Assessing Global and Regional Trends in Spatially Co‐Occurring Hot or Wet Annual Maxima Under Climate Change

Recent years were characterized by spatially co‐occurring hot and wet extremes around the globe, raising questions about the contribution of human‐induced global warming to the changing likelihoods of such extreme years. To characterize spatially co‐occurring extremes we investigate recent trends in global and regional land area that is concurrently affected by hot or wet annual maxima taking observational uncertainty into account. Observed trends in the land area affected by extreme events are assessed in the context of Earth System Model (ESM) simulations for present‐day and early‐industrial climate conditions in a detection and attribution setting. We compare different reanalysis and station‐based observational products to account for observational data uncertainty. At the global scale, trends of spatially co‐occurring hot or wet annual maxima in all observational products can be explained by ESM simulations driven by historical radiative forcing that accounts for human‐induced changes in the composition of the atmosphere but cannot be explained by ESM simulations that account for an early‐industrial radiative forcing. At the regional scale, trends in spatially co‐occurring hot annual maxima are in general coherent among observational products and can in most cases be attributed to human influence on the climate system. Trends in spatially co‐occurring wet annual maxima show differences in some regions, highlighting the importance of a multi‐dataset approach to overcome observational product dependencies. Despite observational uncertainty, we find robust detection and attribution results for many regions. These results can complement previous assessments on regional exposure to hot and wet events from the new IPCC AR6 report.


Introduction
Recent years were characterized by widespread spatially co-occurring hot or wet extreme events across various regions worldwide (NOAA, 2021;WMO, 2022) leading to questions about the contribution of human-induced global warming to the changing likelihoods of such extreme years.Spatially compounding climate extremes are defined as the co-occurrence of hazards in multiple locations (Vogel et al., 2019;Zscheischler et al., 2020), whereby such events can strongly amplify global and regional socio-economic impacts (Vogel et al., 2019).For instance, the increasingly inter-connected global food system is becoming more vulnerable to production shocks (Fraser et al., 2013;Gaupp et al., 2019;Levermann, 2013;Puma et al., 2015) due to a climate change-driven increase of hot and wet extremes affecting multiple breadbasket regions during a growing season (Gampe et al., 2021;Gaupp et al., 2019;Haqiqi et al., 2021;Kornhuber et al., 2020;Sarhadi et al., 2018;Seneviratne et al., 2021).Here it is important to note that such extreme events do not have to occur at exactly the same time in order to impact the underlying system.For example, extreme precipitation might damage a years' worth of harvest in a large region, irrespective of the exact timing of the events within the growing season.Moreover, spatially compounding events can amplify socio-economic impacts also on smaller scales.Referring to the example of food security, regions recognized as a climate "hot spot" might have to face an increase in agricultural import dependence due to an increase in concurrent agricultural land being affected by climate extremes (Mouël et al., 2018).
While certain dynamic patterns can lead to an increased probability of concurrent extremes in food regions (e.g., Kornhuber et al., 2020), human-induced climate change can further amplify the resulting temperature extremes through thermodynamic processes (Wehrli et al., 2020).When assessing the contribution of humaninduced global warming to the changing likelihoods of climate extreme events, heat extremes have been the main subject of attribution research in recent decades (e.g., Eyring et al., 2021;Seneviratne et al., 2021).This is in part the case because the signal-to-noise ratio of changes in heat extremes is particularly strong compared to trends in some other extreme events (Seneviratne et al., 2021).For trends in wet extremes, several studies also investigate their attribution to anthropogenic climate change at regional to global scales, with results generally showing a dominant signal toward increases at global scale (Seneviratne et al., 2021), but partially inconsistent results on regional scales (e.g., Chagnaud et al., 2023;H. Chen & Sun, 2017;de Vries et al., 2023;Dey et al., 2018;Dong et al., 2022;Easterling et al., 2016;Giorgi & Ciarlò, 2022;Guo et al., 2022;Hoerling et al., 2021;Kirchmeier-Young & Zhang, 2020;Li et al., 2017;Ma et al., 2017;Mukherjee et al., 2018;Otto et al., 2013;Paik et al., 2020;Seneviratne et al., 2021;C. Sun et al., 2021;Q. Sun et al., 2022;Xu et al., 2022).
Studies investigating the attribution of changes in spatially co-occurring hot or wet events are more limited.A recent study investigated the attribution of concurrent hot extremes during the 2018 summer (Vogel et al., 2019).However, to our knowledge, there have not yet been any studies investigating the attribution of spatially cooccurring hot events at regional scales or the attribution of spatially co-occurring wet extremes in general.
The IPCC AR6 (Seneviratne et al., 2021) assessed based on the available literature that there is high confidence that "the land area affected by concurrent extremes has increased" and that "concurrent extreme events at different locations, but possibly affecting similar sectors (e.g., breadbaskets) in different regions, will become more frequent with increasing global warming, in particular above +2°C of global warming."However, no specific assessments were provided for specific extreme types.In particular, the IPCC AR6 Chapter 11 noted that "very few studies investigate which types of concurrent extreme events could occur under increasing global warming" (Seneviratne et al., 2021).
We investigate changes in spatially compounding hot or wet annual maxima on global and regional scale characterized by annual trends of Concurrent land Area Affected by Hot annual maxima (in the following referred to as CAA-H) or by Wet annual maxima (in the following referred to as CAA-W).Observed trends are put into the context of Earth System Model (ESM) simulations accounting for historical and early-industrial climate conditions in a trend detection and attribution setting.To account for observational uncertainties in the representation of extreme temperature and precipitation (AghaKouchak et al., 2011;Easterling et al., 2016), we compare different observational products.Results are synthesized in order to robustly detect and attribute signals despite observational uncertainties.We first investigate the trends on CAA-H and CAA-W on a global scale.In a subsequent analysis we divide the globe into subregions defined by the IPCC AR6 report (referred to as AR6 regions) to investigate regional differences in the trends of CAA-H and CAA-W.This allows an investigation of regiondependent contributions to the global signal and sheds light on possible regional socio-economic impacts.The findings in global and regional CAA-H or CAA-W are compared to recent published literature for different extreme indicators focusing on overall trends in temperature or precipitation extremes on regional scale rather than affected area.The purpose of this study is therefore to extend existing literature with the perspective of global and regional spatially co-occurring hot or wet extremes, and to complement the previous assessments on regional CAA-H or CAA-W.

