Review

Climate, ecosystems, and planetary futures: The challenge to predict life in Earth system models

See allHide authors and affiliations

Science  02 Feb 2018:
Vol. 359, Issue 6375, eaam8328
DOI: 10.1126/science.aam8328

Integrating the biosphere into climate models

High-quality climate predictions are crucial for understanding the impacts of different greenhouse gas emission scenarios and for mitigating and adapting to the resulting climatic changes. Bonan and Doney review advances in Earth system models that include the terrestrial and marine biosphere. Such models capture interactions between physical and biological aspects of the Earth system. This provides insight into climate impacts of societal importance, such as altered crop yields, wildfire risk, and water availability. Further research is needed to better understand model uncertainties, some of which may be unavoidable, and to better translate observations into abstract model representations.

Science, this issue p. eaam8328

Structured Abstract

BACKGROUND

Earth system models (ESMs) simulate physical, chemical, and biological processes that underlie climate and are the most complex in a hierarchy of models of Earth’s interacting atmosphere–land–ocean–sea ice system. As terrestrial and marine ecosystems have been added to ESMs, the distinction between the physical basis for climate change, mitigation, and vulnerability, impacts, and adaptation (VIA) no longer necessarily holds. The same global change stresses that affect terrestrial and marine ecosystems are critical processes that determine the magnitude and trajectory of climate change, and many of the interventions that might lessen anthropogenic climate change pertain to the biosphere. Here we describe environmental changes that are stressing terrestrial and marine ecosystems. We discuss how these stressors are being included in ESMs, initially with an emphasis on climate processes, but also show their emerging utility for VIA analyses and examine them in the context of Earth system prediction.

ADVANCES

Terrestrial ecosystems face stresses from changing climate and atmospheric composition that alter phenology, growing season length, and community composition; these stresses enhance productivity and water-use efficiency in some regions, but also lead to mortality and increased disturbances from wildfires, insects, and extreme events in other regions. The addition of reactive nitrogen, elevated levels of tropospheric O3, and anthropogenic land-use and land-cover change stress ecosystems as well. The terrestrial biosphere models included in ESMs simulate the ecological impacts of these stresses and their effects on Earth system functioning. Ocean ecosystems and living marine resources face threats from ocean warming, changing large-scale circulation, increased vertical stratification, declining oxygen, and acidification, which alter nutrient supply, the light environment, and phytoplankton productivity; result in coral bleaching; and produce novel marine communities. Three-dimensional ocean models simulate the carbon cycle and associated biogeochemistry. Plankton ecosystem models both drive biogeochemistry models and characterize marine ecological dynamics.

OUTLOOK

The untapped potential of ESMs is to bring dispersed terrestrial and marine ecosystem research related to climate processes, VIA, and mitigation into a common framework. ESMs offer an opportunity to move beyond physical descriptors of atmospheric and oceanic states to societally relevant quantities such as habitat loss, water availability, wildfire risk, air quality, and crop, fishery, and timber yields. To do so, the science of climate prediction has to be extended to a more multifaceted Earth system prediction, including the biosphere and its resources. ESMs provide the means not just to assess the potential for future global change stresses, but also to determine the outcome of those stresses on the biosphere. Such Earth system prediction is necessary to inform sound policy that maintains a healthy biosphere and provides the food, energy, and fresh water needed for a growing global population without further exacerbating climate change. Substantial impediments that must be overcome include advancing our knowledge of biosphere-related climate processes; reducing model uncertainty; and effectively communicating among, rather than across, the disparate science communities of climate prediction, global biosphere modeling, VIA analyses, and climate change mitigation.

The various models used for climate projections and mitigation and VIA analyses overlap in scope and would benefit from a broad perspective of Earth system prediction.

Shown are the domains of ESMs, mitigation models, and VIA models along axes from VIA to climate processes (horizontal) and from primarily serving the research community to informing societal needs (vertical). Panels show forests and agriculture (left) and marine ecosystems (right) as represented across modeling domains.

Abstract

Many global change stresses on terrestrial and marine ecosystems affect not only ecosystem services that are essential to humankind, but also the trajectory of future climate by altering energy and mass exchanges with the atmosphere. Earth system models, which simulate terrestrial and marine ecosystems and biogeochemical cycles, offer a common framework for ecological research related to climate processes; analyses of vulnerability, impacts, and adaptation; and climate change mitigation. They provide an opportunity to move beyond physical descriptors of atmospheric and oceanic states to societally relevant quantities such as wildfire risk, habitat loss, water availability, and crop, fishery, and timber yields. To achieve this, the science of climate prediction must be extended to a more multifaceted Earth system prediction that includes the biosphere and its resources.

