Using the experiments undertaken by 36 climate models participating in the Paleoclimate Modeling Intercomparison Project (PMIP), this study examines annual and seasonal surface temperatures over China during the mid-Holocene. Compared to the present or preindustrial climate, 35 out of the 36 PMIP models reproduced colder-than-baseline annual temperature, with an average cooling of 0.4 K, during that period. Seasonal temperature change followed closely the change in incoming solar radiation at the top of the atmosphere over China during the mid-Holocene. Temperature was reduced (elevated) in boreal winter and spring (summer) in all of the PMIP models, with an average of 1.4 K (1.0 K) at the national scale. Colder (warmer)-than-baseline temperatures were derived from 14 of the 16 atmosphere-only (18 of the 20 coupled) models during the mid-Holocene boreal autumn. Interactive ocean was found to lead to a warming effect on annual (0.3 K), boreal winter (0.5 K), and boreal autumn (0.7 K) temperatures, with reference to the atmosphere-only models. Interactive vegetation had little impact in terms of six pairs of coupled models with and without vegetation effects. The above results are in stark contrast to warmer-than-present annual and winter climate conditions as derived from multiproxy data for the mid-Holocene. Coupled models generally perform better than atmosphere-only models.
The mid-Holocene was around 6000 yr before present, when the climate and environment differed significantly from the present day. Much effort has been devoted to investigating the response of climate models to the different seasonal distribution of incoming insolation due to changes in the earth’s orbital parameters during the mid-Holocene. In particular, model–model and model–data comparisons have been widely performed to identify and understand the consistencies and inconsistencies between each other.
A few simulations of the mid-Holocene climate over China have been conducted using atmospheric general circulation models (AGCMs) (Wang 1999, 2000, 2002; Chen et al. 2002), regional climate models nested within AGCMs (Zheng et al. 2004; Liu et al. 2010a), an asynchronously coupled atmosphere–ocean general circulation model (Wei and Wang 2004), and a synchronously coupled atmosphere–ocean general circulation model (AOGCM) (Zheng and Yu 2009). Surface warming and the intensification of monsoon circulation over China during the mid-Holocene summer (June–August) have been well documented in those earlier studies. However, both the spatial pattern and the magnitude of summer climate change are different among those simulations. Such model-dependent results imply a large degree of uncertainty in the summer climate over China during the mid-Holocene, particularly at the subregional scale, which stresses the need to investigate the mid-Holocene summer climate from the perspective of multiple climate models. Second, little or no attention has hitherto been paid to climate change for the annual mean and the other seasons during that period. Based on a variety of proxy data, Chinese climate has been found to undergo dramatic changes during the mid-Holocene. Among these includes an annual temperature increase of 1–5 K and a warmer winter (December–February), with respect to the present day (see section 4 of this study). Compared to the proxy data, the extent to which annual and seasonal temperatures over China respond to the mid-Holocene forcings in the climate models remains an open question. Previously, Chen et al. (2002) used an AGCM to reproduce a colder-than-present winter temperature of around 2.0 K over China for the mid-Holocene, which was opposite in sign to the proxy data. Investigating how winter temperature changed during that period by the use of multiple climate models is particularly interesting. Third, ocean and vegetation dynamics are typically neglected in most of the earlier simulations for the East Asian climate during the mid-Holocene. It is therefore important to examine what the Chinese climate was like during that period in the simulations of state-of-the-art AOGCMs and fully coupled atmosphere–ocean–vegetation general circulation models (AOVGCMs).
Within the framework of the Paleoclimate Modeling Intercomparison Project (PMIP), a hierarchy of climate models has been used to simulate the mid-Holocene climate. Their results have been compared to proxy data in the tropics (Braconnot et al. 2007a), Africa (Peyron et al. 2006), Europe (Guiot et al. 1999; Masson et al. 1999; Bonfils et al. 2004; Hoar et al. 2004), the high northern latitudes (Zhang et al. 2010), and Greenland and Antarctica (Masson-Delmotte et al. 2006). It is found that climate models are able to reproduce many of the robust qualitative large-scale features of reconstructed climate change, consistent with our understanding of orbital forcing (Jansen et al. 2007). Recently, the mid-Holocene warmer summer climate in East Asia was obtained from 12 AOGCMs within the second stage of the PMIP (PMIP2), which was found to generally agree with annually resolved records at 19 sites over China (Wang et al. 2010). However, such a comparison between summer temperature in the models and the annual mean for the proxy data is not matched. A fair model–data comparison should be annual versus annual and summer versus summer. The spatial coverage of the proxy data is also extremely sparse across the country. Additionally, Wang et al. only used part of the PMIP2 AOGCMs, which makes it impossible to evaluate the role of the ocean and vegetation through the comparison of the different PMIP sets of simulations. Taken together, of special interest now are the questions regarding the degree to which the mid-Holocene annual and seasonal temperatures as derived from all of the PMIP models are consistent with proxy data over China, and the degree to which interactive vegetation and ocean impact the mid-Holocene climate over this region. It is also equally important to quantify the spread of the temperature change over China from the PMIP simulations. Insights gained from this regional-scale case study will contribute to a global perspective concerning model–data comparisons during the mid-Holocene, as such a comparison between multiple climate models and proxy data is still absent in the scientific literature.