Observational Data
For each extreme event type, two products from each of the respective observational categories, corresponding to reanalysis products and quasi-global gridded station observations products, are used to assess the annual trends in concurrent extreme event area for the period 1981-2018.The data sets are interpolated to a common 2.5°× 2.5°g rid using second-order conservative remapping to facilitate comparison with ESM simulation data.

Temperature Data
Hourly temperature data from the ERA5-Land global reanalysis product (Muñoz-Sabater et al., 2021)

Precipitation Data
Hourly precipitation data from the ERA5-Land global reanalysis product (Muñoz-Sabater et al., 2021) and the MERRA-2 global reanalysis product (Global Modeling and Assimilation Office, 2015) are resampled to daily total precipitation sums.The annual maximum precipitation Rx1day is extracted for each data set, gridcell and year.
The Rx1day variable of HadEX3 (Dunn et al., 2020) and GHCNDEX (Donat et al., 2013) are used in order to account for gridded data derived from station observations.The HadEX3 and GHCNDEX data sets do not have a global coverage and for each timestep within each data set coverage differs.To generate a time-dependent mask, grid cells for which more than 3 time steps (years) are missing are not included in the analysis and the land area fraction which is affected by wet annual maxima is set relative to the available grid cells.

ESM Simulations
The observed changes in concurrent land area affect by hot annual maxima and wet annual maxima are put into context of simulations from 35 Earth System Models (ESMs) contributing to the Scenario Model Intercomparison Project ScenarioMIP (Eyring et al., 2016;O'Neill et al., 2016) of CMIP6.All available climate models' and ensemble members' historical simulations (1850-2014) (Eyring et al., 2016) and future simulations (2015-2100) under the highest emission scenario (Shared Socioeconomic Pathway 5-8.5 SSP5-8.5)(O'Neill et al., 2017) are used.For detailed information on the considered simulations see Table S1 in Supporting Information S1.For all model and simulation runs daily maximum temperature and daily total precipitation are analyzed.All CMIP6 data are prepared through a centralized pre-processing (Brunner et al., 2020), that ensures consistency and includes interpolation to a common 2.5°× 2.5°grid.The period 1851-1888 reflects early-industrial climate conditions (EIND) and thus a climate without anthropogenic forcing and the period 1981-2018 reflects the current climate under anthropogenic forcing (HIST).

Study Regions
We consider the global land area (excluding Antarctica and Greenland) and the reference regions defined for the AR6 (in the following referred to as AR6 regions, Figure 1) (Iturbide et al., 2020;D. Chen et al., 2021).

Yearly Times Series of Concurrent Land Area Being Affected by Hot and Wet Annual Maxima
For each grid cell, hot annual maxima are calculated based on the 5-day moving average temperature and the subsequent extraction of its annual maximum.Subsequently, we calculate the 90th percentile climatology of the extracted maxima for the reference period 1981-2010.Years in which the annual maximum 5-day temperature exceeds the threshold are defined as years in which hot annual maxima occurred.Hot annual maxima are calculated for the summer months.For grid cells within tropical latitudes (≤23.5°)hot annual maxima are calculated based on all 12 months of each year, whereas in extratropical latitudes (>23.5°)hot annual maxima are identified from the period June-August in the Northern Hemisphere and from the period December-February in the Southern Hemisphere.The percentile climatology as well as the occurrence of hot annual maxima is calculated for each gridcell of the observed and simulated data.
For each grid cell, wet annual maxima are defined based on the annual maximum 1-day precipitation.We calculate the 90th percentile climatology of the extracted maxima for the reference period 1981-2010.Years in which the annual maximum 1-day precipitation exceeds the threshold are defined as years where wet annual maxima occurred.Wet annual maxima are calculated based on the whole year.The percentile climatology as well as the occurrence of wet annual maxima is calculated for each gridcell of the observed and simulated data.
To characterize spatially co-occurring hot or wet annual maxima we define the metric of Concurrent land Area Affected by Hot annual maxima (CAA-H) or Wet annual maxima (CAA-W).The relative land area that is concurrently affected by a hot annual maximum or a wet annual maximum is obtained by dividing the areaweighted mean of affected grid cells by the global land area or the AR6 regions.The annual concurrent area impacted by the studied event type is then defined as the annual spatial extent of locations that experience a hot or wet annual maximum.

Trend Estimation
Trends in observed and simulated time series of relative concurrence of extremes are estimated based on 5-year means using Sen's slope (Sen, 1968).Observed trends are computed for the 1981-2018 time window.The quinquennial trends of CMIP6 simulations are calculated for the period 1851-1888 reflecting early-industrial climate conditions (EIND) and for the period 1981-2018 reflecting the current climate under anthropogenic forcing (HIST).