Human activities are transforming Earth’s atmosphere, ocean, and land surfaces at a scale and magnitude not previously seen during the past several thousand years of human history. These changes threaten healthy planetary functions and socioeconomic well-being (1, 2). Fossil fuel combustion, industrialized agriculture, urbanization, and other facets of modern human societies are changing climate and atmospheric composition; melting permafrost, glaciers, ice sheets, and Arctic sea ice; raising sea levels; warming and acidifying the oceans; polluting air, water, and soils; altering biogeochemical cycles and freshwater availability; increasing the cycling of reactive nitrogen; reducing forest cover and degrading land; and destroying habitats and reducing biodiversity (35). The ecological consequences of these changes are apparent in individual organisms, the communities they inhabit, and the ecosystems in which they function (68).

The interconnectedness and global scope of this changing environment have transformed the scientific study of Earth as a system. It is now understood that climate change must be studied in terms of a myriad of interrelated physical, chemical, biological, and socioeconomic processes. This broadening basis for climate change research underlies the transformation from global climate models to Earth system models (ESMs). These models have shown that the biosphere not only responds to climate change, but also directly influences the direction and magnitude of climate change. Terrestrial and marine ecosystems, and their uses by humans, are fundamental to addressing the climate change problem. How do we provide the food, energy, and fresh water needed for a growing global population without further exacerbating climate change? Can terrestrial and marine ecosystems be managed to reduce greenhouse gas emissions? With the advent of ESMs, climate science is no longer limited to the physical basis for climate projections, but also includes projections of the biosphere—for example, regarding carbon storage on land and in the ocean, forest dieback, wildfires, crop yield, and fisheries and marine resources.

However, the study of climate change is still often parsed into separate activities of observing changes and deducing causes (3); assessing the vulnerability, impacts, and adaptation (VIA) of natural and human systems to these changes (6, 7); and determining the socioeconomic transformations needed to mitigate them (9). The untapped potential of ESMs is to bring these dispersed activities into a common framework. There has been success, for example, in coordinating climate projections with the integrated assessment models that identify the societal transformations needed to mitigate climate change (10) and even some initial attempts at directly coupling ESMs and integrated assessment models (11). As terrestrial and marine ecosystems have been added to ESMs, the distinction between the physical basis for climate change, VIA, and mitigation no longer necessarily holds. Land-use and land-cover change, for example, is driven by socioeconomic needs for food, fiber, and fuel, but is also an ecological problem that alters habitat and biodiversity and a means to mitigate anthropogenic CO2 emissions (4).

In this Review, we discuss the treatment of the biosphere in ESMs, considering terrestrial and marine ecosystems as they are now represented in the models, exploring how ESMs can be used to study the biosphere, and highlighting opportunities for future research. We then describe environmental changes that are occurring globally and that are stressing terrestrial and marine ecosystems and show how these stresses are included in ESMs, in the past primarily with an emphasis on climate processes, but now with additional utility for VIA and mitigation research. Last, we examine these stresses in the context of Earth system prediction. Our list of stressors is not meant to be exhaustive. Rather, we highlight several key stressors and their coincidence among climate processes, VIA, and mitigation with the goal of initiating a dialog among the scientific communities that study climate change. This Review is timely because it identifies synergies across the climate and ESM research communities involved in the next Coupled Model Intercomparison Project (CMIP6) (12), which provides an unparalleled opportunity to model and analyze the Earth system.

Earth system models

ESMs simulate physical, chemical, and biological processes that underlie climate. They are the most complex in the ongoing evolution of global models of Earth’s atmosphere, ocean, cryosphere, and land (Fig. 1). Climate models focus on the physical climate system, as represented by atmosphere, ocean, and sea ice physics and dynamics and land surface hydrometeorology. In climate models, land and ocean are coupled with the atmosphere through energy and momentum fluxes and the hydrologic cycle. ESMs have the same representation of the physical climate system but additionally include the carbon cycle, terrestrial and marine ecosystems and biogeochemistry, atmospheric chemistry, and natural and human disturbances. ESMs typically couple distinct component modules for land, atmosphere, and ocean physics, and ecosystem dynamics and biogeochemistry are embedded into these modules.

Fig. 1 Representation of the biosphere in Earth system models (ESMs).