a. Models and data
This study is based on all of the available simulations for the mid-Holocene in the PMIP, including experiments using 16 AGCMs from the first stage of the PMIP (PMIP1) and 14 AOGCMs and 6 AOVGCMs from the PMIP2. For the mid-Holocene experiment, the most important change in boundary conditions is the change in the earth’s orbital parameters, which intensifies (weakens) the seasonal distribution of the incoming solar radiation at the top of the atmosphere in the Northern (Southern) Hemisphere, by about 5% (Berger 1978). In addition, atmospheric CO2 concentration varies from 345 ppm for the present period to 280 ppm for the mid-Holocene in the PMIP1. In the PMIP2, atmospheric CO2 concentration is held constant at 280 ppm, while atmospheric CH4 concentration is set to 760 ppb for the preindustrial period and 650 ppb for the mid-Holocene. Sea surface temperatures (SSTs) are fixed at the present-day values in all the experiments of the 16 AGCMs in the PMIP1, while they are computed by oceanic general circulation models in all the experiments of the 20 AOGCMs and AOVGCMs in the PMIP2. All other conditions are kept constant throughout the experiments in the PMIP1 and PMIP2 simulations. Basic information about these 36 climate models has been provided in Table 1. More details relating to the models and experimental designs were given by Joussaume and Taylor (1995) and Braconnot et al. (2007a), and are available online (http://pmip.lsce.ipsl.fr/).
Data used to assess the ability of the models to reproduce the modern temperature climatology over China are taken from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis of surface temperature for the period 1979–2000 (Kalnay et al. 1996). Because there are differences in horizontal resolution among the models, all model and reanalysis data are aggregated to a relatively midrange resolution of 96 × 48 (longitude × latitude) using a linear interpolation or extrapolation approach. The ensemble mean of multiple climate models is obtained using the same weights across the models of concern.
b. Evaluation of the models
To measure the ability of the models to simulate the baseline temperature climatology over the target region, spatial correlation coefficients (SCCs) and root-mean-square errors (RMSEs) excluding systematic model error of annual and seasonal temperatures between each baseline simulation and the NCEP–NCAR reanalysis data are calculated on the basis of 77 grid points over China, respectively. As given in Table 2, SCCs of annual temperature vary from 0.75 [the Yonsei University Model (YONU] to 0.99 [the Met Office model (UKMO), the Met Office model run at Bristol University–third climate configuration of the Met Office Unified Model (UBRIS-HadCM3M2), and the preindustrialized vegetation version of the UBRIS-HadCM3M2 (UBRIS-HadCM3M2-veg)], and RMSEs of annual temperature vary from 1.2 K (UBRIS-HadCM3M2) to 5.4 K [version 1 of the Center for Climate System Research Model (CCSR1) and YONU]. SCCs (RMSEs) of spring (March–May), summer, autumn (September–November), and winter temperature vary from 0.58 to 0.98 (1.5 to 7.6 K), from 0.49 to 0.98 (1.5 to 6.1 K), from 0.78 to 0.99 (1.4 to 5.3 K), and from 0.85 to 0.99 (2.4 to 5.8 K) across the models. That means the ability of the models to simulate the baseline annual and seasonal temperatures over China varies from model to model. Of importance is that all of the SCCs are positive and statistically significant at the 99% confidence level.
Where the ensemble mean of the 36 models is concerned, Fig. 1 shows clearly that the observational meridional annual temperature gradient in eastern China and a large extent of the low annual temperature over the Qinghai–Tibetan Plateau are well reproduced by the models. Quantitatively, the SCC (RMSE) of annual temperature is 0.96 (2.3 K), and the SCCs (RMSEs) of seasonal temperature vary from 0.95 to 0.97 (2.1 to 2.7 K). Taken together, all of the models can successfully reproduce the geographical distribution of the baseline annual and seasonal temperatures over China. The 36-model ensemble mean has a higher ability with reference to most, but not all, of the individual models in terms of the values of SCCs and RMSEs. The ensemble mean result of the 36 climate models is therefore emphasized in the following analysis.
