The observatory combines open plant registers, national generation series and published degradation studies. This page sets out where each layer comes from, what we do to it, which licence applies, what is observed and what is modelled, and what we do not claim.
Last revised 12 October 2026. Each figure carries its source and locator. Where we could read only an abstract, the locator says so.
Each published number names its source, the locator inside that source, the publication or retrieval date and the licence. A number without a source is not published.
VERIFIED is published by its source. ESTIMATED is our calculation, with its assumptions stated. ILLUSTRATIVE is a template or an example, not a real asset.
Residential systems never appear as individual points. Homes enter the observatory only as area totals, and only in cells of at least five systems.
A degradation rate from a study describes a population of modules or systems. We never present it as the measured behaviour of a named plant.
Each row in the data files carries its source. Point and area rows also carry a licence, so any figure traces back to one source and one licence.
National and regional registers of PV installations, where the register publishes plant-level records. They feed the C&I and utility-scale points and the residential area totals. Each point keeps the register's own identifier in its uid field. The URL of each source is in sources.json.
Plant-level open databases, such as the WRI Global Power Plant Database and the NREL PVDAQ system datasets. The OpenStreetMap and Wikidata connectors exist, but they add no rows to the current release. A record enters only with coordinates and a capacity above 10 kWp. Records whose licence is not on the publishable list are excluded, including non-commercial and unverified licences.
Annual national series for solar generation and, where the provider gives it, installed capacity. The current release uses Ember. They give the country context for the observatory's totals. A country series is never allocated to individual plants.
Planned, not in the current release. Module and inverter reference data: power at standard test conditions, efficiency ranges and technology class. It will describe equipment types and will say nothing about how any installed unit performs.
Every record is assigned a class from its power in kWp. The class decides what the site may show.
| Code | Class | Nameplate power | What the site may show |
|---|---|---|---|
| 1 | Residential | up to 10 kWp | Area totals only, per administrative unit, with at least five systems per cell. No points, names or addresses. |
| 2 | Commercial and industrial (C&I) | above 10 kWp, up to 1 MWp | Points where an open plant register or open plant record gives coordinates; area totals otherwise. |
| 3 | Utility-scale | above 1 MWp | Points where the coordinates come from a source whose licence allows publication; area totals otherwise. |
Basis. Power is stored in kWp, and the basis field records how each value was obtained: dc (the source reports DC), est_from_ac (AC × 1.2, ESTIMATED), source_kw_dc_assumed (the source reports kW, taken as DC), reported_gross_mw (gross MW as the Chilean register reports it) or unknown (the source does not say). Totals mix bases. A boundary value falls in the lower class: 10 kWp is residential, and 1,000 kWp is C&I.
AC-only sources. Where a source reports only AC power, the US EIA and UK REPD connectors store kWp = AC × 1.2, with the basis marked est_from_ac. This is ESTIMATED. The ratio of 1.2 is a convention, not a measurement of any plant. The Japanese FIT and FIP register reports AC capacity as published, and that capacity is not converted.
Residential systems can be linked to a household. The rules below are applied when a record is written, before anything is published, and not only when a page is displayed.
Residential data is never published point by point, in any file or view.
Sources describe plants in different units and with different gaps. These steps make the rows comparable. Where a step estimates a value, the basis field says so.
Each normalised point keeps its source identifier, source URL, licence and retrieval date. Raw downloads are stored unchanged, with a checksum, so every normalised row can be traced back to its download.
kW is kept as it is, MW is multiplied by 1,000, GW by 1,000,000, and W or Wp by 0.001. A capacity that cannot be read as a number with a unit is left empty, and a point without a capacity is not published.
Status is coded 0 operating, 1 commissioning, 2 planned, 3 retired or unknown. The WRI 2019–2020 snapshot is coded 3, because its status is not current. Technology and mounting are "unknown" when the source does not say. In the current release, technology is unknown for every point.
