C4MIP Background, Motivation and History

1. Introduction

Over the industrial era since about 1850, it is estimated that cumulative carbon emissions from fossil fuels and cement (390 PgC) have been largely partitioned between the atmosphere (225 PgC), and ocean (150 PgC), with a small net carbon uptake by land (10 PgC) (need REF). The net carbon uptake of 10 PgC over land is estimated as the difference of gross uptake of 155 PgC that is offset by emissions associated with anthropogenic land use change (145 PgC). Had the land and ocean not provided this “ecosystem service” the atmospheric carbon burden at present would have been much higher. The manner in which the land and ocean will continue to provide this ecosystem service is of both scientific and policy relevance. While the role of ocean carbon uptake is clearly large, the role of the land in modulating the atmospheric carbon burden remains highly uncertain. Understanding the future partitioning of anthropogenic CO2 emissions into the atmosphere, land and ocean components, resulting climate warming, and biogeochemical feedbacks requires a full earth system approach to the carbon cycle.

The primary focus of the Coupled Climate-Carbon Cycle Model Intercomparison Project (C4MIP) is to understand and quantify future century-scale changes in land and ocean carbon storage and fluxes. In order to achieve this, a set of ESM simulations has been devised. Due to the very high computational demands on modelling centres of performing many simulations for many different intercomparison studies as part of CMIP6, we have carefully chosen a minimum set of targetted simulations to achieve C4MIP goals. They comprise: 

  • idealized experiments which will be used to separate and quantify the sensitivity of land and ocean carbon cycle to changes in climate and changes in atmospheric CO2 concentration
  • historical experiments which will be used to evaluate model performance and investigate the potential for observational constraints on future projections
  • future scenario experiments which will be used to quantify future changes in carbon storage and hence quantify the atmospheric CO2 concentration and related climate change for given CO2 emissions, or diagnose the emissions compatible with a prescribed atmospheric CO2 concentration pathway

The simulations are designed to partner those prescribed in the CMIP DECK (Eyring et al ref??) and the CMIP6 Historical simulation (Eyring ref). They also align closely with simulations performed as part of ScenarioMIP (Ref?) and LUMIP (Ref?) and OMIP (Ref?).

In this paper we document the scientific rationale and motivation for the C4MIP simulations and carefully document the experimental protocol. Modelling groups intending to participate in C4MIP should follow the design laid out here as closely as possible. Particular attention should be paid to use set-up of boundary conditions in terms of atmospheric CO2 concentration or emissions and which aspects of the model experience changes. Output requirements (diagnostics) are also carefully documented. It is vital for accurate analysis and model intercomparison that every model adheres to the definitions of each output variable in order for like-for-like comparison to be made.

Initial plans for carbon cycle analysis are also listed. Modelling groups will be invited to contribute to central C4MIP analysis papers. We anticipate, and hope, that many further studies and analyses will also be conducted throughout the climate/carbon cycle research community and that these simulations provide a valuable resource to further carbon cycle research. 

2. Background and science motivation

2.1 C4MIP history

Some early studies suggested the potential for a climate feedback onto the carbon cycle whereby carbon released due to warming would further elevate atmospheric CO2 and amplify climate change (Jenkinson et al., 1991; Schimel et al., 1994; Kirschbaum, 1995).

Dynamic global vegetation models were used to study the impact of rising CO2 and climate on the carbon cycle (Kicklighter et al., 1999; Cramer et al., 2001; McGuire et al., 2001). There was a strong model consensus that rising CO2 would stimulate additional vegetation growth and storage of carbon in terrestrial ecosystems, likewise warming climate would accelerate decomposition of dead organic matter and may also reduce vegetation productivity in some (mainly tropical) ecosystems.

In the ocean there was also a model consensus that warming would lead to reduced carbon uptake (Prentice et al., 2001). This was due to both reduced solubility in warmer waters and reduced rate of transport of carbon to the deep due to more stratified surface waters. The processes behind the former (carbonate chemistry and solubility) were reasonably well understood (Bacastow, 1993), but the latter was much more uncertain being sensitive to the underlying ocean model circulation (Sarmiento et al., 1998; Joos et al., 1999). The role of ocean biology and the buffering capacity of the ocean were also seen to be important and not well constrained or represented in models (Sarmiento and Le Quere, 1996). Intercomparison of ocean-only carbon cycle simulations was performed under the OCMIP activity (Orr et al., 2001).

