Simon Frederick Barnard Tett

Despite global warming, China continues to experience record‐breaking cold extremes. However, the extent to which internal atmospheric circulation variability offsets anthropogenic warming effects on cold extremes remains poorly quantified. Here, a nonstationary GEV is used to quantify the relative influences of anthropogenic warming and the Siberian High (SibH) variability on cold extremes across China during 1961–2020. A decade of anthropogenic warming, represented by a one‐decade rise in global mean surface temperature (GMST), increases the GEV location parameter and induces spatially heterogeneous scale changes, generally weakening cold extremes. In contrast, a one‐standard‐deviation strengthening of the SibH produces stronger cooling: it lowers location parameters nationwide, particularly over northwestern China, where the magnitude is about twice that associated with GMST. A strengthened SibH also enhances the scale parameter over northwestern and central China, implying higher spread and increased severity of rare cold events in these regions.

High-impact extreme rainfall events, often unprecedented in the observational record, can trigger severe flooding and cascading socio-economic impacts. Accurately understanding and projecting such events remains challenging due to their rarity and the limitations of existing modelling approaches.Here, we develop an integrated framework to investigate both the evolution and impacts of high-impact extreme rainfall events, using the July 2021 Zhengzhou flood as a case study. The framework combines high-resolution convection-permitting simulations with a hydrological–hydrodynamic model and an AI-based surrogate model. The surrogate model is trained and validated against hydrodynamic simulations to efficiently predict inundation dynamics, including flood extent, depth, and duration. It adopts a two-stage design, first predicting inundation occurrence and then estimating flood depth, enabling improved representation of the sparsity and spatial heterogeneity of flooding.By coupling the surrogate model with convection-permitting simulations, we project future flood responses under the SSP5-8.5 scenario (2091–2100). Inundation outputs are further integrated with land use and building data through a depth–damage function to estimate associated economic losses.Our results show that, beyond total rainfall accumulation, the temporal structure and spatial heterogeneity of precipitation play a critical role in shaping flood responses. Accounting for these characteristics significantly improves the representation of inundation dynamics compared to approaches based solely on aggregated rainfall metrics. Under future scenarios, inundation depth is projected to increase by up to 27%, while economic losses may rise by approximately 10% (around USD 550 million).Overall, the proposed framework provides a computationally efficient and scalable approach for linking rainfall processes to flood impacts, supporting more robust and process-informed assessment of future flood risk.

Child undernutrition is a major global concern, contributing to both mortality and lifelong morbidity. In low-income countries heavily reliant on subsistence and rainfed agriculture, low productivity agriculture affecting both food availability and income is one crucial factor influencing child nutrition. An increasing number of attribution studies have quantified impacts of human-induced climate change on crop yields in rainfed agriculture. Tracing the effect of these climate change impacts further to observed nutrition-related health outcomes is a gap in the attribution literature. Here, we show an approach to attribute undernutrition-related child health impacts of anthropogenic climate change using a multidisciplinary array of data and methods. As a case study, we use India, which has one of the highest global burdens of undernutrition-related child health impacts. As observational data, we use 1) climate reanalysis data (W5E5), evaluated against observational products and weather station data; 2) district-level seasonal crop yield and production data for major staple crops; and 3) individual-level anthropometric and socioeconomic data from two waves of the Demographic and Health Surveys (DHS).These data are used to bias-correct and statistically downscale large-ensemble climate model output from the CMIP6-DAMIP project, to calibrate perturbed-parameter ensembles of three different process-based crop models (APSIM, DSSAT, InfoCrop), and to estimate exposure-response functions linking crop yield anomalies to child health. Factual (with climate change) and counterfactual (without climate change) child health outcomes are then derived by applying this analysis chain to factual and counterfactual CMIP6-DAMIP data, respectively, and analysed in an event attribution framework to quantify the contribution of anthropogenic climate change to the 2014-2015 crop yield deficits and resulting child health impacts across India.The 2014-2015 period saw substantial crop yield deficits across India, with seasonal yields falling more than two standard deviations below long-term trends. Epidemiological analysis reveals that children exposed to such deficits during prenatal and infancy periods face elevated stunting risk, while positive yield anomalies show no corresponding benefit. Attribution findings will be presented and wider implications for climate-food-health attribution and for applications of impact event attribution frameworks be discussed.