The top panel shows land and ocean as included in climate models, and the bottom panel shows the additional processes included in ESMs. ESMs simulate atmospheric CO2 in response to fossil fuel emissions and terrestrial and marine biogeochemistry. Some ESMs also simulate atmospheric chemistry, aerosols, and CH4. Terrestrial processes shown on the left side of the diagram include biogeophysical fluxes of energy, water, and momentum; biogeochemical fluxes; the hydrologic cycle; and land-use and land-cover change (13). The carbon cycle includes component processes of gross primary production (GPP), autotrophic respiration (RA), litterfall, heterotrophic respiration (RH), and wildfire. Carbon accumulates in plant and soil pools. Additional biogeochemical fluxes include dust entrainment, wildfire chemical emissions, biogenic volatile organic compounds (BVOCs), the reactive nitrogen cycle (Nr), and CH4 emissions from wetlands. Ocean processes are shown on the right side of the diagram. Physical processes include sea ice dynamics, ocean mixing and circulation, changes in sea surface temperature (SST), and ocean-atmosphere fluxes. The gray shaded area depicts the marine carbon cycle, consisting of the phytoplankton-based food web in the upper ocean, export and remineralization in the deep sea and sediments, and the physiochemical solubility pump controlled by surface-deep ocean exchange (100).

A prominent feature of ESMs is their inclusion of the biosphere and abiotic interactions that together make up an ecosystem. On land, terrestrial ecosystems are represented in ESMs by the type of vegetation, the amount of leaf area, the stomata on leaves, and carbon and nitrogen pools (13). Similarly, ESMs simulate ocean phytoplankton production of chlorophyll that influences the vertical profile of light absorption in the upper ocean, which in turn affects model sea surface temperature and mixed layer dynamics, as well as large-scale ocean circulation, heat transport, and climate variability (14, 15).

Biogeochemical cycles were added to ESMs because of the potential for large climate feedbacks arising from the carbon cycle. Terrestrial ecosystems and the ocean together absorb about one-half of the annual anthropogenic CO2 emissions (16), but the future efficacy of these sinks is uncertain (17). Biogeochemical processes on land encompass spatial scales from leaves to plant canopies and from ecosystems to landscapes to biomes (13). Temporal scales include near-instantaneous physiological responses (e.g., stomatal conductance, photosynthesis, and respiration) to prevailing environmental conditions; the seasonal emergence and senescence of leaves; and changes in ecosystem structure and biogeography over decades and centuries in response to natural disturbances (e.g., wildfires), anthropogenic disturbances (e.g., land-use transitions), and climate change. Ongoing model development aims to more authentically represent plant demography and life history characteristics using cohorts of individuals of similar functional traits in vertically structured plant canopies (18).

The three-dimensional carbon cycle models used to estimate ocean uptake of anthropogenic CO2 evolved from model tracer studies of ocean physical circulation. Biogeochemical models additionally track natural cycling of inorganic carbon, alkalinity, macronutrients (nitrogen, phosphorus, and silicon), and often O2; net organic matter and CaCO3 production and export from the surface ocean; particle sinking and respiration and remineralization at depth; and air-sea CO2 (and O2) gas exchange (19). Plankton ecosystem models that simulate interactions of phytoplankton, zooplankton, nutrients, and detrital pools arose both to drive biogeochemistry models and to characterize marine ecological dynamics, such as seasonal phytoplankton blooms. Recent biogeochemical developments include incorporation of iron and other trace elements (20), iron limitation being a major controlling factor for phytoplankton growth in much of the ocean; more sophisticated treatment of marine biological nitrogen fixation and denitrification (21); coastal inputs of nutrients; and ocean acidification. Major model advances under way involve expansion of plankton biological complexity to incorporate functional groups, trait-based dynamics, and biodiversity (2224) and efforts to integrate simulated plankton productivity with fisheries catch (25).

An active research frontier for ESMs is incorporating more extensive chemistry-climate interactions. Additional reactive nitrogen affects climate through enhanced terrestrial carbon storage, emissions of N2O, and chemical reactions that determine the amount of tropospheric O3, CH4, and aerosols (5, 26). Atmospheric deposition of nitrogen to the surface ocean can enhance biological productivity in low-nutrient subtropical regions; globally, however, marine biogeochemistry may be more sensitive to anthropogenic iron deposition (27, 28). Increased concentrations of tropospheric O3 decrease plant productivity and reduce the terrestrial carbon sink, but the appropriate way to parameterize this in models is uncertain (29, 30). Emissions of biogenic volatile organic compounds from terrestrial ecosystems influence atmospheric concentrations of O3, CH4, and secondary organic aerosols (31). Wetlands are an important source of CH4, as are permafrost soils and hydrates. Global models of wetlands and CH4 emissions are being developed (32), and some ESMs include methane chemistry in their climate projections (33). In the ocean, more diverse biogeochemistry is needed (e.g., trace gases such as dimethyl sulfide) in models to link ocean-atmosphere chemistry.

