On the additivity of radiative forcing between land use change and greenhouse gases

Authors

  • Andrew D. Jones,

    Corresponding author
    1. Earth Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California, USA
    • Corresponding author: A. D. Jones, Earth Sciences Division, Lawrence Berkeley National Laboratory, One Cyclotron Rd., Mail Stop 84-0171, Berkeley, CA 94720, USA. (adjones@lbl.gov)

    Search for more papers by this author
  • William D. Collins,

    1. Earth Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California, USA
    2. Earth and Planetary Sciences, University of California, Berkeley, California, USA
    Search for more papers by this author
  • Margaret S. Torn

    1. Earth Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California, USA
    2. Energy and Resources Group, University of California, Berkeley, California, USA
    Search for more papers by this author

Abstract

[1] In scientific and policy contexts, radiative forcing—an external change in Earth's mean radiative balance—has been suggested as a metric for evaluating the strength of climate perturbations resulting from different climate change drivers such as greenhouse gases and surface physical effects of land use change. However, the utility of this approach has been questioned given the spatially concentrated and sometimes nonradiative nature of land use climate disturbances. Here we show that when negative forcing from agricultural expansion is approximately balanced by a radiatively equivalent increase in atmospheric carbon dioxide, significant changes in temperature, precipitation, and the timing of climate change result. These idealized experiments demonstrate the nonadditivity of radiative forcing from land use change and greenhouse gases and point to the need for new climate change metrics or the development of climate policies and assessment protocols that do not rely on single dimensional metrics.

1 Introduction

[2] Radiative forcing—a change in Earth's mean radiative balance—is currently used as a metric for comparing the effect of different anthropogenic activities on climate. It underlies the global warming potential metric used to compare the climate effects of various greenhouse gases (GHGs) to one another [Forster et al., 2007], and it forms the basis of the Intergovernmental Panel on Climate Change (IPCC) fifth assessment scenarios [van Vuuren et al., 2011] and “parallel” process for coordinating climate impacts and adaptation research [Moss et al., 2010]. An assumption behind the radiative forcing approach in policy is that equivalent levels of forcing from different activities lead to equivalent climate outcomes and therefore equivalent societal impacts [Fuglestvedt et al., 2003]. A corollary to this assumption is that positive forcing from one agent combined with negative forcing from another agent leads to roughly neutral climate effects. To the extent that this forcing additivity assumption is violated, policies and assessments based on radiative forcing could lead to unintended outcomes or erroneous conclusions.

[3] Existing and proposed climate change mitigation measures (e.g., cap-and-trade programs or low-carbon fuel standards) often rely on taxes, payments, or credits expressed in units of CO2 equivalents to comprehensively address multiple greenhouse gas emissions. Such policies could play a role in promoting terrestrial carbon dioxide removal [Kauppi and Sedjo, 2001] (e.g., biofuels or afforestation). However, a metric of equivalence would be required for these policies to also account for the significant nongreenhouse gas climate effects [Bala et al., 2007; Bonan, 2008] of land use change [Marland et al., 2003]. Radiative forcing has been suggested as such a metric in scientific [Betts, 2000], economic [Thompson et al., 2009], and life cycle assessment contexts [Muñoz et al., 2010; Schwaiger and Bird, 2010; Bright et al., 2012].

[4] However, the forcing additivity assumption between land use change and greenhouse gases has not been explicitly evaluated. It is questionable on theoretical grounds because not all climate changes associated with land use change are radiative in nature [Anon, 2005; Anderson et al., 2010], for example, vertical heat distribution changes within the atmosphere [Boucher et al., 2004; Forster et al., 2007], and because the spatial scale of climate forcing from land use change differs from that of well-mixed greenhouse gases [Pielke et al., 2002; Marland et al., 2003; Anon, 2005]. Questions are also raised by simulations of regional and global climate impacts of past [Findell et al., 2007; Kvalevåg et al., 2009; Pongratz et al., 2009] and future [Fuglestvedt et al., 2003; Feddema et al., 2005; Arora and Montenegro, 2011] land use change.