3. Mid-Holocene temperature over China from the models
a. Annual temperature change
Thirty-five out of the 36 PMIP models reproduced colder-than-baseline annual temperatures over China during the mid-Holocene (Fig. 2a). The magnitude of the regionally averaged annual temperature change at the national scale was different with the models, ranging from −1.0 K [the 11-layer GCM from the University of Illinois at Urbana–Champaign (UIUC11)] to 0.2 K [L’Institut Pierre-Simon Laplace Coupled Model, version 4 (IPSL-CM4-V1-MR)], with a standard deviation of 0.3 K across the models. Averaged across the whole country and all the models, the annual temperature was reduced by 0.4 K during the mid-Holocene with reference to the baseline climate. On the other hand, the geographical distribution of the mid-Holocene–baseline anomalies in annual temperature varied with the models, particularly at the subregional scale. Where the 36-model ensemble mean is concerned, the annual temperature was slightly elevated by less than 0.1 K over the northern part of northeastern China but reduced by 0–0.8 K over the rest of China during the mid-Holocene (Fig. 3a). Annual temperature cooling was generally greater in the south than in the north over China.
The most important difference between the PMIP1 and PMIP2 models is that ocean dynamics is taken into account in the latter, allowing for an evaluation of the effect of oceans. When viewed in terms of these two model classes, the annual temperature was reduced both in the AGCMs and coupled models (excluding IPSL-CM4-V1-MR), but with a larger magnitude in the former as a whole (Fig. 2a). The regionally averaged annual temperature over China was reduced by 0.6 K from the 16 PMIP1 models, which was greater than the 0.3 K from the 20 PMIP2 models. Meanwhile, the geographical distribution of the changes in the mid-Holocene annual temperature was also different between the model classes (Figs. 3b,c). Over the northern part of northeastern China and northern Xinjiang, the annual temperature was reduced in the PMIP1 models but raised in the PMIP2 models. Over the rest of China, the annual temperature cooling was stronger overall in the PMIP1 models than in the PMIP2 models. According to available SST data from 17 PMIP2 models, Fig. 4a shows that the mid-Holocene annual SSTs differed from the preindustrial period values in the Northern Hemisphere. On one hand, the annual SSTs increased in response to the mid-Holocene orbital forcing in most parts of the northern mid- and high latitudes, with an increase in 16 of the 17 PMIP2 models [excluding the Meteorological Research Institute Coupled General Circulation Model, version 2.3.2a (MRI-CGCM2.3.4nfa)] and an average warming of 0.2 K across all the models within the region of 45°–90°N, 0°–180°. Accordingly, surface temperature increased due to ocean feedback over the high latitudes of Eurasia (Braconnot et al. 2007a). In turn, this favored a surface temperature increase over China, as the warmer air from those areas influenced the regional climate of China, particularly in cold months when northwesterly winds prevailed over the country (e.g., Lau and Li 1984; Gong and Ho 2002). On the other hand, annual SSTs decreased overall in the oceans adjacent to the East Asian continent. In the western North Pacific (0°–40°N, 105°E–180°), for example, annual SSTs were generally colder than the baseline values in 13 of the 17 PMIP2 models [excluding Commonwealth Scientific and Industrial Research Organisation Mark version 3.0 (CSIRO-Mk3L-1.0), CSIRO-Mk3L-1.1, the Fast Ocean Atmosphere Model (FOAM), and MRI-CGCM2.3.4nfa-veg], with an average cooling of 0.2 K for the ensemble mean of the 17 models, during the mid-Holocene. This favored a surface temperature decrease over China, because the colder SSTs in that area could lead to larger losses of surface heat in the East Asian region during boreal warm months and smaller gains in surface heat during boreal cold months. In this connection, the response of SSTs to the mid-Holocene forcings needs to be specifically investigated to understand the role of the interactive ocean on the East Asian climate during that period.
