Global climate models need to represent the intricate workings of the climate system and they need to provide information on the ways climate change and climate variability will impact society, including sea-level rise, regional climate trends and extremes, food security and ecosystem health, and abrupt climate change.
Ideally, global climate models would simulate climate dynamics at a spatial resolution high enough to resolve features like cities, river drainages, and mountain ridges, as well as convective storms and ocean eddies, minimizing the need for further downscaling of the model output—a grid spacing of 1-5 km would suffice for many purposes and is achievable within 10-20 years. This resolution is expected to improve representation of critical climate processes such as clouds and cumulus convection, mesoscale ocean eddies, and land-surface processes. Such models would provide information that meets the needs of society and would include fully interactive Earth system components (i.e., atmosphere, ocean, cryosphere, biosphere, land-surface, and human systems). Models should have seasonal and decadal predictive skill, should be able to replicate historical trends and modes of variability (e.g., El Niño/Southern Oscillation [ENSO]; decadal-scale Atlantic and Pacific variability), and should be able to capture the processes and feedbacks involved in major paleoclimate events, such as the last glacial cycle and decadal-scale climate transitions that occurred during the glacial period (i.e., Dansgaard-Oeschger events).
While this idealistic vision is clear, some of this may not be realistic because of intrinsic limits in predictability and practical limits to resolution, physical understanding, and observational constraints. Substantial improvements in model resolution are expected and important (Chapter 3), but the challenges of simulating climate physics are not magically resolved as models go to high resolution and increased complexity. It takes time to add and properly validate new processes and components to a model. Extensive testing and sensitivity experiments are required, involving hierarchical regional climate models and global climate models with a variety of scale-sensitive parameterizations.
However, these challenges and limits should not constrain ambition and exploration. It is difficult to foresee the advances in technology, observational capacity, and process understanding that will extend modeling capability in the coming decades. There needs to be a strategic research agenda for climate science, observations, and model-
ing as the climate modeling community keeps pace with the information needs of a changing climate system, while at the same time improving climate model capabilities and skill.
In this chapter specific scientific targets for advancing climate science and climate modeling in the coming decades are identified. They require modeling efforts at both global and regional scales, or a fusion of these efforts. This chapter emphasizes problems where (i) progress is likely, given appropriate strategic/scientific investment, and (ii) progress would directly benefit societal needs with respect to weather and/or climate impacts and investments in climate change mitigation and adaptation.
STRENGTHS OF CLIMATE MODELS
Bader et al. (2008) provide a detailed discussion of strengths and weaknesses of the current generation of global climate models. Current models have demonstrable skill in many aspects of climate dynamics, including their ability to simulate large-scale features of ocean and atmospheric circulation, planetary Rossby waves, extratropical cyclone dynamics and storm tracks, radiative transfer, and global temperatures (Chapter 1). Climate models conserve energy, mass, and momentum; can be integrated for multiple centuries; and have demonstrated the ability to simulate the broad features of 20th-century climate, both the mean state and historical climate change. The rich array of models and expertise, nationally as well as internationally, allows for extensive testing and model intercomparison activities. This cooperation within the global community provides further insight and confidence into the capabilities of climate models. No other global scientific endeavor enjoys this level of international cooperation, or is subject to the same degree of scientific and public scrutiny; although this presents some challenges, this has helped to drive climate modeling forward.
Several considerations underlie the reliability of climate models for many aspects of climate change. It is important to recognize that climate projections are not forecasts of the specific state of the climate system at a particular place and time; rather, they should be interpreted as a realization of the mean statistics of weather for a period of time in the future (commonly taken as the average over multiple decades). Constructing the statistics of future climate conditions is a different problem from predicting what the weather will be like on a given day or month in the future; it is less sensitive to nonlinear dynamics and initial conditions, as the statistics of short-lived weather systems average out over many years. The average climate of a location depends on the relative frequency of different weather systems, which is governed by large-scale features of atmospheric circulation that are reasonably robust in climate models.
Although the magnitude of climate change that will occur this century is uncertain, all climate models indicate that the planet will warm. The suite of global climate models deployed in IPCC (2007c) report a mean climate sensitivity1 of 3.2°C, with a standard deviation of 0.7°C (IPCC, 2007c, Table 11.2); this indicates broad agreement, with some scatter, about the effects of carbon dioxide on global mean temperature. Other largescale aspects of climate change are also robust, such as water vapor feedbacks (increasing atmospheric moisture), thermosteric sea-level rise, ocean acidification, Arctic amplification of climate warming, warming feedbacks due to reductions in seasonal snow cover, and a poleward shift of circulation systems.
Despite these confirmations of the value of climate models, a number of longstanding and emerging problems require improvements and developments in model capability. Bader et al. (2008) provide a detailed summary of weaknesses of the current generation of climate models. The next section examines some of these weaknesses and outlines several high-priority scientific frontiers that can be better addressed through advances in climate models.
GRAND CHALLENGES FOR CLIMATE MODELS
Climate change is expected to affect society in many ways, including impacts on health, infrastructure, food and water security, ecological integrity, and geopolitical stability. Climate models are essential tools to inform planning and policy development surrounding these issues, but advances are required on a number of research fronts to improve the information that climate models can provide. High-priority questions include the following:
• Climate sensitivity: How much will the planet warm this century?
• How will climate change on regional scales? How will this affect the water cycle, water availability, and food security?
• How will climate extremes change?
• How quickly will sea level rise?
• How will Arctic climate change?
• What is the potential for abrupt change in the climate system?
• How will marine and terrestrial ecosystems change?
• How will society respond to and feed back on climate change?
• Can the evolution of the climate system over the next decade be predicted?
1 Climate sensitivity: the equilibrium, global mean temperature change associated with a doubling of CO2.
It is not straightforward to prioritize these scientific questions, because they operate on different time scales (and, hence, are of varying urgency); some are more “basic” in nature, and the importance and societal cost of climate change impacts such as drought, sea-level rise, increased tropical storm frequency, and Arctic sea-ice loss depend on the specific regional or national context (i.e., vulnerability of lives and infrastructure in different parts of the world). This list of grand challenges is therefore not ranked, but the first four questions are flagged as “high-priority issues” for climate modeling that have the most impact, require the most attention, are globally important, and/or limit progress on other important issues. The sections below discuss the state of the modeling for these issues and provide ideas for potential ways forward.
Climate Sensitivity: How Much Will the Planet Warm This Century?
The severity of future warming affects most aspects of climate change, and mitigation and adaptation strategies hinge on this question, so better constraints on this question are one of the highest priorities in the climate modeling enterprise. If climate models cannot capture the mean state and main features of atmosphere and ocean circulation, they cannot provide meaningful insight regarding regional details. Although all climate models project global warming in response to increasing greenhouse gas concentrations in the atmosphere, there is uncertainty as to the magnitude and rate of expected warming for a given radiative forcing. This uncertainty in climate sensitivity is due to a range of internal feedbacks in different climate models, particularly with respect to how clouds are expected to change, as well as from a lack of observational constraints.
Although there is some irreducible uncertainty in projections of future climate change, improved confidence in climate sensitivity is important if climate models are to provide more useful guidance to planning and policy decisions. For a given emissions scenario, much of the uncertainty arises from the treatment of cloud processes, the carbon cycle, and aerosols within climate models. The brief discussion of these processes below includes an analysis of likely improvements in these aspects of climate models over the next 10-20 years.
