Executive Summary
UK mean winter precipitation has intensified in a way that becomes clearer only after separating circulation-driven (“dynamical”) variability from thermodynamic (“non-dynamical”) change: the non-dynamical component scales at 7.6% per °C of UK winter warming (24.4 mm/°C, p < 0.001) over 1901–2023, consistent with near–Clausius–Clapeyron moisture scaling (~7%/°C) under high humidity (Carruthers et al., 2026). In contrast, CMIP6 models simulate a weaker intensification—an ensemble-mean 4.0%/°C—with most model members below the observed scaling rate, implying that precipitation–temperature scaling (and therefore flood-risk escalation) could be biased low in many model-based risk workflows (Carruthers et al., 2026).
Key Results In Numbers
Across observations and model comparisons, five numbers dominate the interpretation: (1) observed non-dynamical scaling 7.6%/°C (24.4 mm/°C; standard error 6.93 mm/°C; R² = 0.09) (Carruthers et al., 2026); (2) CMIP6 ensemble-mean scaling ~4.0%/°C (Carruthers et al., 2026); (3) the dynamical component explains ~62% of interannual variance in UK mean winter precipitation but shows no robust long-term trend, leaving the long-term wetting signal primarily non-dynamical (Carruthers et al., 2026); (4) detectability metrics show signal-to-noise (S/N) = 1.23 for the non-dynamical component in 2023 (vs 0.91 for total precipitation), with emergence in 2012 under the study’s thresholds (Carruthers et al., 2026); and (5) spatially, statistically significant wetting trends occur across 22% of UK land for total winter precipitation and 42% for non-dynamical precipitation (Carruthers et al., 2026).
Why This Matters For Flood Risk And Losses
Even small percentage differences in precipitation–temperature scaling can translate into large, compounding impacts on flood probability, infrastructure exceedance frequency, and insured losses—especially in catchments where antecedent wetness and storm sequencing are dominant risk multipliers (Carruthers et al., 2026). This matters in economic terms: UK flood damage and management are measured in billions of pounds per year, with physical damage estimated at £2.4 billion annually and projected to rise to £3.6 billion by 2050 in England, plus wider economic impacts around £6.1 billion per year (House of Commons Environmental Audit Committee, 2025). On the insurance side, weather-related damage to homes reached a record £585 million (US$724.7 million) in 2024, alongside £5.7 billion (US$7.1 billion) in total Q4+annual property claims reported in the same dataset, underscoring how precipitation-linked hazards already map into material cash flows (Insurance Journal, 2025; Association of British Insurers, 2025).
The Core Scientific Question
The paper asks a detection-and-attribution question framed in process terms: are observed changes in UK winter precipitation primarily driven by dynamical shifts in atmospheric circulation (e.g., storm-track and regime frequency changes) or by thermodynamic intensification in a warmer, moister atmosphere (Carruthers et al., 2026)? This distinction is not semantic—UK winter rainfall has large natural variability tied to North Atlantic circulation, so attributing a secular trend requires an approach that can isolate forced signals from circulation “noise” (Carruthers et al., 2026).
Data Sets, Spatial Scale, And Study Window
Observational precipitation is derived from HadUK-Grid, a gridded dataset derived from interpolated in situ rain-gauge measurements, analyzed over 1901–2023 to align with model availability and long-term detectability requirements (Carruthers et al., 2026; Hollis et al., 2019). For the detection metric, precipitation is regressed against a smoothed global-mean surface temperature series from HadCRUT5, using a 41-year loess filter to emphasize low-frequency, forced variability (Carruthers et al., 2026; Hawkins et al., 2020).
Dynamical Adjustment Using Weather Patterns
To isolate circulation-driven precipitation, the authors apply a dynamical adjustment based on European weather patterns: the MO30WP classification consists of 30 daily mean sea level pressure (MSLP) anomaly patterns spanning 30°W–20°E and 35°N–70°N, originally derived via k-means clustering on the EMULATE MSLP dataset and extended through 2023 using ERA5 reanalysis (Carruthers et al., 2026; Neal et al., 2016; Hersbach et al., 2023). The method generates a synthetic annual time series for the dynamical component (including 1,000 realizations), enabling the residual (total minus dynamical) to be interpreted as the non-dynamical component (Carruthers et al., 2026).
What Counts As Dynamical Versus Non-Dynamical Rainfall
In this framework, the dynamical component represents precipitation variability explained by the frequency of circulation regimes (weather patterns), while the non-dynamical component captures the remainder, including the thermodynamic effects of warming (Carruthers et al., 2026). Importantly, the dynamical component explains the majority of interannual swings—~62% of the variance in UK mean winter precipitation—yet the long-term wetting signal is dominated by the non-dynamical residual, illustrating why trend detection is difficult without explicitly constraining circulation variability (Carruthers et al., 2026).
