Search

Detection and attribution (D&A) of forced precipitation change are challenging due to internal variability, limited spatial, and temporal coverage of observational records and model uncertainty. These factors result in a low signal-to-noise ratio of potential regional and even global trends. Here, we use a statistical method - ridge regression - to create physically interpretable fingerprints for the detection of forced changes in mean and extreme precipitation with a high signal-to-noise ratio. The fingerprints are constructed using Coupled Model Intercomparison Project phase 6 (CMIP6) multi-model output masked to match coverage of three gridded precipitation observational datasets - GHCNDEX, HadEX3, and GPCC - and are then applied to these observational datasets to assess the degree of forced change detectable in the real-world climate in the period 1951-2020. We show that the signature of forced change is detected in all three observational datasets for global metrics of mean and extreme precipitation. Forced changes are still detectable from changes in the spatial patterns of precipitation even if the global mean trend is removed from the data. This shows the detection of forced change in mean and extreme precipitation beyond a global mean trend is robust and increases confidence in the detection method's power as well as in climate models' ability to capture the relevant processes that contribute to large-scale patterns of change. We also find, however, that detectability depends on the observational dataset used. Not only coverage differences but also observational uncertainty contribute to dataset disagreement, exemplified by the times of emergence of forced change from internal variability ranging from 1998 to 2004 among datasets. Furthermore, different choices for the period over which the forced trend is computed result in different levels of agreement between observations and model projections. These sensitivities may explain apparent contradictions in recent studies on whether models under- or overestimate the observed forced increase in mean and extreme precipitation. Lastly, the detection fingerprints are found to rely primarily on the signal in the extratropical Northern Hemisphere, which is at least partly due to observational coverage but potentially also due to the presence of a more robust signal in the Northern Hemisphere in general., Earth System Dynamics, 14 (1), ISSN:2190-4987, ISSN:2190-4979

Because long-term observations are sparse and heavily affected by inter-annual variability, the detection of forced trends in land carbon uptake has been remarkably difficult. While changes in aggregated monthly or yearly carbon uptake have been heavily studied, changes in the diurnal cycle of carbon uptake remain uninvestigated. Here, we evaluate long-term changes in three-hourly Net Ecosystem Production (NEP) and the contribution of different times of the day to long-term changes in the terrestrial carbon sink. We first show that between 1950 and 2014, five CMIP6 models show a significant increasing trend in the diurnal NEP amplitude (DNA). Second, we show that DNA trends have a much higher signal-to-noise ratio compared to trends in NEP itself, and linearly scale with long-term annual NEP. Evaluating these model-based results against observations, we show that positive DNA trends are also found at a majority of observational eddy-covariance sites between 1990 and 2022. The positive correlation between DNA and long-term NEP is also confirmed by these observations. Our results reveal a widespread but previously undocumented emergent climate signal in terrestrial carbon exchange at the diurnal scale with the potential to serve as a new observational constraint for Earth system models.

The Intergovernmental Panel on Climate Change (IPCC) assesses the sensitivity of the climate system to increases in greenhouse gas concentrations using multiple lines of evidence, covering paleoclimate data, historical observations, and numerical Earth system model (ESM) simulations. Within IPCC’s latest Assessment Report (AR6), there is, for the first time, a non-negligible difference between the most likely rate of warming estimated in the report and the average warming rate simulated by the ESMs that participated in the Coupled Model Intercomparison Project (CMIP6). This discrepancy occurs because a large number of CMIP6 models have projected future warming rates that are higher than previously reported but quite unlikely according to historical observations. The consequence is that using a random selection of CMIP6 simulations is likely to overestimate historical and future warming (compared to what is assessed in the IPCC report), potentially leading to avoidable inconsistencies when compared to observations or greater projected changes compared to what could be inferred from CMIP5.As this constitutes a wide-spread obstacle and limitation to using CMIP6 simulations ‘out of the box’, we propose here a simple model weighting method with the objective to address this problem. Our approach can be used to 1) evaluate the extent to which any given set of CMIP6 simulations is consistent with IPCC-assessed warming rates and 2) calculate the appropriate model weights so that potential inconsistencies are reduced as much as possible. The calculation of the weights is solely based on the user’s selection of a CMIP6 subset and does not require any data manipulation. The weights can then be easily implemented in existing analyses to calculate weighted (i.e. instead of just arithmetic) multi-model means, weighted quantiles, etc. We demonstrate the interest and flexibility of the method with some examples, including global to regional assessments of historical and projected changes in temperature and precipitation. We illustrate the extent to which applying model weights can reconcile otherwise divergent scientific results and provide assessments that are more robust across CMIP generations.