A year is taken from the source as published, and is left empty where the source gives none. It is not estimated. Planned and commissioning plants can carry years after the data cut-off year, and a few records carry years before 1990, as published. Coordinates must fall between 56° S and 74° N and between 180° W and 180° E, and 0,0 is rejected.
Within one source, the connector skips a source identifier it has already read, so one record is not counted twice.
Across sources, two points within 150 m of each other, with capacities within a factor of two, are treated as one plant. The point from the higher-priority source is kept, and the other sources are listed in its also field. Matching uses distance and capacity only, so one plant that two sources place more than 150 m apart is counted twice.
Each row sets the operating capacity in the open registries against the national capacity in the Ember series. Where Energy-Charts publishes a country, a second generation series gives a check. The grade says how far the data can be relied on for that country. It is not a ranking of solar health or of the market.
The country table needs JavaScript. It reads the coverage file in your browser. The method below applies to every row.
Coverage ratio = operating kWp in the registry sources ÷ Ember capacity in kW. Ember capacity in kW is GW multiplied by 1,000,000 kW per GW. ESTIMATED
Operating means status 0 only. Commissioning (status 1) is shown in its own column and is never added to the operating figure. Planned rows (status 2) are excluded. Unknown rows (status 3) are shown as unknown, never as retired. Wikidata rows that give only a nominal capacity are excluded from every sum and ratio. OpenStreetMap rows are excluded from ratios and grades.
A ratio above 1.10 means the registries hold more operating capacity than the national series, by more than a tenth. We do not publish that total as a sum. The row is marked non-additive. The same mark is set when two sources overlap in one country, even when the ratio falls inside the band.
A non-additive row keeps its figures and its flag. Its grade cannot be better than C.
The United States is counted on operating capacity only. Adding the national survey's net-metering totals would move its ratio past the band, so no United States total across sources is published until that overlap is resolved.
Implied CF = Ember solar generation (TWh) ÷ (Ember capacity in GW × 8.76). The factor 8.76 is 8,760 hours a year divided by 1,000. The result is the share of the year that the installed capacity would have to run at full output to produce the generation. ESTIMATED
A CF below 5 % or above 35 % is flagged. A flagged value points to a basis or year mismatch between the two Ember series, or to a stale capacity figure. Check it before you use the country.
A country's capacity is stale when it is identical in two consecutive years. A stale value is not carried forward. Each row shows the year of its own Ember value. A country without a capacity figure for the latest year shows that year as missing, and does not borrow an earlier one.
A stale flag caps the grade at C.
Generation agreement for 2024 = 1 − |Energy-Charts − Ember| ÷ Ember. A value of 100 % means the two series match exactly. ESTIMATED
Capacity gap = (Ember − Energy-Charts) ÷ Energy-Charts, shown for 2024 and 2025. A gap above 10 % in either direction is flagged. Energy-Charts (Fraunhofer ISE) covers a subset of countries. A country it does not cover has no agreement value. Licence and credit for each source are in sources.json.
Counts, kWp, and the Ember and Energy-Charts values are VERIFIED as their sources publish them. Ratios, implied CF, agreement, gaps and grades are ESTIMATED from those values. The Ember capacity basis is not verified.
DC or AC. Registry kWp mixes bases. US EIA rows are DC where the source reports DC, and AC × 1.2 otherwise. UK REPD rows are AC × 1.2. Chilean rows are reported gross. IEC 61724 does not specify the rating basis for AC yield, and the Sandia guide uses DC power at STC (std-sandia-rating-basis-not-specified). Every ratio here therefore compares mixed bases. Read a ratio as a consistency check, not as an exact share of the market.
The grade is the worst of two checks, a coverage band and an agreement band, and any flag can lower it. ESTIMATED
Most countries have no Energy-Charts value, so most cannot reach A. This is deliberate. An A needs a second source that agrees closely with the registries and the national series.
