In the virtual pages of Science Advances today appears a study by Tom Matthews, Radley Horton, and me reporting advances in knowledge of the historical record of heat-humidity extremes. In this post I'll discuss some of the key findings, my thoughts about how they fit into the broader scientific context, and the kinds of future work that they help incentivize.
The study is making some headlines probably because the concept of a wet-bulb temperature so high that it physiologically cannot be withstood for prolonged periods of time captures people's imaginations, but we have been careful to point out that the bulk of the work (as well as the novelty) rests in the story about values that are still extreme but decisively below this level. For instance, 33C corresponds to a temperature of 40C accompanied by a dewpoint temperature of 31C, and even 31C (which has occurred about 7,500 times globally over 40 years) exceeds the all-time maximum in the notoriously humid Washington, DC metropolitan area. These values preferentially occur in South Asia and select portions of Mexico, the Middle East, Australia, and East Asia. As published, in fact, the station dataset HadISD contains significantly more such extreme values (including dozens of exceedances of 35C), but we progressively eliminated many of these with additional quality-control measures ranging from temporal and spatial consistency checks (against the station itself, other stations, and reanalysis) to removing any dewpoint temperatures above the widely-reported 35C dewpoint recorded in Dhahran, Saudi Arabia in July 2003. We also dug deep into possible instrumental and observer errors. The result is what we consider to be a conservative estimate of global frequency and intensity -- keeping in mind that many of the stations record at 3- or 6-hourly intervals, rather than hourly, and also the paucity of good-quality observations in many global hotspots such as the Sahel; Iran and Pakistan; and East Africa.
Every threshold value except for the exceedingly rare 35C has experienced at least a doubling in frequency since 1979, underscoring the tight and nonlinear relationship between global mean temperature and extreme heat. This creates an alarming prospect in a world where temperatures are rising rapidly. Expanding the reach of artificial cooling (in the form of air conditioning) serves as an obvious solution in the short term, but looking forward, the inevitability of further increases is a sobering reminder of the importance of both mitigation of the temperature rise, and adaptation to what's already in store. Architecture, urban & regional planning, and national policies around farming and land use could play important roles on the adaptation side, but need to be guided by highly region- and season-specific knowledge in order to be effective. For instance, in South Asia prior to the onset of the summer monsoon, wet-bulb extremes are driven primarily by excessive temperatures (~45C or higher), whereas afterwards they occur in conjunction with greater relative humidity but lower temperatures (more like 37C). This difference could lead to policies that prioritize reducing heat in spring, and reducing humidity in summer, while remaining cognizant of the many constraints on behavioral changes, such as the need of farmers to irrigate at specific times within the growing season.
Global trends in extreme humid heat. (A-D) Annual global counts of TW exceedances above the thresholds labeled on the respective panel, from HadISD (black, right axes, with units of station-days), and ERA-Interim grid points (gray, left axes, with units of grid-point-days). We consider only HadISD stations with at least 50% data availability over 1979-2017. Correlations between the series are annotated in the top left of each panel, and dotted lines highlight linear trends. (E) Annual global maximum TW in ERA-Interim. (F) The line plot shows global mean annual temperature anomalies (relative to 1850-1879) according to HadCRUT4, which we use to approximate each year's observed warming since pre-industrial; circles indicate HadISD station occurrences of TW exceeding 35°C, with radius linearly proportional to global annual count.
This study does not so much increase our knowledge about the future as it does about the past. Specifically, it improves our understanding of the historical baseline and the details of the brief 'spikes' that it contains. While extremes of any variable are more intense when looking at small spatial and temporal scales, we were particularly surprised by the steepness of the horizontal and vertical gradients of wet-bulb temperature in places like the Persian Gulf and Gulf of California. During the most severe events, the dropoff as seen by radiosondes ascending from 995 to 975 mb (150 to 310 m) averages 5C or more! This is consistent with the expectation from the two powerful competing forces that shape extreme heat in these arid regions: a source of moisture, itself at high temperatures, sitting underneath a near-constant large-scale high-pressure system which acts to trap heat and moisture close to the surface. Under the right meteorological conditions (e.g. a continuous onshore flow, plus maybe some other anomalies which future work will need to establish), the sun beats down and moist static energy builds and builds in a shallow boundary layer with no means of escape. The small scales and brief times at which this occurs -- before being dissipated by some modest amount of horizontal or vertical mixing, for example -- make it very difficult for models or reanalyses to capture it with appropriate severity. ERA5 does a much better job than ERA-Interim, but the underestimates still generally exceed 2 deg C. Where the spatiotemporal scales of the extremes are larger, such as in the interior eastern United States, these reanalyses are right on the money.