Precipitation (P) and evapotranspiration (ET) play a crucial role in the water cycle and have a significant impact on water resources and the energy balance of the Earth's surface. However, it remains a challenge, in particular on regional scales, to detect changes in hydrological variables and attribute them to anthropogenic or natural influences. Traditional studies that aim to detect or attribute changes in atmospheric variables often consider only a single variable at a time. This makes detecting changes in hydrological variables challenging due to large internal variability, the lack of long-term observational coverage and partly poor mechanistic understanding of land-atmosphere coupling processes in a changing climate. However, because P and ET are related to various other atmospheric variables, such as temperature, humidity, and sea level pressure, the detection of anthropogenic influences may be conducted in principle within a broader multivariate space. Here, we aim at exploiting multivariate relationships to more robustly detect anthropogenic changes to the hydrological cycle at the regional or up to continental scale. We train statistical models from coupled Earth system models to learn the relationships between relatively well observed variables and more poorly observed ones, like P and ET. We demonstrate that such models can predict and extract patterns of forced change in P and ET, albeit somewhat contingent on the realism of the simulation of the Earth system model. The main advantage is that the method does not rely on sparse observations of P and ET, and instead relies on covariates which are more abundantly and reliably observed. We demonstrate the effectiveness of this approach in a climate-model-as-truth framework, showing that it can capture a wide range of possible hydrological responses produced by the different climate models. We also apply the statistical model to observations to identify forced changes in P and ET that have already occurred. For example, we see an increase in ET in the northern hemisphere likely induced by a reduction in aerosol emissions. Our results show that this method can infer changes in P and ET that may have taken place, in principle even without the need for direct observations of those variables and can provide constrained projections of future water resources and energy balance.

Recent unprecedented extremes such as the 2023 New Year’s winter warm spell in Europe, the 2022 summer heat and dryness in China and Europe, the 2021 Pacific Northwest heatwave, the 2021 floods in northwestern Germany and the 2020 Siberian heat anomaly broke previous observed record intensities by large margins. Based on the observations up to the year before the event, some of these record-shattering extremes were inconceivable. Could the scientific community have better quantified the potential for such unseen events based on the current generation of climate models?Here, we demonstrate how a new ensemble boosting approach can be used to generate physically coherent storylines of such unseen events. In ensemble boosting a fully-coupled free-running climate model (CESM2) is used to develop physical storylines of very rare extremes. To this end, the most extreme events in an initial-condition large ensemble for the near future are re-initialized with slightly perturbed atmospheric initial conditions to efficiently generate events with magnitudes unprecedented in the observational record.We demonstrate that, with this approach, CESM2 can simulate events that e.g. substantially exceed the magnitude of the 2021 Pacific Northwest (PNW) heatwave anomaly. Even though the most extreme ensemble members were only selected based on the local temperature anomaly over the PNW region, they show a very similar temporal evolution and spatial pattern as the observed event. The associated 500hPa geopotential height anomaly is remarkably similar to the observed event with a strong anticyclone over PNW that is part of a wave pattern extending over much of the Northern Hemisphere. We further show that in some of the storylines pre-conditioning dry soils strongly contributed to the heatwave intensity, whereas in others, heatwaves of similar magnitude occur even at average land surface conditions.We further develop storylines for heatwaves in the Greater Paris and Chicago regions of much greater intensities than ever observed. Particularly the US Midwest, where no intensification of heatwaves has been observed in recent decades, has not experienced anything close to the heatwave intensities possible in the coming years.Finally, we demonstrate examples how ensemble boosting can also be used to generate storylines for multi-year drought events and large-scale heavy precipitation extremes and compare the findings to other storyline approaches. Event storylines based on ensemble boosting can be used in impact studies that require physical consistency across variables, e.g., for the evaluation of humid heatwaves or compound events, for assessing wildfire risks or for ecosystem modelling.

