A panoply of trifling dissatisfactions

The old version of this post – A multitude of possibly unsatisfying answers – finally broke something in WordPress. It stopped saving the footnotes, thus causing conditions under which I cannot work. Nevertheless, the congregation of unsatisfying answers continues to gather.

2024.01.23: A nice thread by Paul Roundy fleshes out the hypothesis that it was all ENSO. The protracted La Niña had associated mid-latitude warming that persisted through a rapid transition to El Niño. This thesis was proffered as a counter to the accelerating-warming argument advanced by Rahmstorf and (at greater and more informative length) Tamino. Roundy’s criticism is (I think) that the simple regression technique doesn’t account for the whole of ENSO-related global temperature variability. Even in the original paper, it didn’t seem to. In addition, the original paper assumed a linear trend, which they are now arguing is no longer the case. How to de-ENSO a time series has always been somewhat contested. My favourite entry in the genre is Compo and Sardeshmukh. While others pettifog over whether to use SOI or MEI and a lag of six months or only five, Compo and Sardeshmukh ask what if “ENSO is best viewed not as a number but as an evolving dynamical process for this purpose?” Good question: wild answer, and one badly in need of an update.

Tim Osborn has a quick twitter thread which looks at 2023 in a bit more detail. There’s an interesting comparison of the first and second halves of the year.

RealClimate have issued their annual update of the ongoing model-obs comparison. 2023 isn’t really that unusual when you look at it this way, which harks back to Philip Brohan’s question early on: how predicted this?

2024.02.02: A brief communication has, as they say, arisen in climate and atmospheric science: “The jump in global temperatures in September 2023 is extremely unlikely due to internal climate variability alone“. Submitted on 17 October, it has only now seen the light of day. They get to their conclusion/title by analysing a large set of model runs and looking at the margin by which records are broken in Septembers. They find that the observed deviation of 0.5°C is above the 99.94th percentile, 99.95th percentile or completely outside the distribution depending on which of three collections of model runs they look at. Thus they argue that the deviation is unlikely to be due to variability alone, which conclusion is aided and abetted by the fact that models generally have a higher standard deviation than has been observed. They then do some back of the envelope calculations to quantify the effects of volcanic eruptions (Hunga-tonga and Raikoke) and the reduction in sulhpate aerosols from global shipping. Their chosen metric here is the difference between September 2020 and September 2023, which is oddly specific, but they find that the combined effects of volcanoes and shipping fuel might amount to 0.1°C. The probability of a natural fluctuation which is 0.1°C smaller is much larger. The oddly specific metric here is relevant because counter-arguments to the volcano-and-shipping hypothesis have pointed out that the magnitude is likely small (as here) but also that the “shape” is wrong. Focusing on 2020 vs 2023 ignores the “shape” question, but as it’s a small effect, that may be beside the point.

2024.02.08: Copernicus have put out their January numbers. Based on these, the BBC declared that this is the first 12 month period to top 1.5C, conveniently forgetting that Berkeley Earth said the same thing last month based on a different 12 months of data. Also, the explanation for the heat has simplified. It’s climate change and El Niño now.

2024.02.27: Hobbs et al. have published a sort of long-form version of Purich and Doddridge, 2023. They note that, statistically speaking, that sure is a big anomaly in Antarctic sea ice. They look at things like the variance and autocorrelation of the series, as well as a linear model that includes various atmospheric indices and lagged sea ice anomalies. With studies like this, I find it very hard to assess what each new step in the analysis tells me that is new. We see that sea ice dropped rapidly around 2016 having risen gently from around 2007. So, of course the variance has increased. The autocorrelation has increased too, but to what extent it has to based on an increase in variance is hard to say – large anomalies have to accumulate (maybe a different time series model would be better than a simple AR?). Their linear model is rubbish (sorry, less skilful) until there is some variance to explain and then autocorrelation explains it. To what extent these are new pieces of information and to what extent they are different ways of presenting the same piece of information (which was already out there in Purich and Doddridge) is not something I can work out. The longer term persistence is interesting, but also why we noticed it in the first place. They suggest that the increase in variance and autocorrelation are harbingers of a tipping point (al la Lenton) but also that this already constitutes a regime shift. Absent a physical mechanism this is all rather floaty and suggestive, which is why the discussion is so interesting – it goes beyond the statistical manipulation and proposes physical mechanisms for some of their statistical regularities.