Trend Detection and Attribution
The methodology used for the detection and attribution of trends in CAA-H or CAA-W is based on the definition of climate change detection and attribution (G.C. Hegerl et al., 2006), whereas "Detection" refers to "the process of demonstrating that an observed change is significantly different (in a statistical sense) from natural internal climate" (G.C. Hegerl et al., 2006) and "Attribution" is the "Demonstration that the detected change is consistent with simulated change driven by a combination of external forcings, including anthropogenic changes in the composition of the atmosphere and internal variability" (G.C. Hegerl et al., 2006).We note that formal detection and attribution studies often also test trends against historical natural forcings that exclude human influence on the climate system.In our study we refrain from testing against historical natural forcings and restrict the testing to the early-industrial historical forcing, since our methodology relies on data-driven estimates of distributions representing internal climate variability.This means that we are formally not able to isolate the influence of natural radiative forcings (e.g., volcanic eruptions).
To assess if observed trends can be sufficiently explained by EIND simulations or if human influence on the climate system needs to be accounted for we adapt the approach of Gudmundsson et al. (2022) who put observed trends into the context of empirical distributions derived from an ensemble of climate models simulations.Three conditions must be fulfilled to suggest a detection and attribution.
The first condition refers to the "Detection" process, where it needs to be demonstrated that (I) the observed trend is unlikely given the empirical trend distribution implied by the EIND simulations.The likelihood of the observed trend given the early-industrial climate conditions is quantified using approximate probabilities (p) defined as the fraction of simulated trends under early-industrial climate conditions that are smaller than the observed trend.The probabilities are categorized according to the calibrated language of the IPCC (D. Chen et al., 2021), whereas p > 0.9 indicate a very likely, p > 0.95 indicate a extremely likely and p > 0.99 indicate a virtually certain detection.We also add a probability of p > 0.66 indicating a likely detection according to the IPCC calibrated language in order to show emerging signals.In the subsequent analysis however condition (I) is only considered to be fulfilled in regions where trends are at least very likely (p > 0.9) detectable.The Supporting Information S1 files contain figures related to at least likely (p > 0.66) detectable trends.
If condition (I) is met, the data are investigated regarding a second and third condition, referring to the "Attribution" process.To claim attribution we require that (II) the observed trend is consistent with the HIST simulations and that (III) the expected HIST trend is significantly different from the EIND trends.Here, the term consistency refers to the observed trend lying within the 5th to 95th percentile range of the empirical trend distribution of HIST.To check whether the mean of HIST is significantly different from EIND the following formula is applied: whereas EIND (i) refers to the empirical trend under early-industrial climate conditions of the ensemble member i and HIST refers to the expected value (mean) of the empirical trend distribution under external forcing.The probabilities are estimated as the fraction of models being greater than 0. Probabilities greater than p > 0.9 indicate an attribution result, whereas if the probability is less than p < 0.9 no attribution result can be found.Similar to the detection process, the Supporting Information S1 contains figures referring to the attribution statements based on a probability of greater than p > 0.66.

Regional Synthesis and Comparison to Recent Published Literature for Overall Trends
We synthesize the results on regional scale in Section 5, highlighting both when the results from the different considered data sets converge and where there is evidence of divergence.Results regarding trends in the area affected by concurrent extremes drawn from the different observational products are synthesized based on the following criteria: Inconsistent evidence Data sets do not agree on the direction of the detectable trend

No evidence
No data set suggests a detection or attribution result Emerging evidence Data sets do not disagree on the direction of the detectable trend and at least one data set suggests a detection or attribution result

Strong evidence
All data sets suggest a detection or attribution result In this synthesis, we also compare the evidence on the attribution of concurrent extremes at regional scale with the IPCC sixth assessment report (AR6) Chapter 11 (Seneviratne et al., 2021) (hereafter also referred to as AR6Ch11) regarding overall trends in the respective IPCC regions.Note that we do not necessarily expect detection and attribution results regarding regional trends in the concurrent area affected by hot or wet annual maxima to be identical to those provided for overall trends over the whole region for that extreme.
Based on a semi-systematic literature review, we examine in more detail regions in which (a) our results show an inconsistency in detectable trends based on the different data sets investigated and which have been subject to recently published literature, (b) we could gain new evidence in comparison to the AR6Ch11 and which have been subject to recently published literature, (c) the detectable trends from the current analysis indicate a different tendency than the AR6Ch11 assessment for overall trends or also event attribution.The semi-systematic literature review contains peer-reviewed articles published after the AR6 editorial deadline of 31 January 2021.We selected publications dealing with trend estimates or trend detection and attribution of hot or wet extremes, which were published on the "Web of Science" platform until 30 April 2023.A list of the literature used within this contextualization, the indicators, data sets and time periods investigated in the selected studies can be found in Tables S2 and S3 in Supporting Information S1.The selection of publications for this review does not mean to be completely exhaustive.The synthesis is a rough contextualization of the results with the existing literature instead of a direct comparison, in particular because we look at a new metric (areal trends) which was not assessed in the AR6Ch11 or recent published literature.Note that studies considered in AR6Ch11 and the more recent published studies cover very different time periods, spatial scales and indicators and are based on different methods and data sets.The AR6Ch11 also considers both publications related to event and trend attribution.In our analyses, on the other hand, we limit our consideration to trends in the area affected by hot or wet annual maxima.Finally, in our contextualization based on the literature review, we consider only studies related to mean trend attribution, not event attribution.

Global Trends in the Land Area Affected by Hot Annual Maxima and Wet Annual Maxima
Within the period 1981-2018 the global concurrent land area affected by hot annual maxima (CAA-H; Figure 2a, left) increases steadily on a global scale, independent of the selected observational data set.The global CAA-H averaged over the last five observed years spans from 18% (BEST) to 25% (CPC).The observed trend in global CAA-H (Figure 2b, left) is estimated between 3.3% (BEST), 3.6% (MERRA-2), 5.3% (ERA5-Land) and 5.5% (CPC).Simulated EIND trends of global CAA-H spread around zero.
The right column of Figure 2a shows the global concurrent land area affected by wet annual maxima (CAA-W) between 1981 and 2018 for different observational products.Despite large differences in year-to-year variability, all products show an overall increase in global CAA-W.The global CAA-W averaged over the last five observed years spans from 15% (HadEX3) to 19% (MERRRA-2).Observed and simulated global trends of CAA-W are shown in the right column of Figure 2b).The observed trend in global CAA-W is estimated between 1% (ERA5-Land), 1.4% (HadEX3), 1.6% (MERRA-2) and 2.1% (GHCNDEX).Simulated EIND trends of global CAA-W spread around zero.
Earth's Future Figure 3 (middle column) displays the likelihood of trend detection in concurrent land area affected by hot annual maxima (CAA-H), that is, the approximate probability of the observed trends given the trends in the EIND ensemble.For most regions a detection result can be found.Independent of the selected data set, most observed trends are virtually certain (p > 0.99) larger than the simulated EIND trends.The reanalysis data sets suggest a 95% likelihood of detection over WSB.ERA5-Land and MERRA-2 agree on very small non-detectable trends over TIB, SAS and CAU.Trends over ENA, TIB and SEAF estimated based upon BEST are not detectable.Based on CPC, the estimated trend over NWS is also not detectable.
Figure 3 (left column) displays the attribution results for regions, in which the observed trends are very likely (p > 0.9) not explained by internal climate variability.An attribution is suggested if at least 90% of the investigated HIST simulations are different from the simulated EIND simulations.Of the 43 considered regions 36 (ERA5-Land and MERRA-2), 37 (BEST) and 38 (CPC) regional trends are explainable when the forcing accounting for anthropogenic emissions is taken into account.Within the remaining regions, either no detectable trends were estimated or the observed trend in CAA-H does not agree with the simulated HIST trends.In these regions the observed trend is mainly weaker than simulated HIST trends, except for EAU where the observed trend is stronger than the simulated trends.The analysis referring to a likelihood of detection and attribution of at least 66% is displayed in Supporting Information S1.