Vegetation feedback has been regarded as an important process in the mid-Holocene climate system (Jansen et al. 2007). More specifically, reconstructed paleovegetation was found to be able to cause an overall warming of the annual temperature over China during the mid-Holocene in the earlier experiments using an AGCM (Chen et al. 2002) and two regional climate models (Zheng et al. 2004; Liu et al. 2010a). Also found in those studies were the differences, both in sign and magnitude, of vegetation-induced annual and seasonal temperature changes on a large scale. Such model-dependent results imply a level of uncertainty. Within the PMIP2, 12 simulations were performed by six pairs of climate models, namely, the Climate de Bilt–coupled large-scale ice–ocean model and the Vegetation Continuous Description Model (ECBILTCLIOVECODE), ECHAM53–Max Planck Institute Ocean Model (MPIOM127)–FOAM–Lund Postdam Jena (LPJ), FOAM, MRI-CGCM2.3.4fa, MRI-CGCM2.3.4nfa, UBRIS-HadCM3M2, and their AOVGCM counterparts (Table 1). Importantly, the same AOGCM version was used for both AOGCM and AOVGCM simulations for each pair of the models (T. Fichefet, R. Gladstone, S. Harrison, A. Kitoh, U. Mikolajewicz, and I. Ross 2011, personal communication). These two kinds of parallel simulations provide an opportunity to estimate the role of vegetation within a fully coupled climate model context during the mid-Holocene, as a dynamic global vegetation model component is the main difference between each example.
First, it can be seen in Fig. 5 that interactive vegetation gave rise to additional annual temperature changes during the mid-Holocene. Compared to the values of −0.3 K from ECBILTCLIOVECODE, −0.59 K from ECHAM53–MPIOM127-LPJ, −0.4 K from MRI-CGCM2.3.4fa, −0.6 K from MRI-CGCM2.3.4nfa, and −0.2 K from UBRIS-HadCM3M2 (−0.1 K from FOAM), the corresponding regionally averaged annual temperature changes of −0.02, −0.57, −0.3, −0.04, and −0.1 K (−0.5 K) as derived from their AOVGCM counterparts were weaker (stronger) over China during the mid-Holocene (see Fig. 2a). In this sense, the vegetation effect reduced (amplified) the annual temperature cooling over China in five (one) of the six pairs of the models during that period. Second, Fig. 5 also shows a large spatial variability of vegetation-induced annual temperature changes between each pair of the models. In most parts of China, for example, interactive vegetation led to a strong cooling effect from the FOAM pair, which was opposite to a strong warming effect from the MRI-CGCM2.3.4nfa pair during the mid-Holocene. It is a pity that such results cannot be further investigated in a cause-and-effect manner, as the outputs of vegetation, albedo, leaf area index, and so on, are not yet available in the PMIP database. Third, when averaged over the whole country, the annual temperature change due to the vegetation effect was 0.3 K from the ECBILTCLIOVECODE pair and 0.6 K from the MRI-CGCM2.3.4nfa pair, but −0.4 K from the FOAM pair, respectively. It was around 0–0.1 K from each of the other three pairs of the models. The difference in the mid-Holocene–baseline anomalies in annual temperature between the ensemble mean of the six AOVGCMs and that of the six AOGCMs was less than 0.3 K over China (Fig. 5), with a regionally averaged value of 0.1 K. Taken together, interactive vegetation only gave rise to a weak annual temperature warming, and it did not appear to be an important component in contributing to the annual temperature change over China during the mid-Holocene.
Note that the modern vegetation was used in the preindustrial experiments of ECBILTCLIOVECODE, FOAM, MRI-CGCM2.3.4fa, and MRI-CGCM2.3.4nfa (T. Fichefet, S. Harrison, A. Kitoh, and I. Ross 2011, personal communication). For each pair of these models, part of the aforementioned vegetation-induced temperature changes during the mid-Holocene is due to the differences between the AOGCM and AOVGCM preindustrial experiments. To investigate vegetation feedback more transparently, the preindustrial vegetation simulated in the AOVGCM preindustrial experiment should be fixed as boundary conditions in the AOGCM experiments so that the AOGCM and AOVGCM mid-Holocene experiments share the similar control experiments (Braconnot et al. 2007b). Averaged over China, the annual temperature difference was 0.04 K between ECHAM53–MPIOM127–LPJ-veg and ECHAM53–MPIOM127–LPJ and –0.001 K between UBRIS-HadCM3M2-veg and the UBRIS-HadCM3M2 preindustrial experiments for which the preindustrial vegetation was used in the AOGCM control experiments (R. Gladstone and U. Mikolajewicz 2011, personal communication). Comparatively, as discussed before, the mid-Holocene annual temperature change was 0.02 K from the ECHAM53–MPIOM127–LPJ pair and 0.1 K from the UBRIS-HadCM3M2 pair. Pure vegetation feedback still has a slight influence on the mid-Holocene annual temperature over the country, which is consistent with the conclusions drawn from all six pairs of the models. Moreover, two recent sets of strict numerical experiments also indicated that the effect of the vegetation feedback on the annual climate over China was quite limited during the mid-Holocene (Dallmeyer et al. 2010; O’ishi and Abe-Ouchi 2011).