Meteorological conditions when TW=35°C. (Top) ERA-Interim composite of 10-m winds and 2-m TW on the n=4 days when TW=35°C was recorded at Ras al-Khaimah, UAE (blue square). Resolution of plotted data is 0.5°x0.5° and 6-hourly. (Bottom) Same as top but for ERA5. Resolution of plotted data is 0.25°x0.25° and 1-hourly. Mean TW daily maximum near station location is 28.7°C for ERA-Interim and 30.8°C for ERA5.
Vertical profiles of coastal extreme humid heat. Radiosonde temperature (diamonds) and TW (circles) for Abu Dhabi International Airport, United Arab Emirates at 12Z on all days between 1983 and 2019 with a lowest-level TW value greater than the annual 99.75th percentile (red, 10 days); between the annual 97.5th and 99.75th percentiles (orange, 90 days); between the annual 90th and 97.5th percentiles (green, 298 days); and between the annual 50th and 90th percentiles (blue, 1593 days). Profiles are truncated at 850 hPa for visibility. Vectors on the right-hand side indicate composite wind speed and direction on these days for each height bin, where available; the map on the left-hand side is a 2.5x2.5 box indicating the location of the launch site.
There remains much we do not know about the ingredients and impacts of these exceptionally rare events. How are they affected by irrigation or other land-cover change? By urbanization? By short-term variations in sea-surface temperature? By meteorological conditions such as monsoon progression or passing weather systems? Given the generally larger importance of humidity in producing the 'spikes' that we see, should more attention be focused on limiting extreme humidity rather than extreme temperature? Our paper hints at some hypotheses about these factors, but does not definitively establish or quantify them, especially in terms of how they may work together in varying combinations to affect extreme wet-bulb temperatures in one region versus another.
Ultimately, speaking for Radley and Tom as well as myself, we see our paper fitting neatly within the structure of the running conversation begun by hot-water-immersion tests in the early and mid 20th century, emerging definitively in the climate literature with Sherwood and Huber 2010, and continuing recently with papers by Pal and Eltahir, Im et al., and others. We hope that this robust conversation broadens and continues, evolving such that even as the leading edge of these extremes push beyond levels seen in the Holocene, human ingenuity and compassion find ways to forestall disasters in the near-term, and to develop policies and technologies in the medium- and long-term, that look holistically at this 'wicked' problem and chip away at the socioeconomic vulnerabilities, consumption habits, and development geographies that exacerbate it.
A personal, subjective attempt at summarizing the top climate stories and advances of the decade that has been:
Climate models continued to prove very good at predicting global temperature change from greenhouse-gas forcings, and total emissions continued to track the top-line emissions scenario. Temperature increases led to 8 of the decade's 10 years ranking in the top 10 warmest years since 1850.
A recent review paper looking back at studies from the late 20th century found that even the simpler, coarser-resolution models then in use predicted temperature changes entirely consistent with what has since been observed. The similarity across generations of models gives further confidence to global-average statistics such as mean annual temperature changes. It is sobering but not at all surprising, given the strength of the economic, social, and political status quo, that efforts such as the Paris Agreement of 2015 have not yet made any appreciable dent in the irrepressible upward track of greenhouse-gas *emissions* (not to mention concentrations), and thus that global-average temperature records continue to be broken left and right.
A variety of severe extreme events, often distinguished by their long durations, inspired new efforts to understand and mitigate them.
Several such events made their mark on the arc of history by striking wealthy and/or populated areas, rather than by their geophysical rarity. These included the 2010 Russian heatwave, 2010 Pakistan floods, 2010-11 Queensland floods, 2011 Thailand floods, 2012 Midwest drought and heatwave, 2012 Hurricane Sandy, 2014 and 2015 US Midwest and Northeast cold snaps, 2015 and 2018 European heatwaves, 2017 Hurricanes Harvey, Irma, and Maria, 2017-2019 California wildfires, and ongoing 2019 Australia bushfires. Some were notable more for highlighting to physical scientists aspects of climate variability or change that were previously underappreciated (such as the much larger rainfall amounts associated with slow-moving tropical cyclones, or the mid-latitude effects of Arctic sea-ice melt), while others made headlines for their dramatic economic or ecological effects (such as the vulnerability of international supply chains to floods or storms). The now-ubiquitous Internet, and in particular social media, enhanced the power of some events by making the visual evidence of them compelling and unavoidable. Areas from agriculture to international trade to urban planning were increasingly shaped by the recognition that these kinds of extreme weather and climate events pose major (and in many cases growing) risks which it is imperative to address.
