Solar panel hotspot and degradation mapping from airborne and satellite thermal infrared
Thermal infrared sensors detect the excess heat that faulty photovoltaic cells radiate, but satellite resolution sets a hard ceiling on what can be diagnosed remotely. This page explains where ECOSTRESS, Landsat 9 TIRS, and airborne sensors each earn their place.
Sensors
- ECOSTRESS (ISS-mounted): Thermal infrared radiometer with roughly 70 m ground sampling distance and a spectral range of 8–12.5 µm across five bands. Revisit is irregular, tied to ISS orbital precession, typically every few days to a few weeks at mid-latitudes. Useful for identifying large thermal anomaly zones across an entire farm rather than individual strings or cells.
- Landsat 9 TIRS: Two thermal bands centred at 10.9 µm and 12.0 µm, 100 m native resolution (resampled to 30 m in distributed products). 16-day revisit. At this scale a single anomalous pixel covers more area than most utility-scale inverter blocks, so TIRS is best used for farm-level temperature trend monitoring and seasonal degradation baselines rather than fault localisation.
- Sentinel-3 SLSTR: Sea and Land Surface Temperature Radiometer with 1 km resolution in thermal channels. Daily revisit. Resolution precludes any per-panel or per-string analysis; useful only for coarse, farm-averaged surface temperature time series to detect gross operational failures or confirm that a large installation is generating heat consistent with normal operation.
- Airborne thermal cameras (FLIR-class, fixed-wing or helicopter): Uncooled or cooled microbolometer arrays flown at 100–500 m AGL routinely achieve 5–20 cm ground sampling distance in the 8–14 µm band. At this resolution individual cell strings, bypass diodes, and junction-box faults become visible. Inspection rates of 50–200 MW per flight hour are achievable depending on altitude and overlap.
- Drone-mounted thermal sensors: Sub-5 cm resolution possible at low altitude. Slower areal coverage than fixed-wing aircraft but lower mobilisation cost for sites under roughly 10 MW. Standard for IEC 62446-3 compliant thermographic inspection. Not a satellite product; included here because satellite data should direct drone deployment rather than replace it.
What a hotspot actually is, and why it radiates
A photovoltaic cell that cannot convert incident irradiance into current must dissipate that energy somewhere. It goes into heat. The mechanisms vary: cell cracking, soiling concentrated on one area, partial shading causing reverse-bias conditions, manufacturing defects in the p-n junction, or a failed bypass diode forcing current through a resistive path. In each case the result is a localised temperature rise, sometimes only 5–10 °C above ambient for a mildly degraded cell, sometimes exceeding 80 °C for a hard short-circuit fault.
Thermal infrared cameras detect the emitted longwave radiation from this excess heat. The physics is straightforward: Stefan-Boltzmann emission scales with the fourth power of absolute temperature, so even modest temperature differences produce measurable radiance contrasts at 8–14 µm wavelengths. The diagnostic challenge is not sensitivity; modern sensors are more than capable. The challenge is spatial resolution relative to the size of the fault.
The resolution ceiling that satellites cannot climb over
ECOSTRESS pixels are roughly 70 m × 70 m. A standard 72-cell commercial PV module is approximately 1 m × 2 m. A single ECOSTRESS pixel therefore covers the equivalent of around 2,500 modules. Even a cluster of 50 severely degraded modules within that pixel contributes only 2 percent of the total radiance signal. The thermal contrast vanishes into noise.
What satellites can detect is the aggregate thermal signature of a large fault zone. If an entire inverter block, a full string combiner circuit, or a section covering several thousand square metres is underperforming, the area-averaged surface temperature will be measurably elevated compared with functioning sections of the same farm. ECOSTRESS and Landsat 9 TIRS are genuinely useful for that: flagging which zones of a 100 MW or 500 MW installation warrant closer attention, and tracking whether those zones worsen across months and seasons.
Sentinel-3 SLSTR at 1 km resolution is a coarser instrument still. Its value here is limited to confirming gross operational state, such as whether a farm that should be generating heat on a clear afternoon is showing any thermal signature at all, which can indicate a catastrophic shutdown.
Where airborne surveys take over
Cell-level and string-level diagnosis is an aircraft or drone problem. IEC 62446-3, the international standard for photovoltaic thermographic inspection, specifies conditions under which airborne thermal surveys should be conducted: irradiance above 600 W/m², wind speed below 4 m/s, and a minimum temperature differential between fault and background. Under those conditions a fixed-wing aircraft carrying a cooled FLIR sensor at 200 m AGL can image individual cell strings at roughly 10 cm resolution, sufficient to classify hotspot severity and type.
The practical workflow is a two-stage triage. Satellite data, processed monthly or quarterly, identifies which blocks or rows within a large farm are running anomalously warm relative to the farm average and relative to the same period in prior years. Those flagged zones receive airborne or drone inspection. This concentrates expensive flight time on the highest-probability fault areas rather than flying the entire installation blind.
For farms below about 5 MW, a single drone flight may be economical enough to cover everything. For a 300 MW desert installation, satellite-guided triage can reduce the required flight area by a factor of five or more, depending on fault prevalence.
Tracking degradation rates over time
Individual PV modules degrade at published rates of roughly 0.5–0.8 percent per year in power output under standard conditions, though real-world rates vary with climate, soiling regime, and module technology. Thermal infrared time series can serve as a proxy for degradation acceleration. A zone that shows progressively higher relative surface temperatures across two or three years of seasonal comparisons is likely accumulating faults faster than the rest of the installation.