b. Seasonal temperature change
According to the work of Berger (1978), the regionally averaged incoming solar radiation at the top of the atmosphere over China was reduced by 11.0 W m−2 in winter, 8.2 W m−2 in spring, and 3.6 W m−2 in autumn, whereas it increased by 21.3 W m−2 in summer during the mid-Holocene with respect to the present day. The PMIP experiments indicated that the seasonal temperature change generally followed closely the above insolation change over China during that period. In winter and spring, all 36 models reproduced colder-than-baseline mid-Holocene climates over China (Figs. 2b,c). Winter (spring) temperature cooling varied from 0.6 K in ECBILTCLIOVECODE-veg to 3.1 K in the LMCE model from the Laboratoire des Sciences du Climat et de l´Environnement (LMD, LMCELMD4) [0.5 K in Community Climate Model 3 (CCM3) to 2.3 K in Canadian Centre for Climate Modelling and Analysis (CCCma) Coupled General Circulation Model, version 2.0 (CCC2.0)], with an average of −1.4 K and a standard deviation of 0.5 K (0.4 K) across the 36 models. For the ensemble mean of the 36 models, the mid-Holocene winter (spring) temperature was 0.6–1.9 K (0.9–1.9 K) colder than the baseline climate, with a weaker magnitude over northern China (Figs. 6a,b).
In summer, temperature increased in all of the 36 models during the mid-Holocene (Fig. 2d), ranging from 0.6 K (ECHAM5–MPIOM1, ECHAM53–MPIOM127–LPJ, and ECHAM53–MPIOM127–LPJ-veg) to 1.6 K (IPSL-CM4-V1-MR). Averaged over the country, a warming of 1.0 K, with a standard deviation of 0.2 K, was obtained from the 36 models. Based on the 36-model ensemble mean, the mid-Holocene summer temperature increased by 0–2.3 K over China, except for over the southernmost Tibetan Plateau where weak cooling was presented (Fig. 6c). In addition, warming was more pronounced at high latitudes, which was consistent with the insolation change (Berger 1978).
There was a large discrepancy in the simulated autumn temperature change between the atmospheric and coupled models during the mid-Holocene (Fig. 2e). Fourteen of the 16 PMIP1 AGCMs (excluding LMCELMD4 and LMCELMD5) reproduced colder-than-baseline climates, whereas 18 of the 20 PMIP2 AOGCMs and AOVGCMs [excluding Community Climate System Model, version 3 (CCSM3.0) and MRI-CGCM2.3.4nfa] reproduced warmer-than-baseline climates during the mid-Holocene. The autumn temperature was reduced (increased) by an average of 0.3 K (0.4 K) in terms of the ensemble mean of the 16 PMIP1 (20 PMIP2) models. Averaged across the 36 models, it was increased by 0.1 K, with a standard deviation of 0.5 K. As further seen in Fig. 6d, the mid-Holocene autumn temperature changes were no more than 0.3 K over China, except for over most parts of the Tibetan Plateau where a warming of 0.3–0.7 K occurred.
Compared to the results of the 16 PMIP1 AGCMs, the mid-Holocene seasonal temperature changes as derived from the 20 PMIP2 AOGCMs and AOVGCMs were 0.5 K in winter, 0.03 K in spring, −0.1 K in summer, and 0.7 K in autumn (Fig. 2). In this regard, the interactive ocean led to a warming effect on the winter and autumn temperatures over China during the mid-Holocene, which explained why the above-mentioned annual mean cooling of 0.3 K in the PMIP2 models was weaker than the 0.6-K result found in the PMIP1 models. Similar to the annual mean (Fig. 4a), the mid-Holocene winter SSTs slightly increased overall in the northern high latitudes but decreased in the oceans adjacent to mainland China, with reference to the baseline values, in terms of the 17-model ensemble mean (Fig. 4b). The former favored a temperature increase, but the latter favored a temperature decrease in winter over China. During the mid-Holocene autumn, all 17 PMIP2 models reproduced warmer-than-baseline SSTs in the Northern Hemisphere, with an average warming of 0.4 K within the region of 0°–90°N, 0°–180° (Fig. 4c). Such SST changes were undoubtedly in favor of autumn warming over China, particularly for those that occurred in the oceans north of Eurasia and adjacent to the East Asian continent.