Concurrent Land Area Affected by Wet Annual Maxima
Figure 4 (left column) shows both decreasing and increasing decadal trends in regional CAA-W, depending on the region and the data product.Independent of the investigated data set, negative trends in CAA-W are much weaker than positive trends.In many regions the data products agree on the direction of trend.However, in some regions there are inconsistencies in the direction of trends (i.e., over Europe and Asia).There is a tendency that the South American and African tropics show the highest increases in the CAA-W of up to 8.3% per decade based on the reanalysis products.In addition, the east coast of North America and some Asian regions are particularly affected by an increase in the CAA-W.Figure 4 (middle column) displays the likelihood of trend detection in concurrent land area affected by wet annual maxima (CAA-W), that is, the probability that the observed trend cannot be explained by EIND variability.Whereas the reanalysis data sets suggest very likely (p > 0.9) a detection of change in CAA-W for many regions, the station-based data sets suggest emerging trends (p > 0.66) over many regions.Independently of the investigated data set, an increase in CAA-W can be detected over most regions of North America with a few regional exceptions.GHCNDEX suggests an emerging negative trend (p > 0.66) over CNA.Over NWN a negative trend can be at least extremely likely (p > 0.95) detected, independently of the selected data set.Over Central America (SCA and CAR), the reanalysis data sets report a detectable increase in CAA-W.Over South America the direction of detectable trends depends on the observational data set.While MERRA-2 detects virtually certain (p > 0.99) positive trends over whole South America, ERA5-Land (HadEX3) diverge from the trend direction over SES and SWS (SAM and SWS).HadEX3 categorizes the decreasing trends over Central South America as emerging signals.Over Europe mixed detectable signals can be seen.Whereas over MED the data sets agree on a virtually certain (p > 0.99) detectable positive trend, the data sets disagree over the rest of Europe.The reanalysis data sets report a virtually certain (p > 0.99) detectable negative signal over WCE and NEU, whereas the stationbased data sets report increasing trends in CAA-W over these regions.HadEX3 however, suggests only emerging trends (p > 0.66) over WCE and NEU whereas the likelihood of detection based upon GHCNDEX is higher.Over EEU all data sets agree on an increase in CAA-W, except for ERA5-Land which suggests a virtually certain (p > 0.99) decrease in CAA-W.The estimated trends based upon the station-based data sets however can be only categorized as emerging signals (p > 0.66).All available data sets, report increases in CAA-W over the African continent, with high likelihoods of detection (p > 0.99).Over the Asian continent, the direction of detectable trends is regionally scattered and highly differ between the investigated data sets.Only regarding ESB and SEA (WCA) the available data sets agree on an increase (decrease) in CAA-W, suggesting however different Earth's Future 10.1029/2023EF004114 BIESS ET AL. likelihoods of detection.In these regions, the reanalysis data sets suggest in general a higher likelihood of detection whereas the station-based data sets categorize the change in CAA-W often only as emerging signals.
Over the Australasian region the investigated data sets mainly agree on increasing detectable trends in CAA-W over NAU and SAU, whereas they differ over CAU and EAU.
Figure 4 (right column) displays the attribution results for regions, in which the observed trends are very likely (p > 0.9) not explained by internal climate variability.An attribution is suggested if at least 90% of the investigated HIST simulations are different from the simulated EIND simulations.It can be seen, that only detectable positive trends in CAA-W can be attributed to human-induced climate change.Based upon the reanalysis data sets, within more than half of the regions, the change in CAA-W can be attributed to human-induced climate change.Within the remaining regions, either the simulated trends under EIND and HIST forcing do not differ strongly enough, or the observed trend does not agree with the simulated HIST trends.In these regions the observed trends are generally stronger than the simulated trends.Regarding the station-based observational products, only over NEU the signal is attributable to human-induced climate change based upon GHCNDEX.HadEX3 reports no attributable trends over any region.Within the regions where the station-based observational products did not suggest an attribution of the signal, the simulated trends under EIND and HIST forcing do not differ strongly enough.

Synthesis of Regional Trends in Concurrent Extremes and Contextualization With Recent Literature for Overall Trends
In this chapter, the findings regarding global and regional CAA-H and CAA-W are compared to the AR6Ch11 and recent published literature.On a regional scale, the AR6Ch11 and recent published literature did not assess changes in concurrent extremes but focus on overall trends on regional scale.However, we can expect changes in the area affected in a given type of extreme to have a similar sign as that identified for overall trends or event attribution of single events.We restrict the contextualization of regional results to those regions, where (a) our results show an inconsistency in detectable trends based on the different data sets investigated and which have been subject to recently published literature, (b) we could gain new knowledge in comparison to the AR6Ch11 and which have been subject to recently published literature, (c) the detectable trends from the current analysis indicate a different tendency than the AR6Ch11 assessment for overall trends or also event attribution.