Natural and human disturbances are a continuing research priority for ESMs. Wildfires affect climate and air quality through emissions of long-lived greenhouse gases, short-lived reactive gases, and aerosols and by altering surface albedo (34). Wildfires are included in ESMs, but our ability to model the precise details of fire regimes is limited (35). The mountain pine beetle epidemic in western North American forests has reduced terrestrial carbon uptake (36), increased surface albedo (37), and warmed the surface by reducing evapotranspiration (38). Efforts to represent insect outbreaks in ESMs are promising but still nascent (39). Human disturbances include land-use transitions (e.g., deforestation, reforestation, farm abandonment, and shifting cultivation) and wood harvest (40). Global models of crop growth are included in ESMs, but specific cultivars, time of planting, crop rotation, and other management practices are lacking (41, 42). Nor is forest management and commercial timber production included, despite having an effect on temperature similar to that of land-cover change (43).

Planetary stresses and climate feedbacks

The inclusion of terrestrial and marine ecosystems in ESMs enables study of the global change stresses on the biosphere and feedbacks with climate change (Table 1). A prominent signal on land is a “greening” of the biosphere, although this is partially countered by increased tree mortality and disturbance. Novel community assemblages are likely to emerge that depend on the magnitude and rate of climate change and the ability of species to adjust to these changes through dispersal. Marine ecosystems also face numerous threats from climate change (4446). Surface ocean waters are warming globally and freshening at high latitudes; together, these trends act to increase vertical physical stratification, resulting in altered regional patterns of nutrient supply, light environment, and phytoplankton productivity. Ocean warming leads to shifts in plankton seasonal phenology and poleward migration of plankton, invertebrate, and fish species; coral bleaching; and sea ice loss in polar marine ecosystems. Climate change is projected to alter the spatial patterns and size of marine wild-caught fisheries (47) and may potentially change marine disease outbreaks (48). Marine communities and ecosystems also may be reorganizing into novel assemblages, requiring more sophisticated ESMs that can track in more detail the effects of warming on plankton community composition and trophic interactions (49).

Table 1 Planetary stresses faced by terrestrial and marine ecosystems.

View this table:

Human activities imperil ecosystems and biota in ways other than direct climate change (Table 1). Additional reactive nitrogen alters biodiversity; terrestrial, freshwater, and marine biogeochemistry; and water and air quality. Anthropogenic aerosols increase the amount of diffuse solar radiation, which can enhance terrestrial productivity. High concentrations of tropospheric O3 can cause stomata to close and thus decrease plant productivity and transpiration. Vast areas of forests have been cleared over the industrial era, and many of the remaining forests are managed or secondary rather than old-growth primary forests. About one-third of the ice-free land is covered by cropland or pastureland, and much of the terrestrial productivity has been appropriated for human uses.

Marine impacts arise from ocean acidification owing to increasing atmospheric CO2 and deoxygenation from climate-related circulation changes. Regional stresses, particularly on continental shelves and some parts of the open ocean, are occurring from overfishing, anthropogenic noise, seabed habitat destruction, pollution, and coastal eutrophication, as well as loss of coastal wetlands, mangrove forests, and seagrass beds owing to development and sea level rise (50).

Vulnerability, impacts, and adaptation

Agriculture and food security, forest and water resources, terrestrial ecosystems, and fisheries and marine ecosystems are facets of the biosphere that sustain socioeconomic well-being. Assessing the impact of climate change on these goods and services, their vulnerability to disruption in a changing climate, and the adaptations needed to maintain their future availability is critical for informing sound climate policies (6, 7). Such assessments are commonly obtained by using climate projections to drive models of terrestrial ecosystems (51), crop yield (52), water availability (53), and fisheries and marine ecosystems (54). This indirect, two-step approach has deficiencies because archived climate model output may not capture variable types or temporal resolution needed for some VIA models and restricts the ability to study feedbacks on climate and biogeochemistry.

With inclusion of the biosphere in ESMs, VIA can be investigated directly. For example, the ESMs used to quantify future carbon-climate feedbacks can also be used in retrospective studies to assess responses of terrestrial and marine ecosystems to historical anthropogenic forcings and climate variability (16, 55, 56). The ocean component of ESMs provides a tool for reconstructing the variability, trends, and mechanisms of historical ocean biogeochemistry (57) and plankton dynamics (58). ESMs with high-resolution ocean circulation models can be used to track larval dispersal and connectivity among coral reefs subject to bleaching (59).