Simulation of clouds and how they will respond to future greenhouse gas and aerosol changes is a central challenge in climate modeling. Small changes in cloud cover, thickness, altitude, and cloud particle size and type (liquid versus ice) affect the radia-
tive energy balance significantly. The differences from these small changes are enough to explain the majority of model-to-model differences in global warming over the next century.
The problem is challenging for several reasons. First, clouds are quite variable on all time and space scales. Second, many clouds (e.g., cumulus cloud systems) are not well resolved by the grid of a typical climate model. Third, clouds often result from the small-scale interaction of multiple physical processes, which are separately represented in the climate model. Cumulus clouds, for instance, involve turbulent updrafts usually initiated by surface-driven turbulence in which small droplets condense into rain or freeze into small ice particles, some of which fall as snow and some of which are ejected or detrained into the surroundings as cirrus clouds of various thicknesses. There is still considerable controversy about how to best represent some of these processes (e.g., cumulus convection and ice cloud microphysics) and how to best handle complex interactions between parameterizations.
“Low cloud feedbacks” from marine boundary-layer clouds in the lowest 1-2 km of the atmosphere are the largest source of spread between predicted global temperature change in leading climate models (Soden and Held, 2006; Soden and Vecchi, 2011). These clouds are hard for climate models to vertically resolve, and they involve tight interactions of turbulence, cloud and precipitation formation, radiation, and aerosol at subgrid scales. Low clouds are particularly sensitive to human-induced aerosol increases, which change their typical droplet size and albedo, so they are also the principal contributor to intermodel differences in simulating the effect of human-induced aerosols on climate change.
Inaccuracies in the representation of organized tropical cumulonimbus cloud systems contribute to systematic errors made by many climate models in the mean geographical and seasonal distribution of tropical precipitation (e.g., monsoons and “double-ITCZ [Intertropical Convergence Zone]” biases) and its variability on diurnal, intraseasonal (e.g., the Madden Julian Oscillation), and interannual scales (e.g., El Niño). Through their effects on latent heat and rainfall, these errors lead to circulation biases and generate planetary-scale waves in the upper troposphere that disperse to the midlatitude storm tracks, affecting simulations of the entire Earth system.
Full cumulonimbus-permitting (“cloud-resolving”) global simulations with no deep cumulus parameterization require a horizontal resolution of 4 km or less, with vertical resolutions of 200-500 m. While this may not be commonplace for multicentury global climate simulations, it is already feasible for global simulations of a few weeks or for longer simulations with regional models and will likely become attractive within the next decade for some types of global climate modeling. Such simulations give much
more realistic descriptions of the diurnal cycle of deep convection over land and of the Madden-Julian Oscillation but still may include biases in seasonal-mean tropical precipitation or cloud statistics, due to residual parameterization uncertainties in processes that are still unresolved such as ice processes, boundary-layer turbulence, and small-scale land-surface inhomogeneity. In particular, the boundary-layer cloud and cloud-aerosol uncertainties in climate models will not automatically go away in atmospheric models of cloud-resolving resolution, although they may become easier to reduce. Although these are short-term processes, they have a potentially large spatial and cumulative effect on modeled tropical circulation; systematic biases can influence overall climate sensitivity in decadal to centennial predictions in climate models.
The cumulative extent of greenhouse gas emissions, primarily the amount of carbon dioxide (CO2) and methane (CH4) released into the atmosphere, are of first-order importance to future climate. About half the CO2 from fossil fuel combustion remains in the atmosphere and is the principal forcing of climate change; the remainder is absorbed by the land and oceans. There are numerous feedbacks in the carbon cycle however, both positive and negative, that influence the amounts of CO2 and CH4 that remain in the atmosphere versus those which are taken up in the ocean and the land surface. These carbon sinks need to be included in climate models to provide the best possible estimate of future greenhouse gas forcing in the atmosphere.
Feedbacks are two-way processes, however; climate change affects land cover and the ocean by modifying ecosystem structure and function, as well as the physical controls on gas exchange (e.g., solubility of CO2 in the ocean, and soil respiration rates). These changes can in turn have important impacts on climate. Ecosystem models predict the distribution of natural land cover on the basis of local temperature, precipitation, and other factors. These ecosystem models are now being coupled with general circulation models (GCMs). Efforts so far have focused on feedback loops involving the biogeochemical cycle of carbon. For example, increasing soil respiration and tropical forest dieback resulting from expected 21st-century changes in temperature and precipitation patterns could produce a major positive feedback on CO2. There is also the potential of large positive feedbacks involving increased emission of methane from warming wetlands and thawing permafrost.
CO2 exchange between the land and the atmosphere is via the processes of photosynthesis and decomposition, whose rates vary with sunlight, atmospheric CO2, temperature, precipitation, and ecosystem distribution. Where not water or nutrient
limited, photosynthetic uptake in vegetation can increase in a high-CO2 environment, providing a negative feedback to CO2 accumulation in the atmosphere. Currently the imbalance between these processes results in net carbon storage on land, but the first generation of the Earth system model results suggest that this could switch to a net carbon loss to the atmosphere with shifts in ecosystems (e.g., Cox et al., 2000) and as soil respiration rates increase with warming.
Earth system models for the next decade will include multiple processes that interact with carbon cycling, and feedbacks that occur between these processes and climate change. These include the major biogeochemical cycles providing nutrients important for life (e.g., nitrogen and phosphorus). The establishment and mortality of ecosystems will change in response to the changing climate and in turn influence carbon fluxes, atmospheric CO2, and climate. The transient dynamics of this interaction depend on the time scales of growth, senescence, and mortality intrinsic to ecosystems (including ephemeral and invasive species) as well as on the rate of climate change. Furthermore, variations of structure and functioning within ecosystems, as a result not only of local climate variations but also of age, health, and other differences, must be central components of the next-generation carbon-cycle models. These models need to include models of disturbances beyond fires and land use and include pests, infestation, and other processes that could influence the survival of and competition among ecosystems.
A major advance in the next decade must be in the representation of carbon-climate feedbacks via subsurface processes, for which there are only sparse observations. Most important is soil water, the critical determinant of photosynthesis and decomposition rates, as well as the health and survival of ecosystems. For example, the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) generation of climate models do not agree on whether soil moisture near the end of the 21st century will increase or decrease with global warming (see IPCC, 2007c, Chapter 10). Carbon-rich permafrost soils are particularly vulnerable to climate change. Models of the next decade should include the dynamics of permafrost, as well as functional classification of microbe communities and mechanistic representation of soil biogeochemistry. As an example, a shift between populations of methanogens and methanotrophs as a consequence of warmer, drier soils would have first-order importance for methane flux to the atmosphere.
CO2 exchange between the oceans and the atmosphere is driven by the difference in CO2 partial pressures in the surface waters and the atmosphere, with the oceanic value dependent on the ocean circulation, marine biology, and carbonate chemistry. Ocean biogeochemistry is a central determinant of the uptake of CO2 from the atmosphere
and will change as the climate and ocean change. Ocean biogeochemistry models currently include climate-sensitive carbonate chemistry, rudimentary representation of different classes of phytoplankton and zooplankton, and multiple nutrient cycles (nitrogen, phosphorus, silica, and iron). They will continue to be improved with observations and understanding of their responses to macro- and micronutrient variations. New modeling directions need to include the cascading impacts on the entire marine biota from ocean acidification and purposeful and inadvertent additions of macronutrients (e.g., from rivers) and micronutrients such as iron, and their impact on surface CO2 concentrations. Better resolution of coastal circulation and biogeochemistry will be helpful, as well as improved coupling with continental hydrology models.