[5] At the global level, forcing additivity requires that the climate sensitivity (i.e., the global mean temperature response to a unit of forcing) of different forcing agents be equal. Yet, literature on the relative climate sensitivities of CO2 and land use change is inconclusive [Hansen et al., 2005; Davin et al., 2007]. To the extent that these quantities differ, it may be appropriate in some contexts to scale forcing estimates by an efficacy factor [Hansen et al., 2005] to provide global-scale equivalence. However, the impacts and costs of climate change are experienced at the regional scales where the stress of climate change is felt by human systems [Hibbard and Janetos, 2013]. Thus, in policy and assessment contexts, forcings would ideally be both regionally and globally additive in order to convey actionable signals.

[6] The present study quantifies the potential magnitude of the departure from perfect forcing additivity between land use change and greenhouse gases—and its implications for climate mitigation and assessment. We do this by examining an idealized scenario in which land use change forcing is approximately offset by a radiatively equivalent change in greenhouse gas concentrations. Such an outcome would occur by design under any carbon trading or offset policy that accounts for the biophysical effects of land use change in CO2 equivalent units based on radiative forcing. Thus, our analysis is critical to understanding the outcomes of such policies. To the extent that forcing additivity holds, we would expect little climate change (at any spatial scale) in this nearly net neutral forcing scenario.

2 Methods

2.1 Climate Simulations

[7] We consider three scenarios in which stepwise changes in land use (LUC), CO2 concentrations (CO2), or both (TRADE) are imposed on preindustrial conditions for 60 model years, resulting in the radiative forcing values summarized in Table 1. We also consider a preindustrial control simulation with no forcing. Simulations are conducted at approximately 1° resolution with release version 8 of the Community Climate System Model 4 (CCSM4) [Gent et al., 2011; Bitz et al., 2012], using standard configurations for the atmosphere and sea ice components, a slab ocean model, and prognostic carbon-nitrogen biogeochemistry within the land model. The timescale of equilibration for the slab ocean model is on the order of 10–30 model years, so the final 30 years of each perturbed forcing simulation reflect a new equilibrium climate.

Table 1. Global Forcing and Temperature Responses for the Three Perturbed Forcing Simulations Relative to Control
ScenarioLand Use ForcingCO2 ForcingNet ForcingGlobal Temperature ResponsePercent of Globe With Significant Temperature ChangeMean Absolute Temperature Change
 (W/m2)(W/m2)(W/m2)(°C)(%)(°C)
LUC−0.920−0.92−0.57650.60
CO200.900.900.74990.76
TRADE−0.920.90−0.020.17780.55

[8] In the land use change (LUC) scenario, large areas of natural vegetation in each continent are replaced with crops, resulting in approximately 50% loss of global forestland. This pattern of land use change is drawn from Jones et al. [2013] and reflects the relative difference between two integrated assessment model projections of 21st century land use under alternative climate policies. For a small number of grid cells (<5%), we scaled the pattern of land use change down to accommodate the abundance of certain vegetation types within the preindustrial data set (e.g., if the land use change pattern called for more deforestation than there were trees). We impose land use change in the land component of CCSM, the Community Land Model [Lawrence et al., 2012], by changing the fractional coverage of various plant functional types within each model grid cell. Plant functional types differ in their photosynthetic, aerodynamic, optical, and biogeochemical (e.g., carbon-nitrogen ratios of different organ pools) properties. Thus, land use change in our simulations influences the entire surface energy and water budget, not just albedo. We note that changes in terrestrial carbon stocks associated with land use change do not translate into changes in atmospheric greenhouse gas concentrations in these idealized simulations. We configure the model in this manner by design in order to isolate the biogeophysical aspects of land use change from the biogeochemical aspects.

[9] In the CO2 scenario, CO2 levels increase by 57 ppm relative to the preindustrial level of 285 ppm in order to approximately offset the negative forcing present in the LUC scenario. A third simulation, which we call the TRADE scenario, is designed to achieve net neutral forcing. It examines the combined effect of land use change from the LUC scenario and increased CO2 concentrations from the CO2 scenario.

2.2 Radiative Forcing and Equivalent CO2 Calculations

[10] We obtain the radiative forcing from albedo change via a two-step process designed to isolate the first-order effect of land use change on the net shortwave flux at the tropopause. First, we perform an offline land model simulation for 20 model years in which the atmospheric forcing variables passed to the land model are held at preindustrial conditions. This step eliminates atmospheric feedbacks on snow cover and vegetation growth dynamics that might influence land surface albedos. The surface albedos from this simulation are then used to drive the Parallel Offline Radiative Transfer Model [Conley et al., 2012] where, again, atmospheric state variables (e.g., water vapor, GHG, and cloud distributions) are held at preindustrial values drawn from the first 20 years of our control simulation. We compute radiative forcing as the difference between the net shortwave flux at the tropopause (200 hPa) from this simulation and the control simulation.

