Previous: Chapter 3. The wheel is turning
Chapter 4. Accuracy
an observer’s refrain (The Cure, “Accuracy”)
Stockholm, March 2035
Nur was an excellent cook, even though she was only fifteen. She certainly hadn’t learned it from Eza, who didn’t enjoy preparing food and usually went for quick and easy meals. But Norman, Eza’s husband, simply loved cooking. I liked Norman a lot. He was a civil engineer specialising in urban regeneration projects in Malaysia and so he was still over there, planning to move when his major projects had finished. He was a Malay Chinese from Butterworth in Penang. They had met early in her PhD, when he was doing a postdoc at Strathclyde. They’d married soon after. I had gotten to know him quite well when I helped them looking after baby Nur. He was a rather atypical husband as he did his share and more of the housework and childcare; and he was a fantastic cook and a dedicated teacher. He had even taught little Nur to speak Penang Hokkien. I knew she missed him a lot.
It was my turn to do the kitchen cleaning up. I loaded the dishwasher and turned it on. As usual, after the initial gurgling sounds, it started to sing, some rather pleasing harmonic resonance. I started scrubbing the hob. But Nur not only knew how to cook, she was very tidy as well, so it was done in no time and I moved to the washing up of the cutting board and knives. Doing the dishes always makes my mind wander. I thought back to that exciting period when the first results had started to come in.
Glasgow, February 2026
It was early in 2026, barely two years since that fateful winter morning in Helensburgh, when Li-Zhen’s team at Canada’s CCCma managed to confirm the results of the seminal 2023 work by the Ditlevsens and the follow-on work from 2024 by Dijkstra, then narrow down their estimate. Soon after, Mirza’s team at the APEC Climate Center Busan followed suit, and then it was like crocuses in spring, everywhere confirmation studies started to crop up: a team led by Eza in Malaysia and Sarah in Oman, using the ICON model as their starting point; teams from Japan and Australia working with Mirza, and from Chile via Li-Zhen, each with their own models. Rather to my surprise, even the Mofem-based model led by Glasgow managed to deliver the goods. Not that it was a bad model or a poor team, on the contrary, but Mofem was not a circulation model in origin. Its use as a climate model started out with a simulation of the fictional planet Arrakis from the Dune universe, and that had stimulated researchers to extend it and try it out for other planetary simulation purposes. What this meant was that the codebase was very different from the other circulation models, so the fact that it returned compatible results greatly increased the confidence in the project.
Glasgow, October 2026
By autumn of 2026, there was almost complete consensus. Only the team of Eza and Sarah had a significantly different result, with an earlier collapse but a much larger uncertainty. We discussed what the reason could be. Optimising and parallelising simulation code was my expertise, so I agreed to get hands-on with the codebase, in extended debug sessions with the researchers who wrote and maintained the code. As often with this kind of simulation code, it was built on a core of much older code that had proven reliable and accurate in the past. At first, we left that code alone, as it was more likely that the bug was elsewhere. But we didn’t get anywhere: everything else seemed just fine, we couldn’t find any bug that could have caused that discrepancy. So after a few weeks, we reluctantly tackled that crufty old code anyway, metaphorically rolling up our sleeves and getting our hands dirty. It was truly a mess (the term “spaghetti code” didn’t come even close) and it took us weeks to just start to understand how it worked. The postdocs started to look tired. I suspected they kept banging their heads against the problem long after I’d gone to bed. I told them to look after themselves, but I knew only too well that sometimes programmers can’t find peace until they’ve cracked the problem. One of the perks of age is that you find it easier to let things go.
But then suddenly, one beautiful day, the problem became obvious. Long ago some forgotten scientist had taken a shortcut to make the code easier to parallelise, and so introduced a potential convergence problem. For years it had gone unnoticed, until it cropped up in our complicated AMOC model. And we only really became suspicious because it did not match the results of the other teams. On its own, the results it produced were plausible enough. We were over the moon when we’d finally figured it out. Fixing the bug was comparatively straightforward, meaning that it took only half as long as finding it. And so a few months later, when the daffodils where just starting to show their petals, the simulation results were finally in agreement with those of the other teams.
Glasgow, March 2027
We had narrowed the AMOC collapse down to 2035-2055, with better than 95% certainty, significantly better than any other estimate published so far and with an unprecedented degree of consensus. It was as good as it ever got.
But it wasn’t good enough. To be convincing, the uncertainty would have to be a few years rather than decades. The problem was the accuracy of the observational data. Measuring the AMOC was difficult: the currents flowed really deep in some parts of the ocean, and for accurate modelling, many measurement points of temperature, salinity and density were needed, all across the length and depth of the currents. That required a research ship with a diving bell that could go really deep, and that was hard to come by and very expensive. Plus, the old refrain, it was not a priority so getting funding for it was effectively impossible. It was a serious conundrum.
Eventually we did get the measurements, and that’s how we got where we are now: scientific consensus that, unless emissions are cut dramatically, AMOC will collapse in 2045 with a window of just one year on either side and still better than 95% certainty.
That had been more than a tall order. In fact, it had been pure luck.
Next: Chapter 5. Atomic