These “offline” land and ocean experiments found potentially high sensitivity of the carbon cycle to environmental forcing but were not able to simulate the full effect of this feedback onto climate. Intermediate complexity models were able to simulate this feedback (Joos et al., XXXX;Prentice et al., 2001) but lacked spatial detail in their climate response. So by the end of the 1990s some modelling groups were beginning to couple interactive carbon cycle modules to their climate GCMs. These early studies (e.g. Cox et al., 2000; Friedlingstein et al., 2001; Thompson et al., 2004) were able to recreate an experimental setting more like the real world where a forced climate change affected natural carbon sinks and stores which in turn affected changes in atmospheric CO2 and hence climate.

It soon became apparent from the first publications that there were substantial differences in the sensitivities of these new models. The desire to understand and reduce this uncertainty led to the development of a linearised feedback framework to diagnose the sensitivity of different parts of the system and their contribution to the overall feedback (Friedlingstein et al., 2003), and also of a multi-model intercomparison activity (C4MIP: Coupled Climate–carbon cycle model intercomparison, also known as the “Flying Leap” experiment—Fung et al., 2000). The result was the first C4MIP intercomparison paper, (Friedlingstein et al., 2006) which quantified the feedback components across 11 models for a common CO2 emissions scenario. All models agreed qualitatively that the sign of the carbon–climate feedback was positive—i.e. the interaction of the carbon cycle with climate led to reduced carbon uptake and hence an increase in atmospheric CO2 which amplified the initial climate change. However, there was large quantitative model spread in the total feedback and its sensitivity components.

Initial analysis of the causes of this uncertainty concluded that the land played a greater role than the ocean, in particular its sensitivity to climate (i.e. “γL”). Regionally, the tropics were seen to be particularly different between models (Raddatz et al., 2007), bearing in mind that none of these models included representation of permafrost carbon. As the soil was the biggest terrestrial carbon pool, analysis focused on that and specifically its sensitivity to temperature, often characterised in these models by a Q10 parameter. Models had either single or multiple carbon pools, and some had differing Q10 parameters for each pool (Zeng et al., 2004) although some incubation experiments supported a consistent Q10 across pools (Fang et al., 2005) results and implications from such experiments were often hotly contested (Giardina and Ryan, 2000; Knorr et al., 2005). The global implications (Jones et al., 2003) and potential constraints (Jones and Cox, 2001) of Q10 uncertainty were also explored in the climate–carbon cycle models.

Others noted that changes in soil carbon storage were not necessarily due to soil processes alone (Post et al., 1996), with Matthews et al. (2005) being the first to explore the important role of NPP sensitivity to climate in the C4MIP models. The review by Jones and Falloon (2009) show a strong relationship between changes in SOC and the carbon cycle gain, g, but show this is due jointly to inputs (vegetation processes) and outputs (soil processes).

The initial terminology of the “climate carbon cycle feedback” hid the importance of the sensitivity of the system to elevated CO2 (“β”) and left the focus on “γ”. The work of Gregory et al. (2009) highlighted the two components of the carbon cycle feedback—the climate–carbon and concentration–carbon terms. They showed that β is both larger and more uncertain than γ in models, a result backed up recently by Arora et al. (2013) analysis of CMIP5 simulations.

The CMIP5 experimental design for carbon cycle feedback diagnosis (Taylor et al.) was based closely on C4MIP, informed by Gregory et al. (e.g., around rate dependence and non-linearity of the components). Modelling centres around the world contributed results to CMIP5 and their analysis led to many key papers including a special collection published in the Journal of Climate (http://journals.ametsoc.org/page/C4MIP). C4MIP activity under CMIP5 therefore was central to IPCC AR5 across several chapters as described below. 

2.2 Carbon cycle in CMIP5 / AR5

In this section we review how analysis of coupled climate carbon cycle models contributed to the IPCC 5th Assessment Report (AR5). In this section, figure and table numbers refer to those from the AR5 WG1 chapters (Chapter 6 carbon cycle, chapter 9 evaluation and chapter 12 projections). Some quotes are also taken from the chapter executive summaries which highlight the contributions made by the coupled modelling to the key IPCC conclusions.

For the first time the IPCC WG1 carbon cycle chapter had a section devoted to the feedbacks and future projections from coupled carbon cycle ESMs. Figure 6.20 compiled a synthesis of biogeochemical feedbacks and showed the carbon cycle response to climate and CO2 were the largest and most uncertain components. Comparing these between C4MIP generations showed some interesting features (fig 6.21): notably the mean response of both land and ocean to both CO2 and climate were all smaller in CMIP5 than from Friedlingstein et al. (2006). But differences in experimental design, rate of change of forcing in the scenario used and the set of models make inferences from this result inconclusive.