CL-REG-05). A mean above that limit is treated as an error until it is checked.The table gives each licence as the publisher states it, checked against the publisher's own licence page or, where we could read it, the licence text. The licence text governs. The wording here is a summary for orientation. The table covers the external sources this page names. For each source that the release lists, the licence and attribution text are in sources.json.
| Source | Licence | What use requires | Locator |
|---|---|---|---|
| OpenStreetMap | Open Database License (ODbL). The legal code checked is ODbL 1.0. | Display the OpenStreetMap attribution notice and state that the data is available under the ODbL. A publicly used derivative database must be released under the ODbL, a later similar version or a compatible licence. A produced work, such as a map or chart, carries a notice that its content is available under the ODbL. | OpenStreetMap, Copyright and License: "OpenStreetMap licensing" and "How to credit OpenStreetMap", items 1–2. ODbL 1.0, §4.3(a) and §4.4(a). |
| Ember | CC BY 4.0 | Credit Ember when sharing or adapting. "Ember" is a trademark held at the UK and EU intellectual property offices, so we use the name to credit the source only. | Ember, Creative Commons, first statement. Ember, Data, site footer. |
| Global Solar Power Tracker (Global Energy Monitor) | CC BY 4.0 when shared or adapted. Records with a TransitionZero ID are CC BY-NC 4.0. | Credit with the recommended citation, "Global Solar Power Tracker, Global Energy Monitor, February 2026 release". CC BY-NC records are excluded. | Global Energy Monitor, tracker page, FAQ "How do I cite this data?", paragraphs 2–3 and the final note. |
| Eurostat | Reuse for commercial or non-commercial purposes with acknowledgement, subject to exceptions. Editorial content is CC BY 4.0. | Acknowledge the source with Eurostat's citation template. Data identified as belonging to other sources may not be reused commercially. Data for countries outside the EU, EFTA and the official EU acceding and candidate countries must be removed before commercial reuse. Third-party copyright material is not covered. | Eurostat, Copyright notice and free re-use of data: "General principles", "Exceptions", and the citation template. |
| PVGIS (Joint Research Centre, European Commission) | No PVGIS-specific licence was found. The nearest official notice is the Commission's legal notice: EU-owned content on commission.europa.eu is CC BY 4.0 unless a notice says otherwise, and logos and names are excluded. | Do not use EU or JRC logos as our brand. The API allows 30 calls per second per IP address and returns HTTP 429 above that. Browser-side (AJAX) calls to the API are not allowed. This site does not call PVGIS, from the browser or from a server. | European Commission, Legal notice, "Copyright notice". PVGIS API documentation: warning in the introduction, above §1.3; rate limit, §1.3. |
| NASA POWER | Free access. NASA public information may be distributed and copied with credit. | Cite the service name, the version and the access date. Include both the POWER Reference and the POWER Data Reference. | NASA POWER, homepage, "Access". NASA POWER Docs, Referencing guide. NASA website disclaimer, second paragraph. |
No OpenStreetMap-derived extract is in the current release. If one is published, it carries the OpenStreetMap attribution notice and the ODbL statement. A point or area record is published only if its licence is on the pipeline's publishable list, so records with unverified or non-commercial licences are excluded.
The observatory shows the first column as reported. The second column is labelled wherever it appears.
A band in a projection is the spread of model runs over stated input ranges. It describes model uncertainty. It is not a guarantee, and it is not a measured confidence interval for a real asset. Per-plant degradation is shown only where a published source reports a rate for that plant.
Values are the magnitude of performance loss, in %/year. Sources that print a negative number for loss are shown here without the sign. Medians and means are not interchangeable.
The PV Health Score uses 0.75 %/year as its default system reference. It is the overall median system performance loss rate for US commercial and utility systems (Jordan, Anderson, Perry & Muller, 2022; deg-jordan2022-plr-overall). It is a population median for systems, not a measured rate for any one system.
Split by climate, the same study gives medians of 0.48 %/year for cooler climates and 0.88 %/year for hotter ones (deg-jordan2022-plr-climate). The Health Score shows both bands beside the default.
The module reference stays visible. Jordan and Kurtz (2013) report a median of 0.5 %/year across 1,751 field-measured crystalline-silicon rates (deg-jk2013-median-si). Those rates are pooled from individual modules and whole systems. The figure is shown beside the system default. It is not the system default.
