Polar amplification of Pliocene climate by elevated trace gas radiative forcing

Significance

Warm periods in Earth’s history provide the only empirical evidence of how the climate system responds to raised atmospheric carbon dioxide (CO₂) levels. The Middle Pliocene, 3.3 to 3.0 My B.P., was the last time when CO₂ levels were as high as today. However, climate model simulations of the Pliocene underestimate the warming that has been reconstructed from geological archives. Using a numerical model of the global methane cycle we show that the inclusion of enhanced concentrations of non-CO₂ trace gases could have been responsible for an additional warming of 0.6 to 1.0 °C, with larger increases over northern land masses. These findings demonstrate the importance of trace gas climate forcing for both the Pliocene and potentially warm periods during much of Earth’s recent history.

Abstract

Warm periods in Earth’s history offer opportunities to understand the dynamics of the Earth system under conditions that are similar to those expected in the near future. The Middle Pliocene warm period (MPWP), from 3.3 to 3.0 My B.P, is the most recent time when atmospheric CO₂ levels were as high as today. However, climate model simulations of the Pliocene underestimate high-latitude warming that has been reconstructed from fossil pollen samples and other geological archives. One possible reason for this is that enhanced non-CO₂ trace gas radiative forcing during the Pliocene, including from methane (CH₄), has not been included in modeling. We use a suite of terrestrial biogeochemistry models forced with MPWP climate model simulations from four different climate models to produce a comprehensive reconstruction of the MPWP CH₄ cycle, including uncertainty. We simulate an atmospheric CH₄ mixing ratio of 1,000 to 1,200 ppbv, which in combination with estimates of radiative forcing from N₂O and O₃, contributes a non-CO₂ radiative forcing of 0.9 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ (range 0.6 to 1.1), which is 43% (range 36 to 56%) of the CO₂ radiative forcing used in MPWP climate simulations. This additional forcing would cause a global surface temperature increase of 0.6 to 1.0 °C, with amplified changes at high latitudes, improving agreement with geological evidence of Middle Pliocene climate. We conclude that natural trace gas feedbacks are critical for interpreting climate warmth during the Pliocene and potentially many other warm phases of the Cenezoic. These results also imply that using Pliocene CO₂ and temperature reconstructions alone may lead to overestimates of the fast or Charney climate sensitivity.

Introduction

The Middle Pliocene warm period around 3.3 to 3.0 My B.P. was the last period in Earth’s history when atmospheric CO₂ was comparable to today’s level, at approximately 400 ppmv (1–4). The Middle Pliocene warm period could therefore provide useful information on the response of the Earth system to greenhouse gas (GHG)-induced warming, that is relevant to the future evolution of the Earth system under continued anthropogenic GHG emissions (5–9). According to syntheses of reconstructed surface temperatures from geological archives on land (10) and in the ocean (11, 12), the Earth was globally warmer than the preindustrial, with a significant polar amplification especially in the Northern Hemisphere. The warmer conditions acted to reduce global ice volume, so that sea level was around 20 m higher than present (13–15).

The Pliocene Model Intercomparison Project (PlioMIP) is a coordinated study of climate model responses to Pliocene boundary conditions (16, 17) aimed at quantifying the underlying drivers of warmth during this time and to better understand the Earth System response to an atmospheric CO₂ concentration of $\sim$400 ppmv. The results from the first phase (PlioMIP1) showed that coupled general circulation models (GCMs) failed to reproduce high-latitude warming seen in reconstructions (16). Several hypotheses could explain why the model simulations underestimate warming, including the role of orbitally induced climate variability (16), the configuration of ocean gateways and palaeogeography (e.g., refs. 18 and 19), and radiative forcing from trace gases other than CO₂ (20). While several studies have addressed the first three possibilities, the contributions of the atmospheric methane, nitrous oxide, and ozone (CH₄, N₂O, and O₃) have not been considered together.