[11] We estimate the increase in CO2 concentrations over preindustrial levels required to offset the albedo forcing present in the LUC scenario with the following relationship [Myhre et al., 1998]:

display math

[12] where F is the radiative forcing in units of W/m2, C is the perturbed CO2 concentration, and C0 is the baseline CO2 concentration prior to perturbation. Based on an initial estimate of the land use forcing at −1 W/m2, this yields a required increase of 57 ppm over the preindustrial level of 285 ppm in order to generate 1 W/m2 of positive forcing. We perform a subsequent offline radiative transfer model simulation similar to that performed for the LUC in order to quantify the actual forcing due to this CO2 concentration change within the CCSM4 model. Here, we take forcing to be the change in net radiative flux at the tropopause (200 hPa). After applying a stratospheric adjustment factor of −12% based on Hansen et al. [1997], this analysis yields a CO2 forcing of 0.90 W/m2.

3 Results

3.1 Temperature Response

[13] The radiative forcing and equilibrium temperature responses for each of these perturbed-forcing simulations are shown in Table 1. At the global level, the implied climate sensitivity (i.e., the ratio of mean equilibrium temperature change relative to the applied forcing level) is somewhat higher in the CO2 case (0.82) compared to the LUC case (0.62). This yields an efficacy factor (the ratio of these values) for LUC forcing of 0.76.

[14] Although the TRADE scenario achieves nearly neutral forcing, it has more locations that experience a statistically significant change in mean annual temperature than the LUC scenario (see Table 1). This is because the cooling associated with land use change is highly concentrated in the northern latitudes where the albedo contrast between forest and cropland is most pronounced (see Betts [2000] and Bala et al. [2007] for similar results), whereas the warming associated with CO2 change is more uniform (Figure 1). In the TRADE scenario, which includes both LUC and CO2 forcings, the Southern Hemisphere and tropics warm even as the northern middle and high latitudes cool.

Figure 1.

Spatial pattern of equilibrium surface air temperature change relative to control for each perturbed forcing simulation, as well as the linear combination of the LUC and CO2 changes. Grid cells are colored only if a t test indicates they are significantly different from the control simulation at the 95% confidence level.

[15] In the rightmost column of Table 1, we present an index of the mean absolute temperature change—i.e., the absolute deviation from preindustrial conditions averaged across each of the approximately 1° model grid cells. This metric weighs both positive and negative deviations equally such that for cooling and warming to cancel one another, they must occur in the same location. For the LUC and CO2 cases, in which the temperature change is essentially unidirectional across space, the value of the mean absolute temperature change is similar to the global mean temperature change. However, for the TRADE case, the scale of absolute temperature change is on the order of the other scenarios and not close to zero as the global mean temperature change would suggest.

[16] The mean temperature change in the TRADE case is very nearly the linear combination of changes in the LUC and CO2 cases. In order to determine if the same is true of the spatial patterns of the temperature responses, we perform a statistical analysis comparing the sum of the CO2 and LUC temperature anomalies (relative to control) to the TRADE temperature anomalies (see Figure 1d). A t test reveals a significant difference at only 17% of grid cells (representing 12% of the Earth's surface) at the 95% confidence level. The pattern of statistically distinct differences indicates that the linear combination of responses may overestimate both high-latitude Northern Hemisphere cooling (above 80°N) and Southern Hemisphere warming relative to the TRADE simulation (see Figure S1 in the supporting information). However, it should be noted that given spatial autocorrelation, we expect the t test to characterize more than 5% of grid cells as statistically significant at the 95% confidence level simply due to natural variability [Livezey and Chen, 1983], so our analysis provides only weak evidence of an interaction effect between the spatial patterns of CO2 and LUC temperature responses.

