The National Climate Centre expects its sea temperature index to peak around November at 3.2 to 3.5 degrees, above the 2.8 it assigns to 2014-2016. That is a forecast, not a record already set. Four numbers near 3 degrees are now circulating on four different averaging windows, and the wire copy dropped the one detail that tells them apart.
On 29 September the China Meteorological Administration told a news conference that the National Climate Centre expects the equatorial central and eastern Pacific sea temperature index to peak around November at 3.2 to 3.5 degrees, markedly above the historical maximum, forming the strongest super El Niño since monitoring records began.
The verb in the CMA headline is future tense. No three-month event record has been shown to fall yet. What makes this worth a Media Watch is not that the agency overstated its case, because it did not, but that four numbers close to 3 degrees are now in circulation, measured over three different lengths of time, and the wire copy that carried the forecast around the world dropped the detail that tells them apart while introducing an arithmetic error of its own.
From the CMA newsroom item of 29 September, which is the fullest version. The equatorial central and eastern Pacific entered an El Niño state in May 2026. Since then:
A pentad is five days. A dekad is ten. That qualifier is in the Chinese original and it is doing real work, because the record those numbers are being read against is not a five-day value.
The forecast follows: the index is expected to keep fluctuating upward, peaking around November at 3.2 to 3.5 degrees. The CMA newsroom version says only that this exceeds the historical maximum. The parallel China Weather Net item supplies the figure, 2.8°C, the peak of the 2014-2016 event, which is the value we published in August from Appendix A of China's national standard GB/T 33666.
Here is every figure in that neighbourhood currently in circulation, with the only property that matters attached to it.
| Figure | Source | Averaging window | Observed or forecast |
|---|---|---|---|
| 3.02°C | China NCC, fifth pentad of September | 5 days | Observed |
| 3.11°C | Carbon Brief analysis, 21 September | 1 day | Observed |
| 3.2 to 3.5°C | China NCC, forecast November peak | 3 months, inferred (see below) | Forecast |
| 2.8°C | China NCC, standing record from 2015 | 3 months | Observed |
Shorter windows can reach higher peaks than longer ones, because a three-month mean averages a spike against the weeks either side of it. That is a tendency, not a rule: a three-month mean can exceed the pentads that came before it when a series is still climbing steeply, which is exactly what China is forecasting here. The point is narrower and firmer than a rule about magnitudes. A five-day value and a three-month record are different quantities, so one cannot beat the other, whichever is larger.
The two observed 3-point-something figures are both real and both correctly reported by their sources. Carbon Brief says plainly that it is working with a daily anomaly and that it follows NOAA's ONI convention rather than publishing the ONI itself. The CMA names the pentad and dekad scales explicitly. The problem appears when the four rows above get stacked in a reader's head without the third column.
The 29 September briefing does not say. It uses one term, the monitoring-region sea temperature index, for a five-day value, a monthly value, a three-month seasonal mean and the forecast peak, all within two paragraphs.
The best available answer is an inference, and we are labelling it as one. In its August announcement the same agency attached its super-strong threshold to the three-month running mean, which is also the quantity GB/T 33666 uses to identify and rank events. A peak forecast meant to be compared with a three-month record is most naturally a three-month figure. But the September briefing does not say so, and the difference between an inference and a statement matters more here than usual, because the whole comparison rests on it.
It is also a plausible forecast on that basis. September pentads are already at 3.0, and a three-month mean centred on the late-autumn peak can sit above the pentads that precede it if the warming continues. The trouble is not the number. The trouble is that nothing in the 29 September briefing tells a reader which of four windows it is on, and the record it is being measured against only makes sense on one of them.
China's index is NINO3.4, the same box NOAA uses, on a conventional anomaly against a changing 30-year climatology. NOAA's legacy ONI is built the same way. NOAA's operational index since February 2026 is RONI, which subtracts the tropical-mean anomaly and therefore runs lower. Those three quantities for the same three months of this year:
| Index, June to August 2026 | Construction | Value | Record to beat |
|---|---|---|---|
| China NCC monitoring index | Conventional, changing 30-yr climatology | 2.09 | 2.8, Dec 2015 |
| NOAA CPC ONI | Conventional, changing 30-yr climatology | 1.80 | 2.59, 2015-16 |
| NOAA CPC RONI | Relative, tropical mean removed | 1.36 | 2.40, 1982-83 |
Same ocean, same box, same three months, and a spread of 0.73 degrees between the highest and lowest reading. None of the three is wrong. The nearest counterpart to China's number is NOAA's ONI, not RONI, and the two differ by 0.29 degrees this season. Setting 3.5 beside RONI's 1.36 and concluding that the agencies disagree by two degrees would be a pure artefact of comparing a conventional index with a relative one.
One detail worth carrying forward: 2.09 is both China's July monthly value, reported in August, and its June to August seasonal mean, reported this week. Two different quantities, identical to two decimal places, in the same agency's statements six weeks apart. That is this whole problem in miniature.
Reuters carried the story on 29 September and kept the tense intact, reporting a peak that is projected to occur. It named no index, no averaging window and no baseline, calling the quantity only a sea surface temperature index, which leaves a reader no way to check it against any other published number.
It also converted the range into Fahrenheit, as 37.76 degrees Fahrenheit to 38.3 degrees F. That is the absolute-temperature formula applied to an anomaly: multiply by nine fifths and add 32. The 32 belongs only when converting a reading, not a departure from a mean. As a temperature difference, 3.2 to 3.5°C is 5.76 to 6.3°F. The bracketed figures are neither the anomaly nor the sea surface temperature, which in that part of the Pacific runs nearer 29°C. The Celsius numbers in the story are correct and the reporting around them is accurate; the conversion is not, and it is the kind of error that survives every downstream republication because nobody re-derives a number in brackets.
The syndication chain carried that version onward, through Business Standard, TimesLIVE, MarketScreener, Yahoo and CGTN, which is how a bracketed conversion becomes the version most readers see.
Carbon Brief's piece of 21 September, updated on the 23rd, is a separate line of reporting that predates the Chinese briefing by more than a week. Its analysis is careful and its method is stated in the text: a daily anomaly of 3.11°C against a previous daily record of 3.08°C. The word daily is not in the headline. Put that headline beside the Chinese forecast in a feed and a reader reasonably concludes a record has already fallen.
The CMA deserves credit on both counts where it earned it. It named the pentad and dekad scales rather than letting a 3-degree number float free, and it put the record claim in the future tense. What it did not do is state the averaging window for the forecast peak, which is the one piece a reader needs to use the number at all. The wires then dropped the index entirely and, in the Reuters case, added an arithmetic error. Nobody here invented anything; the information was thinned at each step.
NOAA's own weekly file for the week centred 23 September 2026 reads, on the relative index, Niño 1+2 +3.9, Niño 3 +3.1, Niño 3.4 +2.2 and Niño 4 +0.1. The conventional file for the same week reads +4.7, +3.9, +3.1 and +1.1. Niño 3.4 rose 0.1 on the relative index from the week before, its second consecutive weekly rise.
Both of those are one-week values and neither is the ONI. The seasonal indices that rank events are published with a lag: CPC's ASO figures are due in early October, and China's September monthly value will follow its usual schedule. Those are the numbers that will eventually settle whether a record falls, and none of them is available yet.
China's forecast is a legitimate statement from a national climate agency about its own index, made in the future tense, and it may well verify. It cannot be read against NOAA's operational number without a translation step, and it cannot be read against a daily or a five-day value at all. Anyone quoting 3.5 should be able to say, in one clause, which index and over how many days. If that clause is missing, the number is not yet information.