The Warranty Guard uses a single-system spread. It describes how far one system's measured loss could scatter around its own trend, because of measurement error. It is a setting on that page, and it is not a population standard deviation.
The population spread comes from the 2025 meta-analysis: a standard deviation of 0.99 %/year (deg-sm2025-spread-sd) around a median of 0.94 %/year (deg-sm2025-overall). The Warranty Guard's population prior is a lognormal fitted to that median and to the published spread.
The Health Score's system reference moves from 0.5 to 0.75 %/year. At the same measured loss, a score under the new reference is more lenient. The module reference stays labelled, and the Warranty Guard's spread is labelled as a measurement spread.
A trend is shown only when the record is long enough to separate it from measurement noise. A shorter record is labelled "insufficient window".
| Study | Population | Value (%/year) | Source and locator |
|---|---|---|---|
| Jordan & Kurtz, 2013 | All field-measured rates in the review, modules and systems, flat-plate terrestrial (1,920 rates) | Median 0.5 · mean 0.8 | Photovoltaic Degradation Rates: An Analytical Review, Prog. Photovolt. Res. Appl. 21(1):12–29. Sec. 3.1, Fig. 2(a). |
| Jordan & Kurtz, 2013 | Crystalline-Si technologies (1,751 rates) | Median 0.5 · mean 0.7 | Same paper, Fig. 2(b). |
| Jordan & Kurtz, 2013 | Thin-film technologies (169 rates) | Median 1.0 · mean 1.5 | Same paper, Fig. 2(c). |
| Jordan & Kurtz, 2013 | Module-level medians by technology, installed after 2000 | mono-Si 0.36 (n = 55) · multi-Si 0.64 (n = 36) · a-Si 0.87 (n = 31) · CIGS 0.96 (n = 10) · CdTe 0.40 (n = 6) | Same paper, Table 1, author manuscript p. 18. |
| Jordan & Kurtz, 2013 | Share of all reported rates | 78 % below 1; distribution skewed to high rates | Same paper, Conclusion, p. 16; Sec. 3.1, pp. 5–6. |
| Jordan, Kurtz, VanSant & Newmiller, 2016 | More than 11,000 rates from almost 200 studies in 40 countries; crystalline Si (x-Si) | Median 0.5–0.6 · mean 0.8–0.9 | Compendium of photovoltaic degradation rates, Prog. Photovolt. Res. Appl. 24(7):978–989. Abstract only; full text not retrieved. |
| Jordan, Kurtz, VanSant & Newmiller, 2016 | HIT and micro-crystalline Si | About 1 (qualitative; the abstract gives no median) | Same compendium, Abstract only. |
| Straub-Mück, Geyer-Klingeberg & Rathgeber, 2025 | 80 primary studies, about 70,000 modules, 27 countries, installed 1979–2017; all technologies | Median 0.94 · mean 1.10 · SD 0.99 | Determinants of the long-term degradation rate of photovoltaic modules: A meta-analysis, Renew. Sustain. Energy Rev. 216:115697 (CC BY 4.0). Abstract, p. 1; Sec. 1, p. 2; Fig. 1 inset, p. 2. |
| Straub-Mück et al., 2025 | HIT modules | Median 0.29 | Same paper, Sec. 3.2, p. 6; Fig. 2. |
| Straub-Mück et al., 2025 | CIS/CIGS and CdTe modules | Median 1.5–2 | Same paper, Sec. 3.2, p. 6. |
| Straub-Mück et al., 2025 | Crystalline-Si modules, predicted mean under best-practice assumptions, by climate zone and mounting | 0.43–1.41 (predicted mean) | Same paper, Sec. 3.2, p. 7; Table 3, p. 6. |
| Straub-Mück et al., 2025 | CdTe modules, predicted mean under the same assumptions | 1.56–2.54 (predicted mean) | Same paper, Sec. 3.2, p. 7; Table 3, p. 6. |
| Jordan, Anderson, Perry & Muller, 2022 | US commercial and utility systems, 7.2 GW, 1,700 sites, 19,000 inverters; system performance loss rate | Overall 0.75 · cooler climates (median) 0.48 · hotter climates (median) 0.88 | Photovoltaic fleet degradation insights, Prog. Photovolt. Res. Appl., doi:10.1002/pip.3566. Abstract. |