Global mean surface air temperature (GSAT) is a key diagnostic for understanding and constraining historical climate variability and change, and for climate policy. Yet, global temperature estimates (1) are usually based on blending sea surface temperatures (SST) with near-surface air temperature over land (LSAT), and (2) contain many missing values due to incomplete coverage in the historical record. While these issues are usually accounted for in model-observation comparisons, elucidating the consistency of LSAT and SST records, and their contribution to GSAT variability and change, remains difficult.Here, we present a set of new GSAT reconstructions based separately on either the historical LSAT or SST record. The method is based on regularized linear regression models trained on climate model simulations to optimally predict GSAT from the climate model’s LSAT or SST patterns, respectively. We then predict GSAT from the HadSST4 and CRUTEM5 observational data, respectively, for each month from January 1850 up to December 2020.We demonstrate that the land- or ocean based GSAT estimates show very similar variability and long-term changes, both in the early (1850-1900) as well as in the late record (post-1950). For example, GSAT of the past decade (2011-2020) increased by 1.15°C (LSAT-based) and 1.17°C (SST-based) relative to an early reference period (1850-1900), which is both well within IPCC AR6 estimates.However, the GSAT estimates show pronounced disagreement in the early 20th century (1900 up to around 1930), when the SST-based GSAT estimates appear on average around 0.3°C colder than the LSAT-based estimates. Decadal changes in the LSAT-based estimates are well explained by the multi-model mean of CMIP6 simulations driven with historical forcings, thus implying only a small role of unforced decadal global variability. In contrast, the SST-based estimate highlights pronounced variability during the early 20th century cold anomaly, which may be related to concerns about instrumental cold biases in SST measurements, but overall reasons for the disagreement remain unclear. Further analysis based on physical reasoning, climate models, and proxy reconstructions, indicates that the ocean data may indeed be implausibly cold.In conclusion, our methodology and results may help to constrain the magnitude of early 20th century warming, and thus to better understand and attribute decadal climate variability.

Recent extreme temperature and precipitation events such as the dry and heat events in summer 2022 in Europe and China, the New Year’s warm spell 2022/23 across Europe, the 2021 heavy rainfall extremes in northwestern Germany, Belgium and the Netherlands and the 2021 Pacific Northwest heatwave broke previous observed record levels by large margins. The probability of such unprecedented record-shattering extremes increases with the rapid rate of warming. Thus, there is a crucial need for analyzing the underlying processes leading to these events and quantifying potential intensities of events possible in the coming decades.Here, we evaluate how ensemble boosting (Gessner et al. 2021 and Gessner et al. 2022) can help assess the tail of climate distributions and generate climate model-based storylines more resource-efficiently. In ensemble boosting the most extreme simulated events in an intermediate-size initial condition ensembles are re-initialized in targeted experiments in order to efficiently sample very extreme states of the model climatology. Here, we evaluate different ensemble design choices including lead time, ensemble size and potential iteration choices to most efficiently allocate computational resources to simulate events of very extreme intensity.The resulting boosted events are analyzed through a storyline approach, thus helping to interpret the underlying mechanisms of each physically consistent unfolding extreme event and its consequences. The Pacific Northwest heatwave in June 2021 will be used as a starting point; but ensemble boosting and storylines can be powerful tools for understanding extremes beyond heat. We further discuss how ensemble boosting can also be applied to compound extremes and future climate scenarios.