Aerosol and Atmospheric Chemistry Feedbacks
The role of aerosols in modulating radiative fluxes through the atmosphere, both directly and indirectly through their influence on cloud formation, is a major source of uncertainty in current climate models. Most climate models now include an interactive simulation of aerosols to describe aerosol-climate interactions, but the underlying chemistry and microphysics are only crudely parameterized. This limitation introduces uncertainty in model quantification of aerosol radiative forcing and its dependence on the hydrologic cycle, both through hygroscopic growth and precipitation scavenging. In addition, atmospheric oxidant and nitrogen chemistry are generally not described in climate models, and this limitation stymies a proper description of simple chemical feedbacks, such as methane-hydroxyl radical (OH) coupling, and more complicated feedbacks involving the effects of changing land cover on atmospheric composition. In general, maintaining an appropriate tradeoff between the complexity of aerosol descriptions and chemical mechanisms represented in climate models and their computational cost continues to be an important research topic.
Atmospheric aerosols are greatly sensitive to land cover and vegetation. Increased desertification associated with drying of the subtropics could represent an important source of dust. Changes in ecosystem structure and function would affect the supply of organic aerosol produced by oxidation of biogenic volatile organic compounds (VOCs). The resulting climate feedback loops are potentially important, and they could be either positive or negative depending on the poorly understood radiative properties of dust and the climate dependence of biogenic VOC emissions. The latter emissions depend in a complicated way on vegetation type, temperature, water availability, leaf phenology, and CO2. Current land-cover models disagree on the sign of the change in biogenic VOC emissions in response to 21st-century climate change. Aero-
sol yields from biogenic VOCs may also depend on the preexisting supply of anthropogenic aerosols, further complicating the feedback loops.
Atmospheric chemistry plays a critical role in aerosol formation and contributes to other climate-chemistry feedbacks driven by changes in land cover. Deposition of reactive nitrogen (nitrate, ammonium) may significantly affect carbon uptake by ecosystems, and climate change in turn will affect the terrestrial emission and atmospheric chemistry of nitrogen oxides and ammonia. Biogenic changes in nitrogen oxide and VOC emissions will affect the concentration of the hydroxyl radical (OH), the main sink for methane, and will also affect ozone. Like the land-cover impacts and feedbacks that are involved in the carbon cycle, understanding of these effects requires coupling of sophisticated, dynamic ecosystem and land-surface models.
The advance of coupled land-surface, vegetation, boundary-layer, and aerosol chemistry models promises to be an exciting frontier that may transform aspects of climate modeling, and climate model utility in, for example, air quality and land-use simulations. It may pave the way for unification of current efforts in air pollution modeling and in human-climate interactions, discussed further below. In the context of decadal to centennial climate change, these short-term processes influence climate system sensitivity through cumulative effects on radiative transfer and cloud properties. Aerosol chemistry, through direct and indirect effects on atmospheric absorption and scattering, are one of the greatest sources of intermodel climate variability.
How Will Climate Change on Regional Scales? How Will This Affect the Water Cycle, Water Availability, and Food Security?
Climate change impacts and adaptation activities are most strongly manifest on regional scales, where ecological and human systems are adapted to a specific set of historical climate “normals.” Agriculture, water resource management, transportation, energy systems, recreational activities, wildfire hazards, and biological systems are all vulnerable to shifts from these historical normals, creating a demand for climate models that can provide accurate and detailed regional information. This demand is a challenge for the current generation of models, particularly with respect to simulation of regional precipitation; climate models need improved skill on regional scales to address this need. Issues concerning rainfall and the hydrologic cycle are of foremost concern. Simulation of ecosystems, ice-ocean interactions, and severe weather, among other climate processes of interest, also require model skill at regional scales.
Accurate simulation of regional precipitation patterns and trends is difficult. Current models are generally limited in their ability to simulate regional precipitation pat-
terns (Kerr, 2011), and this is a significant weakness given the importance of drought to agriculture, water resources, food security, and geopolitical stability (Romm, 2011). Regional precipitation is controlled by atmospheric moisture convergence associated with large-scale and mesoscale circulation, but local forcing from the surface related to orography, land-surface heterogeneity, and precipitation recycling in general alter its amount and intensity, thereby modulating its spatial and temporal characteristics.
Projections of 21st-century regional precipitation trends are of particular societal interest. Climate models consistently agree that globally averaged annual mean precipitation will increase poleward of 45° latitude, as well as over the warmest parts of the tropical oceans (IPCC, 2007c). Held and Soden (2006) gave a simple theoretical argument for this behavior as a consequence of the increased water-holding capacity of a warmer atmosphere as well as increased rates of evaporation in a warmer world. In the subtropics and in some midlatitude regions, many models project drying trends, but the location and magnitude of projected drying vary between models. Model differences in regional precipitation trends have multifaceted causes, including grid resolution but also treatments of cumulus convection, air-sea interaction, land-surface processes, upper ocean dynamics, aerosols, cloud microphysics, and the simulated global climate sensitivity.
These factors interact. As discussed above, model representations of cloud physics, convective processes, orographic and frontal forcing, and land-surface exchanges (i.e., evapotranspiration) are still limited by model resolution as well as process understanding. Because hydrologic cycle processes are inherently multiscale, increasing model resolution to more explicitly represent finer-scale processes is important. Partly because of insufficient spatial resolution, models tend to “drizzle” a lot, overestimating the number of precipitation days but underestimating high-intensity precipitation events (e.g., days with rainfall totals in excess of 10 mm) (e.g., Dai, 2006). Spatial precipitation patterns are similarly blurred in climate models because of the limited ability to resolve strong orographic and frontal gradients.
Orography is an important forcing mechanism for precipitation worldwide. There are significant challenges in predicting both cold and warm season orographic precipitation fundamentally because of the myriads of scale interactions involved. For example, mountains can modulate large-scale circulation, causing changes in local moisture convergence, but local condensation and microphysical processes also influence flow stability upstream. In summer convective regimes, orography can induce convective storms that can organize onto larger spatial scales as they are blown downwind, challenging models’ ability to simulate the multiscale precipitation patterns (Houze, 2012). Resolution of snow versus rainfall in mountain regions is also critical for water
resources management and climate change adaptation studies (Leung et al., 2004). Addressing limitations in measurements and data assimilation over mountain regions can provide stronger observational constraints for modeling.
Besides orography, frontal forcing is another precipitation mechanism where increasing model resolution is beneficial. Storm tracks are prominent features of the extratropical regions. A cold front can produce narrow bands of precipitation, sometimes with embedded severe rainstorms or snowstorms, and in the warm sector, squall lines and severe thunderstorms are common. High-spatial-resolution and nonhydrostatic models can better capture the temperature gradients and simulate frontogenesis that produces the upward motion responsible for frontal clouds and precipitation.