2024.03.08: Forbes has an article headlined “Hot Streak: Why February 2024 Was The 9th Consecutive Hottest Month On Record” which sadly neglects to say why February 2024 was the 9th consecutive hottest month on record. Greenhouse gases and El Nino are mentioned, but only  Richard Allan comes close to answering the question. Unfortunately, he skimps on the specifics: “The widespread substantial warmth of the sea surface is consistent with a combination of ocean fluctuations on top of a continued powerful heating effect from rising greenhouse gas concentrations and other smaller factors.“

2024.03.17: The New Yorker has spoken. Why, they ask, is the sea so hot? Bu they admit that they haven’t really been paying attention. “In early 2023,” they start, “climate scientists—and anyone else paying attention to the data—started to notice something strange.” They do bring some new news though – 2024 is hot too. Hotter, in fact, than 2023. Gavin Schmidt is ready with the appropriate word. Last year Zeke set the tone¹ with gobsmackingly bananas, this year Gavin gives us “disquieting“. The last time we knew what was going on, he says, was March last year. All the usual suspects are lined up – Hunga Tonga–Hunga Ha‘apai, ship fuel, solar variations – but Brian McNoldy has a novel explanation: “it’s like the whole climate just fast-forwarded fifty or a hundred years“. Susan Wijffels says it “could end up just being natural variability” and that the “real test will be what happens in the next twelve months” which is something we’ve heard before. But the best quote is saved for last – back to Gavin: “All of these statistics that we’re talking about, they’re taken from the prior data. But nothing in the prior data looked like 2023. Does that mean that the prior data are no longer predictive because the system has changed? I can’t rule that out, and that would obviously be very concerning.”

2024.03.18: Glen Peters pointing out that it wasn’t just the global temperature forecasts that went phut. Carbon dioxide forecasts were also a bust with the annual growth rate at Mauna Loa well above the previous record and well outside the uncertainty ranges of the forecasts. That’s fine; we expect one year in twenty or so to fall outside those ranges, but given how far outside the ranges 2023 was, the question is, do the methods used need to change?

2024.03.19: Gavin Schmidt’s world view has been aired in Nature. He reckons that “taking into account all known factors, the planet warmed 0.2C more last year than climate scientists expected.” It’s something of a form letter at this point: El Nino, eruptions, the sun, and shipping fuels. None of these is big enough to explain why everyone got their 2023 predictions so wrong. He does however break from the tradition by noting that our measurements of aerosol emissions take a while to make it through the pipeline. Consequently, we might be missing an important part of the puzzle. We need that data faster. Alternatively, we can wait till August and if the anomaly hasn’t stabilized by then we “will be in uncharted territory“. Some people have been saying this since almost this time last year. Gavin must have a better chart². The idea that we passed through uncharted territory and came out the other side is kind of interesting though.

2024.03.20: Pielke Jr has picked up on Gavin’s piece over on Twitter. It’s the usual Pielke confection, but it did highlight the Berkeley Earth 2023 roundup (which I’d missed and is excellent as usual) and the difficulty of keeping everything straight. Explaining trends over ten years (or longer) and temperatures in a single year aren’t necessarily the same thing. One can get the right long-term trend for the wrong short-term reasons. We really need to understand it all. The Berkeley analysis suggests that natural variability – a combo of a rapid ENSO shift³ and Saharan dustlessness – contributed to the warmth of 2023 along with some minor buffs from the HTHH eruption, solar variation, and shipping aerosols. Quoting: “it does appear likely that the inclusion unforced natural variability in the North Atlantic or elsewhere is necessary to explain the full magnitude of the warming observed in 2024.” As I’ve said before, this kind of explanation feels unsatisfying, like a poker player raking all the chips towards himself without revealing his cards. Firstly, there’s quite a lot to explain. Secondly, it’s not quantified. Rather, it’s assumed that it can close the gap. Perhaps it can, but like the “it’s just ENSO” explanation, it needs filling out. Was Saharan dust transport exceptionally low, were winds also low, what role did cloud cover play, are these things connected? Did lack of wind mixing lead to strong stratification that was self persisting etc.

2024.03.20: Never mind, Ryan solved it. Or perhaps nevermind that.

2024.03.21: Chris Colose pulls out a chess metaphor. It’s a good one – not everyone is confused on the same level – but his point is “We *probably* understand 2023 at any level of relevance to the public.” I’m not sure that’s entirely true. Most people haven’t a clue, so there is only expert opinion on this and some experts are scratching their heads. The people who think they understand it haven’t – as far as I’ve seen – tried to quantify the effect. Those who have tried to quantify it have got as far as ball-parking the large residual effect that needs to be explained. Chris and Ryan both cite the 1870s El Nino as an analogue with Chris giving more detail here.