Landsat 9 TIRS provides the most useful archive for this purpose among freely available sensors. The Landsat programme has continuous thermal data back to Landsat 5 TM (launched 1984), though sensor characteristics changed across missions. Landsat 8 and 9 TIRS data from 2013 onwards is internally consistent enough for long-term trend analysis. Pairing TIRS-derived land surface temperature retrievals with generation output data from SCADA systems, where a client can share them, substantially improves the ability to distinguish thermal anomalies caused by faults from those caused by soiling, shading from nearby vegetation, or simply higher ambient temperatures.
Cloud cover is an honest constraint. Thermal infrared cannot penetrate cloud. For farms in humid or frequently overcast climates, the number of usable clear-sky acquisitions per year from Landsat 9 may be fewer than ten, which limits temporal resolution for trend analysis.
Honest limits and common misreadings
Satellite thermal data will not tell you which specific module is faulty, which diode has failed, or whether the cause is soiling versus cracking versus a manufacturing defect. Anyone claiming otherwise is overstating the physics. What it will tell you, reliably, is whether a defined zone of a large farm is thermally anomalous relative to its own history and relative to adjacent zones under the same irradiance conditions.
Emissivity variation across panel surfaces, soiling patterns, and differential albedo can all produce apparent thermal contrasts that are not fault signatures. Careful normalisation against irradiance and ambient temperature, and comparison against clear-sky baselines, reduces but does not eliminate false positives. Ground truth from at least a sample of flagged zones is always advisable before committing to large-scale maintenance mobilisation.
Satellize runs this type of anomaly-flagging analysis on ECOSTRESS and Landsat 9 TIRS data for utility-scale installations, producing quarterly zone-level thermal deviation reports that feed directly into maintenance scheduling workflows. The Tonga crop-estimation programme is the company's named public analytics engagement, but the underlying methods for extracting quantitative signals from open-constellation thermal data apply across sectors.
Designing an inspection programme that uses each sensor correctly
The most cost-effective structure puts satellite data at the top of the funnel and aircraft or drones at the bottom. Satellite revisit, even at low resolution, is effectively free once a processing pipeline exists. It provides continuous monitoring across an entire portfolio of farms without mobilising any field resource.
Quarterly satellite analysis sets the maintenance priority list. High-priority zones receive airborne thermal inspection under IEC 62446-3 conditions once or twice a year. Drone inspection of specific flagged strings follows where the airborne survey confirms concentrated faults. This structure means that the expensive, weather-dependent, logistics-intensive drone operations are deployed only where the evidence already points.
For a new installation, establishing a thermal baseline in the first year of operation is particularly valuable. Early-life hotspots often indicate manufacturing defects that may be covered by module warranties. Satellite data alone is insufficient to make that warranty case, but it can trigger the airborne inspection that produces the IEC-compliant thermographic evidence needed.
Typical figures
| ECOSTRESS spatial resolution | ~70 m ground sampling distance |
| Landsat 9 TIRS spatial resolution | 100 m native thermal (30 m resampled product) |
| Sentinel-3 SLSTR spatial resolution | 1 km (thermal channels) |
| Airborne thermal (fixed-wing, 200 m AGL) | 5–20 cm GSD depending on sensor and altitude |
| Landsat 9 revisit | 16 days (combined Landsat 8+9: ~8 days) |
| ECOSTRESS revisit | Irregular, typically 1–5 days at ISS inclination (51.6°); not guaranteed |
| Spectral bands (thermal) | ECOSTRESS: 8–12.5 µm (5 bands); TIRS: 10.9 µm and 12.0 µm; SLSTR: 10.85 µm and 12.0 µm |
| Minimum detectable anomaly (satellite) | Zone of several thousand m² with sustained temperature deviation >2–3 °C above farm average |
| Landsat thermal archive depth | Landsat 8+9 TIRS consistent from 2013; broader archive to 1984 with caveats |
| Cloud limitation | Thermal infrared blocked by cloud; usable clear-sky acquisitions vary from <10/year (humid tropics) to >30/year (arid regions) |
Analytics Satellize can run
| Farm-level thermal anomaly map | Land surface temperature retrieval from TIRS or ECOSTRESS radiance using split-window or single-channel algorithms; spatial deviation from farm-wide mean under matched irradiance conditions | Quarterly GIS layer (GeoTIFF or shapefile) showing relative thermal deviation by inverter block or grid zone, with flagged priority areas |
| Multi-year degradation trend report | Time-series regression of zone-averaged LST anomaly against seasonal and irradiance baselines across Landsat 8+9 archive | Annual PDF report with per-zone trend charts and ranked degradation rate estimates; CSV of underlying time series |
| Maintenance prioritisation ranking | Composite scoring of thermal deviation magnitude, trend direction, and anomaly persistence across multiple acquisitions | Ranked zone list with recommended inspection urgency tier (immediate, next scheduled, monitor), formatted for integration into CMMS or asset management platforms |
| Airborne inspection targeting layer | Satellite anomaly polygons buffered and formatted as flight-planning waypoints or survey boundaries | KML or shapefile flight-zone layer ready for upload to drone or aircraft mission-planning software |
| Clear-sky acquisition calendar | Cloud-mask analysis of Landsat and ECOSTRESS overpasses for a given site over the preceding 12 months; forward projection based on climatological cloud frequency | Site-specific acquisition quality report showing number of usable thermal scenes per quarter and recommended monitoring frequency given local cloud regime |
| New-installation thermal baseline | First-year TIRS and ECOSTRESS time series processed to establish normal operating temperature distribution across farm zones before degradation accumulates | Baseline GIS layer and statistical summary used as reference for all subsequent anomaly detection; retained in client archive |
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.