Attributing observed and projecting future impacts of climate change is key for risk assessment and corresponding action. Climate data provide only one, but crucial, input for these studies. Standardised global climate model output is often practical for this, but needs for many impact modelling studies to be bias-corrected, and ideally available at spatial resolutions that allow consideration of non-climate factors varying at spatial scales higher than typical for, say, CMIP output. Such processed climate data are available typically only for a very small subset of available climate model simulations. Making large-ensemble climate data available for impact modelling can reduce the gap between state-of-the-art climate science and the climate information informing societally highly relevant impact studies.Here, we present a method, code, and data of bias-corrected and statistically downscaled large-ensemble CMIP6 data from the CMIP, ScenarioMIP, and DAMIP subprojects. First, we present a method tweak that allows the Bias Adjustment and Statistical Downscaling (BASD) code used in the Intersectoral Impact Modelling Intercomparison Project (ISIMIP) Phase 3 to be scientifically better applicable to large-ensemble climate data. Specifically, instead of correcting a single ensemble member for the bias of the same-labelled (e.g., r1i1p1f1) member of the historical simulations compared to observationally-derived data, we correct each member for the bias of the respective model’s full ensemble (r*i*p*f*) to preserve ensemble spread. Second, we provide a user-friendly software that implements this ISIMIP3BASD-LE method and the complete pre-/postprocessing at an improved computational speed. Third, we present an overview of the CMIP6 simulations newly processed with this method and code.The method and code are easily applicable to upcoming CMIP7 data, as well as any other climate model data available. Current challenges regard the availability of CMIP6 data in terms of individual ensemble members/variables, as well as decisions regarding the ensemble size for bias correction with incomplete variable coverage compared to what is desired from an impact modelling perspective. Compared to native high-resolution simulations and dynamical downscaling, the thus-processed climate data retain all disadvantages associated with the original method/data. Nonetheless, uptake of the method, code, and data will allow impact attribution to make a step forward, and impact projections to be on a climate-scientifically much sounder basis.

In 2023-2024 austral summer (Dec-Mar), an intense drought caused severe economical and human losses in the South of Africa, resulting in a loss of 1/3 of the total crop harvest. Here we report on a fairly standard attribution study for the drought of 2023/24 summer to assess if human influence increased the occurrence and intensity of droughts in the region. We used HadGEM-GA6 data to assess the likelihood of observing these events in scenarios with/without anthropogenic activity using 3 month Standardized Precipitation Evapotranspiration index (SPEI3) to quantify drought intensity. The sensitivity to region choice was explored using definitions of South of 20S, South of 15S, the region analyzed in the last World Weather Attribution report as well as individual countries. Simulations (with and without human activity) for the climatological period (1970-2010) as well as for 2023-2024 specifically were compared. The influence of the El Niño Southern Oscillation (ENSO) on SPEI3 and associated attribution statements was considered by compositing simulations by year into El Niño and La Niña phases. When using HadGEM simulations for the historical period (1970-2010), results showed that simulations with human activity showed lower SPEI values compared to natural simulations, hence implying that South African is drier compared to a natural scenario. Nonetheless the probability of drought is sensitive to the region chosen for the analysis, for example, for the south of 20S the probability of drought is mostly between 1.1 - 2 times more likely in simulations with human activity, whereas in the WWA area this probability rises to 5.9 - 16.9. By contrast, in HadGEM simulations with the prescribed conditions of 2023-2024, the probability of drought is much higher but also shows more uncertainty.In addition, human activity strengthened the intensity and frequency of the dry periods set up by El Niño conditions in most countries located in the South of Africa, but the occurrence of droughts changes with the region. For example, in Zimbabwe, drought occurrence is 1.8 more likely in simulations with human influence during El Niño events, whereas in South Africa and Zambia the drought occurrence is 1.6 and 3.2 times more likely respectively whereas in Malawi it remains unchanged. In addition, when considering the prescribed conditions of 2023-2024 the probability of drought rises drastically for all countries.

This poster provides an overview of the PROMOTE (Progressing Earth System Modelling for Tipping Point Early Warning Systems) project. One project aim is to develop a version of the UK Earth System Model (UKESM) that is suitable to become the model component of a potential early warning system of Subpolar Gyre or Greenland Ice Sheet tipping. In PROMOTE, we (i) undertake targeted development of UKESM to advance its representation of the Greenland Ice Sheet and North Atlantic Subpolar Gyre, (ii) develop innovative simulation techniques aiming to make the simulation of tipping behaviour in the Subpolar Gyre and Greenland Ice Sheet more controlled and efficient, (iii) use machine-learning and other analysis techniques as well as model simulations to inform the design of observational networks, and (iv) evaluate tipping processes and impacts of tipping in our newly developed model versions. PROMOTE is run by a team of scientists and model developers at 7 UK national research centres and universities during 2025-2030, and is part of the “Forecasting Tipping Points” programme funded by the Advanced Research and Invention Agency (ARIA).

Parameterization in climate models often involves parameters that are poorly constrained by observations or theoretical understanding alone. Manual tuning by experts can be time-consuming, subjective, and prone to underestimating uncertainties. Automated tuning methods offer a promising alternative, enabling faster, objective improvements in model performance and better uncertainty quantification. This study presents an automated parameter-tuning framework that employs a derivative-free optimization solver (DFO-LS) to simultaneously perturb and tune multiple convection-related and microphysics parameters. The framework explicitly accounts for observational and initial condition uncertainties (internal variability) to calibrate a 1° resolution atmospheric model (GAMIL3). To evaluate its performance, two main tuning experiments were conducted, targeting 10 and 20 parameters, respectively. In addition, three sensitivity experiments tested the effect of varying initial parameter values in the 10-parameter case. Both tuning experiments achieved a rapid reduction in the cost function. The 10-parameter optimization improved model accuracy for 24 of 34 key variables, while expanding to 20 parameters yielded improvement for 25 variables, though some structural model biases appeared. Ten-year AMIP simulations validated the robustness and stability of the tuning results, showing that the improvements persisted over extended simulations. Additionally, evaluations of the coupled model with optimized parameters showed, compared to the default parameters settings, reduced climate drift, a more stable climate system, and more realistic sea surface temperatures, despite a residual global energy imbalance of 2.0 W m⁻² (about 1.4 W m⁻² arising from the intrinsic imbalance of the atmospheric component) and some remaining regional biases. The sensitivity experiments further underscored the efficiency of the tuning algorithm and highlight the importance of expert judgment in selecting initial parameter values. This tuning framework is broadly applicable to other general circulation models (GCMs), supporting comprehensive parameter tuning and advancing model development.