Spatial Trends Across The UK Land Surface
The dynamical adjustment changes the map of detectability: statistically significant wetting trends in total winter precipitation appear across 22% of UK land area, but significant wetting trends in the non-dynamical component cover 42%, implying a more spatially extensive forced signal once circulation variability is accounted for (Carruthers et al., 2026). Where total precipitation trends are significant, the dynamical contribution averages up to 31%, with the non-dynamical component accounting for roughly ~70% of the total trend in those grid cells—consistent with thermodynamic intensification as the primary long-term driver (Carruthers et al., 2026).
The Observed Temperature Scaling: 7.6% Per °C
The central quantitative result is a robust scaling of UK mean non-dynamical winter precipitation with UK mean winter temperature: 24.4 mm per °C, corresponding to 7.6%/°C when normalized by the mean UK winter precipitation of 321.96 mm (Carruthers et al., 2026). The relationship is statistically significant (p < 0.001) with R² = 0.09, which is expected for precipitation (high variance) even when a forced low-frequency relationship exists; the point is not that temperature explains most year-to-year variance, but that the forced component exhibits near–Clausius–Clapeyron scaling once dynamical variability is controlled (Carruthers et al., 2026).
Detectability: Signal-To-Noise And Emergence Timing
Using the signal-to-noise framing (forced signal estimated via regression on smoothed GMST; noise defined by residual variability), the non-dynamical component becomes detectably more “signal-dominated” than the raw total: by 2023, S/N = 1.23 for non-dynamical precipitation versus 0.91 for total precipitation, making the non-dynamical signal ~37% larger in S/N terms (Carruthers et al., 2026). Under the study’s time-of-emergence thresholds, the non-dynamical precipitation signal emerges in 2012, and recent values are categorized as “unusual” relative to the longer record—an important result for event attribution and for updating baselines used in design standards (Carruthers et al., 2026).
Attribution: Natural Versus Anthropogenic Forcing In CMIP6
Attribution is assessed by comparing observed non-dynamical trends to CMIP6 distributions under natural-only forcing versus historical forcing (with historical runs extended to 2023 using SSP2-4.5) (Carruthers et al., 2026). Using a 90% confidence-interval classification, observed non-dynamical wetting trends fall outside the CMIP6-natural 90% range while lying near the upper 90% of CMIP6-historical, supporting an attribution of the observed non-dynamical intensification to anthropogenic forcing and simultaneously indicating that many models understate the magnitude of the forced wetting trend (Carruthers et al., 2026).
Model Performance: 4.0% Per °C Ensemble Mean And Large Spread
On the model-evaluation axis, the discrepancy is clear: the CMIP6 ensemble-mean scaling rate is ~4%/°C, versus 7.6%/°C in observations, implying the observed scaling is about 90% higher than the ensemble mean when expressed as a relative difference ((7.6–4.0)/4.0 ≈ 0.90) (Carruthers et al., 2026). Model spread is large: only eight simulations fall in the ~6–8%/°C Clausius–Clapeyron-consistent range, several members show >8%/°C, and at least five members exhibit negative scaling rates; within-model internal variability can be substantial, exemplified by UKESM1-0-LL members spanning roughly −4%/°C to 9%/°C (Carruthers et al., 2026).
Why Coarse Models Can Underestimate Scaling
The authors point to a physically plausible mechanism for the low bias: in coarse-resolution GCMs, key precipitation processes are parameterized rather than explicitly resolved, including convection, land–sea interactions, and orographically enhanced precipitation, all of which are materially important for UK winter rainfall (Carruthers et al., 2026). They further note that convection-permitting and other high-resolution modeling studies often project stronger increases in precipitation intensity (and sometimes seasonal means) than coarser models, implying that model resolution and physics representation can directly influence precipitation–temperature scaling and thus impact-relevant risk projections (Carruthers et al., 2026; Kendon et al., 2020; Chan et al., 2020).
Practical Next Steps For Science And Model Evaluation
Scientifically, this work argues for making “process separation” routine in precipitation detection studies: if dynamical variability explains most variance yet contributes little to the forced trend, then trend detection and model evaluation should be conducted on both total precipitation and the thermodynamic residual, with uncertainty quantified via resampling (e.g., the paper uses 1,000 bootstrap resamples for confidence intervals) (Carruthers et al., 2026). For model evaluation, the implication is that precipitation scaling diagnostics (e.g., %/°C for non-dynamical components) could serve as an additional constraint alongside mean-state and circulation metrics, particularly in regions such as the UK, where storm-track variability can mask forced signals in unadjusted time series (Carruthers et al., 2026).
Climate Risk Intelligence™ Implications For Planning And Finance
For users building Climate Risk Intelligence™ pipelines (flood screening, loss modeling, infrastructure stress testing), this study provides an actionable calibration target: 7.6%/°C scaling of UK mean winter precipitation on the thermodynamic component, with detectability increasing over time and post-2012 conditions flagged as statistically “unusual” relative to early-20th-century baselines (Carruthers et al., 2026). The economic motivation is already visible in observed losses: a single recent year produced £585 million (US$724.7 million) in weather-related home-damage payouts, and national-level assessments anticipate multi-billion-pound annual damages with growth toward mid-century (Insurance Journal, 2025; House of Commons Environmental Audit Committee, 2025). Translating science to decisions, therefore, means stress-testing whether planning standards, flood defenses, and underwriting assumptions implicitly rely on model precipitation–temperature sensitivities closer to 4%/°C than to the observed 7.6%/°C, and quantifying the tail-risk consequences of that gap (Carruthers et al., 2026).
