The sun is constant. The sky is not.
The solar output reaching the top of the atmosphere varies by about a tenth of a per cent over a solar cycle: for planning purposes, the sun is a constant. What varies is the atmosphere between it and the panel, chiefly aerosols and the clouds they help form. The scientific record calls the resulting swings global dimming and brightening: measured surface sunlight fell several per cent across the industrialised world from the 1950s to the 1980s as pollution rose, then recovered as clean-air laws took hold, a history documented in surface-radiation networks and reviewed by Wild and colleagues. The last two decades are the part of that history a solar buyer owns, and it is now measurable market by market from space.
Two instruments, one question
The Solar Analytica yield baseline carries two independent readings of surface irradiance for every market: NASA's MERRA-2 reanalysis, a physics-model reconstruction spanning 1994 to 2024, and NASA's CERES satellite radiometers, a direct measurement record from 2000. Comparing each instrument's early period with its late period asks the same question two different ways. MERRA-2 first:
And the same markets through CERES:
| Market | GHI 1994-2003 kWh/m²/day | GHI 2014-2024 | MERRA-2 change | CERES change | Agree |
|---|---|---|---|---|---|
| Berlin, Germany | 2.72 | 2.91 | +6.8% | +4.0% | yes |
| London, United Kingdom | 2.89 | 3.00 | +3.6% | +3.8% | yes |
| Santiago, Chile | 5.34 | 5.53 | +3.5% | +6.0% | yes |
| Athens (Attica), Greece | 4.74 | 4.89 | +3.3% | +2.4% | yes |
| Perth, Australia | 5.56 | 5.61 | +1.0% | +0.7% | yes |
| Riyadh, Saudi Arabia | 6.21 | 6.11 | -1.5% | -1.5% | yes |
| Shanghai, China | 3.86 | 3.76 | -2.6% | +0.5% | no |
| Jakarta, Indonesia | 4.89 | 4.75 | -2.9% | -3.8% | yes |
| Reykjavik, Iceland | 2.33 | 2.24 | -4.1% | -4.2% | yes |
| Jodhpur (Rajasthan), India | 5.82 | 5.47 | -6.1% | -4.0% | yes |
Solar Analytica yield baseline: MERRA-2 window means (1994-2003 vs 2014-2024) and CERES period means (2000-2009 vs 2020-2024), per market. Full 40-market table in the public dataset. Set averages: +0.4% (MERRA-2), +0.5% (CERES); 26 markets brightening, 14 dimming; direction agreement in 29 of 40.
The pattern maps onto air-quality history. The strongest brightening sits over Europe (Berlin +6.8%, London +3.6%, Paris +3.5%), where aerosol emissions have fallen sharply since the 1990s, with Santiago alongside them. The strongest dimming is over South Asia (Jodhpur -6.1%, the same market whose local temperature trend ran negative in part 01: hazier skies are both dimmer and cooler). Reykjavik and Jakarta dimmed on both instruments. Where the instruments disagree on direction (Shanghai most notably, with Antofagasta and Lhasa marginal), this report treats the trend as unresolved rather than picking the convenient reading.
A per cent of sky, in output
Irradiance converts to output almost one for one: a panel in a sky delivering 5 per cent more light produces close to 5 per cent more energy, before any other term in the yield model moves. That makes these trends directly comparable with part 01's warming penalty, and the comparison is lopsided. Berlin's observed two-decade brightening (+6.8%) is roughly eight times the heat loss that warming is modelled to add there by year 25 (-0.83%), the largest such penalty in the set. In Jodhpur the sign flips: the dimming (-6.1 per cent over two decades) is forty times its modelled warming drift (-0.15 per cent by year 25, part 01), and it, not temperature, is the climate variable a Rajasthan yield model should worry about. Perth's +1.0 per cent lines up with the south-west WA drying trend documented in the series' rainfall work: fewer clouds mean drier panels and brighter skies, the same drying shift appearing in both records. In most of the set the two-decade sky trend outweighs the modelled warming drift several times over, which makes data vintage the largest correctable error in this series' yield model.
40 times
Jodhpur's two-decade sky dimming against its modelled warming penalty: the climate variable yield models ignore, against the one they include.
The practical question follows: which years is the yield estimate built on? An estimate resting on a 1990s-era typical-year dataset understates Berlin by several per cent and overstates Jodhpur by more; a recent basis (2014-2024) prices the sky the buyer is purchasing. In brightening markets an older estimate runs conservative; in dimming markets it runs optimistic, which does more damage. And where the instruments disagree on direction, Shanghai most notably, any single-source trend claim deserves suspicion.
Each instrument is only compared with itself
The figures compare period means within each instrument separately: MERRA-2's 1994-2003 window against its 2014-2024 window, and CERES's 2000-2009 against its 2020-2024. The CERES late window is five years against ten, so it is noisier. A two-decade change is an observation; aerosol policy, which drives much of it, can move in either direction, and Asian clean-air programs could flip today's dimming markets the way Europe's did. Changes of a few points sit within the reach of cloudy-decade chance in some markets, which is why two independent instruments are shown and sign-disagreements are flagged rather than resolved. And these are horizontal-irradiance trends: tilt, temperature, soiling and smoke (parts 01, 05 and 06) sit on top. Every input series is public, and the per-market values can be recomputed from the named NASA products.
A July 2026 process-modelling study now adds evidence that tropical Pacific low-cloud loss amplifies warming, assessed in Pacific Low-Cloud Feedback Under Warming and High CO2. Its simulated regional response is context for why cloud cover may change; it is not substituted for either instrument in this observed surface-irradiance record.