The vegetation effect also led to additional seasonal temperature changes over China during the mid-Holocene. Averaged across the six pairs of the PMIP2 models and the country, it generally gave rise to a weak warming effect on the seasonal temperature, with the values of 0.1 K in winter, 0.1 K in spring, 0.002 K in summer, and 0.2 K in autumn at the national scale. Seasonal temperature changes due to interactive vegetation were not spatially uniform (Fig. 7). For example, a strong warming (cooling) effect occurred over most parts of the Qinghai–Tibetan Plateau in spring (summer). In addition, there were large differences, both in sign and magnitude, in the simulated vegetation-induced seasonal temperature changes between the pairs of the models. Similar to the annual mean, a strong warming effect was obtained from the ECBILTCLIOVECODE and MRI-CGCM2.3.4nfa pairs, whereas a strong cooling effect was obtained from the FOAM pair. The effect of vegetation was quite small in all of the other three pairs of the models. In summary, interactive vegetation contributed little to seasonal temperature over China during the mid-Holocene. Such results also can be seen in the two recent sets of experiments conducted to examine the pure vegetation feedback during the mid-Holocene (Dallmeyer et al. 2010; O’ishi and Abe-Ouchi 2011).
4. Model–data comparison
A great deal of effort has been made to reconstruct the mid-Holocene climate over China through the use of a variety of proxy data. In this study, two preconditions were set to choose “reliable” reconstructions of the mid-Holocene temperature. First, the reconstructions had to be published in peer-reviewed journals, which means that the data quality has been verified by the experts in that field, and second, the reconstructions had to draw a clear conclusion about the mid-Holocene temperature over China. In this manner, the records of pollen, lake cores, paleosol, ice cores, peat, sediment, stalagmites, and fossil fruits at 64 sites were finally chosen for the model–data comparison (Table 3). Paleoenvironmental and paleoclimatic evidence from those records indicated that stably warmer climate conditions prevailed over China during the mid-Holocene (Fig. 3d). The annual temperature was elevated by about 1 K in southern China and by about 2 K in the Yangtze River valley in light of the peat profiles and pollen records at Dahu Swamp, the stalagmite records at Dongge Cave, Xiangshui Cave, and Xianrendong Cave, and the pollen assemblage at the Hanjiang Delta, Heqing Basin, and Qingfeng Section. A number of lake cores and pollen records in Inner Mongolia, northern Xinjiang, and Qinghai Lake, together with a few paleosol and peat records, suggested that the annual temperature was about 3 K higher in most parts of northern China. The largest warming of 4–5 K was recorded on the Qinghai–Tibetan Plateau by the lake cores and ice cores. These results are in general consistent with the proxy data before the 1990s, as summarized by Shi et al. (1993), in which the deviation of the annual temperature from the present-day values was roughly estimated to be 1 K in southern China, 2 K in the Yangtze River valley, 3 K in northern and northeastern China, and 4–5 K on the Qinghai–Tibetan Plateau during the mid-Holocene (contours shown in Fig. 3d).
Based on a dataset of 158 pollen samples across China, Yu et al. (1998, 2000) reconstructed the mid-Holocene vegetation of the country. They concluded that the northward shifts of the tropical, broadleaved evergreen/warm mixed, and cool mixed forest zones in eastern China must imply warmer winters than the present because the poleward boundaries of the affected biomes in China today are associated with winter-temperature isotherms that in turn reflect the typical tolerance limits of tropical, subtropical (broadleaved evergreen), and temperate broadleaved deciduous woody plants. The northern boundary of the temperate deciduous forest, which showed the greatest northward shift of all, is also controlled by winter temperatures, occurring where the winter temperatures become cold enough to satisfy the chilling requirements of boreal needle-leaved evergreen trees. Such changes in vegetation and hence generally warmer winters during the mid-Holocene were confirmed by Ni et al. (2010), in which they used 188 pollen samples to reconstruct the paleovegetation over China through a new global classification system of plant functional types and a standard numerical technique for biome assignment. In addition, a few studies suggested that winter warming was stronger than the annual mean during the mid-Holocene. The existence of Ceratopteris at Baiyangdian and Tancheng during the mid-Holocene, now living in subtropical lakes and swamps, implied that winter temperature was about 6 K warmer than it is today (Xu et al. 1988; Shi et al. 1993). The discovery of nuts from Helicia plants dating back to the mid-Holocene at Baohuashan indicated a 6.3-K warming in January (Kong et al. 1991). Stable forest in the Qinghai Lake area during the mid-Holocene implied a January warming of 8 K (Kong et al. 1990). Pollen records from Bayanchagan, Hidden, Hulun, and Ren Co Lakes also indicated a stronger warming in winter (Tang et al. 2000; Guiot et al. 2008; Jiang et al. 2010; Wen et al. 2010). Using the proxy data before the 1990s, winter warming was estimated to be 70%–100% stronger than that of the annual mean, particularly in eastern China (Shi et al. 1993).