For example, Gregory et al. (2009) showed the response to CO2 (β) to be especially scenario dependent with reduced uptake under faster scenarios. This could explain the reduced concentration–carbon response in CMIP5. Similarly, Schwinger et al. (2013) showed big sensitivity of ocean response to climate (γO) to the definition of the metric—using the RAD forced run (as done for CMIP5 models; Arora et al., 2013) typically resulted in much smaller sensitivities than using the difference between fully coupled and BGC runs (as done previously; Friedlingstein et al., 2006). This could explain the difference in the ocean climate–carbon response in CMIP5. Finally, the small but differing sample of models also makes reliable comparison difficult between C4MIP generations—with the exception of two outliers (UMD and HadCM3LC) in C4MIP and the models with terrestrial nitrogen cycle in CMIP5, the land response between generations was not substantially different. This underlines the need for systematic diagnosis of feedback metrics in a common experimental framework.

Building on the analysis of Roy et al. (2011) spatial maps of land and ocean sensitivities were also presented (fig 6.22) showing general consensus of positive β values everywhere, but changes in the sign of βL towards high latitudes. All models simulate global aggregate γL and γO to be negative (i.e. reduced carbon storage in response to warmer climate).

“There is high confidence that climate change will partially offset increases in global land and ocean carbon sinks caused by rising atmospheric CO2.” [Ch.6]

However, most terrestrial models simulated positive γL (i.e. increased carbon storage in a warmer climate) at high latitudes as enhanced growth in temperature limited ecosystems outpaced enhanced turnover of SOC. The inclusion of nitrogen cycling and permafrost however may change these results in future.

In an analysis of whether the terrestrial carbon cycle feedbacks in the CMIP5 ESMs were driven primarily by changes to input or output fluxes, Koven et al. (2015a) showed that changing inputs were responsible for the bulk of changes to both vegetation and soil carbon pools, with only small changes to vegetation turnover, and with the bulk of soil carbon turnover time changes arising from transient shifts in the soil carbon pool distributions rather than changes to the environmentally-controlled cycling rates. This result contrasts with other model intercomparisons that show a large role in changing vegetation turnover (e.g., Friend et al., 2014), underscoring a common lack in the ESMs participating in CMIP of processes that may govern both changing vegetation turnover times (mortality, allocation), and changing soil turnover times (permafrost, priming and stabilization processes).

Historical land and ocean simulations (figure 6.24) showed considerable spread but the multi-model mean agreed closely with observed cumulative changes in land and ocean carbon since pre-industrial (table 6.12). Projections into the future showed reasonable levels of consensus for ocean carbon uptake across models and scenarios, but a much wider uncertainty for land, with model-spread comparable in magnitude to scenario-spread (Hewitt et al., 2015).

“With very high confidence, ocean carbon uptake of anthropogenic CO2 emissions will continue under all four Representative Concentration Pathways (RCPs) through to 2100, with higher uptake corresponding to higher concentration pathways. The future evolution of the land carbon uptake is much more uncertain …” [Ch.6]

The most important aspect of the projections was to infer the fossil fuel emissions required to achieve the RCP prescribed CO2 trajectory—so called “compatible emissions” (Jones et al., 2013). Especially for low scenarios this aspect of ESMs allows quantification of emissions reductions to meet climate targets.

“Taking climate and carbon cycle feedbacks into account, we can quantify the fossil fuel emissions compatible with the RCPs. … For RCP2.6, an average 50% (range 14 to 96%) emission reduction is required by 2050 relative to 1990 levels.” [Ch.6]

ESMs also quantify the role of future scenarios in determining the fraction of emissions remaining in the atmosphere or entering the land and ocean. Although the airborne fraction (AF) has remained remarkably constant for several decades (Denman et al., 2007; Ciais et al., 2013), Raupach (2013) showed that this is largely an artefact of near-exponential historical emissions. Future AF may depart from current values depending on the rate of emissions in the scenario (figure 6.26) with higher AF for higher scenarios and greater uptake efficiency, especially for the ocean, under lower/slower emissions scenarios.

CMIP5 went a long way in improving the level of agreement among models on past and future ocean carbon uptake. While the important role of the North Atlantic on ocean carbon uptake and climate feedbacks has been recognized for two decades (Sarmiento and LeQuere, 1996), it has only been through CMIP5 that the important and complex role of the Southern Ocean has been recognized (Frolicher et al., 2014), including the potential for enhanced biological uptake of CO2 under warming (Ito et al., 2015). Improved representation of of ocean watermasses and biogeochemical cycling in CMIP5 pointed to enhanced ocean acidification at depth (Bopp et al., 2013; Resplandy et al., 2014). The chapter then went on to consider some processes which are not commonly represented in CMIP5 ESMs.