At present, CH₄ is the second most important anthropogenic GHG after CO₂ (21). Its nearly threefold concentration increase since CE 1750 is responsible for approximately 25% of the GHG radiative forcing. We know that variations in CH₄ are huge in intensity and abrupt during the Anthropocene relative to the past several thousand years. Before widespread direct atmospheric monitoring, CH₄ could only be traced through air bubbles trapped in polar ice cores. These ice core records provide robust evidence that CH₄ is very sensitive to climate, with higher values during warmer periods and with large-amplitude rapid variations during abrupt climate events (22, 23). Nitrous oxide (N₂O) also displays similar characteristics (24), while O₃ and many other important atmospheric constituents (e.g., OH) cannot be reconstructed from ice core gas samples. Before 800,000 y ago, variations in all trace gases with the exception of CO₂ are essentially unknown and must therefore be simulated using models (e.g., ref. 25).

Beerling et al. (26) simulated enhanced wetland emissions during the Pliocene but did not link this with other methane sources (26). Unger and Yue (20) (UY14 hereafter) simulated an increase in CH₄ lifetime by 23 to 31% during the Pliocene (for CH₄ = 1,000 to 2,000 ppbv) and an increase in tropospheric O₃ by 21 to 25% which contributed a radiative forcing of approximately 0.3 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$. However, this CH₄ lifetime increase is partly caused by the self-feedback (e.g., ref. 27), a function of the prescribed CH₄ concentration in the model. Other than a model study of the Eocene and late Cretaceous superwarm periods (28), there is no comprehensive understanding of how trace gases may have affected pre-Quaternary warm periods like the Pliocene.

We address this with an ensemble of terrestrial biogeochemistry simulations for the Pliocene. In combination with climate fields from several GCM simulations of the Pliocene we develop an estimate of the Pliocene global CH₄ source, CH₄ sinks and thereby the CH₄ concentration, and CH₄ radiative forcing, including uncertainty. We combine these with estimates of N₂O and O₃ radiative forcing to evaluate how all of these trace gases may have determined the climate during the Pliocene.

Results

Emissions of CH₄ and Other Trace Gases.

Our approach follows that used for modeling of the Quaternary CH₄ variations (25, 29, 30). We drive the Lund–Potsdam–Jena General Ecosystem Simulator (LPJ-GUESS) dynamic global vegetation model (31, 32) with climate simulations from four models used in the Pliocene Model Intercomparison Project phases 1 and 2 and simulate the key trace gas emissions at the land surface, including CH₄, emissions from wildfires, nonmethane biogenic volatile organic compounds (NMVOCs) from vegetation, and soil nitrogen oxides (NOx) (SI Appendix, Fig. S1). Assuming that the Pliocene is a quasi-equilibrium climate state, we do not account for any increased emissions from marine clathrates or permafrost which are likely more sensitive to abrupt warming (33). We use PlioMIP phase 1 (experiment 2) simulations (34) with Community Climate System Model 4 (CCSM4), Goddard Institute for Space Studies Model E2R (GISS-E2-R), and Institut Pierre-Simon Laplace Coupled Model 5A-LR (IPSL-CM5A-LR) and PlioMIP phase 2 (experiment with Pliocene orography and ice: ${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$) (16) simulations with IPSL-CM5A-LR and Hadley Centre Coupled Model 3 with Met Office Surface Exchange version 2.1 (HadCM3-M21). In all Pliocene climate model simulations the orography, land ice, and vegetation are based on geological reconstructions for the Pliocene (35, 36), and the mixing ratio of atmospheric CO₂ is increased from 280 ppbv in the preindustrial to 400 or 405 ppbv. We also incorporate HadCM3 PlioMIP phase 2 simulations performed with atmospheric CO₂ mixing ratios of 450 (37) and 490 ppmv which are labeled ${\mathrm{E}\mathrm{o}\mathrm{i}}^{450}$ and ${\mathrm{E}\mathrm{o}\mathrm{i}}^{490}$, respectively. All climate simulations are described in Materials and Methods.

We estimate the CH₄ lifetime and the resultant concentration of CH₄ using a simplified offline one-box model of the atmospheric CH₄ chemistry (38) and radiative forcing (39). With this we do not resolve atmospheric transport or detailed chemical pathways, but we are able to quantify uncertainty by sampling the model stochastically. See Materials and Methods and SI Appendix, Table S2, for full details of all models used.

