Over the last two centuries, human civilization has become a geological force. We are inducing planetary environmental conditions like those that Earth has not experienced for millions of years. Without mitigation of emissions, we may generate greenhouse gas concentrations and global temperatures more akin to those of the early Paleogene, over forty million years ago, than those of the current geological period, the Neogene.
The course society steers today – through decisions made in institutions that act on the timescales of news cycles, quarterly reporting periods, and elections – will have consequences that linger into geological time. Yet these decisions must be made under conditions of sometimes-profound uncertainty. The geological record of past environmental changes is a key resource for testing the models used to make future projection. But this resource, too, is riveted by uncertainty; it is noisy, diverse, and full of gaps.
My research group focuses on past and future Earth system uncertainty, and upon the incorporation of uncertain Earth system knowledge into policy decisions. Our research has three major threads. The first thread integrates the geological record of past environmental changes to quantify our understanding of the past and provide data sets against which to test the models used to make projections of future global change. The second thread links uncertain projections of future changes to statistical and process models of human systems to assess the risks posed by climate change. The final thread addresses how climate change risk assessment can usefully inform real-world policy and investment decisions.
Reconstructing past environmental changes using statistics, physics, and geology
Our paleoclimatic research employs statistical frameworks and physical models to understand the geological record of past environmental changes. Currently, the major focus of this thread is on reconstructing past sea-level changes and understanding the implications of these changes for ice-sheet stability and for ocean dynamics.
Ice sheets have waxed and waned throughout Earth history in response to changes in temperature and insolation. For most of geologic time, sea-level proxies provide our best guide to their behavior. When interpreting sea-level records, whether observational or proxy-based, it is critical to be aware of the difference between local sea level and mean global sea level (GSL). The latter is almost linearly related to changes in ice sheet volume; the former, however, is influenced by a range of factors, including atmosphere/ocean dynamics and changes in Earth’s gravitational field, rotation, and crustal and the mantle deformation associated with the redistribution of mass between land ice and the ocean. These factors complicate inference of GSL from records; without uniform geographic coverage, averaging local records does not necessarily produce an unbiased estimate of GSL. But because the sea-level patterns caused by melting different ice sheets differ, these factors may also make it possible to infer how much ice was lost from different ice sheets.
Kopp et al. (2009) took a first step down this road, applying a Bayesian framework to reduce bias in estimates of GSL during the Last Interglacial stage (LIG; ~130-115 thousand years ago). During the LIG, polar temperatures were comparable to those expected to accompany 1-2°C above pre-Industrial levels of future global warming. We found a 95% probability that GSL during the LIG peaked at least 6.6 m higher than today, a 67% probability that it exceeded 8.0 m, and a 33% probability that it exceeded 9.4 m. Subsequent analysis constrained rates of sea-level change during the LIG (Kopp et al., 2013). These paleo-sea level and paleo-rate estimates inform expectations about the ice-sheet response to modest levels of future warming, such as the 1.5-2°C aspirational target of the Paris Agreement. We are now working to go beyond GSL estimates, to the budgeting of ice loss during the LIG and earlier warm periods.
During periods when ice sheets have been relatively stable, such as the last several millennia (the late Holocene), sub-millennial sea-level variability arose primarily from changes in atmosphere/ocean dynamics. We have developed a set of statistical methods to separate signals of global and regional change over such time periods. These approaches led to the construction of the first truly global, statistically robust sea-level record for the last 3000 years (Kopp et al., 2016). This record identified a ~8 cm global sea-level fall associated with the Little Ice Age and established with 95% probability than 20th century global sea-level rise was faster than during any century in at least 2800 years. Using a statistical model calibrated to the relationship between global mean temperature and rates of GSL change over this time period, we are assessing the human role in historic sea-level rise and identifying human ‘fingerprints’ on coastal flood events. We are also investigating connections between regional sea-level changes along the Atlantic coast of North America and past variability of the Atlantic Meridional Overturning Circulation (AMOC) and the North Atlantic Oscillation (NAO), and we are starting to use our paleo-reconstructions to assess climate variability in global climate models.
Our statistical techniques have also enabled a new generation of GSL reconstructions for the historical period, based upon tide-gauge records (Hay et al., 2015). Our analysis has shown that GSL rise from 1900-1990 was less than previously thought, and that the acceleration over the last 25 years has been commensurately greater. We are now working to expand our statistical analyses to combine parallel lines of information relevant to sea-level change, including satellite altimetry of sea-surface height, satellite gravimetry of surface mass redistribution between the ocean and land ice, and Argo float data about ocean heat content.