The land surface, particularly where there is substantial vegetation, plays a significant role in the global hydrologic cycle, but current estimates of evapotranspiration and precipitation are not sufficiently accurate to close the hydrologic cycle, even on an annual basis over relatively large river basins (Lawford et al., 2007; Roads et al., 2003). There are a variety of challenges associated with simulation of the hydrologic cycle in GCMs, some associated with representation of convection and cloud processes (see above), but some connected with issues of resolution and appropriate representation of land-surface processes (e.g., land-surface cover, soil moisture, vegetation, agriculture, and the associated evapotranspiration), as well as feedbacks between the land surface and the atmosphere (Dirmeyer et al., 2012).
Sophisticated regional- and continental-scale models exist for land-surface hydrology, but these models are only coupled with GCMs, through the grid scale, with subgrid variability of essential land-surface processes being forced by grid mean atmospheric forcing. For realistic routing of surface water and representation of land cover, hydrology models require fine resolution (1 km on continental scales, and considerably less in many regional studies). This resolution is essential to predictions of soil moisture and evapotranspiration fluxes to the atmosphere and is also the scale of information needed by water resource managers. Work to couple land-surface hydrology models with atmospheric models is advancing, through direct coupling approaches and through “tiling”’ or “representative land-surface units” (subgrid representation of the landscape), and more sophisticated, energy- and moisture-conserving schemes are needed.
In addition to precipitation, many other processes involving land surface-atmosphere moisture, energy, and chemical exchange at regional scales are expected to be better represented as coupling schemes and resolution improve, for example, influences of land-use changes on the climate, aerosol sources, crop- and biome-specific evapotranspiration rates, and the influence of built structures (e.g., cities, wind farms) on
atmospheric turbulence and forced convection. In return, improved regional-scale climate change forecasts, including, for example, wind, snow, and growing-degree-day forecasts at scales of around 5 km, can feed into climate change impact and adaptation studies for cities, agriculture, tourism development (e.g., ski areas), and renewable energy developments. However, regional projections are more reliable for temperature than for precipitation fields because of the intrinsic scale and complexity of physical processes at play. Improved model fidelity at regional scales is essential to assessment of water resource and agricultural stress and to drought and flood hazards, which are also an element of climate extremes.
Another challenge for global and regional climate models is their representation of patterns or modes of variability, such as ENSO, the Southern Annual Mode, the Arctic and North Atlantic Oscillations, and Pacific decadal variability. Because of their persistence, these ocean-atmosphere patterns strongly influence regional climate variability on time scales of years to decades. If not represented well in models, or if these modes are triggered and sustained at different times in different models, regional climate projections can diverge. Such errors place limits on decadal predictability, particularly on regional scales, and caution is required when interpreting the results from a small number of realizations and/or a small number of models. Work is needed to better understand modes of decadal variability, the underlying ocean-atmosphere feedbacks, and their representation in models.
How Will Climate Extremes Change?
Severe weather events such as tropical cyclones, droughts, floods, and heat waves have tremendous impacts on society, economically, and through loss of life. Extreme events are not predictable years in advance, because most of these reflect an instantaneous state of the weather, with its well-known limitations to prediction. On the other hand, insofar as these events are a function of the mean climate state, statistical probabilities for extreme weather events may be possible to project, which would have great value for decision support and infrastructure design. There are good examples of the ability to extract statistical information on climate extremes from climate models (e.g., Katz, 2010; Kharin and Zwiers, 2005), and experience is growing in the application of advanced statistical methods to assessment of climate hazards and climate change adaptation strategies (Klein-Tank et al., 2009).
For reliable insight from climate models, however, models need to be adept at representing the essential phenomena (e.g., tropical cyclone frequency and strength; tornado development; heavy rain events). Physical arguments and climate models sug-
gest that precipitation extremes are controlled by different physics than time-mean precipitation. Climate models project more frequent floods and droughts in the 21st century, but, as with regional rainfall trends discussed above, intermodel differences in the magnitudes and regional patterns of model trends are substantial, due to many of the same factors. Drought persistence is another example, involving feedbacks between soil moisture, evapotranspiration, atmospheric and surface temperatures, dust aerosols, cloud condensation nuclei, and interactions between regional and synoptic circulation patterns (i.e., blocking). Simulation of these feedbacks requires multiscale modeling with an interactive and sophisticated treatment of land-surface and boundary-layer processes.
Tropical cyclones are only roughly represented in many climate models, primarily because of low spatial resolution of the tight circulation and sharp gradients found in tropical cyclones. Simulations done with very high (25 km or less) resolution models greatly improve the representation of tropical cyclones, even without including the nonhydrostatic effects that are needed to include the vertical component of velocity in the model’s prognostic variables. Some coupled models are now able to simulate interannual variations in the frequency and intensity of tropical cyclones (e.g., National Centers for Environmental Prediction Climate Forecast System [CFS]), and seasonal forecast skill for upcoming hurricane seasons is improving. Seasonal landfall forecasts may be the next frontier.
In most cases, prognoses of severe weather will have to be statistical in nature (i.e., estimation of the likelihood of extreme events in future decades in a specific region). Statistical likelihoods are of great value for many applications, however, such as water resource management, infrastructure and emergency relief planning, and the insurance industry (Box 1.1). It is arguable whether climate models need to generate the full range of behavior and variability that is seen in the real world in order to extract information on extremes. In some cases, probability distribution functions may be constructed and offer appropriate inferences on extremes (e.g., Hegerl et al., 2004). In a nonstationary climate, however, statistical properties of probability distribution functions for some climate phenomena (e.g., the dispersion or shape of distributions) may evolve relative to the historical climate record.
How Quickly Will Sea Level Rise?
Global eustatic (mean) sea-level rise over the past century has been driven by a combination of thermal expansion of the oceans, melting of mountain glaciers and the Greenland ice sheet, and increased dynamical discharge to the oceans in Greenland
and Antarctica (Church and White, 2011). On a regional scale, sea-level change is more complex, involving local land movement (e.g., isostatic, tectonic, or subsidence due to groundwater depletion), atmospheric winds and pressures, regional ocean circulation changes (which influence thermosteric changes), and changes in the gravitational field in response to changing land-surface mass (e.g., Wake et al., 2006). Hazards posed by sea-level rise are most acute when compounded with storm surge events that are superimposed on high tides and the various other factors that drive more gradual, sustained sea-level rise in a region (e.g., Dasgupta et al., 2009). Storm surges arise from tropical cyclones and marine storm events. The challenge of forecasting local and regional sea-level rise and associated hazards is therefore multifaceted. This is a true Earth system problem involving many aspects of climate dynamics and geophysics, including Earth and ice-sheet models that have not traditionally been included in climate modeling efforts.
For global mean sea level, one of the greatest challenges involves simulation of ice-sheet mass balance and ice-ocean interactions. Recent, dramatic changes in the Greenland and Antarctic ice sheets are driven by both increased surface melting and ocean warming at intermediate depths, where marine outlet glaciers and ice shelves are in contact with the sea (e.g., Holland et al., 2008). The current generation of icesheet models and ice-climate models cannot simulate these processes and other aspects of ice-sheet dynamics that give rise to interannual ice-sheet variability. Hence, models are limited in their ability to assess ice-sheet sensitivity to climate change. Most ice-sheet models use only minimal climate data (e.g., temperature and precipitation fields), without interactive or physically based (process-resolving, energy-conserving) coupling with climate models. Coupled ice sheet-climate model simulations to date typically simulate ice-sheet melt as a function of positive degree days, based on interpolated GCM temperature fields (e.g., Huybrechts et al., 2011; Ridley et al., 2005, 2010; Vizcaino et al., 2010); without a proper energy balance at the ice-atmosphere interface, these simulations do not conserve energy. Physical processes at the ice-ocean interface (calving, marine melting) are also neglected or oversimplified in models.