A derailment on 2020-08-12 near Carmont, Scotland resulted in the death of three people and the injury of the other 6 people on the train. The cause of the derailment was gravel washed onto the tracks due to heavy rain and an improperly built drain. The heavy rainfall near the derailment site lasted about 4 h, with a large burst of rain just prior to the derailment. The rainfall event was heavy with return periods between 1-in-20 and 1-in-40 summers. Simple theory suggests that extreme precipitation should increase with changes in saturated humidity (Clausius–Clapeyron; CC). The UK’s convective permitting model suggests, that extreme hourly rainfall near Carmont changes at about 1.5× CC rates while extreme four hourly rates increases below CC rates. This leads to an increased probability, relative to late 19th century conditions, of hourly (4 h) extremes similar to that which occurred near the derailment by about 20 (15)% with a further frequency increase of about 20 (15)% for such events in a +2C warmer world. Some characteristics of the simulated extreme rain evaluate well against radar values, but the simulated rainfall is about 20%–30% larger than the radar rainfall. This likely reflects shortcomings in radar calibration of extreme rainfall though poor model performance is also possible.

Parameterization in climate models often involves parameters that are poorly constrained by observations or theoretical understanding alone. Manual tuning by experts can be time-consuming, subjective, and prone to underestimating uncertainties. Automated tuning methods offer a promising alternative, enabling faster, objective improvements in model performance and better uncertainty quantification. This study presents an automated parameter-tuning framework that employs a derivative-free optimization solver (DFO-LS) to simultaneously perturb and tune multiple convection-related and microphysics parameters. The framework explicitly accounts for observational and initial condition uncertainties (internal variability) to calibrate a 1-degree resolution atmospheric model (GAMIL3). Two experiments, adjusting 10 and 20 parameters, were conducted alongside three sensitivity experiments that varied initial parameter values for a 10-parameter case. Both of the first two experiments showed a rapid decrease in the cost function, with the 10-parameter optimization significantly improving model accuracy in 24 out of 34 variables. Expanding to 20 parameters further enhanced accuracy, with improvement in 25 of 34 variables, though some structural model errors emerged. Ten-year AMIP simulations validated the robustness and stability of the tuning results, showing that the improvements persisted over extended simulations. Additionally, evaluations of the coupled model with optimized parameters showed–compare to the default parameter setting–reduced climate drift, a more stable climate system, and more realistic sea surface temperatures, despite a slight energy imbalance and some regional biases. The sensitivity experiments underscored the efficiency of the tuning algorithm and highlight the importance of expert judgment in selecting initial parameter values. This tuning framework is broadly applicable to other general circulation models (GCMs), supporting comprehensive parameter tuning and advancing model development.

Parameterization in climate models often involves parameters that are poorly constrained by observations or theoretical understanding alone. Manual tuning by experts can be time-consuming, subjective, and prone to underestimating uncertainties. Automated tuning methods offer a promising alternative, enabling faster, objective improvements in model performance and better uncertainty quantification. This study presents an automated parameter-tuning framework that employs a derivative-free optimization solver (DFO-LS) to simultaneously perturb and tune multiple convection-related and microphysics parameters. The framework explicitly accounts for observational and initial condition uncertainties (internal variability) to calibrate a 1-degree resolution atmospheric model (GAMIL3). Two experiments, adjusting 10 and 20 parameters, were conducted alongside three sensitivity experiments that varied initial parameter values for a 10-parameter case. Both of the first two experiments showed a rapid decrease in the cost function, with the 10-parameter optimization significantly improving model accuracy in 24 out of 34 variables. Expanding to 20 parameters further enhanced accuracy, with improvement in 25 of 34 variables, though some structural model errors emerged. Ten-year AMIP simulations validated the robustness and stability of the tuning results, showing that the improvements persisted over extended simulations. Additionally, evaluations of the coupled model with optimized parameters showed–compare to the default parameter setting–reduced climate drift, a more stable climate system, and more realistic sea surface temperatures, despite a slight energy imbalance and some regional biases. The sensitivity experiments underscored the efficiency of the tuning algorithm and highlight the importance of expert judgment in selecting initial parameter values. This tuning framework is broadly applicable to other general circulation models (GCMs), supporting comprehensive parameter tuning and advancing model development.