Contextualization of Trends in Hot Extremes
Our results show that human influence is a contributor to the observed increases in concurrent area affected by hot annual maxima (CAA-H) on a global scale, as also concluded in the AR6Ch11, which assessed that is extremely likely (p > 0.95) that "human influence is the main contributor to the observed increase in the intensity and frequency of hot extremes" (Seneviratne et al., 2021).For temperature signals, the AR6Ch11 reports detectable and attributable trends in most regions, except for CAF, CNA, ENA, and SSA (Seneviratne et al., 2021) (Figure 5).In the following paragraphs, the results from the quantitative analysis above are further discussed below and put into context of recently published research (Figure 6).
In our study, trends in CAA-H are detectable and attributable over CAF which is supported by recent published literature (Engdaw et al., 2023) (see Figure 6).In SSA, CNA and ENA, positive trends in CAA-H are detectable and attributable to human-induced climate change, which is supported by some studies (e.g., Engdaw et al., 2023;Rusticucci & Zazulie, 2021).In other studies, regional detection and attribution results over the Americans depend upon the spatial scale of investigation.Seong et al. (2022) detected and attributed trends in daily maximum temperature to anthropogenic forcing over the North American and South American continent.On a higher spatial scale however, they did not detect a trend for CNA, ENA or SSA (see Figure 6).

Contextualization of Trends in Wet Extremes
Our results show, that human influence is a contributor to the observed increases in CAA-W on a global scale, as also assessed in the AR6Ch11, which concludes that "detection and attribution analyses have provided consistent and robust evidence of human influence on extreme precipitation of 1-and 5-day durations at global to continental scales" (Seneviratne et al., 2021).However, "evidence of human influence on extreme precipitation at regional scales is more limited and less robust" (Seneviratne et al., 2021).The following paragraphs and Figure 7  contextualization of the trend detection and attribution results of regional CAA-W drawn from our study with the AR6Ch11 and recent published literature which focused on regional overall trends in precipitation extremes.
In our study, trends in CAA-W are detectable and attributable over WAF which coincides with Chagnaud et al. (2023), who found that a positive trend in extreme daily precipitation is well reproduced by historical simulations incorporating anthropogenic forcing for the West-African Sahel (see Figure 7b).Over many Asian regions the AR6Ch11 reports with at least medium confidence a detectable increasing overall trend in wet extremes (Seneviratne et al., 2021).In our study detection and attribution results highly depend on the selected observational product.Recent literature focusing on the detection and attribution of wet extremes over the Asian region also depend on the selected data set and the regionalization of results (see Figure 7a).Recent studies report on a mainly increasing trend in precipitation extremes over South-East (SE) and Western (W) China (comparable to ECA, TIB and southern parts of EAS) and mainly decreasing trends from South-Western (SW) Figure 5. Synthesis of detection and attribution results drawn from the investigated observational products and results reported in the AR6Ch11.Colors represent the synthesized results drawn from our study.Regions are colored according to "Emerging evidence" if at least one investigated data set suggests at least a very likely ( p > 0.9) detection (i.e., the trend is very likely not caused by internal climate variability) or attribution (i.e., at least 90% of the investigated HIST simulations are different from the EIND simulations).Regions are colored according to "Strong evidence" if all investigated data sets suggest at least an very likely ( p > 0.9) detection or attribution.Hatched areas refer to regions where the considered observational products disagree in the direction of detectable trends.The dots refer to the confidence in detection and attribution results regarding overall trends reported in the AR6Ch11 for the respective region.
China to North-Eastern (NE) China (similar northern and north-eastern parts of EAS) (Dong et al., 2022;Guo et al., 2022;C. Sun et al., 2021;Q. Sun et al., 2022;Xu et al., 2022).While an increase in precipitation extremes over SE and NW China is attributable to human-induced climate change (Guo et al., 2022), models used in the studies do not reproduce the observed decreases in precipitation extremes over SW and NE China (Guo et al., 2022;C. Sun et al., 2021;Xu et al., 2022).Similar to our study, Q.Sun et al. (2022) did not differentiate between the drying Northern part of EAS and the wetting Southern part of EAS and thus report no evidence for a attributable signal over EAS, since there is less consistency between observations and model simulations and larger internal variability in the investigated region.Over ECA (WCA) our synthesis suggests a detectable decrease in CAA-W based upon the reanalysis data sets (reanalysis data sets and HadEX2, respectively).This suggestion diverges from a medium confidence in an increase in precipitation extremes over ECA and WCA reported in the AR6Ch11 (see Figure 7c).For ECA AR6Ch11 builds upon the study from Q. Sun et al. (2021) which uses HadEX2 and GHCNDEX data to detect signals in RX1day and RX5day over ECA.Recent published literature build their analysis on station-based observational data sets and agree therefore in detectable and partly attributable increases in precipitation extremes over ECA (see Figure 7c).Over WCA studies referred to in the AR6CH11 assessment agree on detectable increases in precipitation extremes over WCA based upon meteorological station-data.Q.Sun et al. (2021) however show an emerging but non-significant decrease in precipitation extremes over WCA based upon HadEX2 and GHCNDEX.
Over the Australasian region new insights could be gained for SAU, where our study suggests a positive detectable and attributable trend.Q.Sun et al. (2022) detected a mixed pattern of positive and negative trends in Rx1day over SAU, which could not be attributed to human influence (see Figure 7a).
Over Europe the AR6Ch11 reports with at least medium confidence an overall increase in wet extremes, which is attributable to anthropogenic forcing over NEU (Seneviratne et al., 2021).Our study shows, that an increase in CAA-W over NEU, WCE and EEU is detectable and attributable to human-induced climate change only based upon the station-based observational products (except for EEU, where MERRA-2 also reports on positive trends).Regional overall trend directions of wet extremes over WCE and EEU assessed in recent literature also depend very much on the selected data product (see Figure 7a).Studies relying on observation-based data sets.also report an increasing trend in wet extremes over NEU, WCE and EEU (de Vries et al., 2023;Q. Sun et al., 2022), whereas studies relying on reanalysis data sets show a more spatial heterogeneous pattern of trends in precipitation extremes (Giorgi & Ciarlò, 2022).-W; (c) findings drawn from our study diverging from the AR6 assessment in the direction of detectable trends.Results drawn from our study are contextualized with recent published literature focusing on regional overall trends in precipitation extremes."T.D."/"D"/"A" refers to "Trend Direction"/"Detection"/"Attribution."Note that in (a) the highest level of confidence assigned to the data products investigated in our study is "Emerging evidence/Medium Confidence" since the investigated products show inconsistencies among themselves.The level of evidence in detection and attribution results of recent published literature is assigned as follows: Limited evidence/Low confidence is assigned if the recent literature source does not suggest a detection or attribution result; Strong evidence/High confidence is assigned if the recent literature source suggest a detection or attribution result.2023), 8) Giorgi and Ciarlò (2022); "N" refers to northern part, "NE" refers to north-eastern part, "SW" refers to south-western part and "SE" refers to south-eastern part of the investigated region.The AR6Ch11 reports with high confidence an regional overall increase in wet extremes over CNA.The trend in CNA is with medium confidence attributable to human influence (Seneviratne et al., 2021).Over CNA no robust detection and attribution result for trends in CAA-W can be found within our study, since the data sets used in our study disagree in the direction of detectable trends among each other.Whereas ERA5-Land, MERRA-2 and HadEX3 agree on a detectable positive trend in CAA-W over CNA, GHCNDEX contradicts these findings.Based on ERA5-Land the increase in CAA-W is attributable to anthropogenic forcing, which coincides with the results from Q. Sun et al. (2022) regarding regional overall trends in precipitation extremes over CNA (see Figure 7a).In our study, a negative (positive) trend in CAA-W over NWN (WNA, respectively) is detectable and attributable to human influence.Q.Sun et al. (2022) does not report on detectable and attributable trends in precipitation extremes over these regions (see Figure 7b).