These simplifications limit the range of behavior in modeled ice-sheet interactions with the climate system. For instance, ice-sheet models have no sensitivity to ocean warming, and the climate processes that may give rise to ice-shelf breakup, grounding line retreat, and marine ice-sheet instabilities are not well represented. Improved models of ice dynamics are also needed. Fast-flowing outlet glaciers and ice streams need to be spatially resolved in ice-sheet models, and the controls on fast flow (e.g., basal
lubrication, calving and basal melting at the ice-ocean interface, and grounding line dynamics) need to be better understood and included in the models (Nick et al., 2009).
The intrinsic resolution of these processes will continue to be a challenge. Ice-sheet models require horizontal resolutions of about 5 km to resolve the snow accumulation and melt (energy balance) processes near the ice-sheet margin, where orographic gradients are high. Even greater resolution may be needed where ice is in contact with the ocean to simulate floating ice dynamics, grounding line migration, and fluxes of energy and mass at the ice-ocean interface. The latter requires coupling with regional and/or coastal models of ocean circulation at and below the ice front (i.e., beneath iceshelf cavities).
Interactive two-way coupling is required for simulation of decadal- to century-scale sea-level rise, including energy- and mass-conserving schemes to simulate melt rates at the ice-ocean interface and in the ice-sheet ablation zone. Much of this is technically feasible, and regional-scale modeling studies show promise for both ice-ocean and ice-atmosphere interactions (e.g., Box et al., 2008; Grosfeld and Sandhäger, 2004; Holland and Jenkins, 1999). To extend this to a global scale, considerable numerical and scientific resources need to be channeled at this problem. However, considerable progress can be expected in the next 10-20 years to improve the realism in ice-sheet and sea-level projections.
Steric changes in ocean height also need to be better constrained. The evolution to eddy-resolving ocean models will improve this aspect of sea-level projections through more detailed representation of mixing processes. Increased observational constraints on evolution of intermediate and deep waters will also inform and improve the models. In combination with increasing attention to geophysical processes and local landscape models (e.g., for coastal geomorphology and relative sea-level considerations), improved projections of regional sea-level rise should be possible in the coming decades.
OTHER SCIENTIFIC PRIORITIES
Several other important scientific questions identified above are discussed here. These are arguably of lower priority because they are of more regional interest or represent basic Earth system science processes, which lay the groundwork for longer-term advances in climate modeling. They are nonetheless pressing questions that require advances in climate modeling.
How Will Arctic Climate Change?
The Arctic is an important player in long-term global climate evolution, and it may also contribute to abrupt climate change. The Arctic sea-ice cover is particularly critical in effecting such change, because it insulates Earth’s relatively warm ocean water from the cold atmosphere, and it strongly influences Earth’s absorption of solar radiation through high albedo (or reflectivity) compared with “dark” absorptive ocean. The multiyear icepack acts as a key indicator of the state of global climate through “polar amplification.” Polar amplification is a self-reinforcing system, which amplifies polar climate warming through positive feedback loops derived from decreasing snow and sea-ice coverage.
The retreating sea-ice cover has powerful feedbacks on regional albedo, ocean warming, and cloud conditions. These influences contribute to a strong amplification of climate warming in the Arctic, making it one of the most sensitive and rapidly changing regions on Earth. Because of the geopolitical and environmental ramifications, there is tremendous interest in reliable climate change and sea-ice forecasts for the Arctic region. The complexity and scale of sea-ice processes and ice-ocean-atmosphere exchanges, as well as the relative dearth of subsurface observational data, make this a challenging problem for climate models.
Over the past decade, various studies have attempted to estimate the future trajectory of Arctic climate and have proposed projections of the disappearance of summer Arctic sea ice ranging from the end of this decade to the end of this century. The majority of GCMs, including those participating in the IPCC AR4, have not been able to adequately reproduce satellite-observed Arctic sea-ice extent variability and trends (Stroeve et al., 2007), however, in particular the extent of late-summer ice loss in the past decade. Model representation of sea-ice thickness presents additional challenges because it involves not only thermodynamic interaction with the ocean below but also the dynamic and thermodynamic effects from the atmosphere above.
The inability of climate models to adequately reproduce the recent states and trends of Arctic sea ice diminishes confidence in their accuracy for making future climate predictions. It suggests a great need for improved understanding and model representation of physical processes and interactions specific to polar regions that are omitted from, or poorly represented in, most current-generation GCMs. These processes include the following: oceanic eddies, tides, fronts, buoyancy-driven coastal and boundary currents, cold halocline, dense water plumes and convection, double diffusion, surface/bottom mixed layer, sea-ice thickness distribution, concentration, deformation, drift and export, fast ice, snow cover, melt ponds and surface albedo, atmo-
spheric loading, clouds and fronts, ice sheets/caps and mountain glaciers, permafrost, river runoff, and air-sea ice-land interactions and coupling. There are also a number of important limitations in the way sea ice and ocean models are coupled in current-generation GCMs, which can contribute to pycnocline displacement via Ekman pumping, freshwater water exchange between ice and ocean, or thermohaline coupling at the ice-ocean interface.
To facilitate a better understanding of interconnectivity within the Earth system (Doherty et al., 2009; Rind, 2008; Roberts et al., 2010) work is under way to (a) improve the fidelity and number of polar-centric processes represented within Earth system models, (b) refine coupling channels between them, and (c) expand the hierarchy of available models and observations to help quantify sources of uncertainty and skill in sea-ice simulations. Model development is being targeted toward physical and biogeochemical processes that are suspected of strong interconnectivity with the surface Arctic Ocean energy and mass budgets. By increasing the number of interconnected processes in models, the degrees of freedom of the simulated Earth system expand, which poses problems for understanding causal climatic links and is likely to increase apparent model uncertainty in the next decade (Hawkins and Sutton, 2009). At the same time the need for high-fidelity regional ensemble projections has grown, especially in the Arctic, where economic, social, and national interests are rapidly reshaping the high north in step with regional climate change (e.g., Arctic Council, 2009; Proelss, 2009).
Roberts et al. (2010) proposed the creation of an Arctic System Model (ASM) based around a core climate model configuration comprising an ocean circulation model, atmospheric model, sea-ice model, and terrestrial model. Such a model has been recently developed (Maslowski et al., 2012), and it is currently being evaluated for physical performance. It will have high spatial resolution (5 -50 times higher than currently practical in global models) to advance understanding and modeling of critical processes and determine the need for their explicit representation in global Earth system models. More opportunities for advancing ASMs are under way with the development of a variable-resolution or unstructured-grid approach (Ringler et al., 2010), which shows great promise for bridging the gap and enabling high-resolution regional Arctic climate change exploration within the context of the global climate system model framework. Subject to further progress with its development, including space-dependent physical parameterizations, an improved framework for robust regional Arctic climate system modeling should become available within the next several years. Overall, these different modeling methodologies and results point to the ongoing need for a hierarchical approach (as discussed in Chapter 3) to better understand the past and present states and estimate future trajectories of Arctic sea ice and climate change.
