In the OSTI Collections: Earth System Models
 |
Article Acknowledgement: |
Experiment, theoretical analysis, and something like both
Perfecting and using earth system models
Enabling technologies, user training
Reports available from SciTech Connect
“The abstract of the proposal for this award states, ‘Our challenge for SciDAC[DoE] is to transform an existing, state-of-the-science third generation global climate model, the Community Climate System Model (CCSM), to create a first generation Earth system model that fully simulates the coupling between the physical, chemical, and biogeochemical processes in the climate system. The model will incorporate new processes necessary to predict future climates based on the specification of greenhouse gas emissions rather than specification of atmospheric concentrations, as is done in present models that make assumptions about the carbon cycle that are likely not valid. We will include comprehensive treatments of the processes governing well-mixed greenhouse gases, natural and anthropogenic aerosols, the aerosol indirect effect and tropospheric ozone for climate change studies. We will improve the representation of carbon and chemical processes.’ All of these goals have been accomplished in that the CESM1 [the Community Earth System Model, version 1] has been used to study many new scientific problems that could not be studied using the CCSM4.”
—“A Scalable and Extensible Earth System Model for Climate Change Science” [SciTech Connect], National Center for Atmospheric Research
Experiment, theoretical analysis, and something like both
We test our ideas about the laws of nature—about how the universe works—by comparing those ideas with experiment. When our ideas or their logical/mathematical consequences don’t match our observations, we know our ideas are in some way incorrect or incomplete, and need to be revised in some way.
The physical process that a real system undergoes in an experiment can look quite different from the logical process of calculating what a theory says that system should do. But as computers have been made more and more powerful, one type of theoretical calculation has become increasingly feasible that more closely resembles real physical processes. In this type of calculation, a physical system is described as a large number of pieces, each of which interacts with its neighbors, so that what happens in any part of the system at a given moment depends directly on what happened in that part’s immediate neighborhood a moment before, and affects what happens in that part’s immediate neighborhood in the next moment. Each segment of the calculation thus represents an interaction between parts of the system, so executing all the segments in sequence represents an entire physical process. Executing this type of calculation on a computer amounts to a simulation of the actual physical process, and the simulation is more precise if the system is represented by a larger number of smaller pieces whose interactions take place over a larger number of smaller time intervals. So while such a simulation is in fact a theoretical calculation, running the simulation on a computer is a lot like conducting an experiment with the real system.
Some of the biggest simulations in the world model the entire geophysical system. These computer models, in effect, allow researchers to experiment with an artificial earth without disturbing the real one. As researchers design more efficient algorithms and use the greater capabilities of newer high-performance computers[DoE Science Showcase], their simulations of the earth can represent the earth in more detail, dividing its land, ocean, and atmosphere into larger numbers of smaller volumes that interact over more numerous smaller time intervals—and also take into account more details about how the various small portions of the planet interact with each other to shape our environment.
One determines how accurate a computer model is in the same way one tests any theory, by comparing its predictions with real observations and experiments. But present theory-based models of geophysical interactions can only be compared to a single actual earth (unless we eventually find some sufficiently earthlike planets elsewhere that we can examine in as much detail as we do this one). The closest we can now come to testing geophysical models against repeatable experiments is to check the models’ predictions against historical geophysical data, and compare what the models say should have resulted from actual conditions with what really did happen. If a model’s projections match actual events closely, we may have more confidence in the model and what it implies about future conditions.
The report quoted at the beginning of this article, about a project supported by the Energy Department’s Scientific Discovery through Advanced Computing program (SciDAC), indicates some of the ways in which the accuracy and usefulness of earth system models have been improved in the last few years; other recent reports discussed below describe some of these improvements in more detail and some things learned from making them.
Perfecting and using earth system models
The Energy Department funding mentioned in the report “A Scalable and Extensible Earth System Model for Climate Change Science”[SciTech Connect] substantially supported the development of two components of the new Community Earth System Model, CESM1. One of these components is a coupler that directs how a computer runs the model’s other components. Whereas the earlier CCSM’s (Community Climate System Model’s) coupler constrained all of that model’s components to run as separate programs on different sets of processors in the computer, the CESM1 coupler allows all the CESM1 components to run on potentially overlapping sets of processors, thus increasing the program’s flexibility for optimal efficiency. The other component developed with the same funding simulates atmospheric chemistry.
