Dust falls everywhere. Rain does not.
Soiling is the layer of mineral dust, traffic particulate, pollen, salt and bird droppings that accumulates on module glass and blocks light before it reaches the cell. Unlike the heat penalty of part 01, it is not a property of the technology: it is a race between deposition and cleaning, and in most of the world the cleaner is rain. The literature converges on a threshold of roughly 1 to 5 millimetres in a single event to wash a lightly soiled array (the modal modelling value is Kimber's 3 mm; the IEA PVPS basic model resets at 1 mm; about 0.5 mm was enough on lightly soiled Atacama arrays), with two honest complications: light rain on heavy dust can bake on a muddy residue rather than clean, and Moroccan red rain deposits dust instead of removing it.
Priced across the 40-market baseline, the result is a spread no single global assumption can cover:
| Market | Annual loss, no cleaning | Dry-period rate %/day | Dominant soiling | Confidence |
|---|---|---|---|---|
| Masdar City / Abu Dhabi, UAE | 20%+ | 0.6 | Desert dust, humidity cementation | Medium |
| Riyadh, Saudi Arabia | 20%+ | 0.5 | Desert dust | Medium |
| Antofagasta (Atacama), Chile | 18% | 0.31 | Desert dust | Medium |
| Jakarta, Indonesia | 8% | 0.06 | Traffic-industrial PM | Medium |
| Santiago, Chile | 7% | 0.6 | Urban pollution PM | High |
| Dubai, UAE | 6% | 0.3 | Desert dust | Medium |
| Perth, Australia | 5% | n.a. | Dust, coastal salt, bird | High |
| Phoenix, Arizona, USA | 3% | 0.25 | Dust, monsoon haboobs | High |
| Shanghai, China | 1.7% | n.a. | Traffic-industrial PM | High |
| Berlin / London / Paris / New York | 1.5% | <0.1 | Urban PM, pollen | High |
Selected from the 40-market soiling baseline. Abu Dhabi and Riyadh carry the NREL Middle East lower bound (20%+, post-storm peaks to 80%); Atacama is the Iquique transect point (range 3-39%); Perth is the Murdoch University field midpoint (4-6%); the temperate 1.5% rows are the NREL regular-rainfall band midpoint. Full 40-row table, with sources, windows and the reason behind every blank, in the public dataset.
Seventeen of the forty markets carry no annual figure at all, and that is deliberate. For markets like Cairo, Lagos and Nairobi the strongest available studies report optical transmittance or seasonal losses that cannot honestly be annualised; for Cape Town, Kabul, Lhasa and Reykjavik no credible field study exists. Sub-Saharan Africa and South and Central Asia are blank across the board despite carrying measured dry-period rates near 0.44% per day, among the highest in the set: the largest genuine data gap in this series.
The rate and the level are different numbers
The accumulation rate during rain-free periods spans a factor of fourteen across the measured markets, and it is the number that decides whether a dry season hurts:
Madrid shows why the rate and the annual level must not be conflated. The CIEMAT rooftop campaign fitted a modest 0.16% per day, but after 52 rain-free summer days the loss level reached about 6%, and a 4.5 mm event failed to fully clean low-tilt modules. A low rate with a long dry spell ends in the same dirty glass as a high rate with a short one, which is why the dataset records rate, dry spell and threshold separately.
Where cleaning pays, and where rain does the job
The economics fall into three bands. In the temperate 1 to 2% band (most of Europe, the eastern United States), rain does the job and paid cleaning generally returns less than it costs; a Belgian field study concluded annual cleaning would have negligible effect. In the marginal band around 5% (southern California, Australia's coastal cities), the standard practice is a derate and roughly one clean a year: Perth's measured 4 to 6% sits exactly here, with a series cross-link worth naming, because the structural south-west WA rainfall decline lengthens the summer accumulation window the derate assumes. In the desert band, cleaning is not optional: UAE utilities wash plants 40 to 45 times a year to hold soiling under 3%, Atacama operators clean on a 45-day to 4-month cadence, a Moroccan 44 MWp plant cut losses from 5.8% to 1.9% with an optimised schedule worth about 5.23 US dollars per kilowatt per year, and in Riyadh anti-soiling coatings return a measured 3.2% annual surplus.
The buyer action follows the band. Check which band your market sits in before paying for a cleaning contract; in the temperate band the money is better spent on anything else. In dry-summer markets, tilt matters: rain cleans adequately above about 5 degrees and improves with tilt, so near-flat arrays soil harder and stay dirty longer. And in the desert band, price the cleaning schedule into the quote from day one, because at a 20% uncleaned loss it is the single largest controllable term in the yield model.
Method and limits
The figures come from the series' 40-market soiling baseline: annual energy loss with no manual cleaning, dry-period accumulation rates, rain-cleaning thresholds and season metadata, each row named to its source with its observation window and confidence grade. Ranges are recorded at midpoints and bounds are flagged as bounds; sales-blog figures were located and excluded. Three boundaries matter. Deposition is strictly separated from atmospheric attenuation, so haze and smoke dimming belong to part 06 and are not double-counted here. Snow markets (Oslo, Stockholm, Helsinki, Ulaanbaatar) record snow as the dominant mechanism, a different physical process with a different remedy. And blanks are honest: nothing was interpolated, and no neighbour's value was borrowed.