Geophysical Research Letters

Polar amplification in the mid-Holocene derived from dynamical vegetation change with a GCM

Authors


Abstract

[1] AOGCM simulations of the mid-Holocene tend to largely underestimate annual mean temperature over land in northern hemisphere compared to that of paleodata reconstruction. While the vegetation feedback has not been yet quantitatively reported, its neglect is suggested to be one of the cause of this underestimation. Here, we perform several experiments using an atmosphere-ocean-vegetation coupled model and quantify a vegetation-induced feedback in the mid-Holocene climate using MIROC GCM. Our result indicates an annual warming of +1.3K over land north of 40°N in the mid-Holocene, much larger than the previous GCM results. This warming is due to direct amplification of warming over high latitude land through increases in vegetation and reduced albedo during the summer and indirect amplification through sea-ice feedback in autumn and winter and snow albedo feedback in spring. These feedback were not properly represented in previous GCM analysis.

1. Introduction

[2] General Circulation Models (GCMs) project a CO2-induced global warming of 2–4K by the end of the 21st century [Meehl et al., 2007] with an even higher warming in the northern latitudes referred to as the ‘polar amplification’ [e.g., Boe et al., 2009]. As one of the processes, vegetation feedback is pointed out and quantified. Previous GCM studies revealed the quantitative importance of vegetation feedback under high CO2 [Notaro et al., 2007; O'ishi et al., 2009]. In order to validate model performance, reconstructed paleo climate from geological evidence is used [Braconnot et al., 2007]. Especially, the mid-Holocene (6000 years before present day) was warmer and more humid than present-day [Hoelzmann et al., 1998; Prentice and Webb, 1998]. The annual mean warming over land in the northern hemisphere ranged from +0.5K to +4K compared to pre-industrial levels on the regional scale [He et al., 2004; Jansen et al., 2007] and +2 ± 0.5K in the northern high latitude [Sundqvist et al., 2010a, 2010b]. The annual warming at the Greenland summit is also estimated to be +2 ± 0.5K [Masson-Delmotte et al., 2006].

[3] In PMIP2 (Paleoclimate Modeling Intercomparison Project 2) [Braconnot et al., 2007], various atmospheric-ocean coupled GCMs (AOGCMs) predicted an annual mean warming of about +0.5K [Braconnot et al., 2007] using mid-Holocene orbital forcing [Berger, 1978], without vegetation change. Though experiments indicate that polar amplification appears in the northern high-latitude region [Ganopolski et al., 1998; Claussen et al., 2006; Braconnot et al., 2007] as in future projections [Meehl et al., 2007], warming in model results is far smaller than that indicated by paleo-data (+2 ± 0.5K) [Mann et al., 2009; Sundqvist et al., 2010a, 2010b].

[4] By using an Earth system Models of Intermediate Complexity (EMICs), which include interactive vegetation, warming is predicted to be +1.2K over northern hemisphere land [Ganopolski et al., 1998] annually, and seasonally +4K (summer) and +1K (winter) at latitudes north of 60N [Crucifix et al., 2002]. Wohlfahrt et al. [2004] showed an annual warming of +1K over land, north of 40°N [Wohlfahrt et al., 2004] by using an AOGCM asynchronously coupled vegetation change with bias correction for temperature and precipitation. Even the AOVGCMs (atmospheric-ocean-vegetation coupled GCMs; without bias correction) are used, +2K over northern hemisphere land in summer [Gallimore et al., 2005] and annual averaged +0.5K warming over >40° N northern hemisphere [Otto et al., 2009] do not explain the paleo-data.

[5] In the present study, we try to improve the coupling procedure in our existing atmosphere-ocean-vegetation GCM MIROC-LPJ [O'ishi and Abe-Ouchi, 2009] which includes interactive vegetation. We intend to predict the correct present-day vegetation, so as to detect the correct strength of “vegetation feedback” in the mid-Holocene climate. We define “vegetation feedback” as the total change in climate-vegetation system which includes non-linear processes, induced by the introduction of interactive vegetation [Bony et al., 2006; Randall et al., 2007]. We especially focus albedo feedback because it is larger than other aspects (roughness, transpiration, interception. etc) due to difference of vegetation albedo, snow masking effect and interaction between ocean ice coverage in the high latitude. By using this modified MIROC-LPJ, we quantify the influence on the climate-vegetation system induced by vegetation change in the mid-Holocene.