2024.03.22: The Sidney Morning Herald treats us to the dinner party chatter of climate scientists and portrays the confusion about 2023 as a battle of personalities with Jim Hansen on the one side and Mike Mann on the other.

2024.03.25: The FT has an Editorial saying “The world is warming faster than scientists expected“. It quotes Jim Skea, chair of the IPCC, who declared 2023’s temperature surge as “completely unusual” – the kind of phrase where you know what they mean so long as you don’t think about it – “quicker than we all anticipated“, and “a surprise“. The short piece also says that the surprise came “out of the blue” and was “off the scale” and the “climate threat is now moving into uncharted territory” for those who haven’t filled out their cliché bingo card yet. The pull quote from the piece is “A full explanation remains elusive which underlines a compelling echo of history“, original, but the kind of thing you might say in the moment after receiving an unexpected electric shock. There’s not a lot here of interest, and it elides 2023 and long-term warming.

2024.03.26: Tangentially related to the 2023 question is the what exactly happened with Antarctic sea ice: regime shift or blip. A lengthy Carbon Brief article quoting Mark Serreze can’t do much better than “we must wait and see” a notion that, as a climate monitorer, I can fully endorse. An unnamed⁴ expert said what happened was “completely outside the bounds of normality” but now it’s happened so normal must mean something different.

2024.03.30: A spat over Hansen choosing his favourite part of the overall future warming pdf. Put that way, it’s not so different from what the lukewarmers do. Whether it makes sense to substitute vibes for a reasoned survey of the available evidence is above my epistemological pay grade⁵. Leon accuses his (and Hansen’s) detractors – Michael Mann in this case – of ignoring key evidence. The word he uses is “denying” which isn’t at all inflammatory. The spat has flared up around the issue of shipping aerosols. The evidence Mann points to doesn’t have the shipping aerosol changes factored in and for Leon it’s the most important part. At the same time, Mann points out, reasonably enough, that something that started happening a couple of years ago isn’t an obvious cause for a sudden spike in 2023. Right direction, wrong shape. Back to Gavin’s Nature world view: we need better and more timely measurements of aerosols.

2024.03.31: Last year, we heard that not only was a particular day the warmest in the 85 year long daily record from ERA5, but [insert probability qualifier] the warmest in 125,000 years. The people I’d trust to know such things were making quietly sceptical noises⁶ about these statements on twitter and elsewhere, but the caveats were drowned out by hooting from the more hysterical end of the spectrum. Now there’s a paper which takes aim at the ultimate source of these claims, the IPCC. It points out that proxy records aren’t really up to the task of saying whether a decade was the warmest in that long (let alone a day or week). At best the proxies over such time periods have a temporal resolution measured in centuries and constitute a real grab bag of sources that don’t cover the whole annual cycle or whole globe and have all manner of interesting issues. It’s not entirely clear who the paper is for. It’s sufficiently technical that it reads like a sermon to the converted rather than a necessary corrective to the ignorant, but at the same time admits it says little that’s new. It takes aim at the IPCC (and other targets for that matter) saying their statement is meaningless while also admitting that it’s probably right. Anyway, it would have been handy to have this six months ago and written for a broader audience. Meanwhile, the 1870s El Nino remains the go to analogue for 2023, but still no one provides any details (there’s not a lot of detail back then).

2024.04.02: Andrew Dessler over at the Climate Brink is having another go. He finds that 2023 just isn’t that unusual if you look at CMIP6. The chosen metrics are the margin by which the record was broken and the first year to exceed 1.5C. 2023 isn’t unusual by either metric. He also compares to the CMIP6 ensemble but doesn’t mention that it runs hot. The margin by which 2023 broke the record isn’t that exciting even if we only look at the observed time series of global temperature. Other big El Ninos broke the contemporary records by similar margins. What’s unusual about 2023 is the individual months, particularly September. I also have reservations about just comparing what happened to CMIP6 without considering whether the comparison or the models constituting CMIP6 are, in any way, reasonable. For example, Dessler notes that “the extreme warmth was definitely surprising given the state of the climate in 2022” but none of the chosen metrics actually show this. To what extent we can make a model-obs comparison without conditioning the model distribution on the observations, or derive meaningful comparisons if we don’t is an interesting question and therefore one I won’t answer here. The article ends, once again, with the suggestion that we should wait and see what happens and hope that we haven’t broken the climate.