Nitrogen cycle. Nutrient limitations, especially of nitrogen, play an important role in moderating the amount of carbon which can be stored in biomass and soils. Future changes in climate and deposition of reactive nitrogen from human activity will both alter the amount of available nitrogen for plant growth but this is not represented in most ESMs participating in CMIP5. The two models which do incorporate treatment of this process (CESM1-BGC and NorESM-ME which both do so by using the CLM4 land surface model) show clearly much smaller sensitivity to both climate and CO2 (grey dots in fig 6.21). These results are in agreement with offline model results from independent models, however, the magnitude of the effect appeared to be strongest in the CLM4 model (Zaehle et al., 2010; Zhang et al., 2014) The other, non-terrestrial-nitrogen, models risk taking up much more carbon than would be possible given nutrient constraints on photosynthesis. This was assessed at a gridpoint level and the reduction in possible land carbon storage calculated as a result (fig 6.35; see also Zaehle et al., 2015; Wieder et al., 2015). Across all scenarios the CMIP5 mean land carbon uptake was reduced by 100–200 PgC.

“It is very likely, based on new experimental results {} and modelling, that nutrient shortage will limit the effect of rising atmospheric CO2 on future land carbon sinks” [Ch.6]

Permafrost. Northern permafrost-affected soils contain 1100–1500 Pg of carbon (Hugelius et al., 2014), which is stabilized by being frozen and thus vulnerable to decomposition with warming. None of the models in either C4MIP or CMIP5 considered the dynamics of carbon in permafrost layers, and thus these ensembles may systematically underestimate the potential carbon losses from these layers and the northern region generally. Inclusion of simplified permafrost carbon dynamics have now been included in several models (reviewed in Schuur et al., 2015), and lead to potentially larger carbon–climate feedback. However, the timescale of these releases may be slower than other terrestrial feedbacks, with plant uptake dominant during initial warming and slower loss from deeper soil layers overtaking the increased productivity later. The timing of this transition, and magnitude of eventual carbon losses is not well known, and may be strongly dependant on high-uncertainty processes such as: nitrogen dynamics of permafrost soils, whether thawing soils become wetter or drier upon thawing, the relative rates of decomposition in shallow and deep soils, and microbial heat release in decomposing soils (Koven et al., 2011; 2015b; Burke et al., 2014).

One of the fundamental uncertainties in ocean carbon uptake is the physiological role of ocean acidification on plankton biodiversity through growth and calcification. While several studies have pointed to a relative advantage for nitrogen fixation under high CO2 (Hutchins et al., 2007), ocean models have yet to incorporate CO2 controls on phytoplankton physiology. While some CMIP5 models included rudimentary control of calcification through aragonite and calcite saturation state, a great diversity in physiological response has been observed that is not incorporated in models (Doney et al., 2009). Similarly, modeled interactions with biogeochemically active coastal systems such as coral reefs and marine sediments are rudimentary at best.

There are still significant gaps in our current understanding of the sensitivity of marine organisms to ocean acidification and warming on organismal and ecosystem scales (Andersson et al., 2015; Kroeker et al., 2013). Although the response of marine biology to warming, ocean acidification, and deoxygenation can be conceptualized in projections of the modern ESMs (Bopp et al., 2013), the combined effects of these environmental stressors on marine ecosystems and elemental cycles remain yet to be explored.

Chapter 12 covered emission driven vs. concentration driven simulations, using the CMIP5 ESMs, following historical and RCP8.5 scenario.

Key results:

  • uncertainty in simulated atmospheric CO2: “By 2100, the multi-model average CO2 concentration is 985 ± 97 ppm (full range 794 to 1142 ppm), while the CO2 concentration prescribed for the RCP8.5 is 936 ppm.”
  • Slightly larger warming when models are driven by emissions: “Global warming simulated by the E-driven runs show higher upper ends than when atmospheric CO2 concentration is prescribed. For the models assessed here, the global surface temperature change (2081–2100 average relative to 1986–2005 average) ranges between 2.6°C and 4.7°C, with a multi-model average of 3.7°C ± 0.7°C for the concentration driven simulations, while the emission driven simulations give a range of 2.5°C to 5.6°C, with a multi-model average of 3.9°C ± 0.9°C, that is, 5% larger than for the concentration driven runs. The models that simulate the largest CO2 concentration by 2100 have the largest warming amplification in the emission driven simulations, with an additional warming of more than 0.5°C.”
  • TCRE estimate in chapter 12, based on multiple lines of evidence, including CMIP5 1% simulations and historical and RCP8.5 simulations with ESMs. : “Expert judgement based on the available evidence therefore suggests that the TCRE is likely between 0.8°C to 2.5°C per 1000 PgC, for cumulative CO2 emissions less than about 2000 PgC until the time at which temperature peaks.”

Those results also described in more depth in Friedlingstein et al. (2014), Hoffman et al. (2014), Arora et al. (2013), and Gillet et al. (2013).


Back to top