2. Experimental Settings

2.1. Models

[6] In the present study, an AOVGCM MIROC-LPJ [O'ishi and Abe-Ouchi, 2009] was used in order to quantify the influence of vegetation change upon climate in mid-Holocene. The MIROC is an atmosphere-ocean general circulation model (AOGCM) which contributed to the IPCC AR4 [Hasumi and Emori, 2004]. The LPJ dynamic global vegetation model (DGVM) [Sitch et al., 2003] predicts vegetation change using process-based terrestrial carbon modules based on ten plant functional types (PFTs). Photosynthesis, respiration, carbon allocation, plant establishment, growth, turnover, mortality and competition among PFTs are simulated. In coupled MIROC-LPJ, vegetation distribution and atmosphere-ocean are coupled interactively; in LPJ-DGVM, vegetation distribution is predicted by using monthly mean surface air temperature, precipitation and cloud cover ratio from MIROC atmosphere. The vegetation distribution is translated into MIROC vegetation type along BIOME3 classification [O'ishi and Abe-Ouchi, 2009] and handed to MIROC land surface scheme once a year. Due to a technical limitation, MIROC-LPJ do not handle vegetation fraction but one vegetation type foe each gridcells.

[7] However, the temperature and precipitation biases of the atmosphere and ocean component cause the predicted vegetation distribution to stray from that seen in present-day. For example, a summer warming bias (compared to ECMWF ERA40 [Uppala et al., 2005]) of around +3K in Siberia shifts the tundra/forest boundary to the north and a perennial warming bias inland of Eurasian and North American continent (averaged around +8K) erroneously predicts temperate forest instead of boreal forest. Wet precipitation bias (compared to CPC Merged Analysis of Precipitation: CMAP [Xie and Arkin, 1997]) reduces desert area in Africa and increases savanna in southern Amazonia. In the present study, we adopted the following correction to remove the temperature and precipitation bias of MIROC before LPJ simulation (hereafter called the ‘anomaly procedure’):

equation image
equation image

where Tmodel and Pmodel are surface air temperature and precipitation predicted in an MIROC-LPJ experiment. Tobs and Pobs are present day observational surface air temperature and precipitation. TPD,model and PPD,model are surface air temperature and precipitation predicted in the present-day MIROC-LPJ experiment which adopts 345ppm atmospheric CO2 without this anomaly procedure. Tinput and Pinput are the corrected surface air temperature and precipitation and serve as input values for LPJ-DGVM in an MIROC-LPJ experiment. This procedure is designed so that the present-day MIROC-LPJ experiment predicts present-day potential vegetation distribution. In the present study, we adopted Tobs from ECMWF ERA40 and Pobs from CMAP. The present-day MIROC-LPJ experiment provided a reasonable potential vegetation distribution compared to observed potential vegetation [Ramankutty and Foley, 1999]. In order to detect the vegetation-induced influence, we also used MIROC with fixed vegetation. This is similar to “the online feedback suppression approach” of Bony et al. [2006]. In these fixed experiments, vegetation in each land grid cell is defined by the type which appears most during the last 50 years in AOV(PI), because in each grid cell, MIROC-LPJ does not handle vegetation fraction in LPJ directly.

[8] All simulations in the present study used a slab-ocean model which predicts sea surface temperature and sea ice extent by assumed seasonal change of ocean heat transport.

2.2. Settings

[9] As listed in Table 1, we performed two sets of experiments with the coupled MIROC-LPJ: a pre-industrial control simulation with present-day orbital elements and atmospheric CO2 concentration set to 285ppm (AOV(PI)) and a Holocene simulation with 6ka orbital elements and atmospheric CO2 concentration set to 285ppm (AOV(6ka)) following the 6ka experiment of PMIP2 [Braconnot et al., 2007]. The simulation is started with an initial condition of no vegetation and run for 390 years until the vegetation distribution reaches equilibrium. In AOV(6ka),the model was run for 110 years, continued from the last state of AOV(PI) until the global vegetation pattern shows an equilibrium state. We performed two additional MIROC simulations without LPJ-DGVM. In pre-industrial and Holocene simulations with ‘fixed vegetation’ (AO(PI) and AO(6ka)), the vegetation was fixed at the pre-industrial equilibrium state from AOV(PI). In the present study, the contribution of vegetation is defined by [AOV(6ka)-AOV(PI)]-[AO(6ka)-AO(PI)]. These AO simulations share all the same settings as those of dynamic vegetation (AOV) simulations, except for the inclusion of vegetation change.

