PVDeg Canary Islands degradation-mechanism atlas#

This contribution imports PVDeg 0.7.4, rather than reproducing its algorithm. The original thermal gallery includes a 100 m archipelago map, a Tenerife detail and parameter sensitivity.

AEMET-adjusted continuous rate maps#

The AEMET/ERA5 follow-up fits seasonal weather corrections from local observations and calculates continuous land rasters for hydrolysis, photodegradation, thermomechanical and combined degradation rates. The weather is locally adjusted; module-rate parameters remain a published mono-Si scenario. Raw and corrected maps and held-out validation are included.

Nine hourly degradation-mechanism maps#

The mechanism gallery adds nine maps, recomputed with PVDeg 0.7.4 from real PVGIS TMY inputs at 11 locations spanning all eight main islands. The original thermal raster alone was not a complete degradation analysis.

Map

Calculation and interpretation

Thermal

Hourly module-temperature Arrhenius exposure

Damp heat / hydrolysis stress

Temperature × surface humidity exposure, n=1

Broadband light + heat

Temperature × POA irradiance exposure, p=1; not UV-specific

Light + moisture + heat

Joint exposure, n=p=1; not total module degradation

Surface moisture

Mean module-surface relative humidity

Front encapsulant moisture

Mean EVA W001 front encapsulant RH

Back encapsulant moisture

Mean EVA W001 back encapsulant RH

Backsheet moisture

Mean PET W017 backsheet RH

Solder fatigue

PVDeg hourly thermomechanical fatigue damage metric, kPa

The first four are dimensionless relative exposures at 60 kJ/mol, normalized against equal-duration exposure at 25 C, 50% RH and 1000 W/m². The other maps have their own units. Do not sum them or interpret them as %/year.

Shared scenario and actual spatial support#

PVGIS selects representative months from 2005–2023 (SARAH3 radiation and ERA5 meteorology); the manifest retains each site’s selected source years and provider metadata. This assembled TMY is not a continuous historical year or an extreme weather record. A 2023 UTC calendar is used only to align month/day/hour. Solar position includes the provider’s irradiance time offset. Model parameters: 28° south-facing fixed tilt, isotropic sky, albedo 0.25, SAPM open-rack glass/polymer temperatures and provider 10 m wind without a height correction.

Moisture uses upstream EVA W001 and PET W017 properties, 0.46 mm back encapsulant and 0.30 mm backsheet. The upstream initial water state is retained (half equilibrium at the first sample); no multi-year moisture equilibration is claimed. The front-encapsulant model uses the full-year weighted exposure. Solder fatigue uses upstream hourly defaults (54.8 C reversal, C1=405.6, n=1.9, b=0.33, Q=0.12 eV), with fixed UTC day boundaries. TMY month joins may introduce artificial transitions, so fatigue is screening, not field validation.

Only the colored points are calculated. Gray island silhouettes come from the existing 100 m climate masks. Weather has its provider’s coarser spatial support: no 100 m humidity, wind, irradiance or degradation accuracy is claimed. Coastal and mountain points do not represent whole-island averages. Source and processed-input SHA-256 hashes are retained. Small negative PVGIS wind speeds (from -0.5 to 0 m/s) are capped at zero with per-site correction counts; larger negative speeds fail validation. Original provider bytes remain unchanged; the gallery runs without networking.

Mechanisms not yet quantified#

  • Spectral UV degradation / discoloration needs UV spectral irradiance and material action spectra; broadband POA is not an equivalent input.

  • LeTID and LID need cell technology, initial defect states, lifetime and kinetics. A generic cell assumption would not establish the islands’ actual loss rates.

  • PID needs voltage/polarity, electrical topology and technology-specific leakage and degradation calibration. Moisture alone is not PID.

  • Corrosion, delamination and cracking require mechanism/material calibration; the moisture and solder maps are stress indicators, not those failure rates.

  • Annual power loss, remaining lifetime and combined loss need field calibration and an interaction model. None is fabricated from the nine screening metrics.

