Long-term panel degradation tracking via satellite reflectance time series
Encapsulant browning, delamination, and potential-induced degradation all shift a panel's optical signature over years. Dense Landsat and Sentinel-2 archives, properly normalised, can detect those shifts at the farm scale before they appear in financial audits.
Sensors
- Landsat 8/9 OLI: 30 m ground sample distance, 16-day repeat per satellite (8-day combined), Collection 2 surface reflectance products with atmospheric correction applied; archive extends to 1972 across the broader Landsat programme, giving the longest continuous multi-spectral record available.
- Sentinel-2 MSI: 10 m in visible and NIR bands, 20 m in red-edge and SWIR; 5-day revisit at the equator with both Sentinel-2A and 2B operating. Level-2A surface reflectance available from 2017. Finer spatial detail resolves sub-array variation within large farms.
- MODIS MCD43A BRDF/Albedo product: 500 m resolution, daily composites. Not useful for resolving individual farms, but the Ross-Thick/Li-Sparse BRDF model it applies is the published reference for correcting directional reflectance effects that would otherwise swamp a multi-year trend signal.
- Landsat Collection 2 surface reflectance archive: USGS-reprocessed archive applying a consistent atmospheric correction (LaSRC) and terrain correction across all Landsat 4–9 scenes. Consistency across sensors is essential for trend analysis; earlier ad-hoc corrections introduced artefacts that mimic degradation signals.
Why a degrading panel looks different from orbit
A healthy crystalline-silicon panel is designed to absorb. Its anti-reflective coating and dark cell material keep broadband reflectance low, typically in the 3–6 % range in the visible. When encapsulant browning begins, the EVA polymer yellows and scatters more shortwave radiation back upward. Delamination introduces air gaps that act as partial mirrors. Potential-induced degradation shunts current but also alters the surface chemistry of the cell metallisation. Each mechanism shifts the panel's reflectance signature, and in aggregate across thousands of panels in a utility-scale farm, that shift becomes detectable from 700 km altitude.
The signal is small. We are talking about reflectance changes of a few tenths of a percent per year, compounding over a decade. That is why this is a multi-year problem, not a single-pass one. It is also why it must be kept clearly separate from soiling: dust and bird deposits raise reflectance within days and fall back to baseline after rain or cleaning. Degradation does not fall back. The analytical task is to separate a slow, monotonic trend from a noisy seasonal cycle of soiling and recovery.
The normalisation problem is not optional
Raw top-of-atmosphere reflectance is useless for this application. Solar zenith angle alone changes the apparent brightness of a tilted panel surface by several percent between a summer and a winter overpass, swamping any degradation signal. Atmospheric path length varies with season and aerosol loading. The sensor view angle shifts the bidirectional reflectance distribution function (BRDF) of the panel surface. All three effects must be corrected before any trend is credible.
The standard approach applies the MODIS BRDF model coefficients, or a functionally equivalent kernel-driven model, to normalise each observation to a fixed sun-sensor geometry, typically nadir view at 45-degree solar zenith. This is published practice in the surface-albedo literature and is the method underpinning USGS Collection 2 processing. After BRDF normalisation, a cloud and shadow mask is applied. What remains is a time series of geometrically and atmospherically consistent surface reflectance values that can be fitted with a linear or piecewise trend model.
A minimum archive depth of five years is not a conservative preference. It is a statistical necessity. With Sentinel-2's 5-day revisit, a cloud-free site at a mid-latitude location might yield 60–80 usable observations per year after masking. That sounds generous, but the seasonal soiling cycle has an amplitude that can exceed the annual degradation increment by a factor of ten. Separating the two requires enough full seasonal cycles to fit and subtract the periodic component reliably.
Which spectral bands carry the degradation signal
Encapsulant browning preferentially absorbs in the blue and green, so the red-to-blue reflectance ratio rises as browning progresses. Landsat OLI Band 2 (450–510 nm) and Band 4 (640–670 nm) are the relevant pair. Sentinel-2 Band 2 (490 nm) and Band 4 (665 nm) cover the same physics at finer resolution.
Delamination and soiling both raise broadband reflectance, but delamination tends to raise SWIR reflectance disproportionately because the air-gap interfaces scatter longer wavelengths differently from the encapsulant-glass interface. Landsat OLI Band 6 (1570–1650 nm) and Sentinel-2 Band 11 (1610 nm) are therefore diagnostic. A farm showing rising SWIR reflectance without a corresponding rise in visible bands is a candidate for delamination rather than simple soiling.
Thermal infrared adds a complementary view. Landsat 8/9 TIRS Band 10 (10.6–11.2 µm) at 100 m resolution can detect changes in surface emissivity associated with delamination and cell cracking, though the spatial resolution limits this to large homogeneous arrays. TIRS data is noisier than OLI and requires careful stray-light correction; the USGS has published known calibration issues with TIRS Band 11 that make Band 10 the more reliable choice.