All CH₄ emissions increase in response to the simulated Pliocene climate conditions (SI Appendix, Table S1). Despite warmer and in many places wetter conditions (SI Appendix, Fig. S2), wetland area increases only marginally, but the emissions increase by 20 to 46%. Emissions increase notably in the Sahel region of North Africa in all models and in South Africa in all models except GISS. Elsewhere, emissions increase in Southeast Asia in all models except GISS and Australia in all models except CCSM4 and GISS. GISS shows the largest emissions increase of all models over the Amazon, whereas both IPSL simulations show a slight reduction in this region. A factor separation approach summarized in SI Appendix, Table S3, demonstrates that increased Pliocene temperature enhances emissions in all models. In HadCM3 and GISS, wetter soils also promote emissions but are less important in the other GCMs. Soil carbon stocks act to reduce emissions in HadCM3 and increase emissions in CCSM4 but have much less impact in the other models. Wildfire and termite emissions also increase by 50 to 83% and 21 to 37%, respectively and soil uptake (at pCH₄ = 715 ppmv) increases by 8 to 27%. Isoprene emissions increase by 5 to 40%, and monoterpene changes by −9 to +7%. These ranges span the different climate models used to drive each emissions scheme.

CH₄ Chemical Lifetime and Radiative Forcing.

Although increased emissions will translate into elevated atmospheric concentration, the concentration is also dependent on the lifetime, which is influenced by other trace gas emissions and the physical state of the atmosphere. The change in CH₄ lifetime, Pliocene minus preindustrial, is summarized for each GCM in Fig. 1. Increased CH₄ emissions from wetlands, wildfires, and termites contribute a positive lifetime anomaly through the self-feedback effect. Additional NMVOC emissions (CO and isoprene) from wildfires and vegetation enhance the lifetime slightly. The higher global temperatures and associated humidity increase act to offset some of this lifetime increase in each model. This is because the reaction rate of CH₄ with OH scales with temperature, and the production of OH, which oxidizes CH₄, increases with water vapor availability and hence temperature (e.g., refs. 38 and 40). NMVOCs are the most model-dependent term, with a strong increase in emissions during the Pliocene in IPSL and HadCM3 but not in the other two GCMs. This is most likely a result of the differences in the response of the tropical hydrological cycle. Soil NOx shows agreement across the four climate models. The results are also in agreement with the results of UY14. Lightning NOx increase and increased O₃ in the troposphere (both taken from UY14) have opposing influences on the Pliocene lifetime.

Fig. 1.

Simulated mean change in lifetime of CH₄ with respect to OH for the Pliocene minus the preindustrial. Net values and individual terms are shown for the five different sets of climate drivers (HadCM3-M21-${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$, IPSL-CM5A-PlioMIP1, IPSL-CM5A-${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$, GISS-E2-R-PlioMIP1, and CCSM4-PlioMIP1).

As the CH₄ lifetime is dependent on the CH₄ level, there is a nonlinearity which means that the total change in lifetime does not equal the sum of these individual terms. The net change is positive for HadCM3 and both IPSL simulations and very weakly negative in the other two models. This is mostly due to the much stronger BVOC emission term in the former two models.

The results of the sampling of the total Pliocene minus preindustrial CH₄ radiative forcing are shown in Fig. 2. The mean $\pm$ 1 SD anomalies are 0.16 $\pm$ 0.02, 0.20 $\pm$ 0.03, 0.21 $\pm$ 0.03, and $0.33\hspace{0.17em}\pm \hspace{0.17em}0.05$ ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ for CCSM4, GISS, IPSL, and HadCM3, respectively. The combined mean and 1$\sigma$ range is 0.22 $\pm$ 0.07 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$. The total uncertainty in this estimate is dominated by the choice of climate simulation.

Fig. 2.

Estimated radiative forcing expressed as a probability density function. The uncertainty stems from 1) assumed $\pm$20% uncertainties in lightning NOx and tropospheric O₃, 2) uncertainty in the parameters used to estimate OH lifetime of CH₄ (38), and 3) uncertainty in the total radiative forcing due to CH₄ (38, 39). The $\pm$1 $\sigma$ uncertainty is shown above each curve. Dashed lines represent the first two factors, and solid bars underneath include all three sources of uncertainty.