3.2 Precipitation Response

[17] As shown in Table S1, the percent change in global mean precipitation per degree of global warming ranges from 1.4 to 2.1, in line with results from other general circulation models [Vecchi and Soden, 2007]. However, these global metrics again mask significant compensating changes at the regional scale. Comparing the zonally averaged change in precipitation between perturbed-forcing and control simulations (Figure 2) reveals that whereas CO2 forcing leads to an increase in precipitation across most latitudes, the LUC and TRADE scenarios decrease in northern tropical and increase in southern tropical precipitation. This dipole pattern is consistent with a southward shift in the Intertropical Convergence Zone (ITCZ), which has been shown to follow changes in the interhemispheric temperature gradient [Chiang and Bitz, 2005; Broccoli et al., 2006; Frierson and Hwang, 2012; Swann et al., 2012] such as those that occur in the LUC and TRADE scenarios.

Figure 2.

Zonally averaged change in annual precipitation for each perturbed forcing simulation, as well as the linear combination of the LUC and CO2 changes. Dark solid lines indicate changes that are statistically different from the control simulation at the 95% confidence level.

[18] Despite much lower global averaged forcing, the mean absolute precipitation change (analogous to mean absolute temperature change described above) in the TRADE scenario is on par with that of the LUC scenario, and both of these scenarios exceed the magnitude of response in the CO2 scenario (Table S1). Approximately one third of the globe experiences a statistically significant change in mean annual precipitation in the LUC and TRADE cases compared to 18% in the CO2 case.

[19] Adding the LUC and CO2 responses reproduces the TRADE response at most latitudes, with only 9.4% of latitude bands differing statistically between the combined LUC + CO2 response and the TRADE response. The most notable point of differentiation is the overestimate of the precipitation decrease near 10°N (see Figure 2). The longitudinally specific precipitation responses (analogous to Figure 1) can be found in Figure S2.

3.3 Timescales of Change

[20] Critically, we find that land use change causes rapid climate changes relative to CO2. In the LUC case, more than 50% of the equilibrium temperature change occurs in the very first year, indicating that fast timescale feedback processes dominate this response. In contrast, only 12% of the equilibrium temperature change occurs in the first year of the CO2 scenario. These differences are likely due to the lower heat capacity of continents versus oceans—all of the LUC forcing is concentrated over land where temperatures are able to adjust rapidly to a given change in energy fluxes. The fast temperature response in the LUC case is driven by changes over land, whereas the ocean response is both smaller and slower (see Figure S3). The TRADE scenario (not shown) exhibits both the rapid temperature decline evident in the LUC scenario and the slower buildup evident in the CO2 scenario.

4 Discussion

[21] The utility of radiative forcing as a policy metric for judging the climate value of different activities relies on there being a strong relationship between forcing, climate response, and societal impacts. However, to the degree that climate responses vary as a function of the activities that generate forcing, this relationship breaks down.

[22] As we have demonstrated, equivalent LUC and CO2 forcings can lead to substantially different climate change patterns, not just in temperature but also in atmospheric and hydrological responses. In fact, the patterns of climate change from these forcing agents differ so much that when they are combined to yield nearly neutral forcing, a new, hybrid pattern of global temperature and precipitation change emerges.

[23] Thus, the forcings from LUC and CO2 are far from additive in the sense that would be required for them to be traded against one another using a single, simple metric. Specifically, for LUC forcing fLUC and CO2 forcing fCO2, there is not a single spatially resolved climate response function C such that

display math

[24] Furthermore, we note that the use of an efficacy factor (i.e., a coefficient on the fLUC term) does not correct this problem. However, our results do lend support to the notion that the independent climate responses of these different forcing agents are themselves additive. That is, the joint climate response to the two forcing agents is approximately equal to the sum of the two responses:

display math

[25] We refer to this latter property as response additivity to distinguish it from the former forcing additivity. Response additivity is a key assumption underlying both pattern scaling [Mitchell, 2003] and climate detection and attribution studies [International ad hoc Detection and Attribution Group (IDAG), 2005]. It is also confirmed by an idealized experiment similar to ours examining the climate responses to low cloud albedo forcing and greenhouse gases [Ramaswamy, 1997], as well as a comprehensive examination of several forcing agent combinations [Shiogama et al., 2012]. Current climate impact analysis and mitigation market design, however, are based on the incorrect forcing additivity assumption.