Table 1. Summary of Experiments; T, Global Mean Temperature; ΔT, Change in Global Mean Temperature in 6ka; ΔTland, Change in Average Temperature Over Land North of 40°N in 6ka
Name of RunVegetation TreatmentOrbitCouplingIntegrationT(K)ΔT(K)ΔTland (K)
AOV(6ka)dynamic6kaanomaly110yrs287.94+0.36+1.32
AOV(PI)dynamic0kaanomaly390yrs287.58--
AO(6ka)fixed as AOV(PI)6ka-90yrs287.72+0.13+0.38
AO(PI)fixed as AOV(PI)0ka-75yrs287.59--

3. Results

3.1. Vegetation Distribution

[10] The predicted vegetation distribution of the pre-industrial experiment (Figure 1a) is similar to that of present-day potential vegetation [Ramankutty and Foley, 1999] except for the expansion of tundra in northern high latitudes. The predicted 6ka vegetation distribution (Figure 1b) indicates a significant northward expansion of boreal forest and reduction in tundra (total reduction of 5.83 × 106 km2 in tundra across Siberia and North America) compared to pre-industrial experiment. The results also show a slight increase in grassland north of the Caspian Sea and northeast of China. These vegetation changes are consistent with observed vegetation changes based on paleo-data (BIOME6000 [Prentice and Webb, 1998]).

Figure 1.

Equilibrium vegetation distribution obtained in (a) AOV(PI) and (b) AOV(6ka), (c) annually averaged 2m temperature change in 6ka compared to pre-industrial (AOV(6ka)-AOV(PI)) and (d) annually averaged contribution of vegetation change to 2m temperature ([AOV(6ka)-AOV(PI)]-[AO(6ka)-AO(PI)]). (e–h) Same as Figure 1d but averaged over June-July-August, September-October-November, December-January-February and March-April-May, respectively. Shaded area represents 95% confidence interval in the Student-t test.

3.2. Change in Annually Averaged Temperature

[11] The difference between the temperatures of the 6ka and pre-industrial AOV experiments are shown in Figure 1c. The globally averaged surface air temperature change is +0.36K. On the other hand, the contribution of vegetation change (Figure 1d) is a globally averaged warming of +0.23K to the total warming. The inclusion of dynamic vegetation indicates huge amplification of warming, especially in northern high latitudes. The amplification ratio of temperature is generally larger than +40% (more than +100% at most) in northern high latitudes. Total warming averaged over land north of 40°N is +1.32K and the contribution from vegetation feedback is +0.94K. Replacement of tundra by boreal forest causes a reduction in local annually averaged albedo from 0.51(PI) to 0.40(6ka).

3.3. Seasonality of Temperature Change

[12] In summer (June, July and August; Figure 1e), the amplification of warming over the northern hemisphere is caused by a significant reduction in land surface albedo up to −0.1 (Figure 2a) due to the expansion of boreal forest and decrease in tundra. Reduction of sea-ice and sea surface albedo due to vegetation change cause warming at the coastal region of Alaska, Greenland and Eastern Siberia. On the other hand, they show no significant impact on temperature at the central region of the Arctic Sea.

Figure 2.

(a–d) Difference of surface albedo due to vegetation change ([AOV(6ka)-AOV(PI)]-[AO(6ka)-AO(PI)]) averaged over June-July-August, September-October-November, December-January-February and March-April-May, respectively. (e–h) Same as Figures 2a–2d but sea ice cover ratio. Shaded area represents 95% confidence interval in the Student-t test.

[13] In autumn (September, October and November; Figure 1f) and winter (December, January and February; Figure 1g), the amplification of warming is far greater over the Arctic Ocean than over the Eurasian and the North-American continents. This is caused by the difference in sea-ice extent (Figures 2f and 2g) and the resultant albedo reduction (Figures 2b and 2c) around high latitude coastal areas, as noted in previous studies [Ganopolski et al., 1998; Wohlfahrt et al., 2004; Claussen et al., 2006].

[14] In winter, compared to the PMIP2 studies [Braconnot et al., 2007], our experiments show a warming in the northern hemisphere of the mid-Holocene which is larger than that of any other PMIP models which are performed without dynamical vegetation. Without vegetation feedback, the NH winter temperature is lower than present day, because of low insolation.

[15] In spring (March, April and May; Figure 1h), amplification of warming is significant over northern high latitudes due to a strong snow melt albedo feedback (Figure 2d). The reduction of albedo more then −0.2 is caused by snow-albedo feedback induced by vegetation change [Bonan et al., 1992; Foley et al., 1994].

[16] Overall, our results indicate that vegetation change directly amplifies warming over high latitude land through land surface albedo change due to forest expansion in summer and vegetation-snow-albedo feedback in spring. On the other hand, in autumn and winter, warming is amplified due to sea-ice albedo feedback induced by vegetation change over land [Berger, 2001]. This result also indicates that the summer radiation change in 6ka not only directly causes warming in summer but also indirectly causes warming in other seasons by promoting vegetation growth in summer, accelerating snow melt in spring and reducing sea-ice in autumn and winter through known feedback processes noted in previous studies [Ganopolski et al., 1998; Wohlfahrt et al., 2004; Claussen et al., 2006].