It comes as no surprise that the future holds record-breaking weather and climate events. As global warming continues, temperature records will continue to be broken. Also heavy precipitation records are likely to be broken due to the increased water holding capacity of the atmosphere, in combination with changing atmospheric stability and circulation patterns. Improved estimates on the range of possible record-breaking precipitation events – now and in the future – are a first step to inform adequate adaptation policies for heavy precipitation. Of particular interest are events that break records by large margins – record-shattering events –, since these are likely to incur most damage and losses. In order to improve estimates of record shattering precipitation events in the present and future climate we use initial condition large ensemble simulation data (CESM2, SSP370) and statistical models. We evaluate record-shattering events in Rx1d (day with most precipitation per chosen time period (year or season)). In a stationary climate, the probability of Rx1d record-breaking is known to decrease with the number of data points since the start of measurements (inversely proportional). We find, however, that in our nonstationary climate, the decay in Rx1d record breaking and shattering probability is slowed down and even reversed in most world regions. Regional changes in record shattering probability are attributable to a changing underlying probability distribution of Rx1d, which also is region specific. We elucidate the contributions of changes in mean (distribution shift), and in variability (distribution widening/narrowing) to increasing record shattering probability by using a statistical model to create counterfactual realities representative of the regions of interest.We focus on regions of a size relevant for national and cross-border policy that show differently driven changes in record shattering precipitation probabilities. For example, the annual probability of a record shattering precipitation event somewhere in the Benelux-Germany region which was hit by severe floods in summer 2021 increases from ~2% now to ~4.5% at the end of the century in CESM2. This increase results from a non-linear interaction between mean and variability increases, and is primarily driven by increasing variability. At lower latitudes, for example in Central America, the effect of variability is even stronger, where we find increasing record shattering probability despite a negative long-term trend in Rx1d levels.Very unlikely events are, paradoxically, arguably the most important to know about, since their unimaginability often means that critical infrastructure is not sized to withstand these events. Our results may thus prove invaluable for regional policy.

Proceedings of the National Academy of Sciences of the United States of America, 120 (4), ISSN:0027-8424, ISSN:1091-6490

Heatwaves constitute a major threat to human health and ecosystems. Projected increases in heatwave frequency and severity thus lead to the need for prediction to enhance preparedness and minimize adverse impacts. In this Review, we document current capabilities for heatwave prediction at daily to decadal timescales and outline projected changes under anthropogenic warming. Various local and remote drivers and feedbacks influence heatwave development. On daily timescales, extratropical atmospheric blocking and global land–atmosphere coupling are most pertinent, and on subseasonal to seasonal timescales, soil moisture and ocean surface anomalies contribute. Knowledge of these drivers allows heatwaves to be skilfully predicted at daily to weekly lead times. Predictions are challenging beyond timescales of a few weeks, but tendencies for above-average temperatures can be estimated. Further into the future, heatwaves are anticipated to become more frequent, persistent and intense in nearly all inhabited regions, with trends amplified by soil drying in some areas, especially the mid-latitudes. There is also an increased occurrence of humid heatwaves, especially in southern Asia. A better understanding of the relevant drivers and their model representation, including atmospheric dynamics, atmospheric and soil moisture, and surface cover should be prioritized to improve heatwave prediction and projection. ISSN:2662-138X

Recent climate extremes have broken long-standing records by large margins. Such extremes unprecedented in the observational period often have substantial impacts due to a tendency to adapt to the highest intensities, and no higher, experienced during a lifetime. Here, we show models project not only more intense extremes but also events that break previous records by much larger margins. These record-shattering extremes, nearly impossible in the absence of warming, are likely to occur in the coming decades. We demonstrate that their probability of occurrence depends on warming rate, rather than global warming level, and is thus pathway-dependent. In high-emission scenarios, week-long heat extremes that break records by three or more standard deviations are two to seven times more probable in 2021–2050 and three to 21 times more probable in 2051–2080, compared to the last three decades. In 2051–2080, such events are estimated to occur about every 6–37 years somewhere in the northern midlatitudes. Changes in extreme heat are often calculated as anomalies above a reference climatology. A different definition—week-day heatwaves surpassing the current record by large margins—shows that their occurrence probabilities depend on warming rate, not level, and are higher than during recent decades.

Heat waves such as the one in Europe 2003 have severe consequences for the economy, society, and ecosystems. It is unclear whether temperatures could have exceeded these anomalies even without further climate change. Developing storylines and quantifying highest possible temperature levels is challenging given the lack of long homogeneous time series and methodological framework to assess them. Here, we address this challenge by analysing summer temperatures in a nearly 5000-year pre-industrial climate model simulation, performed with the Community Earth System Model CESM1. To assess how anomalous temperatures could get, we compare storylines, generated by three different methods: (1) a return-level estimate, deduced from a generalized extreme value distribution, (2) a regression model, based on dynamic and thermodynamic heat wave drivers, and (3) a novel ensemble boosting method, generating large samples of re-initialized extreme heat waves in the long climate simulation.All methods provide consistent temperature estimates, suggesting that historical exceptional heat waves as in Chicago 1995, Europe 2003 and Russia 2010 could have been substantially exceeded even in the absence of further global warming. These estimated unseen heat waves are caused by the same drivers as moderate observed events, but with more anomalous patterns. Moreover, altered contributions of circulation and soil moisture to temperature anomalies include amplified feedbacks in the surface energy budget. The methodological framework of combining different storyline approaches of heat waves with magnitudes beyond the observational record may ultimately contribute to adaptation and to the stress testing of ecosystems or socio-economic systems to increase resilience to extreme climate stressors.

