Picture a map of China with sparse green patches scattered across a yellow desert. Those patches are where the GloSea5 climate model can genuinely look three months ahead, while the desert is everything else, where its forecasts are no better than a coin toss. Researchers examined twenty years of hindcasts from this UK system to identify where it can be trusted for energy planning.
The second oasis is summer temperature in Beijing. Even after removing the long-term warming trend, the correlation remains at 0.48, meaning the model captures a genuinely predictable signal rather than simply repeating "warmer than the historical average." Winter solar radiation in eastern China is also predicted reasonably well, with a correlation of 0.62, likely due to its link with cloud cover, which is influenced by El Niño and Indian Ocean temperatures.
But beyond these oases lies desert. Summer winds and winter temperatures show no significant skill almost everywhere. In the same South China Sea where winter correlation is 0.78, summer drops to 0.13, and the forecast loses all strength. It is like a compass that works perfectly in clear weather but goes haywire in fog.
For China's energy sector, this is not an abstract issue. The country is simultaneously expanding solar and wind capacity while facing growing demand for cooling in summer. Knowing in advance whether winter will be windy and summer hot means better planning for fuel purchases, maintenance, and grid load. For now, only the oases can be trusted, but they are already sufficient for pilot services in the South China Sea and Beijing.
The predictability map produced by this study is not static. Increasing the ensemble from 24 to 42 members and statistical calibration based on large-scale atmospheric processes could expand the green patches. The desert of randomness will retreat if we learn to understand its landscape better.
The correlation of 0.78 for winter winds in the South China Sea is comparable to the skill of El Niño forecasts, which is considered the gold standard of seasonal prediction. Yet in the same region in summer, the correlation drops to 0.13, meaning the forecast loses all skill.