By Rhidian Thomas, Research Scientist (NCAS) in the Department of Meteorology
What will the weather be tomorrow? Advances in weather prediction mean we can now make skilful forecasts not just of tomorrow’s weather, but next week’s. But what about the weather 50 or 100 years from now?
Regardless of advances in forecasting, predicting the weather that far ahead is not possible. Never mind 50 years, under most circumstances predictions of individual weather systems are difficult beyond a few weeks. Improved models and machine learning-based approaches may extend forecast skill, but predictability will always be limited in chaotic systems such as the atmosphere that depend on processes occurring over a wide range of length scales [1].
If we can’t predict the weather in a month’s time, what can we possibly say about the far future? Using theory and models, we can estimate statistical properties of the future climate, though not the precise weather patterns on a given day. We can also explore the conditions a particular weather pattern might lead to in different climates. An analogy is to think about the changing seasons. We can’t be certain that it will be colder on the 25th of December than the 25th of May – in fact, in 1863 it was 0.1C warmer on Christmas day in Central England than 7 months prior [2]. The temperature I will experience eating my Christmas lunch depends on whether it’s cloudy or sunny, a warm front passing over or clear, frigid skies (in 1863, high pressure over Greenland delivered cold Arctic air in late May, while Christmas was unusually mild owing to a strong North Atlantic jet stream). But we do expect that the same weather patterns, if they happened in December, would lead to cooler temperatures than in May.
In this spirit, our paper, published recently in Earth System Dynamics [3], asked: what would it look like if weather patterns that occurred in the past had instead happened in a warmer climate? Would every day be warmer and rainier, or would the change depend on the baseline conditions for that day? This idea of pasting weather patterns onto different climate states is not a new one, having developed over the last few decades within the field of “extreme event attribution” [4]. Event attribution tries to quantify the impact of a changing climate on the likelihood or intensity of a particular extreme weather event such as a storm or heat wave. Various approaches exist to do this; fundamentally, they all involve reconstructing the weather pattern of interest in a different climate to the one in which it occurred in reality. We used a reanalysis-based method first introduced by Hawkins et al (2023) [5], who focused on a devastating storm that hit NW Europe in 1903 (Storm Ulysses, so named as it was recalled in James Joyce’s novel). In our paper, we extended this to look not just at a single extreme event, but at changes in the day-to-day weather across all 12 months of 1903.
To transport past weather patterns across time, we first reconstructed the atmospheric state in a reanalysis model called 20CRv3, which uses observations of surface pressure to constrain atmospheric circulation patterns in a numerical weather forecast model. We then re-ran the reanalysis model, ingesting the same pressure observations but with a warmer ocean surface and higher atmospheric CO2 concentration – thus transporting those weather patterns into a warmer world. As a proof of concept, we chose simple perturbations to the sea surface temperature (SST) – which was increased by +2K everywhere (+2K) – and atmospheric CO2, which was additionally increased by 80% in a separate experiment (+2K+CO2). In reality, observed trends in sea surface temperature show a lot of regional variation, but our perturbations are sufficient to illustrate the point; we intend to apply more realistic perturbations in upcoming experiments. We focused most of our analysis on four regions with enough pressure observations from 1903 for us to be confident in the reconstruction: Northwestern Europe, the Western Mediterranean, the US East Coast, and Southeastern Australia.
Fig. 1: Change in daily-mean 2m-temperature in the warmer world experiments versus the 1903 reanalysis (20CRv3+), as a function of temperature percentile in the reanalysis. The vertical dotted lines show absolute temperatures in 20CRv3+. Most days are warmer by roughly 2°C, but larger increases are seen on days below freezing and days above about 20°C.
So when comparing our reconstruction of 1903 with the warmer world version, did all days show a similar change in temperature? Actually, we found that the greatest warming occurred on very cold days – those below freezing – and warm days, above 20 °C or so (Fig. 1). The reasons for this varied from region to region, but often involved land surface processes. For example, the cold day response was because the warmer world experiment had less snow and ice, and thus absorbed more solar radiation at the surface. On the other hand, on the hottest days in the US East Coast region the land surface was very dry, and so less able to lose heat through evaporation; this effect was amplified in the warmer world.
Fig. 2: Panels (a) & (d) show the number of dry days in the experiments for two of the four study regions. Panels (b) & (e) show daily precipitation frequencies in the 20CRv3+ and +2 K + CO2 experiments. The blue line shows the percentage change, which crosses zero around the 90–95th percentile of the 20CRv3+ distribution (vertical dashed lines). The other two regions are shown in the paper’s supplementary.