ESMs provide further opportunity to move beyond physical descriptors of atmospheric and oceanic states (such as temperature and precipitation) to societally relevant quantities related to food, energy, and fresh water. For example, the croplands in ESMs allow direct study of the impacts of climate change on agricultural production, the vulnerability of the food supply to future climate disruption, and adaptations to make food production more resilient (42). Some ESMs include urban land cover, which allows study of extreme heat waves and facilitates assessment of heat-stress mortality in cities (60). ESMs are particularly useful for assessing air quality and human health issues because of the interactions among agriculture, wildfire, nitrogen gaseous fluxes, and biogenic volatile organic compounds that affect regional air quality.

Achieving this potential requires effective communication between the scientists developing ESMs and those using climate projections to study VIA (61). One result of better collaboration would be to identify and reconcile discrepancies between ESMs and VIA models, such as are evident in their assessments of water availability in a future climate. ESMs account for the effects of elevated atmospheric CO2 on stomatal conductance and evapotranspiration, but many VIA models do not, resulting in an inconsistency in projections of water availability (62). Other examples of processes included in ESMs but not VIA models are the effect of O3 on stomata (29, 30) and the effect of vegetation greening and land use on runoff (63). Closer collaboration between the communities would help to identify capabilities relevant to VIA, define impact-relevant metrics that ESMs should produce, and develop data sets and protocols for validation of simulated impacts (61).

ESMs remain just one of several means to study VIA. A suite of specialized research tools including statistical models and process-based crop, ecosystem, and hydrology models is required (6, 7). These models have the advantage that they can be run at the fine-scale spatial resolution needed to inform decision making. Also, they are less computationally expensive than ESMs and can therefore be used in an ensemble of simulations to assess uncertainty.

ESMs cannot yet represent the rich ecological detail needed to capture spatial heterogeneity at local scales. Similarly, the ocean ecosystem models used in ESMs typically incorporate the lowest trophic levels of the marine food web (phytoplankton, herbivorous zooplankton) and have only a limited representation of biodiversity. Often, ESMs lack the ecological complexity required to predict outcomes in higher trophic levels and fisheries. The spatial resolution of global models is too coarse to capture regional dynamics of highly productive coastal ecosystems and coral reefs, and models are just beginning to incorporate adequate land-ocean connectivity to assess nutrient eutrophication, water quality, and harmful algal blooms (64). Variable-resolution global models with a horizontal resolution that refines from a 1° global grid to a regional 0.125° (14-km) grid help to bridge the gap between coarse-scale ESMs and the finer scales needed for VIA research (65, 66).

Climate change mitigation

Reducing the sources and enhancing the sinks of long-lived greenhouse gases are the most direct means to mitigate anthropogenic climate change (67). However, many interventions that might reduce greenhouse gas emissions affect the biosphere and have other effects on climate and ecosystem services. Afforestation, reforestation, or avoided deforestation, for example, enhance the terrestrial carbon sink, but also warm climate annually by decreasing surface albedo, cool climate through evapotranspiration and turbulent mixing with the atmosphere, and have additional effects through atmospheric chemistry and aerosols (13, 68). These biogeophysical effects can counter the carbon mitigation benefits of forests so that even more extensive forested land may be required to achieve climate stabilization at a target that avoids dangerous climate change (e.g., 2°C). ESMs are an imperfect but necessary tool to study the net climate effects of forests (68).

Agriculture is another example of the need to consider mitigation in an ESM context. Efficient application of nitrogen fertilizer, tillage, and other management can enhance carbon storage and reduce N2O emissions (69). Crops also affect climate through biogeophysical coupling with the atmosphere; it is likely that expansion of agricultural lands over the industrial era has cooled climate because of these changes (70). Intensification of agriculture is thought to have cooled summer temperatures in the Midwest United States (71). No-till agriculture can increase surface albedo and cool climate (72), and other increases in surface albedo may have geoengineering potential (73). Production of bioenergy for carbon capture and storage (BECCS) can also mitigate climate change, but land use for BECSS must be balanced by arable land for food production (74). ESMs provide a necessary tool to investigate the multidisciplinary outcomes of BECCS for climate, food, energy, and fresh water. ESMs are also being used to determine the effects on ecosystems of geoengineering techniques involving solar radiation modification such as stratospheric aerosol injection, cloud brightening, and surface albedo manipulation (75, 76).

At present, the ocean removes roughly a quarter of anthropogenic CO2 emissions to the atmosphere, with the magnitude modulated by chemical dissolution into surface seawater and the physical rate of exchange between surface and deep waters (16). Over the centuries-long time scales of ocean-overturning circulation, an increasing fraction of anthropogenic CO2 will be stored in the deep ocean reservoir. Several geoengineering approaches have been proposed to increase ocean carbon uptake by injecting CO2 directly at depth, fertilizing phytoplankton to speed up the marine biological pump that transports organic carbon from the surface layer into the deep ocean, and accelerating weathering processes to add alkalinity to seawater (67). A number of studies with ocean-only models and full ESMs have explored the possible efficacy of these approaches, as well as the potential for ecological impacts and biogeochemical feedbacks (75). Substantial alterations to marine ecosystems could also arise from solar radiation modification.

