In the early hours of November 4, the built-in Weather app on iPhones in many Chinese cities showed the same impossible low for the following Friday, November 13: minimums between −28°C and −32°C in Beijing, Shanghai, Qingdao and across Henan Province [1]. That is a one-day swing of nearly 50°C, in cities hundreds of kilometers apart [1].
The fault was not in anyone's phone, and the cold was not weather. Apple's own customer service, after an initial check, named the cause as a "system port" problem (系统端口), explicitly not a hardware fault, and said a technical team had already completed an emergency fix [2]. China's official forecaster had the simpler answer: the China Meteorological Administration never issued any extreme-low-temperature forecast for that period, and its medium-range outlook for November 13–16 points the other way, with most of the country warmer than normal [3][4].
If you plan more than a few days out on a phone-weather number, that morning is the argument for checking the official forecast first.
One date, many cities: the shape of the fault
The anomaly had a precise shape, and the shape tells you more than the number does. Screenshots began circulating around 1 a.m. Beijing time. By a Jimu News reporter's own check of the app, Beijing, Jinan, Wuhan and Shanghai all showed November 13 minimums below −30°C, every date before the 13th read normal, and the temperature-trend graph bent into a sudden plunge on the night of the 13th [5]. One forecast day, wrong everywhere at once, with the days around it untouched. Around 7 a.m. the official @中国天气 (China Weather) account replied that no such temperature swing exists, and by 8 a.m. the reporter's app was showing normal values again [5]. The topic was on Weibo's hot search before mid-morning [1].
Two offices answered, and the forecaster's answer convicts
Apple's first response that morning was hedged: the customer-service line said it could not rule out a display or data anomaly, that it was not a hardware problem, and that it was worth watching for now [4]. By late morning the hedge was gone. Speaking to Zhongxin Jingwei (中新经纬), Apple's customer service had moved to a named cause and a completed fix, and added what the earlier answer lacked: an apology for the large number of user reports, and a denial of location drift. The denial came with its own logic: if the phone's location had jumped to some frozen place, the adjacent days would have gone wrong with it, not one Friday in isolation [2]. All of that is Apple describing its own investigation.
The meteorological side did not hedge. CMA staff, speaking to The Beijing News (新京报), said a real cold event of that scale would have been forecast and announced in advance, and no such extreme-low-temperature forecast had ever been issued [3]. The China Weather account of the CMA Public Meteorological Service Center was blunter: "Minimum −32°C? Move along, move along, there is no such temperature swing at all, it now displays normal. Don't worry: even with El Niño now the strongest on record, it would not be that cold now" [5]. The numbers on record agree with the debunk: the Central Meteorological Observatory's medium-range outlook for November 13–16, issued November 3 and quoted by Modern Express (现代快报), has most of China above normal, by 2–4°C across a broad band from the Northeast through North China and Inner Mongolia to the Huai and Yangtze regions. The observatory's bulletin series kept publishing on schedule through the incident: the November 4 issue, signed off at 10 a.m., covers November 4–13 [4][6].
Why an app can be wrong where the bureau is right
Two facts explain how a gap this wide opens between an app and its national forecaster. The first is where the number comes from. As the Shanghai Meteorological Bureau has explained, the temperatures a weather service publishes come from standardized observation stations measuring to a uniform technical standard, while some phone weather apps do not use actual observations: they use estimates derived from other data and methods [4]. The second is the horizon. The Central Meteorological Observatory, as quoted by Modern Express, puts three-day forecast accuracy above 80%, and at seven days it doubts accuracy even reaches 50% [4]. The corrupted value sat nine days out, in the band where nobody, the app included, can promise much.
How the bad number entered Apple's pipeline is not officially known. The closest thing to a mechanism on record is the meteorology blogger @风云梦远's reading, reported by Jimu News: the anomaly was confined to roughly 20:00 Beijing time on November 13, corresponding to 12:00 UTC, a key time node for numerical weather models, so the likely cause is a severely wrong forecast value at that model node, interpolated by the app into a plunge and rebound [5]. That is an expert's inference, not an official finding.
Twice in five months
It is also not the app's first extreme-value error this year. In June, Apple Weather showed nearly 50°C for Lin'an in Hangzhou; the local meteorological authority checked and concluded it was the app's own system data error, with no extreme-heat warning in effect [2]. Two confirmed errors in five months, with opposite signs: one absurd high in a single city, one absurd low across many cities at once. Two incidents with different shapes are a data point, not a pattern; what the record supports is a repeat problem with extreme values, nothing stronger. Apple's public response this time was an apology and a promise to keep improving the data service [2].
The verdict, and the three-step check
So the answer to the question iPhone owners across China were asking that morning comes in three parts. The phones were fine: Apple's own customer service explicitly ruled out a hardware fault [2]. The sky was fine: the only office entitled to forecast such cold says it never did [3], and the observatory's outlook for those same days points the other way [4]. The wrong numbers were Apple's to fix: a "system port" problem, in the company's words, not a hardware fault, with the repair completed by its technical team [2]. No impact on other Apple services has been reported.
The general lesson is checkable in three steps the next time an app shows you impossible weather. First, compare against the official forecaster's own forecast: where the bureau and the app disagree at range, the bureau wins, as it did here [3][4]. Second, look at the horizon. Inside three days, official forecast accuracy runs above 80%; at seven days, the observatory itself doubts it even reaches 50%; the bad number sat nine days out [4]. Third, look at the shape. One date wrong in many cities at once is the signature of a data fault, not of weather, and it is the same logic Apple's own service used to rule out location drift [2].
The −32°C was never a forecast. What November 4 actually demonstrated is an asymmetry of trust: the app's number broke and was fixed within a morning [2][5], while the official forecaster's word for the same days, most of the country warmer than normal, stood uncorrected [4]. When an app and the bureau disagree at range, that morning says which one to wait on.