Partitioning uncertainty in projections of future climate change into contributions from internal variability, model response uncertainty and emissions scenarios has historically relied on making assumptions about forced changes in the mean and variability. With the advent of multiple single-model initial-condition large ensembles (SMILEs), these assumptions can be scrutinized, as they allow a more robust separation between sources of uncertainty. Here, the framework from Hawkins and Sutton (2009) for uncertainty partitioning is revisited for temperature and precipitation projections using seven SMILEs and the Coupled Model Intercomparison Project CMIP5 and CMIP6 archives. The original approach is shown to work well at global scales (potential method bias < 20 %), while at local to regional scales such as British Isles temperature or Sahel precipitation, there is a notable potential method bias (up to 50 %), and more accurate partitioning of uncertainty is achieved through the use of SMILEs. Whenever internal variability and forced changes therein are important, the need to evaluate and improve the representation of variability in models is evident. The available SMILEs are shown to be a good representation of the CMIP5 model diversity in many situations, making them a useful tool for interpreting CMIP5. CMIP6 often shows larger absolute and relative model uncertainty than CMIP5, although part of this difference can be reconciled with the higher average transient climate response in CMIP6. This study demonstrates the added value of a collection of SMILEs for quantifying and diagnosing uncertainty in climate projections.

Variations in the global water cycle are among the most impact-relevant consequences of climate change to human society. However, large uncertainties exist in regional precipitation trends due to the presence of internal variability, limited coverage of observational data, and model deficiencies. Dynamical adjustment methods have been developed to quantify the role of circulation-induced variability in relevant climate variables such as precipitation and temperature. By subtracting the latter, the thermodynamic signals are expected to remain in the residuals, therefore increasing the signal-to-noise (S/N) ratio of trend estimates. However, dynamical adjustment for high-resolution spatial and temporal data remains challenging, especially over heterogeneous terrain such as the Alps.In this study, we build upon the Latent Linear Adjustment Autoencoder (LLAAE), a new statistical model introduced by Heinze-Deml et al., (2021, Geoscientific Model Development 14(8), pp. 4977-4999, doi:10.5194/gmd-2020-275.) that combines dynamical adjustment techniques with elements of deep learning. By combining a linear model with a nonlinear deep neural network, the method has shown great promise in extracting the forced thermodynamic components of high-resolution winter precipitation over Europe. So far, this model has only been trained and tested on different members of the Canadian Regional Climate Model (CanESM2). Here, we illustrate and evaluate the LLAAE in the context of transfer learning. The LLAAE, trained on 9 members of the CanESM2, is evaluated on 10 different high-resolution (EUR-11) Regional Climate Models of the EURO-CORDEX ensemble. We show, despite large climatological model differences, that the LLAAE is capable of reconstructing up to 50 % of the daily precipitation variance across the spatial domain of the CORDEX-data. Furthermore, trend estimates can be improved when retraining the linear model with relatively few additional data points from the target distribution: Retraining the model with 4000 individual days further increases the explained variance by about 30%, improving the S/N ratio of long-term and spatial trend estimates. This could be particularly useful for future real-world applications such as the detection and attribution of anthropogenic forced regional changes in precipitation and the contribution of the circulation-induced variability.

