The panel cools by convection, and wind runs the fan
A solar module in sunlight is a hot plate. It absorbs far more energy than it converts, and the surplus leaves as heat by three routes: radiation to the sky, conduction through the mount, and convection into the passing air. Convection is the big adjustable term, and wind is what drives it. Still air lets a boundary layer of heat sit against the glass; moving air scrubs it away, and the faster the wind, the cooler the cell and the smaller the temperature penalty that part 01 measured. The relationship is captured by the industry-standard Faiman model, in which module overheating above air temperature falls steeply as wind rises: from about 32 degrees in still conditions toward the low teens in a steady breeze.
The wind resource itself, measured the same way across all 40 markets, varies more than threefold:
| Market | Air temp °C | Wind at 10 m (m/s) | Module overheat °C | Output recovered vs still |
|---|---|---|---|---|
| Cape Town, South Africa | 17.0 | 6.32 | 11.7 | 6.9% |
| Perth, Australia | 19.2 | 6.29 | 11.8 | 6.9% |
| New York, USA | 12.2 | 5.31 | 13.0 | 6.5% |
| London, United Kingdom | 10.4 | 5.01 | 13.5 | 6.3% |
| Cairo, Egypt | 21.8 | 4.04 | 15.2 | 5.7% |
| Riyadh, Saudi Arabia | 26.0 | 3.91 | 15.5 | 5.6% |
| Jodhpur (Rajasthan), India | 26.8 | 3.56 | 16.2 | 5.4% |
| Jakarta, Indonesia | 27.3 | 3.11 | 17.3 | 5.0% |
| Los Angeles, USA | 17.3 | 2.83 | 18.0 | 4.8% |
| Bogota, Colombia | 18.7 | 1.81 | 21.4 | 3.6% |
Selected from the 40-market wind pillar; full table in the public dataset. Wind is the NASA POWER 20-year mean at 10 m. Module overheat is the Faiman model at 800 W/m² reference irradiance; output recovered is that market's cooling relative to a still-air (zero-wind) baseline, via the 0.34%/°C PERC coefficient. Set mean wind 3.81 m/s; mean recovery 5.5%.
Wind does not follow the heat
The temptation is to assume hot places are calm and cool places are breezy, so that climate roughly cancels out; the record says otherwise. Across the 40 markets the correlation between mean wind and mean air temperature is essentially zero (r = -0.09), so wind is statistically independent of the heat map. The consequence is visible in two markets at almost the same air temperature. Bogota sits at 18.7 degrees of air and barely moving 1.8 m/s wind, so its modules overheat by more than 21 degrees; Cape Town sits at 17.0 degrees with a windy 6.3 m/s, so its modules overheat by under 12. The air temperatures are nearly identical, the panels run nine degrees apart, and wind is the entire difference.
9°C
The module-temperature gap between Bogota and Cape Town, two markets with near-identical air temperatures, decided entirely by wind.
Because wind is independent of heat, the hottest markets are not compensated with extra breeze, and some of the hottest are among the stillest. The humid tropics pair heat with stillness: Jakarta, Singapore and Lagos are all above 26 degrees of air yet sit below the wind average, so they take the full heat penalty with little convective relief. The hot deserts fare somewhat better on average but unevenly, with windy coastal Perth (6.3 m/s) cooled hard while inland Bamako and Jodhpur, the two hottest-running markets in the whole baseline, get only middling wind. Wind cooling helps most where a sea breeze happens to coincide with the sun, and that coincidence is a matter of geography, not thermodynamics.
Every market gains, but not equally
Translated into output, every market gains from its wind, but the gains are unequal:
Read against a dead-still baseline, wind recovers between 3.6 per cent of output in calm Bogota and 6.9 per cent in windy Cape Town, with a 40-market mean near 5.5 per cent. The baseline is an illustrative bound, since no location has zero wind; the observable quantity is the spread, the roughly three percentage points separating the windiest markets from the stillest at otherwise similar conditions. The practical consequence: a market-level yield estimate built on irradiance and air temperature alone overrates the still markets and underrates the windy ones by up to about three percentage points, the spread this table measures. It is the reason a coastal site can outproduce an inland one at the same latitude and irradiance.
Leave the ventilation gap behind roof-mounted panels open: a module clamped tight to hot tiles loses much of the convective cooling this report measures, while a free-standing, ventilated rack keeps it. A coastal or exposed site runs cooler, and therefore yields slightly better, than an inland site with the same sun and air temperature; that difference is wind.
In the still, humid tropics the module's temperature coefficient matters more, which strengthens the case part 01 made for a low-coefficient HJT panel where the air is hot and still. One caution runs the other way: the sea breeze that cools also carries salt, and on exposed coasts it drives the cleaning and corrosion questions of part 05.
Two measurement heights, one climatology
Wind speed is the NASA POWER 20-year climatology (2001 to 2020) at 10 metres and at 2 metres, the same MERRA-2-derived source family as the irradiance profiles, pulled for each market's coordinates. The cooling effect uses the Faiman module-temperature model (IEC 61853-2) with standard free-standing coefficients (U0 = 25, U1 = 6.84 W/m²K) at a reference plane-of-array irradiance of 800 W/m², and the output figure applies a PERC temperature coefficient of 0.34 per cent per degree Celsius; the tabulated 10-metre wind series feeds the model, with the 2-metre series carried alongside it in the dataset. The clawback is quoted against a still-air baseline, an illustrative bound rather than a real site, so the meaningful comparison is between markets rather than against zero. The Faiman coefficients assume a well-ventilated, free-standing module; a roof-hugging or building-integrated array cools far less, which matters most for the roof-mount guidance above. And these are annual means: wind has a daily and seasonal rhythm, and the cooling matters most when it blows during the hot, high-irradiance hours. Every input is public, and the per-market inputs and model settings sit in the public dataset.