4. Discussion

[17] In the present study the boreal forest in 6ka expands northward significantly, compared to the control experiment (Figures 1a and 1b). This qualitative change in vegetation is the same as in previous GCM studies [Wohlfahrt et al., 2004; Gallimore et al., 2005; Otto et al., 2009]. However, our results show the largest 6ka warming. The amplification of warming due to this vegetation change is also qualitatively similar to previous coupled AOGCM studies [Gallimore et al., 2005; Otto et al., 2009], although, the quantity of amplification (+180% global amplification) is much larger in our studies.

[18] The warming amplification due to dynamical vegetation can explain the reconstructed ground surface temperature (GST) data by Huang et al. [2000], Sundqvist et al. [2010a, 2010b] and Zhang et al. [2010]. They estimated the annual averaged GST warming over extratropics in the mid-Holocene relative to the pre-industrial value to be +2.0 ± 0.5K from borehole measurement. From estimates of GST by Mann et al. [2009], our results indicate that the mid-Holocene GST warming in the same region is amplified from +0.38K to +1.25K due to the inclusion of dynamic vegetation. The inclusion of dynamic vegetation in our experiments substantially improves in the underestimation of GST warming in the mid-Holocene simulation as suggested by Mann et al. [2009]. The warming at the Greenland summit in the present study is about +2K which explains the observed warming (+2.0 ± 0.5K) [Masson-Delmotte et al., 2006] very well, due to the inclusion of both vegetation dynamics and bias correction.

[19] Compared to our study, Gallimore et al. [2005] predicted a smaller northward expansion of boreal forest in the 6ka experiment. On the other hand, their results show an expansion of grassland at mid latitudes, which is very small in our results. The impact of vegetation feedback in spring in their study is the same as this study (warming). As by Otto et al. [2009] the warming in the northern hemisphere is generally lower than ours. Perhaps their control experiment is a warmer one, so that the response of vegetation is smaller. Comparisons with the results of Wohlfahrt et al. [2004] suggest other discrepancies among the models. Their control vegetation is similar to our present-day vegetation (not shown) because both studies adopt bias correction in the prediction of vegetation. On the other hand, their results show a far larger expansion of grassland at mid latitude in 6ka. This difference between the 6ka vegetation in the models suggests that the response of climate to 6ka orbit differs in each model, so that the resultant vegetation feedback also differs. In their experimental setting, the precipitation bias is corrected by subtraction which may not correct the vegetation distribution in regions of low precipitation efficiently. The difference between distribution of grassland in their result and ours may be caused by the difference in bias correction method. Wohlfahrt et al. [2008] have shown that most GCMs overestimate grassland in the mid-latitudes of Asia in the mid-Holocene like our non bias correction experiments, and that they underestimate warming. This suggests that GCMs (including MIROC) tend to overestimate grassland in mid-Holocene, and that this tendency should be reduced to simulate the proper vegetation-climate feedback in the mid-latitudes. Another possible explanation for the discrepancy is the schematic bias of individual DGVMs. Our results show a larger northward shift of the treeline in Eastern Canada compared to results of Wohlfahrt et al. [2008]. This may be caused by overestimation of NPP in LPJ-DGVM. Differences in limiting factors (except for climate and CO2) assumed in individual DGVMs also lead to different response in vegetation.

[20] EMIC studies predict a far larger 6ka warming over land in the northern hemisphere [Ganopolski et al., 1998; Crucifix et al., 2002] compared to GCM studies (including the present study). Generally, EMICs assume a lower resolution grid size in land and a formulation for heat and energy transportation in the atmosphere which is simpler than that in GCMs. Such coupled systems may show large scale vegetation changes and less well represented energy transport, thus stronger vegetation feedback.

[21] Our additional set of experiments, without the anomaly procedure, indicated that the average warming over land north of 40°N is +0.15K as a result of vegetation change which has a cooling effect. This behavior is caused mostly by the summer atmospheric warming bias of MIROC which affects the southern and northern limits of the boreal forest. This result suggests that all PMIP models may improve on the warming in the mid-Holocene by the use of DGVM. However, it may also depend on the accuracy of the vegetation distribution in the control experiment.

[22] Moreover, uncertainties and discrepancies among models may increase further in paleo-climate modeling and future climate projection due to the different vegetation distribution in the control experiment when all GCMs are coupled with dynamic vegetation. Careful treatment of the simulation and the projection of vegetation could be crucial for the estimation of the polar amplification both in the past and the future.

Acknowledgments

[23] CMAP Precipitation data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.cdc.noaa.gov/. ECMWF ERA-40 data from their data server at http://data.ecmwf.int/products/data/archive/.

[24] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.

Ancillary