Discussion

Total Radiative Forcing.

The global mean radiative forcing due to the CO₂ concentration of 400 ppm during the Pliocene is around 2.0 $\pm$ 0.3 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$. We approximate N₂O radiative forcing based on natural variations in these three GHGs through the late Quaternary as 15% of the combined CO₂ and CH₄ radiative forcing (41). Radiative forcing by O₃ is 0.29 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ as simulated for CH₄ = 1,000 ppbv by UY14. This positive radiative forcing is partly due to a simulated increase in emissions of O₃ precursors, consistent with our simulated emissions (SI Appendix, Table S1). The uncertainty on the O₃ radiative forcing is calculated by scaling with the CH₄ concentration, which in three sensitivity simulations by ref. 20 shows an approximately logarithmic dependence. O₃ is also created as a by-product of CH₄ emissions; thus, it is partially included in the O₃ radiative forcing and in our indirect CH₄ radiative forcing. Hence, we reduce our total forcing using the combined factor for O₃ radiative forcing from CH₄ as in SI Appendix, Table S4, to avoid double counting. Together with a CH₄ radiative forcing, we derive an approximate radiative forcing due to other non-CO₂ well-mixed GHGs of 0.8 (0.72 to 1.02) ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ as summarized in Table 1.

Table 1.

Pliocene radiative forcing, CH₄ cycle sensitivity, and predicted non-CO₂-forced warming from this study and past work

	Mean	Range	Reference
Radiative forcing (${\mathrm{W}\cdot \mathrm{m}}^{-2}$)
CO2	2.0	–	Haywood et al. (17)
CH4	0.22	0.10–0.43	This study
N2O	0.33	0.32–0.36	This study
O3	0.29	0.28–0.36	Unger and Yue (20)
Total GHG	2.80	2.67–3.02	This study
Total non-CO2	0.80	0.67–1.02	This study
With self-feedback	0.86	0.72–1.12	This study
% of CO2	43%	36–56%	This study
CH4 cycle sensitivity to
Temperature ($\times 10^{-3}{\mathrm{W}\mathrm{m}}^{-2}{\mathrm{K}}^{-1}$)
Pliocene (3.3–3.0 Ma BP)	100	89–114	This study
Early Eocene (55 Ma BP)	109	97–120	Beerling et al. (28)
Cretaceous (90 Ma BP)	89	–	Beerling et al. (28)
LGM (21 ka BP)	68	57–85	Loulergue et al. (22) and Annan and Hargreaves (43)
Global mean temperature change (°C)
Warming (non-CO2)	0.7	0.6–0.9	This study
With self-feedback	0.7	0.6–1.0	This study

The rate of methane radiative forcing increase per degree of global mean temperature change (see also ref. 42) is a useful property to compare time periods or models. We denoted this as ${\gamma }_{fCH4}$, and it ranges from 89 to 113 $\times \hspace{0.17em}10^{-3}{\mathrm{W}\mathrm{m}}^{-2}{\mathrm{K}}^{-1}$ across the five climate model simulations, with the lowest value simulated by CCSM4 and the highest by HadCM3. This range slightly exceeds the observed value of 68 (57 to 85) $\times \hspace{0.17em}10^{-3}{\mathrm{W}\mathrm{m}}^{-2}{\mathrm{K}}^{-1}$ derived from the last glacial maximum to the late-Holocene preindustrial (22, 43). It is consistent with simulated values of 97 to 119 $\times \hspace{0.17em}10^{-3}{\mathrm{W}\mathrm{m}}^{-2}{\mathrm{K}}^{-1}$ for the Early Eocene and 89 $\times \hspace{0.17em}10^{-3}{\mathrm{W}\mathrm{m}}^{-2}{\mathrm{K}}^{-1}$ for the Late Cretaceous calculated with a coupled climate–chemistry modeling framework (28), for which the direct radiative forcing values have been augmented here following a recent update (39).