Further components for the new model were developed with other funding. CESM1 improves on CCSM in several ways, including the addition of new prognostic equations for aerosols in the atmosphere, so that the new model accounts for how cloud formation depends on the quality and size of the aerosols, thus indirectly accounting for aerosols’ effect on the climate. Among other improvements are the atmosphere component’s inclusion of a full chemistry subcomponent and a fast chemistry subcomponent, biogeochemistry modules in both the ocean and land components, and the atmosphere component’s modeling of how bulk motion of the air[Wikipedia] transports biogeochemistry tracers[Wikipedia]. With these improvements, the model predicts the concentration of greenhouse gases in the atmosphere from specified emissions; the earlier CCSM required the atmospheric concentration to be specified separately rather than deduced from emission amounts. The new model has been run with greenhouse-gas emissions specified for a simulated 20th century.
A new component, also developed with other funding, accounts for the interaction of ice sheets[Wikipedia] with the rest of the geosystem. Control and 20th-century runs of the new model show a warming atmosphere melting the Greenland ice sheet, with the resulting water entering the surrounding ocean as shown by the model’s river routing scheme. The report notes: “This is one of the first occasions where an interactive ice sheet model has been run as part of a fully interactive, quite high resolution climate model.” CESM1 allows new things to be investigated, such as “future air quality in very large cities, the effects of recovery of the southern hemisphere ozone hole, and effects of runoff from ice melt in the Greenland and Antarctic ice sheets.” Results from a series of future climate projections are available by Web from the Lawrence Livermore National Laboratory’s CMIP5 archive[LLNL], and have been described in many published research papers, several of which are listed in the report. The report ends with a list of a list of websites associated with the Energy Department award.
One reason that earth system models are being revised in this manner is indicated by the report “Winter storms and the Spring Transition over the western U.S.: Quantifying discrepancies between coarse and high-resolution simulations and observations”[SciTech Connect] from the Scripps Institution of Oceanography at the University of California at San Diego. Version 4 of the Community Climate System Model (CCSM4) improved over version 3 in several ways, such as the spatial structure and seasonal agreement of its precipitation predictions with observations; the report’s authors suggest that one reason for this may be version 4’s slightly finer spatial grid, which lets that version account for smaller terrain features that affect the weather. Version 4 also indicates that wet-season precipitation is most intense in north-central California, but has the greatest accumulation in the Pacific Northwest—a phenomenon actually observed. However, CCSM4 inaccurately portrays some climactic features, such as its exaggeration of the relationship between extreme precipitation in the western United States and the phases of the ocean-temperature extremes known as the El Niño-Southern Oscillation[Wikipedia] and the Pacific Decadal Oscillation[Wikipedia].
The authors also note that different climate model simulations disagree about whether California’s annual precipitation will increase or decrease by the 2060s, a result “which has impeded efforts to anticipate and adapt to human-induced climate change”. A group consisting of these authors with others have together examined 25 different model projections and found that most models predict fewer days of rain, but more intense average rainfall on the days rain will occur. Different models give different projections of how the frequency and intensity will balance out, so that “12 projections show drier annual conditions in the state by the 2060s and 13 show wetter.” However, most of this disagreement can be explained by differing projections of how often the most intense rains (more than 60 millimeters/day) will fall; “when such events are excluded, nearly twice as many projections show drying as show wetter conditions.”
The projections of future rainfall in California were made by using different fine-grid models of the entire earth’s atmosphere, and refining the models still further to account for information about even smaller local details—a process known as downscaling. A model used in some of the Scripps Institution’s downscaling, the Weather Research and Forecasting (WRF) model, was also used by researchers at Brookhaven National Laboratory, Stony Brook University, and Pacific Northwest National Laboratory to downscale the Community Climate System Model and the newer Community Earth System Model in order to model the global climate while simulating a region accurately. They describe an interactive technique for integrating WRF with CCSM and CESM in the report “Two-Way Integration of WRF and CCSM for Regional Climate Simulations”[SciTech Connect]. Their interactive method offers several advantages over other downscaling techniques, such as not needing to calculate massive intermediate model outputs at high frequency offline, providing higher temporal resolution for the interaction between global and regional model components, and allowing studies of feedbacks of regional weather to the large-scale global atmospheric circulation. The integrated system was able to simulate the development of a mid-latitude low-pressure area that the CESM alone, with its lower spatial resolution, would miss—though interestingly, this success had little to do with the higher resolution of terrain features.