| Deline et al., 2025 | PV Fleet Performance Data Initiative: fleet median system performance loss rate, N = 4,915 inverters | 0.75 (median) | PV Fleet Performance Data Initiative Final Technical Report, NREL, OSTI 2522744. Subtask 3.1, report p. 17 of 34; Fig. 15. |
| Deline et al., 2024 | Expanded US fleet; median unchanged from 2022. Soiling-aware (CODS) analysis shown separately | About 0.75 · CODS 0.50 (median) | Availability and Performance Loss Factors for U.S. PV Fleet Systems, NREL/TP-5K00-88769, OSTI 2318550. Sec. 6, pp. 19–20; Sec. 7, p. 22. |
| Bolinger et al., 2022 | 631 US utility-scale plants, 31.0 GWdc, commissioned 2007–2018. Plant level, not module level | 1.2 (± 0.1), fleet average | Plant-level performance and degradation of 31 GWDC of utility-scale PV in the United States, LBNL, OSTI 1855368. Abstract. |
Plant rates are higher than module rates. The LBNL authors state that their plant-level figure is larger than module- and cell-level studies. For this reason, we show a plant rate only when a source publishes one for that plant.
Spread. The 2025 pooled rates have a population standard deviation of 0.99 %/year, a maximum of 7.92 %/year and a skew of 2.39 (Fig. 1 inset, p. 2). The single-system measurement spread on the Warranty Guard is a separate setting and is not this figure. Interquartile ranges are not given in the sources we read, so we do not quote them.
Measurement quality. The 2013 review notes that a high percentage of references rest on one or two measurements, usually against nameplate STC ratings, which can add significant error (Fig. 6–7 discussion, p. 10). The 2016 compendium reports sampling bias tied to study size and measurement accuracy.
Climate. Polar and alpine sites showed the highest rates in the 2013 review, attributed only tentatively to snow and wind loads (literature discussion, p. 14). In the 2025 meta-regression, hot-humid and desert climates carry significant positive coefficients against the moderate base group, meaning faster loss (Sec. 3.2, p. 6).
Nameplate power of the modules under standard test conditions, in kilowatts peak. It is the basis of every capacity figure on the observatory. Sandia's performance-metrics guide notes that IEC 61724 does not specify the rating basis for AC system yield, and defines it as the DC power rating at STC (Performance Ratio section, YfAC note).
The ratio of the energy a system delivers to the energy expected from its rated power and the available irradiation. IEC 61724-1:2021 (Edition 2.0, published 2021-07-21) covers PR in clause 14.3.1, according to its table of contents. We did not verify the clause text, so the formula below is taken from Sandia's guide. Daily AC form: PRAC = YfAC / Yr, where YfAC is daily AC energy divided by rated power, and Yr is plane-of-array insolation divided by the reference irradiance of 1,000 W/m², expressed in hours (Performance Ratio section).
Rating conditions of 1,000 W/m² irradiance, 25 °C cell temperature and the AM1.5 spectrum, as Sandia describes them (Performance Ratio section). IEC 60904-3:2019 (Edition 4.0, published 2019-02-15), titled "Measurement principles for terrestrial photovoltaic (PV) solar devices with reference spectral irradiance data", is the standard whose title covers these measurement principles. We did not verify its STC wording against the standard text.
The average relative loss of performance per year, as each study defines it. We show loss as a positive number. Studies differ in whether they report a mean or a median, and in whether they measure modules, systems or plants. The figures are not interchangeable.
These statements define the limits of every page on LIFETIMEBESS Solar.
Send us the number and the page it appears on. We reply with its source, its locator and its licence.
Ask about the method