The Pliocene climate may have induced changes in the emissions and lifetime of natural aerosols. UY14 simulated a net cooling by nitrate, particulate organic matter, and biogenic secondary organic aerosols (SOA) and a warming effect from black carbon, leading to a negative aerosol forcing of −0.4 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ (20). We simulate smaller changes in emissions from biomass burning (+50 to 84% versus +101% by UY14) and BVOCs (precursors of SOA, +5 to 40% versus +50% by UY14), and so this would likely equate to a smaller net forcing. Other aerosol changes may have a warming effect. For example, mineral dust levels are generally lower in warm climates, and deserts contracted during the Pliocene (10). Interactions of CH₄ with nitrate and sulfate can also enhance radiative forcing (44). Pliocene aerosol effects therefore require further study.

The climate simulations could be benchmarked with Pliocene climate reconstructions. Though no global compilations of precipitation change are available for the Pliocene, individual records can provide informative constraints. The West African monsoon was probably stronger during the Middle Pliocene even when compared to the early to middle Holocene (45). There was also a general drying trend in East Africa during the Pliocene (46), which, if continued to the present, implies wetter conditions during the Middle Pliocene. Both of these features are most faithfully reproduced in HadCM3 and GISS and the ${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$ IPSL simulation (SI Appendix, Fig. S2). In Asia, a reconstructed near-doubling of precipitation in southern China (47) is replicated in IPSL (${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$ simulation) and by HadCM3 but not the other models. Only HadCM3 and CCSM4 simulate wetter conditions reconstructed for the early to middle Pliocene in Australia (48). One of the regions of largest intermodel spread in precipitation is tropical South America, but to our knowledge, quantitative precipitation records have yet to be produced.

HadCM3 has the strongest warming globally and in particular over most land masses and over the North Atlantic, where most other models significantly underestimate reconstructed warming (6). HadCM3 model also shows among the best agreement with the available temperature reconstructions (10). This lends support to the stronger methane forcing that is simulated with the HadCM3 climate drivers since the enhanced warming, particularly over land and associated hydrological cycle changes, promote trace gas emissions. A global biome reconstruction could also provide some measure to discern between simulations (49), but this is initially derived from a simulation with HadCM3 and so is very likely biased toward the climatic response of this model.

Likely Non-CO₂ Climatic Response.

Our modeling study probably underestimates the true magnitude of the Pliocene methane cycle feedback because the climate and methane cycle are not coupled. The feedback of methane radiative forcing via climate to methane emissions is not complete, and all of the simulated climates underestimate midlatitude and high-latitude warming over land considerably (10). Furthermore, our study has a relatively modest change in wetland area compared with modeling studies of earlier warm periods (26, 28, 50), and so the resultant radiative forcing is potentially a conservative prediction. Any additional warming resulting from the higher concentrations of trace gases including methane would further perturb the sources and atmospheric chemistry. Using the upper and lower limits for the ${\gamma }_{fCH4}$, we can approximate this. The trace gas forcing-induced warming would lead to an additional methane feedback bringing the total Pliocene methane radiative forcing to 0.72 to 1.1 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$. These total non-CO₂ values are 36 to 56% of the CO₂ radiative forcing, with a central estimate of 0.9 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ or 43% of the CO₂ forcing. This would cause a warming of 0.6 to 1.0 °C (Table 1).

This is an important additional warming signal given that PlioMIP phase 1 GCMs forced only with increased CO₂ show a response of 2.7 $\pm$ 0.8 K (16). To better understand the regional impacts of this additional radiative forcing we analyze Middle Pliocene simulations from HadCM3 with prescribed levels of atmospheric CO₂. The CO₂ radiative forcing differences relative to the default ${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$ are approximately 0.66 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ for the increase by 50 ppmv and 1.1 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ for the increase to 490 ppmv. These encompass the range of non-CO₂ GHG radiative forcing we calculated (0.72 to 1.1 ${\mathrm{W}\mathrm{m}}^2$). The effective radiative forcing (or the temperature change per unit increment of radiative forcing) is actually higher for CH₄ and N₂O than for CO₂, and it is lower for O₃ (51). For simplicity here we assume that they are all equal.