Predicting climate variation over decades and centuries is the subject of work reported by researchers at the University of Wisconsin—Madison, Argonne National Laboratory, and the University of Bristol in “Towards the Prediction of Decadal to Centennial Climate Processes in the Coupled Earth System Model”[SciTech Connect] and by Scripps Institution of Oceanography researchers in “Towards a Fine-Resolution Global Coupled Climate System for Prediction on Decadal/Centennial Scales”[SciTech Connect].
The first group’s report briefly describes several of their numerous simulation studies and mentions significant implications of these, including:
|
|
|
|
|
|
The report ends with a summary of a review article by one of the report’s authors, which describes the history of how interdecadal climate variability is becoming understood.
Major goals of the second group’s project were to contribute to a fully coupled high-resolution earth system model in which “a weather-scale atmosphere is coupled to an ocean in which mesoscale eddies are largely resolved”, and “to understand how the explicit resolution of ocean mesoscale eddies, the realistic resolution of narrow mean ocean currents, and eddy-mean flow interactions, change the depiction of processes that are important to climate in the high-resolution regime.” Mesoscale eddies (i.e., eddies larger than thunderstorms but smaller than features typically depicted on weather maps[Wikipedia]) were found important in the spatial distribution of water masses that are almost vertically homogeneous, which cause heat uptake into the ocean. Seasonal variability and extreme events in the Bering Sea ice cover during the 1980s—the decade before the rapid acceleration of summer sea-ice loss in the 1990s and 2000s—were also explored to see how winds, surface heat fluxes, and ocean currents affected them. Also investigated were the mechanisms responsible for initiating, maintaining, and terminating the western Cosmonaut polyna[U. Toronto; Wikipedia]—an open-water area in the ice pack[Wikipedia]. Polynas are often where heat and moisture are exchanged between the ocean and atmosphere, and so are important to the heat budget of the polar surface atmosphere.
As this last investigation suggests, surface ice has significant interactions with other components of the geophysical system—whether the ice is in the ocean or on land. The role of land ice in geophysics has received a lot of attention, as shown by four sets of slides about the incorporation of land-ice submodels into earth system models. The titles of these slide presentations from Los Alamos National Laboratory indicate some of the progress made in the two years between the first two presentations and the last one:
|
|
|
|
Other institutions credited in these author and contributor lists at the beginning of these presentations include Florida State University, the Chinese Academy of Sciences in Beijing, Sandia National Laboratories, and the Institute for Computational Engineering and Sciences at the University of Texas at Austin.
The first of these presentations, “Predicting Land-Ice Retreat and Sea-Level Rise with the Community Earth System Model”[SciTech Connect], points out that sea-level predictions are of considerable interest because a rise in sea level greatly increases the probability of damaging floods from storm surges, and because so many people live near coasts: in the U.S., about 53% of the people live in coastal counties, and 3.7 million live within one meter of high tide. The presentation illustrates historic sea-level rise, mass loss from glaciers[Wikipedia], ice caps[Wikipedia], and the ice sheets[Wikipedia] of Antarctica and Greenland, and projections of sea-level rise in the 21st century, noting that the projections could be improved by using physics-based ice-sheet models coupled to global climate models. The rest of the presentation describes work being done to make those improvements and the kinds of interactions, e.g. between ice and sea water, that the improvements are expected to model. The author noted that while some contributions to sea-level rise were “fairly well constrained by models,” other contributions were highly uncertain.
How efficiently physical processes are simulated by a model based on a grid of small volumes depends partly on the characteristics of the grid. One modeling method, known as “Model for Prediction Across Scales” (MPAS), involves the use of “unstructured” grids whose volumes are larger in some regions and smaller in others, the smaller volumes being reserved for regions whose details are most important to account for, such as regions that have a lot of variation within short distances. (See right side of Figure 1.) By not having equally small volumes covering the regions with little variation, the number of grid cells can be about one tenth what they’d be otherwise. A presentation made at the same conference as “Predicting Land-Ice Retreat and Sea-Level Rise with the Community Earth System Model”, entitled “Development of a land ice core for the Model for Prediction Across Scales (MPAS)”[SciTech Connect], describes progress in constructing an MPAS simulation of land ice.