What Is the Potential for Abrupt Change in the Climate System?
Various mechanisms have been identified for abrupt climate change, where the climate state undergoes a regime shift over a period of a decade or less on regional to global scales. Candidate processes include large-scale destabilization and release of methane hydrates from shallow marine and permafrost environments, disruption or reorganization of ocean circulation patterns, loss of sea ice, loss of coral reefs, and desertification (i.e., sustained regional droughts, dieback of tropical rainforests, etc.). These events are thought to be threshold processes where, beyond a certain point, gradual climate change might trigger a nonlinear response. It is not known exactly where the thresholds lie, and whether 21st-century climate change is likely to incite such nonlinear responses, but climate models are the best available tool to address this question.
Many of the abrupt climate change instabilities identified here involve Earth system interactions and feedbacks as discussed in Chapter 3. Examples include cryosphereclimate interactions (permafrost thaw, sea-ice retreat) and the combined impacts of changes in the hydrologic cycle, ocean temperature and salinity, sea-ice formation and melt, and freshwater runoff from rivers, glaciers, and ice sheets on ocean stratification and deepwater formation. The expansion of model complexity and improvements in two-way coupling strategies in Earth system models will help to address and quantify some of these feedbacks and threshold processes. For instance, increasingly more sophisticated sea-ice and Arctic Ocean models allow a better assessment of interannual to interdecadal sea-ice variability and the “reversibility” of recent, dramatic reductions in late-summer sea ice (Armour et al., 2011). Similarly, the addition of more sophisticated models of permafrost thermodynamics, including soil biogeochemistry and vegetation, will enable a better assessment of methane release from thawing permafrost.
Other aspects of abrupt climate change involve improvements to the fundamental representation of tropical convection and rainfall patterns, as discussed above with respect to climate sensitivity. Of particular concern here are patterns of tropical and subtropical aridity, including those of North Africa and the Amazon Basin. Agricultural, ecological, and water resource stresses in these two regions have the potential for global-scale impacts (e.g., Betts et al., 2008). Sustained, systematic drying of the Amazon Basin is predicted in some modeling studies, and the likelihood of such highimpact climate shifts needs to be quantified and constrained, requiring improvements in modeled tropical convection, representation of the ITCZ, and possibly land-surface coupling (i.e., for transpiration fluxes and land-cover changes).
How Will Marine and Terrestrial Ecosystems Change?
Ocean warming, acidification, and changes in salinity all affect biogeochemical cycles and marine ecology on local to global scales, threatening ecological integrity and biological diversity in the oceans, which are intrinsically valuable to the planet. This threat has significant implications for the fishing industry and global food security. Marine biological activity also plays a large role in carbon uptake from the atmosphere, with important feedbacks to climate warming. Climate models capable of assessing marine ecology are needed to examine this.
Models of ocean biogeochemistry have been developed and coupled in GCMs, but details of ocean mixing and coastal upwelling are integral to nutrient cycles; these need to be resolved to enable consideration of marine ecosystems and ecological response to changing ocean temperature, salinity, and pH. The anticipated progression to eddy-resolving and multigrid ocean modeling will improve model simulations of mixing, mesoscale eddies, and coastal ocean dynamics, permitting coupling of models of ocean dynamics, ocean biogeochemistry, and marine ecology.
Terrestrial ecosystems are important in the Earth system because they influence the climate through physical, chemical, and biological processes that affect the hydrologic cycle and atmospheric composition. Warming and drying of the climate will potentially induce a shift of plant zones to more drought-resistant varieties and species, alter pest and predator patterns, and shift forest fire regimes in time and space. Climate change will also interfere with the timing of various temperature-related events (e.g., blooming or egg laying) and the cold end of species’ ranges (e.g., toward the poles or higher elevations; NRC, 2011b). Linkages between species that are temperature, moisture, or annual cycle dependent will also be disrupted.
Climate models capable of assessing terrestrial ecology are also needed. These models should represent the drivers and feedbacks from global and regional interactions of climate, ecosystem processes, plant function (e.g., photosynthesis and respiration), carbon and nitrogen dynamics of soils, and ecosystem disturbances (e.g., drought, flooding, and insect outbreak; NRC, 2010b).
How Will Society Respond to and Feed Back on Climate Change?
Future climate evolution will be impacted by human choices in a number of ways, including future emissions scenarios (e.g., through population, energy intensity, and sources of energy), land-use changes, agricultural activities, and potentially through
deliberate interventions in the climate system, so-called geoengineering activities (e.g., injection of reflective aerosols in the stratosphere to reduce insolation). Emissions, land-use changes, and patterns of development are presently prescribed in climate simulations through predetermined scenarios, without allowing for feedbacks or societal “reactions” in response to the patterns and extent of climate change. A great deal of thought goes into these scenarios (e.g., the Representative Concentration Pathway scenarios of the Coupled Model Intercomparison Project, Phase 5 (CMIP5)/IPCC AR5 [Moss et al., 2010]), but they are not exhaustive and are not always consistent with the internally modeled land-surface changes and atmospheric chemistry. The prescribed scenarios also neglect interactive feedbacks with respect to climate mitigation policy or societal choices concerning things like land use or energy systems.
There is increasing interest in introducing interactive human influences in climate models. Increasingly more sophisticated dynamic vegetation models are now being employed in GCMs, but it is difficult to accommodate the influence and impact of human land-use choices in future climate projections. Agricultural practices (i.e., crop selection) depend on the climate, but they also feed back on climate and hydrologic conditions. Forestry and fishery practices, urbanization, and energy systems all have similar two-way implications within the climate system. Many of these effects are implicitly included in future emissions scenarios, but there is an opportunity to develop coupled, dynamic models of human interactions with the climate system to better capture these feedbacks and interactions. Early attempts in that direction are currently under way, including the addition of algorithms for different crop types that simulate changes in crop planting, growth, and harvesting due to human land-surface management in a changing climate (Levis et al., 2012).
Can the Evolution of the Climate System over the Next Decade Be Predicted?
It is not yet known whether climate models can predict climate system evolution on annual to decadal time scales (Meehl et al., 2009). Sensitivity to radiative forcing is reasonably well modeled, but climate evolution is also sensitive to initial conditions and internal variability. This is a challenging problem because of sensitivity to imperfectly known initial conditions, and because internal, natural variability that occurs within models (e.g., ENSO) does not necessarily arise at the same time as similar variability that occurs in nature. The future timing of other climatic influences, such as volcanic events, is also unknown. Thus, the extent to which annual to decadal predictive skill can be reasonably expected in climate models is limited, and at present it seems unlikely that, even in a decade, climate models will have high skill in predicting soci-
etally relevant deviations from “normal” climate over lead times of 2-10 years (i.e., the interval between ENSO and the effect of climate change trends). However, ensemble forecasts that span a statistical space of possibilities are not precluded. Work is needed to understand and quantify the uncertainty associated with such forecasts. To improve forecasts, specific research goals should be set for improving understanding of sources of predictability (NRC, 2010c).