What the method cannot do
Spatial resolution is the hard ceiling. At 10–30 m, a single pixel covers between 100 and 900 square metres. A utility-scale farm of 50 MW might span 100 hectares, giving hundreds to thousands of pixels. That is enough to detect farm-level or zone-level trends. It is not enough to locate a failing string, a single inverter's worth of panels, or a small section of early-stage delamination covering less than a few hundred square metres.
Cloud cover is the data-availability constraint, not an analytical one. Tropical and maritime sites with persistent cloud cover may yield fewer than 20 usable observations per year per sensor. Combining Landsat and Sentinel-2 archives partially compensates, but a site in a persistently overcast climate may not accumulate enough clean observations to separate trend from noise within a five-year window. This is an honest limit that should be assessed site-by-site using historical cloud-fraction records before committing to the method.
The method also cannot distinguish between degradation mechanisms without ground-truth. A rising red-to-blue ratio is consistent with encapsulant browning, but also with a change in the soiling composition (iron-rich dust absorbs blue). Confirmation of mechanism requires electroluminescence imaging or I-V curve testing on the ground. The satellite time series is a screening tool that prioritises where to send the inspection team, not a replacement for it.
Building a defensible baseline for asset owners
The practical output is a per-zone reflectance trend map, expressed as annual reflectance change in each spectral band, with uncertainty bounds derived from the residual variance of the trend fit. Zones where the trend exceeds two standard deviations of the site's historical noise floor are flagged for ground inspection. This is not a pass/fail system; it is a ranked list of concern.
For asset owners managing a portfolio of farms across multiple geographies, the value is comparability. A consistent processing chain applied to Landsat Collection 2 and Sentinel-2 Level-2A data produces reflectance trends on the same scale across sites in Spain, South Africa, and Australia. Internal benchmarking across a portfolio is only meaningful if the underlying radiometry is consistent, which is precisely what the reprocessed archives provide.
Satellize applies this processing chain as part of its satellite-data analytics work, running on open constellation data and delivering zone-level trend reports as GIS layers and structured data feeds. Clients wanting to understand the method before commissioning a multi-year monitoring programme can request a single-site retrospective analysis using the existing archive, which requires no new satellite tasking and produces results within weeks.
Typical figures
| Spatial resolution (optical) | 10 m (Sentinel-2 visible/NIR), 20 m (Sentinel-2 SWIR), 30 m (Landsat OLI) |
| Spatial resolution (thermal) | 100 m (Landsat 8/9 TIRS Band 10, resampled to 30 m in products) |
| Revisit cadence | 5 days (Sentinel-2A+B combined at equator), 8 days (Landsat 8+9 combined) |
| Usable archive depth | Sentinel-2: 2017–present; Landsat OLI: 2013–present; broader Landsat: 1972–present (Collection 2) |
| Minimum analysis period | 5 years of cloud-free observations required for statistically separable trend |
| Key spectral bands | Blue (450–510 nm), Red (640–670 nm), SWIR (1570–1650 nm) for optical; 10.6–11.2 µm for thermal |
| Minimum detectable reflectance trend | Approximately 0.1–0.3 % per year at farm scale after BRDF normalisation (site- and cloud-dependent) |
| Minimum detectable farm area | Roughly 5–10 ha for statistically stable zone-level trend estimates at 10–30 m resolution |
| Atmospheric correction standard | USGS LaSRC (Landsat Collection 2), ESA Sen2Cor (Sentinel-2 Level-2A), BRDF normalisation via MCD43A kernel model |
| Deliverable formats | GeoTIFF trend maps, CSV time-series per zone, GIS polygon layer with flagged zones, structured JSON data feed |
Analytics Satellize can run
| Per-zone annual reflectance trend map | Linear trend fitting on BRDF-normalised, cloud-masked surface reflectance time series (Landsat Collection 2 + Sentinel-2 Level-2A) | GeoTIFF and GIS polygon layer showing annual reflectance change per band per zone, with uncertainty bounds |
| Encapsulant browning index time series | Red-to-blue reflectance ratio (OLI Band 4 / Band 2 or equivalent MSI bands) computed per observation and fitted for monotonic trend | CSV time-series per farm zone, flagging zones where ratio trend exceeds two-sigma threshold |
| SWIR anomaly map for delamination screening | SWIR reflectance trend decomposed from visible trend; zones with disproportionate SWIR rise identified as delamination candidates | Ranked list of suspect zones as GIS layer, with supporting spectral trend plots |
| Thermal emissivity change layer | Landsat TIRS Band 10 multi-year compositing and trend analysis after stray-light correction; compared against OLI visible trend to separate thermal from optical anomalies | GeoTIFF showing zones of statistically significant emissivity change, flagged for ground inspection |
| Soiling-versus-degradation separation report | Harmonic regression to isolate seasonal soiling cycle; residual monotonic component attributed to permanent degradation | Annual report with decomposed trend components, distinguishing recoverable soiling from permanent loss per zone |
| Portfolio-level benchmarking dashboard | Consistent Collection 2 / Level-2A processing chain applied across multiple sites; normalised reflectance trends compared on a common scale | Structured data feed and summary report ranking farms by degradation rate, suitable for asset-management review |
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.