 |  |  |
| Figure 1. Improvement of the accuracy of ice-movement models illustrated for the Greenland ice sheet. Different colors indicate different rates of movement. Left: results of a calculation with SEACISM (“a Scalable, Efficient and Accurate Community Ice Sheet Model”). Middle: slightly different results from FELIX, a newer mathematical model. Right: a small portion of Greenland as mapped by a grid whose volumes are larger in some regions and smaller in others. (P. 47 of 71, “Predicting Land-Ice Retreat and Sea-Level Rise with the Community Earth System Model”[SciTech Connect].) | ||
The third presentation “Progress in coupling Land Ice and Ocean Models in the MPAS Framework”[SciTech Connect] illustrates numerous items to be taken into account in both models so they can be combined, e.g.: hydrostatic pressure at ice-shelf boundaries; mass balance; details of how ice masses break away and move into the ocean; how the ocean circulates under ice shelves; and how heat, salt, momentum, and mass are transported in the ocean boundary layer near ice surfaces. The title of the fourth presentation, “Optimal initial conditions for coupling ice sheet models to earth system models”[SciTech Connect], addresses a key problem: existing methods for specifying the initial state of ice-sheet simulations don’t match present-day observations well without implying that a physically unrealistic “shock” occurs at later times. The presentation describes details of a new mathematical technique to avoid the problem—taking the mass of the ice into account as well as its velocity when determining the initial state of the simulation. Early applications of the technique show that initial model conditions determined by this method match real ice conditions more closely, so projections of future states of the ocean and ice sheets are expected to be less uncertain.
 |
| Figure 2. An application of the mathematical technique illustrated in “Optimal initial conditions for coupling ice sheet models to earth system models”[SciTech Connect]. The goal is to have the model accurately represent the divergence[Wikipedia] of the ice sheet’s flux, which accounts for both the ice’s thickness and its velocity. When the ice thickness is at equilibrium, unchanging, the ice sheet’s flux divergence should equal the ice’s surface mass balance—its accumulation minus its sublimation[Wikipedia] into water vapor and melting into liquid water (right). Earlier methods of specifying the ice sheet’s initial state lead to inaccuracies, including a very different-looking estimate of the flux divergence (left); the new technique (middle) leads to a flux divergence closely matching the surface mass balance. |
Enabling technologies, user training
The existence of earth system models is of no value unless people can access them and use them. The Earth System Grid Federation's mission “is to provide the worldwide climate-research
community with access to the data, information, model codes, analysis tools, and intercomparison
capabilities required to make sense of enormous climate data sets”[SciTech Connect; SciTech Connect]. This was to be accomplished specifically by:
· providing easy, secure data access by the Web,
· presenting data sets in context with other data sets and comparative-analysis tools,
· addressing participating organizations’ requirements for bandwidth, access restrictions, and replication,
· ensuring ready access to data through climate-research analysis and visualization tools,
· transferring advances in infrastructure to other domains.
A team sponsored by the Energy Department, the Earth System Grid Center for Enabling Technologies, led international development and delivered a production environment for managing and accessing “ultra-scale climate data”, which served multiple climate projects like CESM and the Coupled Model Intercomparison Project, provided ocean model and atmospheric data, analysis and visualization tools from multiple sites in DoE national labs as well from foreign and domestic partner sites not funded by DoE. Later in the project, the Center included scientific graphics, animations, and other more detailed information products, secure data-access services, and server-side analysis. Two perspectives on the detailed history of this project and its accomplishments between 2004 and 2011 are available in “DOE SciDAC’s Earth System Grid Center for Enabling Technologies Final Report for University of Southern California Information Sciences Institute”[SciTech Connect], which describes contributions to the project by the University of Southern California, and the Lawrence Livermore National Laboratory report “DOE SciDAC's Earth System Grid Center for Enabling Technologies Final Report”[SciTech Connect], which views the project as a whole.
Two very brief reports from the National Center for Atmospheric Research in Boulder, Colorado (“2011 Community Earth System Model (CESM) Tutorial, August 1-5, 2011”[SciTech Connect] and “2012 Community Earth System Model (CESM) Tutorial - Proposal to DOE”[SciTech Connect]) serve to point out that earth system models will cease to be valuable if people who might use them don’t get to learn how. Each report tells about one such tutorial. According to the first one,
In fiscal year 2011, the Community Earth System Model (CESM) tutorial was taught at NCAR from 1-5 August 2011. This project hosted 79 full participants (1 accepted participant from China couldn't get a visa) selected from 180 applications. The tutorial was advertised through emails to CESM mailing lists. NCAR staff and long-term visitors (who were not eligible to attend) were also invited to 'audit' the climate and practical lectures and to work on the practical sessions on their own. 15 NCAR staff and long-term visitors took advantage of this opportunity. The majority of the students were graduate students, but several post-docs, faculty, and other research scientists also attended. Additionally, many people are using the on-line lessons and practical sessions.