Referring to all of the above, a considerable mismatch existed between the proxy data and the PMIP simulations over China during the mid-Holocene. Contrary to the annual temperature warming as reconstructed by the multiproxy data, 35 of the 36 PMIP models reproduced colder-than-baseline climates, with an average cooling of 0.4 K for all models, over China during the mid-Holocene, particularly in western and northern China where the annual temperature was estimated to be at least 3 K warmer than the present day. Moreover, the winter warming suggested by the proxy data was in stark contrast to the stronger winter cooling, with an average of −1.4 K, as derived from all the models during that period. The annual and winter temperature cooling as derived from the PMIP2 coupled models were weaker overall than those from the PMIP1 AGCMs, and hence they were closer to the proxy data. In this sense, the interactive ocean improved, to a certain extent, the simulation of the mid-Holocene temperature over China. The results of the six AOVGCMs differed little from those of their AOGCM counterparts as a whole. That is, the significant effect of interactive vegetation on annual and seasonal temperatures over China during the mid-Holocene cannot be detected in the present analysis.
Within the protocol of the PMIP experiments, the annual mean incoming solar radiation at the top of the atmosphere was reduced on average by 0.36 W m−2 over China during the mid-Holocene (Berger 1978). Additional negative radiative forcing was derived from the changes in atmospheric concentrations of CO2 in the PMIP1 and CH4 in the PMIP2 (see section 2). Accordingly, it is no wonder that the PMIP models reproduced colder-than-baseline annual mean climate conditions over the region during that period. Although there are model-dependent uncertainties in the PMIP simulations, of importance is that the intermodel variability of annual and seasonal (excluding autumn) temperature changes, represented by the standard deviation of the model results about their mean, was uniformly smaller than the corresponding ensemble mean result during the mid-Holocene. This implies a robust qualitative consistency on the annual and winter cooling between the models at the national scale. Besides the uncertainties in the simulations, another source that may be partly responsible for the model–data mismatch is the uncertainty in the proxy data. This is especially true for Bayanchagan Lake, where the reconstructed normal climate (Jiang et al. 2006) differed from the warming condition obtained both from the same pollen sample but with different methods (Guiot et al. 2008; Jiang et al. 2010), implying that the method used for climatic reconstruction needs to be carefully evaluated, and from the lake cores and pollen records at surrounding sites (Fig. 3d). At Hulun Lake, the normal temperature derived from the lake cores (Wen et al. 2010) was also different from the warming derived from the pollen records (Yang et al. 1995; Fig. 3d). However, given that the mid-Holocene warmer climates were consistently inferred from all of the other 62 sites (Table 3 and Fig. 3d) through multiple proxies and approaches and from proxy data before the 1990s (Shi et al. 1993), and that the mid-Holocene warmer winter climates were consistently suggested by 158 (Yu et al. 2000) and 188 pollen samples (Ni et al. 2010), proxy data at a few sites (Table 3), and proxy data before the 1990s (Shi et al. 1993) across China, they should be reliable, at least in a qualitative manner. Collectively, the aforementioned model–data mismatch in annual and winter temperatures over China during the mid-Holocene appears to be robust.
Responding faithfully to the imposed negative radiative forcing derived from changes in the earth’s orbital parameters and atmospheric concentrations of CO2 or CH4 during the mid-Holocene, 35 out of the 36 PMIP models reproduced colder-than-baseline annual temperature over China, with a larger magnitude overall in atmospheric models than in coupled models. In all the models, seasonal temperature was significantly reduced in winter and spring, whereas it was significantly elevated in summer. In autumn, temperature was generally reduced in atmospheric models but elevated in coupled models, which was related to warmer-than-baseline SSTs in the oceans north of Eurasia and in the western North Pacific as computed by coupled models. Based on the multiproxy data at 64 sites (Table 3 and Fig. 3d), a number of proxy data before the 1990s (Shi et al. 1993), and 158 and 188 pollen samples (Yu et al. 2000; Ni et al. 2010) across China, the mid-Holocene annual temperature of the country was estimated to be around 1–5 K warmer than the present-day values, while winter warming was also warmer than that of today. Taken together, the colder annual and winter climates from the PMIP models are the opposite of those from the multiproxy records.