Integrated assessment of climate change risks
Climate change is a pervasive source of risks to human systems, with potentially large impacts upon the economy, public health, and national security. Quantitative integrated assessment of climate change risks is not always possible, but it can play a key role in informing decisions both about local adaptation and about large-scale mitigation policy. For example, the U.S. government uses integrated assessment models (IAMs) to estimate the benefits of carbon dioxide emissions reductions in the design of regulations, including the Environmental Protection Agency’s Clean Power Plan. But these estimates, and all the IAMs currently capable of performing global benefit-cost analyses, suffer from significant limitations and over-simplifications (see Kopp & Mignone, 2012, for a review).
One critical limitation is a paucity of historical tests of modeled human responses to climate variability and climate change. We and our partners in the Climate Impact Lab (a collaboration between scientists and economists at Rutgers, Berkeley, University of Chicago, and Rhodium Group) are working to address this limitations. Our framework links innovative approaches for (1) generating high-resolution, probabilistic projection of future climate and sea-level changes and (2) empirically identifying robust statistical relationships characterizing how humans have responded to past climate variability and past climate change, in order to (3) project how humans may respond to uncertain future changes.
In projecting climate variables such as temperature, precipitation, and humidity, there is generally a tradeoff between (a) the ability to produce high-resolution projections needed to inform local decisions and model local responses, and (b) the ability to sample uncertainty. Full-complexity Earth system models (ESMs) produce spatial and temporal detail, but an ensemble of ESMs are computationally costly and do not generate probability distributions; instead, they yield ranges of different modeling groups’ semi-independent ‘best estimates’ of climate responses. Importantly for risk assessment, ESMs may fail to capture the low-probability, high-impact end of potential future climate change altogether. Simple climate models (SCMs) can probabilistically project global temperature change and capture tail risk, but lack adequate spatio-temporal resolution. We have developed a suite of weighting, pattern-scaling and downscaling approaches for combining these two different sources of information and are in the process of exploring the uncertainty associated with a range of different approaches.
No single model captures all the processes that contribute to global and local sea-level change. Accordingly, we have developed a framework (Kopp et al., 2014) that merges a range of different sources of probabilistic information regarding the different relevant oceanographic, cryospheric, hydrological, geophysical, and geological processes. This framework has yielded the first global set of fully probabilistic, local sea-level rise projections. These global projections are consistent with an independent set of global projections based upon the relationship between temperature and rate of sea-level change over the last two millennia. While this consistency might be interpreted as an important corroboration, we find it a source of concern, since we know that the processes that will dominate sea-level change in a high-warming scenario will differ from those that were dominant in recent millennia. We are therefore working to integrate recent modeling results regarding instabilities in the Antarctic ice sheet into this framework.
In the Climate Impact Lab integrated framework (piloted in Houser et al., 2015, for the United States), probabilistic physical projections are combined with models of human-relevant responses in areas such as labor productivity, temperature-related mortality, migration, civil conflict, and coastal storm damage. Most of these models are based upon the recent econometric literature identifying human responses to climate. New statistical innovations use cross-country variability in short-term responses to different levels of exposure to estimate how vulnerability will evolve under future socio-economic and climate scenarios. Combining the Climate Impact Lab sectoral models with probabilistic climate projections allows for quantitative risk analysis in the covered sectors.
One important omission that needs to be addressed involves so-called ‘tipping points’ in the climate system and in human responses to climate change. Abrupt state shifts in subcomponents of the Earth system are often poorly captured by Earth system models; sometimes an outlier model produces a state shift in response to projected emissions, and sometimes such state shifts are only hinted at by the geological record. But though their probability is difficult to assess, state shifts are a potentially large contributor to overall climate risk. My research group is in the process of developing a set of experiments using the Climate Impact Lab framework to investigate the consequences of physical ‘tipping points.’ It is our hope that this investigation of state-shift consequences can inform the prioritization of research among different ‘tipping points.’ We are also working to develop an agenda for investigating social tipping points that may be triggered by climate change.