The Pliocene HadCM3 simulations with differing CO₂ concentrations show significant polar amplification, especially in the Northern Hemisphere and around the Atlantic (Fig. 3). The ${\mathrm{E}\mathrm{o}\mathrm{i}}^{450}$ and ${\mathrm{E}\mathrm{o}\mathrm{i}}^{490}$ simulations show relatively muted warming signal in the circum-Atlantic because of a reduction in heat transport in the Atlantic (37). The upper end of our estimated radiative forcing, represented by the ${\mathrm{E}\mathrm{o}\mathrm{i}}^{490}$ simulation, is shown in Fig. 3B. This estimated non-CO₂ GHG radiative forcing (Fig. 3C) causes high-latitude temperatures to increase by around 1 to 2.5 °C over land and by 1 to 2 °C over the ocean surface (Fig. 3 and SI Appendix, Fig. S3). This does not eliminate the large mismatches found over high-latitude Asia (Fig. 3 C and D) and in the North Atlantic (SI Appendix, Fig. S3), but it improves the comparison with the temperature reconstructions by ref. 10 as shown in Fig. 3D. The distribution of errors is shifted from having a significant probability over the range −5 to −1 °C to being approximately centered on 0 °C. For HadCM3, both the smaller and larger increases in radiative forcing improve the model–data agreement, except for sites with very high temperatures at high latitudes. The very high temperatures in high-latitude regions would very likely further enhance trace gas emissions, especially CH₄ from wetlands, and so reinforce the positive radiative forcing.

Fig. 3.

Simulations with HadCM3-M21 and reconstructions of Pliocene near-surface air temperature change ($\mathrm{\Delta }$T) relative to the preindustrial (K). (A) Simulated (37) and reconstructed (10) $\mathrm{\Delta }$T for Eoi⁴⁹⁰ minus preindustrial, (B) simulated $\mathrm{\Delta }$T for ${\mathrm{E}\mathrm{o}\mathrm{i}}^{490}$ − ${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$, (C) latitudinal comparison of the reconstructed and simulated temperature anomalies, and (D) histogram of model minus reconstruction temperature anomalies.

Reasons for the enduring high-latitude discrepancies might include seasonal bias in reconstructions over land or ocean (52) and long-term trends in seawater chemistry (53). If the reconstructions capture peak warmth during orbital cycles, this may not be adequately captured by climate model simulations with preindustrial orbital configuration (54), and global assemblages of reconstructions may capture different phases of Earth’s orbit in different locations (10), further complicating comparisons with simulations. Terrestrial reconstructions are also potentially influenced by the dependence of plant water use efficiency on atmospheric CO₂ and changes in seasonality and the frequency of extreme events (55). However, a general underestimation in polar amplification of past warm states could signify systematic problems with climate model parameterizations of clouds (56, 57) and/or aerosol–cloud interactions (58).

These model results allow an estimate of ${\mathrm{S}}_{LI,GHG}$, the temperature response per unit of radiative forcing due to ice sheets and sea level (LI) and due to GHGs (K/${\mathrm{W}\cdot \mathrm{m}}^{-2}$), where ${\mathrm{S}}_{LI,C{O}_2}$ was possible until now (4). Several studies have considered the Earth system sensitivity, the long-term temperature response to changes in radiative forcing, incorporating vegetation and ice sheet responses (5). Our central estimate of 43% non-CO₂ GHG radiative forcing would reduce values estimated from Pliocene temperature reconstructions from around 9 °C (59) to 6.5 °C. This is still much larger than predicted based on fast feedbacks in the climate system and is therefore potentially consistent with irreversible long-term planetary warming, should we fail to limit warming this century to 1.5 °C (8).