Detection, Attribution, and Process Understanding
In recent years, significant advances have been made in refining our understanding of human-induced climate change (IPCC, 2021) and climate change detection and attribution techniques have played a central role in this effort (Eyring et al., 2021).Detection and Attribution techniques are built to test the hypothesis that changes in the climate system are not consistent with internal climate variability and to check if an emerging signal is consistent with changes in the radiative forcing (e.g., due to greenhouse gas emissions).However, while these techniques rely on climate model simulations and are therefore deeply rooted in physics, they are not designed for expanding our knowledge of the physical drivers.Instead, overall physical reasoning can help to understand the emerging patterns, thereby increasing the confidence in the empirical results.In the context of the present study, changes in spatially compounding climate extremes can be associated with thermodynamic drivers as well as with dynamic drivers related to changes in circulation patterns (Kornhuber et al., 2019;Rogers et al., 2022;Shepherd, 2014).Overall, there is high confidence that anthropogenic forcing directly affects thermodynamic variables.The direct thermodynmaic effect leads to warmer temperatures, increasing the frequency and intensity of warm extremes (Seneviratne et al., 2021).Initial increases in temperature are followed by other thermodynamic responses and feedbacks, such as the soil moisture-temperature feedback amplifying temperature extremes in mid-latitude areas (Seneviratne et al., 2010) or the water vapor feedback (Seneviratne et al., 2021).The water vapor feedback refers to an increase in the water vapor content of the atmosphere, impacting the frequency and intensity of precipitation extremes (Seneviratne et al., 2021).Following the Clausius-Clapeyron (C-C) relationship, water vapor increases by approx.7% for every degree Celsius of global warming (Douville et al., 2021), enabling more abundant and intense heavy precipitation events around the globe (Martinkova & Kysely, 2020).At regional scales, water vapor increases differ from this C-C rate due to dynamical processes (Douville et al., 2021).However, there is less confidence in how dynamic changes affect the location and magnitude of extreme events in a warming climate (Seneviratne et al., 2021).Recent studies highlight the impact of changes in dynamical patterns on spatially compounding climate extremes.For example, Röthlisberger et al. (2019) and Kornhuber et al. (2019) identified synoptic-scale recurrent Rossby waves resulting in simultaneous warm spells in different locations.Further, Ding and Wang (2005) note a relationship between anomalous air temperatures in the Northern Hemisphere summer and circumglobal teleconnections.

Technical Choices and Limitations
While considerable progress has been made in understanding human-induced climate change, challenges remain in detecting and attribution applications, which are related to applying detection and attribution techniques on higher spatial scales, detecting and attributing climate extremes (Easterling et al., 2016) and not at least the detection and attribution of changes in hydrological extremes (Eyring et al., 2016).We aimed to address these major challenges in regional extreme temperature and precipitation detection and attribution.We are able to robustly detect and attribute regional temperature extremes.However some regional inconsistencies between simulated and observed temperature extremes have been highlighted.While our results and recent studies agree on regional detection and attribution of temperature extremes, bigger differences and inconsistencies remain when it comes to detecting and attributing precipitation extremes.Confidence in temperature and precipitation detection and attribution findings are influenced by observational and model data availability and reliability and possible fluctuation from natural internal variability (Sarojini et al., 2016;Stott et al., 2010;van Oldenborgh et al., 2021).These components will be addressed in the following paragraphs.

Methodological Considerations
In our study we put observed trends in context with early-industrial forcing, since our methodology relies on datadriven estimates of distributions representing internal climate variability.Formal detection and attribution studies (e.g., G. Hegerl & Zwiers, 2011;G. C. Hegerl et al., 2006) often test against natural historical forcings.In our study we refrain from testing against natural historical forcing and restrict the testing to the early-industrial historical forcing.This means that we are not able to isolate the influence of natural radiative forcings (e.g., volcanic eruptions).However, temperature and precipitation extremes were attributed to anthropogenic forcing in the light of strong evidence (e.g., Eyring et al., 2021;Kim et al., 2015;Min et al., 2011).Stott et al. (2010), Sarojini et al. (2016), andAlexander (2016) highlight the limited confidence in regional attribution assessments due to the large effect of internal variability resulting from low-frequency variations in atmospheric circulation as well as due to a limited sample size of rare climate extreme events.Thus, the primary analysis of this paper focused on the global land mean trend, intending to increase the signal-to-noise ratio.We used the full ensemble of opportunity within the CMIP6 archive in order to (a) account for the effect of internal variability due to a high number of initial condition ensemble members and (b) to address the potential sensitivity to finite samples due to the applied empirical trend detection and attribution approach.Whereas the ensemble of opportunity within the CMIP6 archive puts more weight on models with more initial condition ensemble members, we intended to increase the sample size used to empirically estimate distributions used for the empirical trend detection and attribution approach.Thus we trade a possibly biased ensemble with possibly biased assumptions on statistical models used in theoretically more sophisticated detection and attribution approaches (Allen & Tett, 1999;Allen & Stott, 2003;G. C. Hegerl et al., 1996;Gudmundsson et al., 2022;Ribes et al., 2016).