2024.04.04: Earth.com does a quick round up of why 2023 was like it was – aerosols, HTHH, and solar cycles – and being based on Gavin’s Nature article, reiterates his concerns but then contrasts those with Michael Mann’s: “The situation is extremely similar to what we saw during the 2014–2016 period,” he says. Which it is (2023 as a whole) and isn’t (just go and stare at September for a while). Channelling his inner Danny, Mike continues “I think it’s unfortunate that so much has been made of the El Niño-spiked 2023 global temperatures, where in my view there is nothing surprising, or inconsistent with model predictions, there.” Well, we know we’ve been spoken to. If he and the article had left it there, it would be fine, but it goes on to link every extreme thing that happened in 2023 to its unusual warmth. Redemption comes a few lines later though by noting that we need better real-time monitoring. Mann’s assertion that would have put him on the wrong side of a recent AP-fact check⁷ and on the same side as Ryan Maue (“meteorologist and internet personality“). As Chris Colose points out, the question “why is 2023 so warm”? can be answered in many ways depending on what you’re comparing it to. 2023 is warmer than every single year in the global temperature record. The change from the 1850-1900 average to 2023 is mostly greenhouse gases, the change from 2022 to 2023 mostly isn’t. Why the changes in temperatures between late 2022 and 2023 were so large is another question. Who was answering what specific question in the fact-check and what they were asked (or thought they were being asked) isn’t clear. It’s not even clear to me what was being fact checked. Alongside Maue’s post, they cite another linking El Nino to flooding in the Congo where the connection is far more tenuous. Anyway, I digress. In his defence Maue cited “El Niño and the AMO Sparked the Astonishingly Large Margin of Warming in the Global Mean Surface Temperature in 2023” a paper which I can’t believe that I missed. Submitted and accepted in December 2023, it’s conclusions are obvious from the title. They use Ensemble Empirical Mode Decomposition⁸ to break the global temperature series down into three components approximately annual, decadal and secular. They find that the annual component explains most of the annual change from 2022 to 2023. They identify the annual component with ENSO but it’s not; EEMD doesn’t work like that. The decadal signal they identify with the “AMO” which is also positive, but slowly varying so not that different in 2022 and 2023. Anyway, the identification with the AMO is hand wavy at best and the warming of the North Atlantic in 2023 was astonishingly abrupt and not obviously a manifestation of multi-decadal variability. Overall, not the most convincing of analyses, but at the same time not necessarily wrong either in some loosely defined sense.

2024.04.04 (a bit later in the day, post-afternoon nap): On the question of whether warming has accelerated, Hansen has doubled down by saying all the same things all over again. Carbon Brief and Real Climate have responded by pointing out that what he’s saying isn’t really any different from what you can find in CMIP6 by (respectively) using SSP2-4.5 or by screening models that match the obs. The Carbon Brief speaks of a reticence about claims of acceleration based on short term trends. The reticence is understandable, and the evidence from global temperature on its own is weak. However, this time we sort of know why. Still, the uncertainties remain large. Meanwhile, a paper in Naturesomethingorother reports “Recent reductions in aerosol emissions have increased Earth’s energy imbalance” which does what it says on the tin but also notes that some important aerosol changes were missing from the inventories used CMIP6. Sadly, the analysis doesn’t extend to 2023. There’s a mismatch here somehow, or the uncertainties are wide enough that they embrace every hypothesis in a group hug.

2024.04.17: The Met Office blog has an article on low Antarctic sea ice extent in 2023. There’s nothing very exciting there, but it did highlight a paper on the low extents – Large-scale drivers of the exceptionally low winter Antarctic sea ice extent in 2023 – by Monica Ionita. It looks in great detail⁹ at the atmospheric drivers of the low ice extent, finding a zonal wave number 3 pattern¹⁰ had a significant impact on sea ice development. It’s interesting also because it cites Purich and Doddridge (2023) and says that they “showed that the unprecedented decline in the Antarctic Sea ice in the first 6 months of the year was mainly driven by subsurface warming of the Southern Ocean, pushing the Antarctic sea ice into a new low-extent state” which is rather stronger statement than I remember Purich and Doddridge making. For example, their abstract says “Here, we show the confluence of Southern Ocean subsurface warming and record minima and suggest that ocean warming has played a role in pushing Antarctic sea ice into a new low-extent state.” One way that sure knowledge develops is the gradual attrition of caveats. On the other hand, they note that Argo floats make few measurements in the critical ocean areas, limiting our ability to understand these changes.

2024.04.26: a paper in nature communications earth & environment has appeared. It has the kind of title that can save you reading the whole thing: Human influence can explain the widespread exceptional warmth in 2023. Aside from reiterating what one can now find in numerous news articles¹¹ (and then ending by suggesting that perhaps someone ought to quantify them), it analyses surface temperatures looking at the area covered by temperatures exceeding 2 (42% of global area) or 3 (13%) standard deviations. It calculates similar using models with and without anthropogenic forcings. Unsurprisingly, warmth as widespread as we saw in 2023 is not the kind of thing that would happen in an unperturbed climate. ENSO might contribute around ±12%.