Earth system prediction

Atmospheric science has long embraced models to make predictions of near-term weather and long-term climate. ESMs enable predictions of the biosphere, but the atmospheric-centric view of prediction needs to be extended to the biological realm. In this section, we introduce terminology and concepts specific to weather forecasts and climate prediction and then show extension to the biosphere.

Forecasting the weather on time scales of a few hours to 2 weeks is a classic prediction problem in which a model that describes the atmosphere–land–ocean–sea ice system is stepped forward in time from initial conditions. The predictability—the capability to make a skillful forecast—is limited by uncertainty in the exact initial conditions, imperfections in the model and understanding of the underlying physics and dynamics, and the degree of randomness or chaotic behavior in the system. The same concept applies to predicting climate at seasonal, interannual, or decadal time scales (7779). Climate projection over several decades must consider additional long-term Earth system processes such as ocean circulation, ice sheet melting, and changes in vegetation, terrestrial and marine biogeochemistry, and human behavior. The lattermost is particularly poorly known and is imposed using anthropogenic forcing scenarios. At multidecadal time scales, the exact initial state is less critical. Instead, uncertainty in climate projections is largely dominated by the choice of an anthropogenic forcing scenario, although uncertainties also remain with regard to climate sensitivity and feedback processes (Fig. 2).

Fig. 2 Schematic depiction of Earth system prediction of the biosphere.

The synergies between climate feedback processes, internal climate variability, and ecosystem impacts determine model outcomes. Subseasonal to seasonal forecasts and decadal climate prediction are initial value problems. Earth system projections are a boundary value problem driven by anthropogenic forcing scenarios. Uncertainty arises from inexactness of initial conditions, model imperfections, and scenarios.

The term Earth system prediction is used to capture this spectrum of temporal scales from subseasonal to multidecadal, mostly in the context of weather and climate (7781). In a broader perspective, however, the scope of Earth system prediction can be expanded to include other facets of the Earth system. The predictability of Arctic sea ice loss is a prominent such example (82). As climate models have evolved into more complex ESMs, the predictability of biosphere states and processes needs to be considered jointly with that of weather and climate.

The predictive capability of a model depends on the sources of errors in the forecast. For climate, these errors are broadly classified into initial conditions, boundary conditions, and model uncertainty, including both model structure and parameters (83). Uncertainty in exact initial conditions manifests in unforced variability internal to the climate system (also known as natural variability), in which small differences in initial conditions produce different climate trajectories. The importance of natural variability can be assessed through a multimember ensemble of simulations with a single model initialized with different states.

The second source of uncertainty is model error, seen in the model response to the imposed forcing scenarios. Models are imperfect and differ in their forced response owing to their spatial resolution and imprecision in their parameterization of the various physical, chemical, and biological processes. Model uncertainty is assessed through multimodel ensemble studies.

The third source of uncertainty is the forcing scenarios and their depiction of the time evolution of greenhouse gases, land use, and other anthropogenic forcings of climate, which are imposed as boundary conditions to the models. For temperature projections at the global scale, model uncertainty and natural variability dominate at near-term decadal time scales (10 to 30 years) (83). Scenarios are the major source of uncertainty at multidecadal lead times. Total uncertainty is larger at regional scales, mostly from natural variability and model structure.

A related concept is time of emergence. Determining the time at which the forced climate change signal emerges from the noise of natural variability is a necessary requirement in assessing when an expected change can be detected or whether observed changes can be attributed to anthropogenic forcings (84, 85). Time of emergence has been mostly studied for temperature and can vary from a few decades in mid-latitudes with low natural variability to several decades or longer in regions with larger natural variability.

Can these concepts of predictability, uncertainty, and time of emergence be extended to study the biosphere in the Earth system? ESMs predict prominent changes in terrestrial and marine biogeochemistry, but only recently have the necessary large, multimodel and multimember ensembles become available to distinguish the forced response from natural variability and model uncertainty. Such analyses give important insights into the use of ESMs to understand changes in Earth system biogeochemistry. In addition to warmer temperatures, the ocean has trends of increased carbon uptake, acidification, lower O2, and reduced net primary production over the next several decades in the absence of mitigation (86). The forced trend in air-sea CO2 flux emerges rapidly in some ocean regions, but large natural variability precludes detection of changes in the rate of carbon uptake until mid-century or later in many regions (87). Other aspects of ocean biogeochemistry such as pH, O2 concentration, and net primary production also have large, regionally dependent natural variability (8890). Ocean acidification has a rapid time of emergence driven by the accumulation of anthropogenic CO2 in the surface layer, and the sea surface temperature warming signal also emerges within a few decades in many regions, but forced changes in O2 concentration and net primary production do not emerge from natural variability until mid- to late-century, if at all (Fig. 3, A to C).