Global-mean surface air temperature (GSAT) is a key diagnostic of climate change and a key metric for climate policies. Yet, global temperature datasets (1) are usually based on blending the sea surface temperature (SST) record with land surface air temperature (LSAT) data, and (2) contain a large number of missing values due to incomplete coverage, particularly in the early record. These issues are usually accounted for in model-observation comparisons via a similar processing of models and/or statistical infilling, but an apple-to-apple comparison between LSAT and SST records in their contribution to GSAT estimates, and also in their spatial consistency, remains difficult. Here, we present a set of novel GSAT estimates based separately on either the LSAT or SST record, and climate model information. The method is based on regularized linear regression models that are trained on climate model simulations to optimally predict GSAT from the climate model’s LSAT or SST predictors, respectively, which are masked to match the observational coverage at any given time step. In a second step, the derived statistical models are used to predict GSAT from the HadSST4 (SST) and CRUTEM5 (LSAT) observational data, respectively, and for any observational coverage from January 1850 up to December 2020. In addition, we generate a variant of these estimates that explicitly take into account the estimated errors as well as bias realizations for HadSST4 and CRUTEM5 data in the GSAT prediction. We show that the resulting independent, land- and ocean based GSAT estimates are remarkably consistent since around 1950: the squared correlation between the land- and ocean GSAT estimates is 0.98 for annual and 0.92 for monthly data, whereas it is only 0.94 (annual) and 0.77 (monthly), for the original CRUTEM5 and HadSST4 global land and ocean time series. In addition, the 1950-2020 long-term trends in GSAT estimates are virtually identical when inferred independently from land- or ocean data (1.14°C or 1.17°C warming per 71 years, respectively), and GSAT of the past decade (2011-2020) increased by 1.18°C (LSAT-based) and 1.15°C (SST-based) relative to an early period (1850-1880). However, the GSAT estimates show a pronounced period of disagreement in the early 20th century (1900 up to around 1920), when the SST-based GSAT estimates appear up to about 0.5°C (0.3°C on average) colder than the LSAT-based estimate, with important implications for the magnitude of the early 20th century warming. This finding is consistent with concerns about biased observed estimates raised in the literature, and is potentially related to instrumental cold biases in SST measurements, but overall reasons for the disagreement remain largely unclear. We show several lines of evidence, based on statistical analysis, physical reasoning and comparison with climate models, which indicate that the ocean data may indeed be implausibly cold. However, the early 20th century ocean cold anomaly, as well as the associated strong early 20th century ocean warming, require further study.

The Pacific Northwest heat wave is one of a series of record-shattering heat extremes that, based on the previous observational record, may have been deemed impossible. Here we address the question of whether the potential for such an extreme heat wave could have been foreseen using simulated physical climate storylines.We use a novel approach, called ensemble boosting, in which a fully-coupled free-running climate model (CESM2) is used to develop physical storylines of very rare heat extremes under present-day conditions. In ensemble boosting, the most extreme events in an initial-condition large ensemble for the near future are re-initialized with slightly perturbed atmospheric initial conditions to efficiently generate events that are even more extreme, with the goal of sampling events with magnitudes that have not been seen before.We demonstrate that, with this approach, CESM2 can efficiently simulate events that reach or even exceed the magnitude and duration of the 2021 Pacific Northwest heatwave anomaly. The atmospheric circulation anomalies associated with the most extreme simulated heat waves in the boosted ensemble are remarkably similar to the observed event. We further evaluate the anomalies in the surface energy and water budgets that contribute to the most intense simulated events. We conclude that based on this approach, heat waves unseen in the observational record can be simulated in models, at least in some regions. After probing this approach for the Pacific Northwest heatwave, we apply it to other mid-latitude regions where extreme heat events of much higher magnitude than has been observed are plausible in the near future.The ensemble boosting approach is computationally efficient, and it preserves physical consistency both in time, in space and across variables. This has the major advantages that the drivers can be directly evaluated against observed events and the generated storylines can be used in impact studies that require physical consistency, e.g. for the evaluation of humid heatwaves or compound events, for assessing wildfire risks or for ecosystem modelling.