Frequently Asked Questions (FAQs)
- Why does UK winter precipitation increase as temperatures rise? It increases because a warmer atmosphere can hold more water vapor, raising the potential for heavier precipitation, and the observed non-dynamical winter precipitation signal scales at 7.6% per °C once circulation variability is accounted for (Carruthers et al., 2026).
- What does “dynamical adjustment” mean in this study? It is a statistical decomposition that uses variability in weather patterns to estimate and remove the circulation-driven component of precipitation, leaving a residual interpreted as the non-dynamical (thermodynamic-dominated) component (Carruthers et al., 2026).
- How much faster is the observed scaling than the CMIP6 ensemble mean? The observed scaling is 7.6%/°C versus a ~4.0%/°C CMIP6 ensemble mean, which is about 90% higher than the ensemble mean on a relative basis (Carruthers et al., 2026).
- When did the non-dynamical precipitation signal emerge, according to the paper? Using the time-of-emergence thresholds applied in the study, the non-dynamical precipitation signal emerged in 2012, and recent conditions are categorized as “unusual” relative to the earlier historical record (Carruthers et al., 2026).
- What is the practical risk if models underestimate precipitation–temperature scaling? If scaling is biased low, projections used for flood planning and risk pricing can understate the rate at which winter rainfall volumes intensify with warming, potentially leading to under-designed defenses and mispriced insurance risk in contexts where losses are already substantial (Carruthers et al., 2026; Insurance Journal, 2025).
Sources
• Association of British Insurers. (2025, February 10). More action needed to protect properties as adverse weather takes record toll on insurance claims in 2024. Association of British Insurers.
• Carruthers, J. G., Fowler, H. J., Bannister, D., & Guerreiro, S. B. (2026). Climate models tend to underestimate scaling of UK mean winter precipitation with temperature. Geophysical Research Letters, 53(3), e2025GL118201. doi:10.1029/2025GL118201.
• Chan, S. C., Kendon, E. J., Berthou, S., Fosser, G., Lewis, E., & Fowler, H. J. (2020). Europe-wide precipitation projections at convection permitting scale with the Unified Model. Climate Dynamics, 55(3), 409–428. doi:10.1007/S00382-020-05192-8.
• Fereday, D., Chadwick, R., Knight, J., & Scaife, A. A. (2018). Atmospheric dynamics is the largest source of uncertainty in future winter European rainfall. Journal of Climate, 31(3), 963–977. doi:10.1175/JCLI-D-17-0048.1.
• Hawkins, E., Frame, D., Harrington, L., Joshi, M., King, A., Rojas, M., & Sutton, R. (2020). Observed emergence of the climate change signal: From the familiar to the unknown. Geophysical Research Letters, 47(6), e2019GL086259. doi:10.1029/2019GL086259.
• Hollis, D., McCarthy, M., Kendon, M., Legg, T., & Simpson, I. (2019). HadUK-Grid—A new UK dataset of gridded climate observations. Geoscience Data Journal, 6(2), 151–159. doi:10.1002/gdj3.78.
• House of Commons Environmental Audit Committee. (2025, October 13). Flood resilience in England: Fourth report of session 2024–26. UK Parliament.
• Insurance Journal. (2025, February 11). Insurers pay record claims for weather-related damage in UK: ABI. Insurance Journal.
• Kendon, E. J., Roberts, N. M., Fosser, G., Martin, G. M., Lock, A. P., Murphy, J. M., et al. (2020). Greater future U.K. winter precipitation increase in new convection-permitting scenarios. Journal of Climate, 33(17), 7303–7318. doi:10.1175/JCLI-D-20-0089.1.
• Neal, R., Fereday, D., Crocker, R., & Comer, R. E. (2016). A flexible approach to defining weather patterns and their application in weather forecasting over Europe. Meteorological Applications, 23(3), 389–400. doi:10.1002/met.1563.
• Newcastle University. (2026, February 11). UK winter rain and warming. Newcastle University Press Office.
• Office for Budget Responsibility. (2024, September). Estimating the direct fiscal costs of flooding (Fiscal risks and sustainability, Box 2.3). Office for Budget Responsibility.
Ready to get started? To learn how ClimaTwin can help you assess the physical and financial impacts of future weather and climate extremes on your infrastructure assets, capital programs, and investment portfolio, please visit www.climatwin.com today.
© 2026 ClimaTwin Corp. All rights reserved worldwide.ClimaTwin® is a registered trademark of ClimaTwin Corp. The ClimaTwin logos, ClimaTwin Solutions™, Climate Risk Intelligence™, Climate Business Intelligence™, Climate Value at Risk™, Future-proofing assets today for tomorrow’s climate extremes™ are trademarks of ClimaTwin Corp. All trademarks, service marks, and logos are protected by applicable laws and international treaties, and may not be used without prior written permission of ClimaTwin Corp.
###