Weather and climate computer models enabled qualitative as well as quantitative improvements in representing the Earth system, across a spectrum from basic research to public-facing operational forecasts.
Probabilistic forecasts of storm surge and fluvial flooding hour by hour and house by house. Continuous global 3-km hourly weather forecasts. Quantifications of how the land surface affects the development of individual severe storms. Near-real-time estimates of the fractional contribution of anthropogenic effects on the characteristics of a natural disaster. Robust partitioning of observed regional climate changes into deforestation, irrigation, urbanization, global greenhouse gases, and major modes of variability. All of these were well beyond the limit of scientific and computational capability before the 2010s, but have now come into their own. A safe bet is that the 2020s will see many more such successes, each of which allows us to see 'around a corner' that had previously been blind. The pipeline from research to operations moves haltingly, but on a decadal basis the progress is clear, even if in ways that don't garner much publicity.
And now, on to the 2020s! They present at the same time the largest-ever opportunity for human development and for furthered understanding of the physical climate (and how the two-way links between it and societies function), as well as the largest-ever risks from the potential misallocation of financial and intellectual resources in the face of rapid ongoing changes. As the world continues to become larger and more complicated, a certain level (even minimal) of harmony and collaboration within and among the international research and policy communities would greatly ease our ability to constructively manage the enormous task that we have effectively set out for ourselves as a species: to be, for the foreseeable future, active and conscientious guardians of an entire planet.
The notion of 'long-term means' is fundamental to climate science. Averaging over time and/or space constitutes the very definition of climate, according to authoritative sources such as the WMO. It's baked into our sense of the world as humans -- that if we wait long enough, all reasonably likely things will occur, and as a corollary, our memories and lived experiences are good proxies for the probability of occurrence of certain events, and the range of possibilities. But in times of rapid change, this sense (which goes by the technical term 'stationarity') can be undermined, and with it associated ideas such as anomalies. The question then becomes, if each decade is different in a statistical sense than the decades on either side of it, how are we to conceptualize the climate system -- as crystallized in, for example, major decisions about where to live, where to make investments, or whether to buy certain kinds of insurance (and how much). Recent evidence points to an approximately 5-year window over which people tend to adjust to climate regimes, which creates some cause for concern that rapid climate change will not be perceived as such, and we will not fully appreciate the environmental damage that it creates.
Although there is clearly inherent tension between physical and psychological realities and the usage of 'averages', there is no good answer for many of these issues, which is likely why 'averages' and their ilk continue to be regularly cited and discussed in many contexts. (The nomenclature of 'climate normal', as used by the National Weather Service and others, is particularly prone to misinterpretation.) An especially striking case in point involves the most recent set of Climate Prediction Center seasonal outlooks. For three-month-average periods going out 15 months, each part of the country is shaded according to whether above-average, near-average, or below-average temperatures are expected, and also the confidence of these predictions. Remarkably, throughout the entire period (out to boreal autumn 2020), every part of the country is expected to see near- or above-average conditions, with above-average accounting for the vast majority of that. These categories are based on a tripartite splitting of the 1981-2010 temperature distribution, which made me curious as to how the probabilities may have shifted in the 20-25 years since the midpoint of that time period. I used the difference between the 2013-2017 and 1981-2010 season-average CPC temperature data as a rough approximation of recent seasonal warming, and added this value to the 1981-2010 33rd percentile, to see just where what used to be a 33rd-percentile day now falls.
Since seasonal averages vary fairly little from year to year, a large percentile change can result from a modest mean warming -- and this is what I find in the below figure. Over much of the country, 20 years of warming have turned what was a 33rd-percentile day into one in the warm half of the 'normal' distribution, and up to the 75th percentile of 'normal' in the most rapidly warming regions and seasons. In other words, a '75' on the map indicates the bottom 75% of the 'normal' distribution now occurs only 33% of the time. I've made a small spreadsheet illustration of this shift, which appears below the map. Clearly 5 years is much too short to make any kind of reliable climate statement; perhaps the next 5 years will bring something a bit different. But the purpose of this analysis is more of a proof of concept, of how much 'normals' are being eroded by ongoing rapid warming, and that the lack of 'below-normal' conditions projected over the entire continental U.S. for the entire next year is not nearly as improbable as it might at first seem.