What about rainfall? Did it become wetter or drier overall? Globally it became wetter, but regionally the answer varied – for example, the Western Mediterranean became drier even without changes in weather patterns. But we found that the day-to-day variability in rainfall increased, even in the Western Mediterranean. All four regions experienced more dry days and fewer drizzly days, though these changes are difficult to distinguish from the noise (two of the four regions are shown in Fig 2). At the other end of the scale, the number of days with very heavy rain – corresponding to the top 5–10% of days in the original reconstruction – increased dramatically. In 2002, Allen and Ingram (2002) [6] found a similar redistribution of rainfall in a climate model experiment, with the rainfall intensity at a given percentile decreasing below the 90th percentile and increasing above it. Although we focused on the frequency of rainy days rather than the total amount of rain, our result is qualitatively similar to theirs, as shown by where the blue line crosses zero in Fig. 2; effectively, the increase in rainfall occurs on around 1 day in 10, with most days showing little change (or even becoming drier). Surprising results like this reflect the highly skewed distribution of daily precipitation, where a small number of extreme rainfall events can dominate the total annual rainfall [7].
Our work shows how the weather patterns of a year from history, 1903, would play out in a warmer climate. While we only looked at a single year, the paired reanalysis experiments offer a high signal-to-noise ratio due to their shared weather patterns. However, to fully understand future changes in our day-to-day weather, it does leave the thorny issue of changes in the weather patterns themselves.
Separating changes in the thermodynamic impacts of the weather patterns – such as temperature and rainfall – from changes in the weather patterns themselves is an increasingly common approach [8]. This is because thermodynamic processes are both simpler theoretically and relatively well represented in weather and climate models. On the other hand, changes in atmospheric circulation are far more uncertain [9] – but this does not mean they are unimportant. For example, changes in the occurrence of persistent weather systems (known as atmospheric blocks) over NW Europe would have implications for future changes in both hot and cold extremes. Model projections tend to produce fewer blocking systems over NW Europe in future climates, but confidence is limited by the fact that these models also struggle to produce the correct amount of such systems in the present climate. Improving our understanding and modelling of these circulation patterns will influence projections of both extreme and everyday weather. That said, it is worth highlighting that some of our results – such as the increase in dry days and shift towards heavy rainfall at the expense of light rainfall – have been seen in multiple generations of climate model experiments where the circulation is allowed to change [10].
Even if we make progress on understanding future changes in the atmospheric circulation, we will never be able to predict with certainty whether it will be mild or freezing, sunny or snowy on December 25th, 50 years from now. Instead, we can estimate how the statistics of our weather may change. By breaking down the problem into changes in the weather patterns themselves, versus changes in the impacts associated with the weather patterns, we can try to make progress in understanding how our everyday weather may change in the future.
References
[1] Lorenz, E. N.: The predictability of a flow which possesses many scales of motion, Tellus, 21, 289–307, https://doi.org/10.1111/j.2153-3490.1969.tb00444.x, 1969.
[2] Legg, T., Packman, S., Caton Harrison, T., and McCarthy, M.: An Update to the Central England Temperature Series—HadCET v2.1, Geoscience Data Journal, 12, https://doi.org/10.1002/gdj3.284, 2024.
[3] Thomas, R., Compo, G. P., George, S., Hegerl, G. C., Schurer, A., Shepherd, T. G., Slivinski, L. C., Thompson, V., and Hawkins, E.: Everyday weather in a warmer world, Earth Syst. Dynam., 17, 1365–1380, https://doi.org/10.5194/esd-17-1365-2026, 2026.
[4] Feser, F. and Shepherd, T. G.: The concept of spectrally nudged storylines for extreme event attribution, Commun Earth Environ, 6, https://doi.org/10.1038/s43247-025-02659-6, 2025.
[5] Hawkins, E., Compo, G. P., and Sardeshmukh, P. D.: ESD Ideas: Translating historical extreme weather events into a warmer world, Earth Syst. Dynam., 14, 1081–1084, https://doi.org/10.5194/esd-14-1081-2023, 2023.
[6] Allen, M. R. and Ingram, W. J.: Constraints on future changes in climate and the hydrologic cycle, Nature, 419, 224–232, https://doi.org/10.1038/nature01092, 2002.
[7] Pendergrass, A. G.: What precipitation is extreme?, Science, 360, 1072–1073, https://doi.org/10.1126/science.aat1871, 2018.
[8] Shepherd, T. G.: A Common Framework for Approaches to Extreme Event Attribution, Curr Clim Change Rep, 2, 28–38, https://doi.org/10.1007/s40641-016-0033-y, 2016.
[9] Shepherd, T. G.: Atmospheric circulation as a source of uncertainty in climate change projections, Nature Geosci, 7, 703–708, https://doi.org/10.1038/ngeo2253, 2014.
[10] Hennessy, K. J., Gregory, J. M., and Mitchell, J. F. B.: Changes in daily precipitation under enhanced greenhouse conditions, Climate Dynamics, 13, 667–680, https://doi.org/10.1007/s003820050189, 1997.
Discussion about this post