Fig. 3 Ocean and land forced trends relative to internal variability and model uncertainty.

Data are from a range of ESMs contributed to CMIP5. (A to C) Multimodel time of emergence for SST, O2, and net primary production (NPP) (89). Time of emergence is defined as the year at which the signal exceeds the noise, which, as used here, includes both internal variability and model uncertainty. The forced SST signal emerges rapidly in many locations, O2 time of emergence is regionally variable, and the forced NPP signal does not statistically emerge by 2100. (D to F) Signal-to-noise ratio for cumulative land carbon uptake in a business-as-usual scenario at 2030 for three different ESMs (92). Positive (negative) values indicate carbon gain (loss). In these panels, the noise is strictly internal variability, and a ratio greater than 2 or less than –2 indicates that the signal has emerged from the internal variability. There are considerable differences among models in the sign of the terrestrial carbon flux and whether the change has emerged from natural variability by 2030. CCSM4, Community Climate System Model version 4; HadGEM2-ES, Hadley Centre Global Environmental Model version 2; CanESM2, second-generation Canadian Earth System Model.

Time of emergence has been less studied in the terrestrial biosphere. Observational and modeling analyses support an enhanced terrestrial carbon sink arising from global change (17, 56, 91). Unforced variability in the land-atmosphere CO2 flux is large and precludes detection of change at decadal time scales (92). There is considerable variability within and among models, and the forced response statistically emerges only after several decades in many regions of the world. The HadGEM2-ES model, for example, shows the forced signal of terrestrial carbon gain as emerging from internal variability in many regions by 2030, but other models show a weaker signal that has yet to statistically emerge, and even carbon loss rather than gain (Fig. 3, D to F).

The various contributions to uncertainty differ depending on the quantity of interest, lead time, and spatial scale. The uncertainty from natural variability is particularly large at small spatial scales and short lead times for pH, O2 concentration, and net primary production in the ocean (89) and air-sea carbon flux (93). By the end of the 21st century, scenario uncertainty dominates total uncertainty for these states and fluxes at the global scale (except for net primary production), but natural variability and model uncertainty remain large at the regional or biome scale. Simulations of the terrestrial carbon cycle are much more variable among models and largely dominated by model uncertainty (94). Comparisons of ocean and land carbon cycle projections point to a markedly different assessment of uncertainty (Fig. 4). For ocean carbon uptake, model uncertainty is initially large, but scenario uncertainty dominates during the latter part of the 21st century. In contrast, model structure contributes 80% of total uncertainty throughout the 21st century for the terrestrial carbon cycle.

Fig. 4 Ocean and land carbon cycle uncertainty.

The percentage of total variance attributed to internal variability, model uncertainty, and scenario uncertainty in projections of cumulative global carbon uptake from 2006 to 2100 differs widely between (A) ocean and (B) land. The ocean carbon cycle is dominated by scenario uncertainty by the middle of the century, but uncertainty in the land carbon cycle is mostly from model structure. Data are from 12 ESMs using four different scenarios (94).

Much of the study of Earth system prediction is focused on climate, and decadal climate prediction is particularly focused on the dynamics and thermal characteristics of the ocean because of its prominent role in climate variability. When soil moisture and vegetation are considered by atmospheric modelers, it is often in the context of how these affect climate predictability, rather than whether they can be predicted (79). In a global change perspective, ecological predictions are as essential as those of the physical climate system. ESMs provide the means not just to assess the potential for future stresses engendered by a changing climate, but also to determine the outcome of those stresses on crop yield, tree mortality, fisheries, and other aspects of the biosphere. For example, characterizing when and where wildfires might occur would be valuable to aid governmental agencies charged with wildfire protection. Particular fire events are nearly impossible to forecast, especially because so many fires are caused by people, but fire behavior on seasonal time scales can be forecast on the basis of relationships with sea surface temperatures (95). Prediction of future fire behavior requires models that accurately depict fire occurrence and severity, but wildfire prediction will also require an understanding of the predictability of the climate that drives fire behavior. A similar argument pertains to crop yield, marine resources, and dust emissions simulated in the current generation of ESMs and to forest mortality and habit loss that will be simulated by the next generation of models. Such forecasts may be particularly relevant at the subseasonal to seasonal time scale (79).