2024.05.08: It’s been a while, but there has been quite some activity which may have clarified some things, but mostly hasn’t. First, people have been marking a year of record SSTs¹². This sparked a very short discussion about SST uncertainty and daily records¹³. There’s little doubt that record global(ish) SSTs have continued uninterrupted for a long time, but a second decimal place is on the verge of acceptability and naming individual days as records is falling off the verge and into the duck pond (for a quick idea compare the Climate Reanalyzer and the series in the BBC article, daily variability is different in the two with one being “sharper” than the other) . Kim Wood has provided an alternative analysis of regional SST for those who would like a second opinion but sadly it doesn’t include a global assessment. Meanwhile, we’re learning some interesting negatives. The extreme warmth in 2023 probably (even more probably than before) wasn’t associated with water vapour from the Hunga-Tonga eruption. The paper isn’t beyond quibbling. Robert Rohde pointed out that one of their assumptions could do with some serious testing. Previous analyses showed water vapour at high altitude had a small neglectable effect, but the quantities put out by Hunga Tonga are far beyond anything previously seen. An old article that passed me by points out that the reduction in sulphur emissions from shipping fuel is small compared to both total sulphur emissions and aerosol anomalies in 2023. Canadian wildfires and a lack of Saharan dust were locally much larger effects. The Daily podcast says something similar¹⁴. While 2023 was exceptionally warm in most places, Fennoscandinavians were pondering why it was so fricking cold there. From October 2023 through January 2024 a cold blob squatted over the north. While it wasn’t exceptionally cold by historic standards, it was very cold in the context of the 21st Century and 2023 in particular. For a short twitter round up on some of these issues, see Ryan Katz-Rosene.

-fin-

1. A rather low tone, one might say. ↩︎
2. Possibly though he’s the kind of party leader who confidently sets off and then says disquieting things like, “we’ll know where we are once we get over this hill. If there’s a stream and a church with a tower then we’re definitely in the right place.” ↩︎
3. Which intriguingly they differentiate from natural variability, not a distinction I’d previously encountered. ↩︎
4. The mighty Zack Labe gets quoted incognito earlier in the article. ↩︎
5. I’m just taking notes. ↩︎
6. Twitter search is misbehaving, so I can’t find specific examples, but it did remind me about that Patrick Brown nonsense. ↩︎
7. My experience of “fact checks” is limited but the few I have read have been scarcely better than the purported facts they’ve been checking. Twitter community notices on the other hand seem to be better. As with things of this kind though, it’s often very hard to tell. ↩︎
8. An aesthetically pleasing by loosey goosey means of extracting modes from time series. ↩︎
9. It uses ERA5 for the atmospheric variables, NSIDC for the sea ice, and OISST for the SSTs. That means that the atmosphere, ice, and sea surface are all somewhat inconsistent with each other. It probably doesn’t matter, but it irks me. ↩︎
10. Three highs and three lows alternating around the continent basically. ↩︎
11. See also the rest of the blog post. ↩︎
12. Which includes the wonderful cosy-Kafka quote “If I were a barnacle I’d be quite worried” uttered by an unnamed (actually named in the video – Dr Nova Mieszkowska, but not in the caption) marine biologist to “BBC News Climate Editor Justin Rowlatt” (named in the caption but not in the video). ↩︎
13. I’m not sure what the uncertainty in daily SST datasets is when averaged over the “globe”. I quarantine the word between bunny ears because it’s not global SST that people are talking about, but SST between 60S and 60N, but let’s roll with it: it gets rid of sea ice and issues associated with that. Uncertainty analysis of daily SST datasets typically focuses on local uncertainty i.e. how accurately SST is known at a particular point on the ocean surface as this is important for weather forecasting. Based on Emma Fiedler’s excellent paper from a few years back and near-real time monitoring, standard deviations relative to validation data (Argo) are of a few tenths of a degree, but some of that will be “noise” and average out when we aggregate the gridded data across the whole globe. How much though is hard to say. Mean offsets from validation data are typically very small in modern daily SST data sets when averaged over whole months, with differences much smaller than a tenth of a degree. That’s phenomenally good to the extent that exactly how you define SST probably starts to become important, particularly under conditions where SSTs are reaching record highs. How good *daily* global mean SST is though can’t be easily worked out from this . ↩︎
14. It also cites the “Day After Tomorrow” which is one of the funniest films ever made but only to a very particular subset of the population. ↩︎