Observational Uncertainty
Observations are our primary source of information about climate change.Long-term observations provide the necessary data to understand key climate processes and our rapidly changing climate.Nevertheless, significant uncertainties in the representation of extreme daily temperature and precipitation exists (AghaKouchak et al., 2011).By considering different observational products within this study, we aim to address observational uncertainties and different advantages and disadvantages of observational product categories.Observational products can be classified into three main categories including gauge-based, satellite-related and atmospheric reanalysis data sets (Nogueira, 2020;Q. Sun et al., 2018).The estimated climate variables are not completely consistent among the different product categories due to their different data sources, estimation procedures and quality control schemes (Easterling et al., 2016;Q. Sun et al., 2018;Wehner et al., 2014).
The reliability of station-based products depend on the density of the station network, the homogeneity of station data, the interpolation algorithms employed, the quality control procedures and the manner in which data coverage that changes over time is handled (Donat et al., 2013;G. C. Hegerl et al., 2015;Q. Sun et al., 2018).Sources of uncertainty in atmospheric reanalysis products include errors in numerical models, errors in observations, data assimilation challenges, and heterogeneity of data sources in space and time (Nogueira, 2020).Even though the number of precipitation stations has increased more than that of temperature stations, the improvements in coverage for precipitation extremes are less impressive (Alexander, 2016), leading to regions affected by high observational sparseness (see Figure 4), making temperature data generally more reliable than precipitation data (Decker et al., 2012;Gleixner et al., 2020;Lindsay et al., 2014).
Both reanalysis products assessed in this study, assimilate surface air temperature from different observational sources.Variables like precipitation are in some regions purely forecasted with no observational input (Gleixner et al., 2020;Lavers et al., 2022;Uppala et al., 2005).For instance, ERA5 does not directly assimilate any raingauge data, although it does assimilate composite radar/rain-gauge precipitation estimated over the United States from the year 2009 onwards (Hersbach et al., 2020).The forcing precipitation in MERRA-2 is primarily based on gauge observations at low and mid-latitudes, but gradually degrades toward a fully modeled precipitation data set over a zonal range from 42.5°to 62.5°latitude (Gelaro et al., 2017).Therefore it is important to be aware of product-dependent disadvantages and advantages, which is also mirrowed in the product-dependent detection and attribution results examined in this study.

Model Uncertainty
For the simulated warming from 1981 to 2018, we extend the CMIP6 historical simulations by the SSP5-8.5 scenario.The warming trend between 2015 and 2018 is very similar across the scenarios (Lee et al., 2021).Attribution studies can highlight differences between models and observations (Stott et al., 2010).Figures 3 and 4 show that CMIP6 models overestimate the concurrent land area affected by hot annual maxima and mainly underestimate the concurrent land area affected by wet annual maxima in several regions.The ability to obtain robust attribution results at the regional scales remains a major challenge, since climate models lack the processes necessary for realistically simulating regional details (Stott et al., 2010).Figure 2 shows a bimodal distribution of the models' trends in concurrent land area affected by hot annual maxima.As Tokarska et al. (2020) note, most models with high climate sensitivity or high transient response overestimate recent warming trends, with differences which cannot be explained by internal variability.Among others, Yazdandoost et al. (2021) analyzed the performance of precipitation estimates from historical runs of CMIP6 over different climatic regions.Yazdandoost et al. (2021) selected Iran as a case study and detected a strong underestimation of precipitation amounts in wet regions.

Conclusion
We applied a simple detection and attribution framework to assess recent global and regional trends in spatially concurrent hot or wet annual maxima taking into account observational uncertainty.Observational data sets agree on the direction and magnitude of increasing trends in CAA-H and CAA-W on a global scale, demonstrating a detectable and attributable increase in the global area affected by these extremes.At the regional scale, trends in CAA-H are in general coherent among observational products and can in most cases be attributed to human influence on the climate system.However, observational data sets differ more when regional trends in CAA-W are assessed.By comparing different observational products within our study and contextualizing our results within a semi-systematic literature review in terms of the observation products used in the literature, a strong observational product dependency became visible.The identified uncertainties for the regional trends in CAA-W are in general consistent with the literature for the overall trends in wet extremes at regional scale.However the specific regions for which uncertainties emerge can be different.We show that there is no general pattern of a specific type of data set outperforming another one, since we are not able to extract any overarching structural or spatial patterns that can explain the discrepancies between the data sets.This highlights the importance of addressing observational uncertainty by using different data products-especially when hydrological variables are investigated.
Overall, there is substantial evidence that on a global scale increasing detectable trends in CAA-H or CAA-W area are attributable to human influence on the climate system.Despite a lower signal-to noise ratio, trends of regional exposure to hot or wet annual maxima emerge in many regions to be robustly detectable and explainable by ESMs when taking human influence on the climate system into account.Independently of the selected data set and variable investigated, it is evident that the southern hemisphere tropical and subtropical regions belong to the most affected regions, showing the strongest trends in CAA-H and CAA-W.Further, pronounced trends in CAA-H and CAA-W appear over many breadbasket regions, including Eastern China, Northern India, Brazil, US Midwest, Northwestern Europe as well as Ukraine and Southern Russia.To avoid a steady evolving climate injustice as well as major impacts associated with human-induced globally occurring hot or wet extreme events, we have to take ambitious mitigation actions to strongly reduce greenhouse gas emissions.