Research needs

As this Review highlights, the biosphere is central to understanding why and how the Earth system is changing and to adapting to and mitigating future changes. Many of the global change stressors that terrestrial and marine ecosystems face need to be understood not only for their impacts on ecosystem services that are essential to humankind, but also as processes that affect the magnitude and trajectory of climate change. A strategy is needed to extend the study of subseasonal and seasonal forecasts and decadal climate prediction to a more multifaceted Earth system prediction, including the biosphere and its resources. The extension of seasonal to decadal climate forecasts to living marine resources, for example, has considerable potential to aid marine management (96). A similar extension to terrestrial ecosystems would aid land resource management.

Toward this goal, this Review has presented several pathways to further define Earth system prediction. First is continued advancement of terrestrial and marine science in light of climate processes and the many ways in which the biosphere influences climate. A prominent example is the carbon cycle, its feedback with climate change, and whether terrestrial and marine ecosystems can be purposely managed to mitigate anthropogenic CO2 emissions. There is considerable uncertainty in ocean carbon cycle projections, particularly at regional or biome scales (89, 93), and land carbon model uncertainty precludes distinguishing among various alternative scenarios (94). Moreover, if planting forests and biofuels are essential to maintaining atmospheric CO2 concentrations within some planetary warming target, how confident are we in our ability to know the net climate outcome of these policies (68)?

The current generation of ecosystem models are abstractions of complex systems. Many ecological and biogeochemical processes are represented, but the challenge of representing the biosphere—with its rich diversity of life forms, their assemblage into communities and ecosystems, and the complexity of ecological systems—is daunting, as is evident in large model uncertainty in terrestrial carbon cycle projections. Theoretical advances are needed, but there may be a limit to how much model uncertainty can be reduced (94). More complexity does not necessarily lead to better predictions or reduce uncertainty.

A second pathway is to better integrate ESMs and VIA models. The gap between models arises from disciplinary expertise (atmospheric and ocean sciences for ESMs and hydrology, ecology, biogeochemistry, agronomy, forestry, and marine sciences for VIA models), but effective communication among, rather than across, disciplines is not trivial. There are also pragmatic considerations, particularly with regard to spatial scale and process complexity, that limit collaboration between global ESMs and VIA models with a more local to regional domain. However, just as the science of Earth system prediction is seen as a means to unite the weather and climate modeling communities (80, 81), so, too, can the broadening of Earth system prediction to include the biosphere stimulate collaborations with the VIA community.

A third promising research pathway is to expand the concepts and methodology of seasonal to decadal climate prediction to include terrestrial and marine ecosystems and to quantify prediction uncertainty at spatial and temporal scales relevant to stakeholders. The predictability of the terrestrial carbon cycle can be considered from an ecological perspective (97), but only recently has it been considered in an Earth system perspective of natural climate variability, the forced climate response, and model uncertainty (92, 94). Analysis of natural variability, model uncertainty, and scenario uncertainty is similarly informing marine biogeochemistry (8790, 93). Whether the biosphere is a source of climate predictability is not necessarily the right question to pose. A more fruitful research pathway may be to investigate how to predict the biosphere and its resources in a changing environment, as identified specifically for marine living resources (96) and considered also for atmospheric CO2 (98). Initial condition uncertainty and the difficulty in separating natural variability from the forced trend likely produces irreducible uncertainty in climate prediction (99). At the regional or biome scale, natural variability is large for the ocean and land carbon cycles (89, 92, 93). Whether a similar irreducible uncertainty manifests in terrestrial and marine ecosystems remains to be explored.

With their terrestrial and marine ecosystems, biogeochemical cycles, and simulation of plants, microbes, and marine life, ESMs challenge terrestrial and marine ecologists and biogeochemists to think in terms of broad generalizations and to find the mathematical equations to describe the biosphere, its functioning, and its response to global change. ESMs similarly challenge geoscientists to think beyond a physical understanding of climate to include biology. The models show much promise to advance our understanding of global change but must move from the synthetic world of an ESM toward the real world. Bridging the gap between observations and theory as atmospheric CO2 rises, climate changes, more nitrogen is added to the system, forests are cleared, grasslands are plowed or converted to pastures, coastal wetlands and coral reefs are degraded or lost, and oceans warm and are increasingly polluted poses challenging opportunities for the next generation of scientists to advance planetary ecology and climate science.

References and Notes

Acknowledgments: We acknowledge funding from the National Institute of Food and Agriculture/U.S. Department of Agriculture (2015-67003-23485) and the NASA Ocean Biology and Biogeochemistry Program (NNX14AL86G). We thank C. Tebaldi and J. Kleypas (NCAR) for comments on the manuscript and figures. NCAR is sponsored by the National Science Foundation.
View Abstract

Navigate This Article