Given the uncertainty in many initial and boundary conditions, particularly with respect to ocean and sea-ice conditions (see section above), model forecasts lay out a range of possible futures, even for a single climate model with the same set of physics and future emissions scenarios (e.g., Laprise et al., 2000; Wu et al., 2005). This manifests particularly strongly in regional climate models, which take large-scale climate fields as boundary forcing. Some of this sensitivity to initial and boundary conditions may be numerical (i.e., model inaccuracies that result in drift), and some is intrinsic to climate dynamics.
Over a long enough period, e.g., 30 years, it may be insignificant that modeled El Niño years differ from reality, because ENSO cycles are relatively short lived. Some patterns of internal climate variability are decadal in nature, however (e.g., the Atlantic Multidecadal Variability [AMV] and Pacific Decadal Variability [PDV)]). Models can reproduce much of this decadal variability (e.g., Meehl et al., 2009; Troccoli and Palmer, 2007), but there is considerable intermodel variability in the timing and duration of such internal variability. Even within the same model, multiple realizations with different initial conditions can give divergent timing of modeled decadal variability, indicating potential limits to decadal-scale regional forecast skill (Meehl et al., 2009; Murphy et al., 2008). Improvements may be possible through data assimilation methods of climate modeling, and through expanded observational data on ocean conditions for model initialization. Such methods show promise for seasonal forecasts using numerical weather prediction models, with demonstrable predictive skill on seasonal time scales for ENSO, for instance (e.g., Tippett and Barnston, 2008).
On a global scale, decadal projections may be less problematic. Patterns of internal variability, such as the AMV and PDV, result in regional-scale redistribution of energy and moisture but lesser impacts on global mean conditions. Predictions of global average temperature depend more on, for example, external forcing; for a given global scenario, however, some regions will warm more than others and some will be less affected, as a result of internal variability and the response of circulation systems to the cumulative climate forcing. The degree of irreducible uncertainty in decadal-scale projections is therefore greater on regional scales than it is for global means. Feedbacks arising from a given circulation pattern (e.g., cloud feedbacks or sea-ice conditions)
can in fact influence radiative forcing and global average temperatures, so the effects of internal, interannual variability have the capacity to influence global conditions.
MECHANISMS FOR CLIMATE MODEL IMPROVEMENTS
As discussed in Chapter 3, model improvements to address these research frontiers will be achieved through three main mechanisms: (i) development of Earth system models (increasing model complexity); (ii) improvements to the existing generation of atmosphere-ocean models through improved physics, parameterizations, and computational strategies, increased model resolution, and better observational constraints; and (iii) improved coordination and coupling of models at global and regional scales, including shared insights and capabilities of modeling efforts in the climate, reanalysis, and operational forecast communities. Progress is likely through a combination of these three mechanisms. The climate modeling community is already pressing on the first two points, advancing Earth system models and refining model physical parameterizations and resolution, and continued progress is needed on both of these fronts, perhaps more strategically focused on high-priority questions. In the committee’s opinion, the third point, coordination of global and regional modeling efforts, as well as “research-oriented” versus operational models, is a weak spot in the U.S. national climate modeling effort, and also an opportunity for advances.
Earth System Models
Several research frontiers can be explored through development of more sophisticated, interactive, and complete Earth system models. A number of examples are discussed above, such as coupling of climate models with models of ice sheets, landsurface hydrology, aerosols, permafrost, and human interactions. In these examples, additional complexity is needed and justified to address high-priority questions. Earth systems model development may yield significant progress in the next 20 years for a number of scientific questions. In some cases this development is a matter of improved coupling between systems (i.e., coupling schemes that conserve energy, mass, and momentum; two-way coupling, where possible, to include feedback processes), as the component models are already quite sophisticated. Model components (e.g., land-surface hydrology and ice-sheet models) need to be resolved and coupled at the natural scale of the relevant processes, where possible.
In general, there is a tension between increasing model complexity and the ability to interpret model results, or even the ability of coupled models to generate meaningful
results. There is experience in this from ocean-atmosphere modeling. For instance, if modeled wind fields are unrealistic in a region, such errors will propagate in the modeled ocean dynamics, including critical features like coastal upwelling, mixing, or ENSO simulations. Such errors can grow and feed back, causing a cumulative drift from reality. Meaningful projections of sea-level rise cannot be justified by adding an ice-sheet model into a climate model; the climate models must be able to generate realistic mass balance fields (snow accumulation and melt) over the ice sheets, and the critical ocean-ice processes have to be understood and included. The building of Earth system models requires extensive testing and adaptive code development, and progress can be slow.
Paleoclimate simulations are one avenue of research to exploit Earth system models and deepen understanding about climate dynamics. Climate variations in the past, such as the Pleistocene glacial cycles, offer insights into the inner workings of the climate system, including important questions such as climate sensitivity, the sign and strength of different climate feedbacks, and processes involving (for example) ice sheets, sea level, aerosols, marine ecology, and the carbon cycle. More subtle climate events in the recent past, such as the Medieval Warm Period and the Little Ice Age, also provide examples of natural variability that can aid in understanding climate dynamics. These events are not fully understood, and they offer exceptional targets for climate modeling studies; lessons from the past can inform process representation in climate models that are used for future projections.
Multimillennial problems such as glacial cycles may be difficult to tackle with full climate models in the next 10 years due to the long integration times, but there are many potential insights from Earth system models of intermediate complexity and reduced Earth system models (see Chapter 3). The last glacial cycle is a particularly good modeling target because it involves numerous important feedbacks and processes, including important fluctuations in the global carbon cycle. Carbon sinks during the glaciation provided an important feedback to the orbitally triggered cooling and icesheet advance, but the exact mechanisms of carbon storage on land and in the ocean are not yet understood. Similarly, there is an incomplete understanding of the roles of permafrost, the hydrologic cycle, and changes in large-scale ocean and atmospheric circulation during glacial-deglacial transitions, millennial-scale climate variability, and potentially abrupt (decadal-scale) climate transitions during the glacial period; improving this understanding is a superb modeling target for Earth system models.
Climate changes over the past two millennia have been more modest, but they are relatively well understood, spatially and temporally, and they provide another good target for Earth system models. Climate variability over this period is largely associated
with fluctuating solar and volcanic activity, but land-use changes and internal (oceanatmosphere-ice-biosphere) climate dynamics may also play a role in both climate forcing and positive and negative feedbacks that amplify or buffer such forcing. Climate models need to be able to provide realistic representations of large-scale events such as the Medieval Warm Period and the Little Ice Age before we can be confident in their ability to replicate natural climate variability. Such representations would provide assurance that the critical processes and Earth system components that give rise to natural variability are adequately represented in future projections, so that natural and anthropogenic forcing can be separated. Model studies of these periods in recent Earth history can also provide an observational constraint on modeled climate sensitivity.
Finding 4.1: Earth system model development over the next 20 years is expected to provide a more complete representation of climate system interactions and feedbacks. This will improve the physical representation of several critical features of climate, such as sea-level rise, sea ice, carbon-cycle feedbacks, ecosystem changes, and the hydrologic cycle.
In addition to new model capacity created through Earth system model development, increased resolution and improved physics in GCMs will drive progress on a number of longstanding scientific problems in the ocean-atmosphere system. Some of this will occur through incremental, “business-as-usual” advances, although progress on some fronts requires strategic investments and prioritization. It is important to recognize that some longstanding problems may not be resolved because of complex, nondeterministic, or poorly understood physics as well as the reality that some essential processes occur at the molecular scale (e.g., cloud physics) and are not amenable to global-scale modeling.