As of August 18, 2011, 407 people had registered to access and use the tutorial from 33 countries all over the world, but a majority from US universities. In fiscal year 2011, the Climate and Global Dynamics Division Information Systems Group (CGD/ISG) built and operated a temporary computer laboratory in a meeting room.
Despite the word “Proposal” in the second report’s title, it too describes a tutorial that had already taken place when the report was written. This report simply points out the URL for the tutorial’s lectures and practical sessions (http://www.cesm.ucar.edu/events/tutorials/073012/), the tutorial’s name and location, the agenda (which also appears at the aforementioned URL), and lists of the participating lecturers, administrators, computer support personnel, and attendees.
Earth system models can play a central role in informing decision-making about climate prediction and future climate change. The formation and first meeting of a working group addressing societal dimensions of earth system modeling is described in another very brief National Center for Atmospheric Research report, “Final Report for proposal "The Interface between Earth System Models and Impacts on Society Workshop, Spring 2011”[SciTech Connect], which documents the use of DoE funding that supported participant travel to the group’s first meeting and the workshop that led to the group’s organization. A white paper from the workshop[NCAR] identified connections of earth system models to societal concerns, specifically models for informing policy and water-climate issues, and gave recommendations on how to establish and pursue projects related to these.
Yet the earth system models we’ve described do have some limits when it comes to informing decisions. As we’ve noted before, an accurate theory may allow us to calculate how a system will behave under various conditions. The calculation will describe approximately the same behavior whether it includes every detail of the system or only includes details that have outstandingly significant effects. This is true whether the calculation’s steps don’t correspond to the steps the behavior takes in time, or whether the calculation simulates the behavior’s steps directly. In many earth system models, the effort to simulate geophysics accurately has led to the inclusion of more and more details of geophysical processes, since all of them appear to have some significance, at least for particular phenomena. But it may turn out that not all of these details are significant for every geophysical phenomenon. Simulations of phenomena for which some details aren’t significant can be made with much simpler calculations than those executed by standard earth system models.
Such simplifications may not only be convenient, but for some purposes essential. The complexity of the earth system models we’ve discussed so far make it difficult to find out whether certain types of events can actually occur. This problem is addressed in the Sandia National Laboratories report “Statistical surrogate models for prediction of high-consequence climate change”[SciTech Connect]. Statistical surrogate models”, or “SSMs”, are intended to provide actionable information that existing earth system models with their full complexity could not. The authors note that designing anything for safety requires designing against high-risk possibilities, even if those possibilities aren’t very likely—as long as they could happen, they need to be designed for. Taking the same approach to climate change, the authors further note that determining possible but low-probability high-risk events by using current earth system models would require such extensive sampling as to be prohibitively expensive; hence their proposed use of specialized SSMs to see how likely certain climate parameters are given some known climate data. “Because of its reduced size and complexity,” the authors note, “the realization of a large number of independent model outputs from a SSM becomes computationally straightforward, so that quantifying the risk associated with low-probability, high-consequence climate events becomes feasible.” The report describes one such model and illustrates its calculation of three probabilities: that the air temperature at each point of the earth’s surface during December will be more than two standard deviations above average; that the maximum average temperature of a region covering much of the Pacific is higher than various specific values; and the average monthly rainfall at every point on earth in June, July, and August will be equal to or less than one standard deviation below average.
Wikipedia
· El Niño-Southern Oscillation
· Atlantic Multidecadal Oscillation
· Brownian noise (aka “red noise”)
· Polynya
· Glacier
· Ice cap
· Sublimation (phase transition)
· National Center for Atmospheric Research
· Lawrence Livermore National Laboratory
· Scripps Institution of Oceanography, University of California at San Diego
· Brookhaven National Laboratory
· Pacific Northwest National Laboratory
· University of Wisconsin—Madison
· Los Alamos National Laboratory
· Sandia National Laboratories
· Institute for Computational Engineering and Sciences, University of Texas at Austin
· University of Southern California
Reports Available through OSTI’s SciTech Connect |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
· Scientific Discovery through Advanced Computing (SciDAC), Office of Science, U. S. Department of Energy
· DoE Science Showcase: High-Performance Computing (January 2014)
· Coupled Model Intercomparison Project Phase 5 (CMIP5) archive
· “The Cosmonaut Sea Polyna”, University of Toronto
· Earth System Grid Federation
· Training materials for the 2012 Community Earth System Model (CESM) Tutorial
· “White Paper on Societal Dimensions of Earth System Modeling, July 5, 2011”, National Center for Atmospheric Research
View Past "In the OSTI Collections" Articles