Reproduce the hourly maps#

uv sync --locked --all-packages --group docs
uv run --no-sync python scripts/prepare_pvdeg_canary_weather.py
cd examples/analysis
uv run --no-sync python plot_pvdeg_canary_mechanisms.py

The acquisition script writes exact raw provider responses under outputs/ and plain CSV inputs plus provenance under examples/analysis/pvdeg_canary_weather/. Existing raw files are reused; remove that output directory explicitly to acquire a fresh provider snapshot.

Original thermal connector, method and analysis#

  • energy_data_retrieval.pvdeg.to_pvdeg_weather validates canonical weather columns and UTC timestamps for PVDeg workflows. The existing PVGIS connector remains the acquisition layer. Missing humidity is rejected, never invented.

  • energy_methods.pvdeg.thermal_stress_index calls pvdeg.degradation.arrhenius with duration weights and normalizes exposure against the same duration at a constant reference temperature.

  • energy_analysis.pvdeg_canary.screen_climate_tile applies that method to twelve monthly temperature bands; write_stress_geotiff exports the results.

The index is sum(w * exp(-Ea/(R*T))) / exp(-Ea/(R*Tref)), with weights summing to one, Kelvin temperatures, R in kJ/mol/K, and Ea in kJ/mol. The default reference is 25 C. A value of two means twice the modeled thermal exposure under the specified assumptions, not twice the annual power loss. Missing any month invalidates a cell; ocean and missing inputs are transparent.

Data and resolution#

CHELSACanaryClim v1.0 monthly air temperature covers 1979–2013, CC0-1.0, Karger et al. (2025), DOI 10.16904/envidat.692. The processor downloads twelve SHA-256-pinned source files through the existing CanaryClim connector, masks encoded nodata, decodes Kelvin and converts to Celsius. Each 100 m EPSG:32628 cell samples the nearest native source pixel; this is a point sample, not an area average. Stored temperatures retain 0.01 C encoding, which does not imply that precision in the underlying climate model.

The 2025 source COG spacing is about 5 m; the published model comparison used 100 m (Patiño et al., 2023). Neither spacing validates local PV degradation accuracy. Input footprints define coastlines; no artificial smoothing or coastal extrapolation is applied. Seven disjoint windows include La Graciosa, Chinijo and Lobos. Provenance records the source URLs, hashes, period, license, processing grid and scenario.

Interpretation limits#

The central scenario uses 60 kJ/mol and air temperature +20 C. These are illustrative assumptions, not fitted material parameters. The gallery varies Ea over 40/60/80 kJ/mol and offsets over 0/20/40 C. Sensitivity curves are not confidence intervals. Humidity and irradiance exponents are zero. Monthly temperatures plus a constant offset do not represent hourly module thermal behavior, solar heating, wind cooling, mounting or nighttime conditions. Nonlinearity means monthly means generally cannot reproduce hourly exposure. This supports regional screening and prioritization of further data collection; it does not predict annual degradation, a 30–50 year lifetime, warranty outcomes, LeTID, UV damage, moisture ingress, corrosion or measured reliability.

For a calibrated follow-up, acquire common-period hourly weather with PVGIS, validate it with the weather adapter, calculate module temperature using measured mounting and wind conditions, and obtain mechanism-specific material parameters and field validation before making lifetime claims.

Reproduction#

uv sync --locked --all-packages --group docs
# About 0.8 GB transferred; one month in memory, no raw disk cache.
uv run --no-sync python scripts/prepare_pvdeg_canary.py
# Compare/copy the resulting tiles and manifest to examples/analysis to refresh.
cd examples/analysis
uv run --no-sync python plot_pvdeg_canary.py

For package use: pip install 'energy-methods[pvdeg]'. The dependency is optional because upstream PVDeg includes substantial geospatial/notebook dependencies. The documentation group and CI explicitly install it; gallery execution uses real imported PVDeg and offline climate fixtures, never a substitute backend.

Credit#

Springer et al. (2026), PVDeg software, National Laboratory of the Rockies / DuraMAT, BSD-3-Clause. PVDeg documentation. The upstream documentation describes its JOSS paper as in review; no published paper DOI is asserted. This is an independent PVDeg integration, not a claim that its mechanism originates in the repository’s foundational thesis.