSTRESS at 70 m is only marginally better in this respect. Sub-panel thermal heterogeneity, relevant for soiling or hotspot analysis, requires airborne or drone-mounted sensors. Satellite data is appropriate for farm-scale and landscape-scale thermal characterisation, not module-level diagnostics.

Building the comparison: control areas and multi-temporal stacking

Quantifying the thermal footprint requires a credible counterfactual: what would the surface temperature be if the farm were not there? The standard approach pairs the farm pixels with spectrally and topographically similar control polygons outside the farm boundary, ideally upwind and at comparable elevation and aspect. Sentinel-2 land-cover classification at 10 m identifies suitable control areas by vegetation type and soil class, reducing the risk of comparing panels with a fundamentally different surface that would be warmer or cooler regardless of the installation.

Multi-temporal stacking of all clear-sky Landsat scenes over a site, separated into pre-construction and operational periods, allows the analyst to test whether observed LST differences exceed pre-construction variability. A decade of Landsat 8 data and the growing Landsat 9 archive (operational since late 2021) make this feasible for most utility-scale sites commissioned after 2013. ECOSTRESS adds value by sampling different times of day, capturing the diurnal LST cycle that a single 10:30 local overpass from Landsat cannot provide.

Downwind microclimate: soil moisture and vegetation stress signals

The heat-island effect does not stop at the fence line. Warmer, drier air advected downwind can reduce soil moisture and increase vapour-pressure deficit in adjacent agricultural or natural land. Sentinel-2 shortwave infrared bands (B11 at 1610 nm, B12 at 2190 nm) are sensitive to canopy and soil water content; a decline in the shortwave infrared reflectance ratio relative to pre-farm baselines in downwind pixels is a detectable proxy for moisture stress. This is a secondary signal and requires careful interpretation: precipitation variability, land management changes, and seasonal phenology all shift these indices independently.

MODIS LST at 1 km provides the regional context needed to separate farm-induced anomalies from regional drought or heatwave signals. If the regional background is warming in a given year, attributing a 2 K anomaly at the farm boundary to the installation requires showing that the anomaly is spatially coherent with the farm geometry and persists across multiple years under varying regional conditions.

From thermal maps to efficiency and permitting numbers

Module conversion efficiency falls with rising cell temperature. The standard temperature coefficient for crystalline silicon panels is approximately minus 0.4% per degree Celsius above the standard test condition of 25 °C. If satellite-derived LST shows the farm microclimate is consistently 4–6 K warmer than the regional background during peak irradiance hours, that translates to a rough 1.6–2.4 percentage-point efficiency penalty relative to a cooler site. This is not a precise engineering figure from satellite data alone, because cell temperature depends on irradiance, wind speed, and mounting configuration as well as ambient air temperature, but it provides an order-of-magnitude correction factor for yield modelling.

Environmental impact assessments increasingly require quantification of the thermal footprint for permitting. A satellite-derived LST differential map, produced from a multi-year stack of Landsat and ECOSTRESS scenes with documented methodology, provides auditable, spatially explicit evidence. Satellize produces these as georeferenced GeoTIFF deliverables with accompanying uncertainty estimates, drawing on the same open-constellation pipeline used in its Tonga crop-estimation programme. The output is compatible with standard GIS environments used by environmental consultants and regulators.

Practical limits a buyer should know before commissioning the analysis

Three constraints dominate. First, cloud: if your site is in a persistently cloudy climate, the number of usable thermal scenes per year may be too small for statistically reliable LST differentials. Fewer than eight clear scenes per season is a warning sign. Second, farm size: installations below roughly 5 km² may not produce a thermally distinct footprint that exceeds the noise floor of 100 m resolution TIRS data, particularly in humid climates where background LST variability is high. Third, age: a farm commissioned in the last 12 months has a short pre-operational baseline, which weakens before-and-after attribution. Sites with at least two years of Landsat coverage post-commissioning yield more defensible results.

None of these constraints are hidden. They are features of the physics and the sensor archive, and an honest analysis will report them alongside the thermal maps rather than paper over them with confident-looking colour scales.

Typical figures

Thermal spatial resolution (Landsat TIRS)	100 m native, 30 m resampled product
Thermal spatial resolution (ECOSTRESS)	~70 m
LST retrieval accuracy	±1–2 K (Landsat TIRS, split-window, clear sky); ~±1.5 K (ECOSTRESS)
Revisit (Landsat 8+9 combined)	~8 days at equator under clear-sky conditions
Revisit (ECOSTRESS)	Variable, typically 1–5 days at mid-latitudes; overpass time varies
Spectral bands used	Thermal: 10.9 µm, 12.0 µm (Landsat TIRS); 8–12.5 µm (ECOSTRESS). Shortwave: Sentinel-2 B4, B8, B11, B12
Minimum detectable LST differential	~2 K at farm scale (>5 km²) under clear-sky conditions
Archive depth	Landsat 8 from 2013; ECOSTRESS from 2018; MODIS LST from 2000
Cloud limitation	Thermal retrieval impossible under cloud; semi-arid sites typically yield 150–250 usable scenes per year
Deliverable formats	GeoTIFF (LST differential maps), CSV (time-series statistics), PDF (EIA-ready summary report)

Analytics Satellize can run

Multi-year LST differential map	Split-window LST retrieval (Landsat TIRS) with emissivity correction from NDVI-based land-cover classification; farm vs. control-area comparison	GeoTIFF showing mean daytime LST anomaly (farm minus control) per season, with pixel-level uncertainty layer
Diurnal LST cycle characterisation	ECOSTRESS multi-overpass compositing across variable acquisition times to reconstruct morning, midday, and afternoon LST at 70 m	Time-of-day LST profile chart and GIS layer showing peak anomaly timing
Downwind vegetation-stress index	Sentinel-2 SWIR ratio (B11/B12) and NDVI change detection in buffered downwind zone, pre- vs. post-commissioning comparison	GIS polygon layer flagging statistically significant SWIR and NDVI decline zones with magnitude and area statistics
Regional background temperature normalisation	MODIS 1 km LST climatology used to detrend regional warming signals before attributing farm-scale anomalies	Normalised anomaly time series (CSV) separating farm-induced signal from regional variability
Efficiency-penalty estimate from ambient temperature offset	LST differential applied to published crystalline silicon temperature coefficient (approx. minus 0.4%/°C) for order-of-magnitude yield correction	One-page technical note with seasonal efficiency-penalty range and stated assumptions and uncertainties
EIA thermal-footprint report	Multi-sensor composite (Landsat, ECOSTRESS, Sentinel-2) with documented uncertainty quantification, formatted for regulatory submission	PDF report with georeferenced maps, methodology annex, and data-quality statement suitable for planning authority review

Sources

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.