The Correlated Extremes Workshop last month catalyzed many fascinating conversations about how impacts-relevant research should be conducted; the conceptual and logistical hurdles that need to be overcome for successful multidisciplinary collaborations; and the emerging ways in which much of climate science is tending toward an Earth-systems-modeling approach where it is the interaction of innumerable preconditioning and initial-conditions factors that shape the ultimate extreme event. I've written up a full summary of the proceedings and some of their key takeaways here.
Below, I elaborate on some themes from the workshop that are specifically relevant to regional-climate scales.
Need for integration with engineering and land use/urban planning communities
These are the groups who decide what gets built, and how. At a time when there is more and more recognition of the importance of these kinds of decisions for extreme climate events themselves, not to mention their impacts, there are large opportunities for making such decisions as sensibly as possible with existing climate information and tools, and for furthering this knowledge with targeted decision-relevant collaborations going forward.
Broadening of data about demographics, economic networks, and the like
Geography in its most elemental sense -- where things are located -- plays a crucial role in affecting vulnerabilities on intra-urban scales. The types of granular data that are needed to make assessments about the true effects of particular combinations of extreme events, however, is largely uncollected or (at least) not compiled into easily usable forms. This encompasses sociocultural resilience (the 'cohesion' of a neighborhood) and micro-level data on trading networks, among many others.
Broadening of metrics to include more-intangible factors
Metrics used to assess the impacts of extreme events typically involve easily measured variables, with the ultimate example being counts of a binary outcome, e.g. numbers of people admitted to hospitals with a certain weather-related condition, or estimates of economic losses from an event. Even governments fall prey to the simplicity trap, based their preparation decisions on metrics that are primarily monetary. Needless to say, these hardly capture the range of damages that people care about, and which consequently affect their decision-making, whether as a prospect or a past experience. Although difficult, making serious efforts to incorporate impacts on quality of life would likely yield a much more faithful representation of the serious qualitative changes in a societal system that could stem from a correlated extreme event.
Variability in correlated extremes is significant, and highly affected by large-scale conditions
On timescales from seasonal to subdecadal, conditions that favor correlated extremes of all stripes vary a lot, up to an order of magnitude or more depending on location and event type. This is often because the timescale of the controlling conditions (e.g. large-scale zonal flow, or a certain SST pattern) is much longer than the timescale of the hazards, so the former can induce multiple of the latter in close succession, with accompanying severe impacts. This situation, and our greater ability to understand when it is in place, raises the possibility of predicting how the probability of correlated extremes varies for a small geographical area in response to known large-scale modulators, anthropogenic or natural, such as ENSO or reduced high-latitude sea ice or snow cover.
Tipping points exist and are important, but aren't always foreseeable
This awareness of the possibility of unforeseen cascading impact incentivizes a somewhat different approach to regional correlated extremes than has previously been practiced. The interconnectedness of regions in terms of dimensions which we care about (economic, cultural, etc.) means they can be conceptualized as tight networks, for which it is possible to estimate a characteristic sensitivity to climate stressors, including feedback responses. Tracing all such causal linkages is essentially impossible, which is where approaches based more on sampling the possibility space, such as 'storylines' and 'wargames', may provide more actionable realizations about what tipping points exist and where they lie. However, another essential aspect of tipping points is their complex and shifting nature, meaning that frequent updates are necessary as average conditions shift. Wealthy regions in the developed world are of course best-equipped to conduct such studies for themselves, but when valuing by lives or livelihoods instead of dollars, crowded and poor areas in the developing world are most deserving of ascertaining tipping points beyond which quality of life would erode, as these are where the greatest vulnerabilities lie. For example, what level of sea-level rise in low-lying regions could set off intense land-based conflicts for the remaining 'high ground'? Needless to say, such conflicts would by no means be confined to their region of origin. Globalization of economy and culture has also tended to produce globalization of climate-related conflict, whether it is the great-power struggle in Syria, Somalian pirates attacking shipping lanes, or immigration due to unprecedented heat creating poor farming conditions in Central America. Conflict thus represents one of multiple pathways through which regional climate has the ability to cause global problems, and consequently this presents a compelling motivation for studying it closely when climate and societal challenges combine in intricate and unfortunate ways.