[26] Climate impacts and adaptation research requires the examination of a wide range of potential future socioeconomic and policy scenarios and their associated patterns of climate change. However, only a handful of scenarios, the Representative Concentration Pathways (RCPs) [van Vuuren et al., 2011], have been systematically evaluated by the international climate modeling community. The IPCC parallel process [Moss et al., 2010] was developed in order to efficiently match a wide variety of socioeconomic scenarios with multimodel RCP climate projections, eliminating the need to simulate each new scenario in global climate models. The assumption underlying this approach is that different scenarios that reach the same forcing target result in the same climate (i.e., forcing additivity). Previous work has shown that the current exclusion of biophysical land use forcing from this measure of total forcing can result in inconsistently matched scenarios with temperature change errors exceeding 3°C in some regions [Jones et al., 2013]. The present study reveals an additional complication: because land use and GHG forcings are not climatically equivalent at the regional scales where human systems encounter the stress of climate change, simply including land use forcing in the calculation of total radiative forcing used to pair scenarios is insufficient to guarantee climate consistency.

[27] Our finding that response additivity may hold suggests an alternative approach. Rather than focusing on a handful of scenarios that blend many different forcing agents, future multimodel intercomparison efforts could instead characterize the unique temporally evolving climate signals associated with each of many different forcing agents. These could then be combined in novel ways through pattern scaling [Mitchell, 2003] to characterize the climate outcomes of a wide range of future socioeconomic scenarios. While the Fifth Coupled Model Intercomparison Project single-forcing simulations are a step in this direction, they do not permit the derivation of unique climate change signals associated with the many potential kinds of land use change (e.g., deforestation, introduction of irrigation) across different regions. For this approach to succeed, additional work is needed to systematically test the assumptions underlying pattern scaling. For instance, it is not clear whether the apparent ITCZ shift in our LUC and TRADE scenarios exhibits threshold behavior or whether incremental land conversions induce incremental circulation changes as linear pattern scaling would assume.

[28] Our finding that fast timescale processes dominate the temperature response to land use change has important implications for society and for the use of appropriate metrics for characterizing the climate consequences of land use change. “Fixed-SST” and “regressed” radiative forcing, which are measures of Earth's radiative imbalance after fast feedback processes have occurred, have been shown to more accurately reflect the long-term mean warming potential of certain forcing agents such as black carbon compared to the standard definition of radiative forcing [e.g., Hansen et al., 2005; Ban-Weiss et al., 2011]. However, in the case of land use change, fast feedbacks have significant implications for surface climate conditions. Indeed, the method of Gregory [2004] for estimating regressed radiative forcing yields a value of only −0.39 W/m2 for our LUC case compared to 0.72 W/m2 for the CO2 case and would therefore underestimate the relative magnitude of global mean temperature change in these two forcing scenarios.

[29] Global-scale metrics and targets, like the ubiquitous 2° warming threshold [den Elzen and Meinshausen, 2006] or the RCP radiative forcing targets, are no doubt useful for benchmarking the overall scale of climate change. However, the present study points to the limitations of using such metrics to drive mitigation policies or to differentiate detailed socioeconomic scenarios, especially when different kinds of forcing agents are combined in meaningful quantities. Of the metrics we examined, the mean absolute difference metric most accurately reflects the relative scale of change present in our three scenarios. However, this metric still ignores important features of the Earth system that would better inform the social costs of climate change. For instance, some of the greatest potential costs of climate change are thought to occur when vulnerable populations and infrastructure are exposed to multiple coinciding extreme conditions [Field et al., 2012].

[30] Given the demonstrated lack of forcing additivity between land use change and GHG, a rational policy approach may be to develop separate targets and regulatory regimes for activities that influence the climate in different ways rather than attempting to bundle them together through the use of credits. Our results show that such credits belie socially significant differences among the climate outcomes of different forcing agents.

Acknowledgments

[31] We would like to thank Michael O'Hare and Anthony Janetos for feedback on early versions of this manuscript. This work was supported by the Director, Office of Science, Office of Biological and Environmental Research, Climate and Environmental Science Division, of the U.S. Department of Energy under contract DE-AC02-05CH11231. The CESM project is supported by the National Science Foundation and the Office of Science of the U.S. Department of Energy. This research used resources of the National Energy Research Scientific Computing Center, which is supported by the Office of Science of the U.S. Department of Energy under contract DE-AC02-05CH11231.

[32] The Editor thanks Steven Running and an anonymous reviewer for their assistance in evaluating this paper.

Ancillary