Figure 1 .
Figure 1.Regions defined for the Sixth Assessment Report of the United Nations Intergovernmental Panel on Climate Change (AR6 regions) (D.Chen et al., 2021).

Figure 2 .
Figure 2. (a) Temporal evolution (1981-2018) of annual global concurrent land area affected by hot annual maxima (CAA-H, left) and wet annual maxima (CAA-W, right).(b) Trend Attribution in global CAA-H (left) and CAA-W (right).Observed trends (black lines) are detected if they are not captured by the ensemble of EIND trends (blue histogram).The consistency of the observed trend with the ensemble of HIST trends (red histogram) indicates an attribution to historical radiative forcing.For each event type the four selected observational products are displayed referring to ERA5-Land, MERRA-2, BEST and CPC regarding the investigation of CAA-H and ERA5-Land, MERRA-2, HadEX3 and GHCNDEX regarding the investigation of CAA-W.

Figure 3 .
Figure 3. Regional Detection and Attribution results for concurrent land area affected by hot annual maxima (CAA-H).Left: 1981-2018 observed trends in CAA-H expressed as % decade 1 .Regions are colored, if their trend in CAA-H is very likely ( p > 0.9) not caused by internal climate variability, otherwise they are hatched.Middle: Likelihood of detection, that is, the probability that the observed trend cannot be explained by EIND variability.Right: Consistency of observed trends with simulations driven with historical radiative forcing.Regions are colored if at least 90% of the investigated HIST simulations are different from the EIND simulation.Regions are hatched if the observed trend does not agree with the simulated HIST trend.

Figure 4 .
Figure 4. Same as Figure 3 but for concurrent land area affected by wet annual maxima (CAA-W).

Figure 6 .
Figure 6.Summary of new findings drawn from our study analyzing trends in concurrent land area affected by hot annual maxima (CAA-H) and their contextualization with recent published literature focusing on regional overall trends in temperature extremes."T.D."/"D"/"A" refers to "Trend Direction"/"Detection"/"Attribution."The level of evidence in detection and attribution results of recent published literature is assigned as follows: Limited evidence/low confidence is assigned if the recent literature source does not suggest a detection or attribution result; Strong evidence is assigned if the recent literature source suggest a detection or attribution result.Sources: 1) Engdaw et al. (2023), 2) Seong et al. (2022), 3) Rusticucci and Zazulie (2021).

Figure 7 .
Figure7.Summary of (a) inconsistencies in the direction of detectable trends in concurrent land area affected by wet annual maxima (CAA-W) due to mixed signals drawn from our study; (b) new findings drawn from our study analyzing trends in CAA-W; (c) findings drawn from our study diverging from the AR6 assessment in the direction of detectable trends.Results drawn from our study are contextualized with recent published literature focusing on regional overall trends in precipitation extremes."T.D."/"D"/"A" refers to "Trend Direction"/"Detection"/"Attribution."Note that in (a) the highest level of confidence assigned to the data products investigated in our study is "Emerging evidence/Medium Confidence" since the investigated products show inconsistencies among themselves.The level of evidence in detection and attribution results of recent published literature is assigned as follows: Limited evidence/Low confidence is assigned if the recent literature source does not suggest a detection or attribution result; Strong evidence/High confidence is assigned if the recent literature source suggest a detection or attribution result.Sources: 1) Chagnaud et al. (2023), 2) Guo et al. (2022), 3) Dong et al. (2022), 4) Q.Sun et al. (2022), 5) C. Sun et al. (2021), 6) Xu et al. (2022), 7) de Vries et al. (2023), 8)Giorgi and Ciarlò (2022); "N" refers to northern part, "NE" refers to north-eastern part, "SW" refers to south-western part and "SE" refers to south-eastern part of the investigated region.
Figure7.Summary of (a) inconsistencies in the direction of detectable trends in concurrent land area affected by wet annual maxima (CAA-W) due to mixed signals drawn from our study; (b) new findings drawn from our study analyzing trends in CAA-W; (c) findings drawn from our study diverging from the AR6 assessment in the direction of detectable trends.Results drawn from our study are contextualized with recent published literature focusing on regional overall trends in precipitation extremes."T.D."/"D"/"A" refers to "Trend Direction"/"Detection"/"Attribution."Note that in (a) the highest level of confidence assigned to the data products investigated in our study is "Emerging evidence/Medium Confidence" since the investigated products show inconsistencies among themselves.The level of evidence in detection and attribution results of recent published literature is assigned as follows: Limited evidence/Low confidence is assigned if the recent literature source does not suggest a detection or attribution result; Strong evidence/High confidence is assigned if the recent literature source suggest a detection or attribution result.Sources: 1) Chagnaud et al. (2023), 2) Guo et al. (2022), 3) Dong et al. (2022), 4) Q.Sun et al. (2022), 5) C. Sun et al. (2021), 6) Xu et al. (2022), 7) de Vries et al. (2023), 8)Giorgi and Ciarlò (2022); "N" refers to northern part, "NE" refers to north-eastern part, "SW" refers to south-western part and "SE" refers to south-eastern part of the investigated region.

Regional Trends in Concurrent Land Area Affected by Hot Annual Maxima or Wet Annual Maxima 4.2.1. Concurrent Land Area Affected by Hot Annual Maxima Figure
For both variables and all investigated observational products detection and attribution results can be found.The observed trend in CAA-H or CAA-W is virtually certain (p > 0.99) larger than the simulated EIND trends.Simultaneously, the mean of simulated HIST trends of CAA-H or CAA-W differ virtually certain (p > 0.99) from simulated EIND trends.Conversely, the observed trends in CAA-H or CAA-W are consistent with simulated HIST trends.Thus, the investigated variables show a detectable climate change signal that is consistent with model estimates of human influence on the climate signal.3 (left column) shows observed trends in regional CAA-H.All observational products agree on the strongest decadal trends in regional CAA-H occurring over the northern part of the South American continent (NSA, SAM, NES) and the Arabian Peninsula with an increase in CAA-H of around 10% per decade.The station-based observational data sets display strong trends in regional CAA-H over the Central Asian regions, which is not mirrowed in the reanalysis data sets.