Progress in modeling clouds offers a good example of how advances may be possible in model parameterizations and scale issues. Such examples are found in many other aspects of climate modeling as well (e.g., sea-ice dynamics). Cloud-related parameterizations, like other major parameterizations in climate models, contain multiple numerical parameters not fully constrained by process modeling and observations, for example, the “lateral entrainment rate” at which air is turbulently mixed into cumulus updrafts, or fall speeds of ice and snow particles. These parameters are typically “tuned” via trial and error to optimize the quality of overall global and regional simula-
tions of cloud cover/depth/thickness, precipitation, and top-of-atmosphere radiative fluxes.
Using clouds as a testbed, some promising new approaches to improving parameterizations are being explored, including perturbed parameter ensembles to explore the range of simulated climates possible by changing parameters within an individual climate model, uncertainty quantification to systematically optimize uncertain parameters, and stochastic parameterization. Traditional parameterizations give a single best-guess estimate of the aggregate effect of a subgrid process such as turbulence or clouds averaged over a grid cell. Stochastic parameterization instead provides a random plausible realization of that aggregate effect, drawn from an appropriate probability distribution function. A conventional parameterization of subgrid fractional cloud cover might specify it in terms of the grid-mean relative humidity, while a stochastic parameterization will randomly choose a cloud cover scattered around that deterministic value. This can help maintain grid-scale variability that conventional parameterizations may artificially damp. Stochastic parameterization has been successfully demonstrated in numerical weather prediction (e.g., Buizza et al., 1999; Palmer et al., 2009; Shutts and Palmer, 2007) and monthly to seasonal prediction (Weisheimer et al., 2011).
A nonstochastic parameterization of a random subgrid process such as cumulus convection cannot produce statistically robust results unless there are many cumulus clouds in each grid cell. As the spatial and temporal resolution in climate models is refined, this “scale-separation” assumption breaks down well before a single cumulus cloud is well resolved by the model grid, creating a “grey zone” in which neither the parameterization nor an explicit simulation of the process is theoretically justified. Many global weather prediction models are approaching that resolution for cumulus convection, and climate models are likely to do so within the next 20 years. Designing parameterizations that can function through this range of resolutions is an important challenge for the next decade. Stochastic parameterization may be a particularly useful strategy in the grey zone.
While a revolution in computational approaches or capabilities is not impossible, in simulating clouds and in the broader challenges of climate modeling, incremental improvements are more likely. Improvements are possible by tapping into model capabilities that already exist in some cases, through strategic cooperation of the sometimes disparate global and regional modeling streams, as well as increased cooperation of global, regional, research-based, and operational modeling efforts. Such improvements will involve unified, scale-invariant physical treatments of key processes, conservative coupling schemes, and, in some cases, two-way coupling.
Finding 4.2: Progress is likely on a number of important problems in climate modeling over the coming decades through a combination of increasing model resolution, advances in observations and process understanding, improved model physical parameterizations and stochastic methods, and more complete representations of the Earth system in climate models.
THE WAY FORWARD
There is generally a tension between different lines of progress in climate modeling. For instance, do we allocate resources to increased resolution or to increased model complexity (i.e., Earth system model development)? There is no one-size-fits-all answer, but instead the approach should be problem driven. Some problems that are of great societal relevance, such as sea-level rise and climate change impacts on water resources, require increased model complexity, and progress is likely through the addition of new model capabilities (ice-sheet dynamics and land-surface hydrology, in these examples). In other cases, such as improved model skill in regional precipitation and extreme weather forecasts, increased resolution and “scalable” physical parameterizations are the highest priorities for extending model capabilities. Other problems, such as water resource management, require both increased resolution and complexity.
The committee finds that an important direction forward is for Earth system models to be developed with realistic representations of ice-sheet dynamics and ice-ocean-atmosphere interactions in order to provide improved projections of sea-level rise. Such models will also improve understanding of glacial-interglacial cycles and millennialscale climate variability during glacial periods. Coupled with sophisticated models of terrestrial and marine carbon cycles, investigations of glacial cycles could shed light on natural carbon sources and sinks and the future evolution of the atmospheric carbon pool.
A number of important scientific and societal questions require detailed and meaningful climate projections at local to regional scales. The committee recommends that the U.S. climate modeling community pursue high-resolution model runs in the coming decades. Specifically, at least one national modeling effort in the next decade should aim to simulate historical and future climate change (i.e., the period 19002100) at a resolution of less than 5 km, to enable eddy- and cyclone-resolving models of ocean dynamics and more realistic representation of land-surface exchanges with the atmosphere. In addition, at least one national modeling effort in the next 20 years should aim for century-scale simulations at resolutions of 1-2 km, to allow cloud-
resolving physics. There is ample evidence that resolving these highly interactive, nonlinear, and thermodynamically irreversible processes provides for qualitatively better simulations of Earth’s climate. Such a resolution will permit improved representation of many features of the climate, such as explicit resolution of mesoscale ocean eddies and the spatial scales of land-surface and hydrologic variability.
The committee recognizes that these suggested efforts are not trivial and will require a substantial investment in manpower, computing power, and financial capital. It is also not certain that increases in resolution will reduce uncertainty. However, improvements in model capability and resolution can be expected to advance understanding of the high-priority climate science questions discussed in this chapter. The “grand challenges” outlined here all refer to societally relevant questions where progress can be anticipated in the next 10-20 years, with highest priority given to the questions of climate sensitivity, regional climate change, climate extremes, and sea-level rise. Each of these is central to provision of critical information for climate policy decisions and climate change adaptation.
Recommendation 4.1: As a general guideline, priority should be given to climate modeling activities that have a strong focus on problems that intersect the space where (i) addressing societal needs requires guidance from climate models and (ii) progress is likely, given adequate resources. This does not preclude climate modeling activity focused on basic research questions or “hard problems,” where progress may be difficult (e.g., decadal forecasts), but is intended to allocate efforts strategically.
Recommendation 4.2: Within the realm where progress is likely, the climate modeling community should continue to work intensively on a broad spectrum of climate problems, in particular on longstanding challenges such as climate sensitivity and cloud feedbacks that affect most aspects of climate change (regional hydrologic changes, extremes, sea-level rise, etc.) and require continued or intensified support. Progress can be expected as resolution, physical parameterizations, observational constraints, and modeling strategies improve.
Recommendation 4.3: More effort should be put toward coordinated global and regional climate modeling activities to allow good representation of landsurface hydrology and terrestrial vegetation dynamics and to enable improved modeling of the hydrologic cycle and regional water resources, agriculture, and drought forecasts. This will require better integration of the various national climate modeling activities, including groups that focus on models of surface hydrology and vegetation dynamics. The annual climate modeling forum discussed in Chapter 13 might provide a good vehicle for a working group with this focus.
Recommendation 4.4: At least one national modeling effort in the next decade should aim to simulate historical and future climate change (i.e., the period 1900-2100) at a resolution of less than 5 km, to enable eddy-resolving models of ocean dynamics and more realistic representation of cumulus convection and land-surface exchanges with the atmosphere. Parallel efforts need to aim for century-scale global atmospheric simulations at 1-2 km, to enable cloud-resolving physics. These national efforts would be facilitated by advances in climate model software infrastructure and computing capability discussed in Chapter 10.