Research on extreme weather typically investigated the physical and dynamical processes involved in the formation of specific meteorological events that occur on time scales of hours to a several days (e.g., heavy precipitation events, windstorms, heat waves). Such events can be extremely hazardous, but for certain socioeconomic sectors the seasonal aggregation of weather is particularly harmful. These sectors include, for instance, agriculture, forestry, energy, and reinsurance. This presentation introduces the concept of “extreme seasons” as an important and not yet thoroughly investigated research field at the interface of weather and climate science. Extreme seasons are defined as seasons during which a particular meteorological or impact-related parameter (or a combination thereof) strongly deviates from climatology. An important conclusion of the presentation will be that large ensemble climate simulations (here using an extended CESM1-LENS data set with 6-hourly output of 3D fields), with about 1000 simulated years per climate period, are an essential resource enabling novel quantitative insight into the processes leading to and characteristics of extreme seasons. The presentation provides examples for the identification of extreme seasons and emphasizes the importance of studying their substructure, including the occurrence of specific weather systems. A first approach to systematically study extreme seasons is to consider the top 10 seasons (for a given metric) in the large ensemble at every grid point, e.g., the 10 wettest winters or hottest summers, or the 10 summers with the largest vapour pressure deficit (as an example for a more impact-related metric). Alternatively, one can look at anomalies in a multi-dimensional parameter phase space, identifying extreme seasons that result from a highly unusual combination of, e.g., surface temperature, precipitation, and surface energy balance. Or, using a pragmatic method based on fitting a statistical model to seasonal mean values at each grid point, spatially coherent extreme season objects can be identified that exceed a local return period threshold of, e.g., 40 years. The same statistical approach can be applied to ERA5 reanalyses to compare characteristics of extreme season objects (e.g., their size and intensity) in climate models with observation-based data. With this approach we can meaningfully estimate how often, e.g., an observed extreme winter like the cold North American 2013/14 winter is expected anywhere in midlatitude regions. The last part of the presentation addresses the substructure and weather system characteristics of extreme seasons. Illustrative results are shown that address the questions: (i) Where are extremely hot summers the result of the warmest days being anomalously hot vs. the coldest days being anomalously mild? (ii) Where are wettest seasons the result of more frequent wet days vs. more intense precipitation on wet days? and (iii) How does the frequency of weather systems and their precipitation efficiency change during the wettest seasons? The answers to these questions reveal interesting and large regional differences.

Detection and attribution (D&A) of anthropogenically forced changes to precipitation is challenging due to the high internal variability of precipitation and the limited spatial and temporal coverage of the observational records. These factors result in a low signal-to-noise ratio of potential regional and even global trends.Here, we use a statistical method – regularised linear regression, or ridge regression – to create physically interpretable detection metrics (fingerprints) for D&A of changes in the precipitation distribution with a high signal-to-noise ratio. The regression coefficients that make up the fingerprints of forced change are based on the CMIP6 multi-model archive data masked to match observational coverage, and are then applied to gridded precipitation observations to assess the degree of forced change detectable in the real-world climate. We show that the signature of forced change is detected and attributed to external forcing in two different observational datasets in global metrics of mean and extreme precipitation (PRCPTOT, and Rx1d, respectively). If the global mean trend is removed from the data, forced changes are still detected, indicating that climate change affects the spatial patterns of precipitation, and increasing confidence in the results of this method for D&A of precipitation, as well as in climate models capturing the relevant processes that contribute to the regional patterns of change. Furthermore, we show the sensitivity of our D&A results to several ‘design choices’, including target metric of forced change, regularisation parameter, season of interest, (spatial coverage of) observational dataset used, the forced trend length and the region of interest (tropics vs. subtropics). The method is largely insensitive to target metric and regularisation parameter, increasing confidence in the robustness of the results. However, we find that June-July-August generally has low forced trend signal-to-noise ratio in both mean and extreme precipitation. Furthermore, the observational dataset choice affects detectability not only through coverage differences but also dataset disagreement, and the chosen trend length can result in different forced trends when comparing observations to model projections. These sensitivities may explain apparent contradictions in recent studies on whether models under- or overestimate the observed increase in extreme precipitation. Lastly, the detection models are found to rely primarily on the signal in extratropical northern hemisphere data, which is at least partly due to observational coverage, but potentially also due to presence of a more robust signal in the northern hemisphere in general. When these regions are excluded, detection of significant forced changes is no longer possible, which may also have implications for the ability to assess risks and inform adaptation policies in the tropics and the global south.Ridge regression is powerful for D&A of the precipitation distribution, opening up possibilities for extension of the method to learn more about mechanisms driving forced changes in precipitation. However, internal variability, limited coverage in time and space, and dataset disagreement in precipitation data continue to play a large role.