Conclusions

The Middle Pliocene is a critically important past time period to understand because levels of atmospheric CO₂ were very likely as high as today’s anthropogenically perturbed levels, at around 400 ppmv. The Pliocene therefore offers unique insight into a warmer Earth system at equilibrium (5, 6). CH₄ is the second most important anthropogenic GHG today, but beyond the ice core era to 800 ka BP, it has been largely ignored. Whereas for CO₂, proxies have been developed to reconstruct its evolution for deep times, no such proxy exists for CH₄. We used vegetation model simulations and an offline CH₄ budget model to produce a comprehensive set of simulations quantifying the likely CH₄ emissions, lifetime, and concentration during this time period, investigating changes of methane sources and sinks but excluding abrupt releases derived from clathrates or permafrost degradation. We show that there is a direct net positive radiative forcing of 0.22 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ (range 0.1 to 0.45), which when combined with estimates of radiative forcing from N₂O and O₃ leads to an additional radiative forcing that is between 36 and 56% of that caused by a CO₂ concentration of 400 pmv. This would likely cause an additional global mean warming of approximately 0.6 to 1.0 °C at the global scale. Terrestrial trace gas radiative forcing is therefore critically important for understanding the reconstructed warmth of the Pliocene. We suggest that the first-order estimates of these additional forcing agents could help to reconcile model and proxy-based estimates of Pliocene warmth and that failing to include such forcings will lead to substantial biases in simulations of past climates. Their omission also has consequences when trying to estimate the fast or Charney climate sensitivity from paleodata (60). Such methods attempt to remove the slower Earth system forcings such as changes in ice sheets and assume the residual changes are due to the fast response to CO₂. However, these estimates have not normally included non-CO₂ trace gases and are therefore potentially overestimating the CO₂ sensitivity.

Materials and Methods

Climate Simulations.

We use climate simulations from four GCMs: HadCM3-M21, IPSL-CM5A-LR, CCSM4, and GISS-E2-R. These GCMs are compared briefly in SI Appendix, Table S2. They span a reasonable range of changes for Pliocene and incorporate either PlioMIP phase 1 experiment 2 (34, 61) or PlioMIP phase 2 boundary conditions for the Pliocene (17). In all Pliocene simulations, the CO₂ mixing ratio is increased from 280 ppmv in the preindustrial to 400 (PlioMIP2) or 405 ppmv (PlioMIP1). The remaining Pliocene boundary conditions are based on geological evidence for orography, land ice, vegetation, and soils or are left unchanged at preindustrial settings. For land ice, the volume of the Greenland ice sheet is reduced, and the West Antarctic ice sheet is removed. Eustatic sea level is 25 m above modern. The global vegetation reconstruction prescribes a northward extension of forests causing a reduction in the area of tundra, while deserts are partly replaced by savannah and forest, particularly in the Sahel (10). In PlioMIP phase 1 simulations, orography increases over the Rockies and East Antarctica, but these changes are smaller in PlioMIP phase 2 for which both the Canadian Archipelago and Bering Strait are closed. The remaining model boundary conditions, including orbital configuration, solar constant, mixing ratios of CH₄, N₂O, and O₃, and aerosols are as used in the relevant preindustrial setup.

In PlioMIP1 phase 1, the boundary conditions are referred to as experiment 2 for coupled atmosphere–ocean simulations (34, 61). In PlioMIP phase 2 the nomenclature used is ${\mathrm{E}\mathrm{x}}^c$, where x can include o and/or i to signify whether or not the orography and land ice, respectively, are based on Pliocene geological reconstructions, and c is the prescribed Pliocene CO₂ mixing ratio (17). Since the default PlioMIP phase 2 CO₂ mixing ratio is 400 ppmv, this simulation is labeled ${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$. We also make use of simulations with HadCM3-M21 in which the CO₂ mixing ratio is set to 450 or 490 ppmv, and these are labeled ${\mathrm{E}\mathrm{o}\mathrm{i}}^{450}$ and ${\mathrm{E}\mathrm{o}\mathrm{i}}^{490}$, respectively. The global mean radiative forcing difference relative to the standard ${\mathrm{E}\mathrm{o}\mathrm{i}}^{400}$ simulation is therefore approximately 0.7 and 1.1 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$, for ${\mathrm{E}\mathrm{o}\mathrm{i}}^{450}$ and ${\mathrm{E}\mathrm{o}\mathrm{i}}^{490}$, respectively. These increases therefore encompass the upper and lower ends of our estimated additional non-CO₂ radiative forcing listed in Table 1.

Vegetation Model Simulations.