Decision-making under uncertainty
An emerging aspect of our research focuses on how to maximize the utility of quantitative, probabilistic information like that we produce in our risk analyses to decision-makers. Some decision-makers can use quantitative risk information directly; this is the case, for example, for financial firms that are accustomed to quantitative risk management. We have, however, found with our probabilistic sea-level projections in particular that, while assessment panels intended to inform governments like having probabilities assigned to outcomes, they also want probability distributions boiled down to a small number of actionable numbers (for example, simple answers to ‘how high should I raise my infrastructure?’). Buchanan et al. (in press) built a sea-level rise allowance framework to help end-users evaluate safety margins for sea-level rise in light of their timescales, acceptable probability of flooding, and level of confidence in expert projections. We are currently working with colleagues in urban planning to test sea-level rise allowances and other ways of communicating to municipalities and real-estate market participants about the risks of sea-level rise. We are also developing plans with social scientists to explore whether such approaches are effective and, if so, how they might be generalized beyond flood risk.
Other projects in deep time
An “Appalachian Amazon” in the early Eocene: A magnetofossil record of stormier times?
Because iron, the fourth-most abundant element in Earth’s crust, is redox-active under surface conditions, it is a critical element in Earth surface processes: both a key player in many abiotic reactions and an essential nutrient for living organisms. The bio-availability of iron is a product of environmental conditions; through its effects on biological productivity, it is also a feedback. Changes in the sedimentary record of iron cycling therefore reflect broader biogeochemical and climatic changes.
Our current research in this area focuses on understanding the paleoenvironmental implications of a radical change in sedimentary iron biogeochemistry in the mid-Atlantic U.S. during the Paleocene-Eocene Thermal Maximum (PETM), a severe global warming event that occurred 55 million years ago. Kopp et al. (2007) and Schumann et al. (2008) found that a clay layer deposited during the PETM in the Salisbury Embayment (which stretches from New Jersey to Virginia) recorded the unusually rich growth of magnetotactic bacteria and of other unique and presumptively eukaryotic iron biomineralizing organisms. Combined with knowledge about the ecological distribution of modern magnetotactic bacteria, this finding suggests that biogeochemical changes during the PETM led to the development of enlarged suboxic zones in the sediments of the Atlantic Coastal Plain.
Because meter-scale suboxic zones occur today within the mobile mud belts of tropical river-dominated continental shelves, such as the Amazon Shelf, we hypothesize that sedimentological and hydrological changes during the PETM fostered the development of analogous conditions on the Eastern seaboard of North America. By mapping the distribution of magnetofossils and the PETM clay, Kopp et al. (2009) found support for an Appalachian-fed, Amazon-like river system located around the modern Potomac or Susquehanna rivers. The development of this river system may be linked to changes in temperature, precipitation, rainwater acidity, and storminess.
Key next steps in this research area focus on testing the “Appalachian Amazon” hypothesis by examining Paleogene geochemical and sedimentological changes across the Salisbury Embayment, expanding the magnetofossil database to other regions and other times and on better understanding the preservation of magnetofossils in modern analog environments.
The rise of oxygen and the co-evolution of life and climate
In the Archean eon, three billion years ago, Earth’s atmosphere was free of oxygen. This absence is reflected in numerous proxies, particularly the record of mass-independent fractionation of sulfur isotopes. By two billion years ago, in the middle of the Paleoproterozoic era, many proxies suggest that oxygen had built up to a small but significant level in the atmosphere. Aside from the origin of life itself, the transition from the anoxic Archean to the oxic Proterozoic is the most radical change to occur in the history of the Earth system.
While all geohistorical data is sparse, the Precambrian eons are times of particular sparsity. In light of this sparsity, Chamberlin’s method of multiple working hypotheses becomes particularly important. While many researchers believe the transition from an anoxic world to an oxic world occurred gradually, Joe Kirschvink (Caltech) and I have suggested instead – based on a critical analysis of the available data, with a strong dose of caution applied to the use of uniformitarian interpretations in such a radically different world – that the transition occurred rapidly (Kopp et al., 2005; Kirschvink and Kopp, 2008). We proposed that the Makganyene Snowball Earth event, which occurred at ~2.3 billion years ago, was a direct consequence of the transition. During the Snowball Earth, the presence of which is indicated by glacial deposits with low paleolatitudes from the Transvaal Supergroup in South Africa, the entire planet may have been sheathed in ice for tens of millions of years.
Simple biogeochemical flux modeling suggests that, if the Archean Earth was kept warm by a methane greenhouse, then the evolution of oxygenic photosynthesis could have triggered a Snowball Earth event on a time scale as short as about a million years (Kopp et al., 2005). Rather than a gradual transition, the anoxic-to-oxic transition, triggered by a chance evolution occurrence, may have been a catastrophic event: the world’s first biologically-caused climate disaster. This hypothesis is an end-member hypothesis, but it is testable and serves to motivate research in this critical interval in Earth history.