We simulate the land surface with the dynamic stand-based global vegetation model LPJ-GUESS v3 (31, 32). LPJ-GUESS was forced with repeating 30 y of monthly fields from the four GCMs listed above. A 500-y spin up to equilibrium was conducting using repeating the 30-y climate data prior to each 30-y simulation. We expand the land–sea mask using coastlines from PlioMIP phase 2 (17), including a partially deglaciated Greenland. The soil type over new land points is extrapolated from nearby existing points using the default soil type in LPJ-GUESS. CO₂ was set to 280 and 400 ppmv for preindustrial and Pliocene simulations, respectively.

Trace Gas Emissions.

Wetland area is calculated from the GCM soil moisture using a TOPMODEL (TOPography based hydrological MODEL) approach (62), in which the fractional area of inundation is calculated from the probability density function within each grid cell as derived from global high-resolution topographic data. CH₄ emissions are a function of microbial activity, i.e., temperature, available substrate, and wetland area (63). Total preindustrial emissions are scaled to 140 TgCH₄/y for each model, consistent with the observed methane isotope and concentration measurements (64).

Termite CH₄ emissions are calculated from the LPJ-GUESS simulated plant functional types coverage and observed emission per biome type (30, 65). Isoprene and monoterpene emissions and biomass burning are represented with process-based schemes coupled within LPJ-GUESS (66–68). We estimate soil emissions of nitrogen oxides (NOx) using a recently developed semiemipirical scheme (69).

CH₄ Chemical Lifetime and Radiative Forcing.

We calculate the uptake of CH₄ in soils with a process-based model (70) driven with climatologies from the GCMs. For the OH lifetime $\tau$, we integrate the Pliocene changes in emissions of all of the above species as well as atmospheric conditions and composition. As a full three-dimensional chemistry-transport simulation is extremely computationally expensive, it would limit the evaluation of uncertainty. Hence, we employ a parametric model (38) which is calibrated with three coupled climate–chemistry transport models. This model is described in SI Appendix.

We combine the CH₄ source estimates with the lifetime calculation in a global mean budget calculation $B=S\times \tau$, where B is the atmospheric CH₄ burden in Tg, S is the global mean CH₄ source, and $\tau$ is the lifetime in years (64). The budget equation is combined with the parametric lifetime model and the soil uptake, which is multiplied by the resultant concentration divided by preindustrial value. This is solved iteratively, to account for the self-feedback of CH₄ on lifetime and soil uptake.

The methane radiative forcing is a combination of direct and indirect components which are given in SI Appendix, Table S4. The total is 0.8 $\pm$ 0.09 ${\mathrm{W}\cdot \mathrm{m}}^{-2}$ per 1,000 ppbv increase in atmospheric CH₄.

To account for uncertainties, we prescribe a standard deviation of $\pm$20% for lightning NOx and stratospheric O₃, which are not simulated directly. We include uncertainties inherent in the CH₄ lifetime and radiative forcing calculations as described in SI Appendix, and an example is shown in SI Appendix, Table S5. The calculations are sampled with 10,000 evaluations of the iterated global CH₄ budget formulation.

Data Availability.

GCM simulation output for the preindustrial simulations are available from Earth System Grid Federation Coupled Model Intercomparison Project Phase 5 archive. Pliocene simulations can be obtained through the PlioMIP project. Pliocene geographic boundary conditions used in this study are available from Pliocene Research, Interpretation, and Synoptic Mapping: https://geology.er.usgs.gov/egpsc/prism/4_data.html. LPJ-GUESS model outputs and all model code developed in this study have been archived on figshare: https://doi.org/10.6084/m9.figshare.12302201 (piControl simulations), https://doi.org/10.6084/m9.figshare.12302216 (PlioMIP1 simulations), https://doi.org/10.6084/m9.figshare.12302228 (PlioMIP2 simulations), and https://doi.org/10.6084/m9.figshare.12344027 (emissions and lifetime code). The source code for LPJ-GUESS v4.0 can be obtained on request through Lund University (http://web.nateko.lu.se/lpj-guess). HadCM3-M21 ${\mathrm{E}\mathrm{o}\mathrm{i}}^{490}$ climate fields are archived on figshare: https://doi.org/10.6084/m9.figshare.12630356.