Interactive ocean gave rise to an additional warming of 0.5 K in winter and 0.7 K in autumn over China, and hence the annual and winter temperatures of coupled models were in better agreement with proxy data than those of atmospheric models during the mid-Holocene. Averaged across the country and the six pairs of PMIP2 models, interactive vegetation was found to have little effect on the mid-Holocene annual and seasonal temperatures. The same conclusion was drawn from the experiments of the ECHAM53–MPIOM127-LPJ and UBRIS-HadCM3M2 pairs for which the preindustrial vegetation was used in the AOGCM control experiments. On the other hand, the spread of vegetation-induced temperature changes between each of the six pairs of the models implies a level of uncertainty in the mid-Holocene vegetation effect in the East Asian monsoon areas, which needs to be specifically investigated within the context of cause and effect. In this respect, the AOGCM and AOVGCM experiments of the mid-Holocene climate should share the same or similar control experiments, as suggested by Braconnot et al. (2007a,b), Dallmeyer et al. (2010), and O’ishi and Abe-Ouchi (2011). In addition, the extent to which the simulated vegetation from AOVGCMs is compatible with the reconstruction (e.g., Yu et al. 2000; Ni et al. 2010) during the mid-Holocene should be evaluated.
Note that the present model–data mismatch in annual and winter temperatures over China during the mid-Holocene was derived from a variety of proxy data and the experiments of 36 climate models. At the moment it is unclear whether the inconsistency arises from the models, from the proxy data, or from both sides. If the interpretations of those proxy data are correct, this raises an important question as to the dynamic mechanism underlying the mid-Holocene East Asian climate change and also poses a challenge for the forthcoming PMIP simulations of the regional climate over China for that period. Examining how, and to what extent, the new generation of climate models and/or earth system models participating in the third stage of the PMIP respond to the mid-Holocene forcing over the country is of particular interest. Whether the missing biogeochemical processes in the previous models affect the mid-Holocene East Asian climate should be examined using earth system models with carbon and nitrogen dynamics. The impact of model biases in simulating the modern climate on model results for the mid-Holocene also should be given special attention at the regional scale (e.g., Braconnot et al. 2002; Ohgaito and Abe-Ouchi 2009). On the other hand, as discussed before, the models seem to do what is asked of them in terms of negative radiative forcing over the country. Is it possible that the paleoarchives are not actually able to record the information that is equivalent to temperature in the models? Interestingly, the mid-Holocene temperature was shown to be colder in part of China in the works of Guiot et al. (2008) and Bartlein et al. (2011) using the modern analog, regression, and model-inversion techniques, in which the general challenge is that there is not a strong direct functional or mechanistic relationship between pollen spectra and climatic variables because pollen production is affected by the interaction of a large number of nonlinear processes (Ohlwein and Wahl 2012). Although there are inconsistencies between their results, for example warmer in the former versus colder in the latter over northeastern China, they at least indicate the uncertainty of the mid-Holocene warming from single-site reconstructions over China. Additionally, it should be kept in mind that the spatial coverage of the proxy data used for the present model–data comparison is still sparse. Collectively, more reconstruction work using multiple proxies and methods is required to narrow the uncertainty of proxy data and then test the model results, which is important to our understanding the mid-Holocene climate in the East Asian monsoon area. In that field, as discussed in the model–data comparison section, the method used for climatic reconstruction should be carefully assessed and validated.
Finally, it is noted that the PMIP2 CCSM3.0 simulations have been used to discuss the East Asian winter monsoon during the mid-Holocene (Zhou and Zhao 2009). The present analysis adds an important caveat to these kinds of studies, since an average cooling of 1.5 K as simulated by this model disagreed qualitatively with the proxy data over China during the mid-Holocene winter.
We sincerely thank the three anonymous reviewers and Dr. Anthony J. Broccoli for their helpful comments and suggestions on the earlier versions of the manuscript; Thierry Fichefet, Rupert Gladstone, Sandy Harrison, Akio Kitoh, Uwe Mikolajewicz, and Ian Ross for information on the PMIP2 AOVGCMs and their AOGCM counterparts; and Fahu Chen, Zhaodong Feng, Wenying Jiang, Bingcheng Li, Hongya Wang, Jüle Xiao, and Xuefeng Yu for information on proxy data. Also, we acknowledge the international modeling groups for providing their data for analysis, and the Laboratoire des Sciences du Climat et de l’Environnement (LSCE) for collecting and archiving the model data. This research was supported by the Chinese National Basic Research Program (2009CB421407), the Strategic Priority Research Program (XDA05120703) and the Knowledge Innovation Program (KZCX2-EW-QN202) of the Chinese Academy of Sciences, and the National Natural Science Foundation of China (40975050 and 41175072). The PMIP2/MOTIF Data Archive is supported by CEA, CNRS, the EU project MOTIF (EVK2-CT-2002-00153), and the Programme National d’Etude de la Dynamique du Climat (PNEDC). The analyses were performed using version 9 August 2010 of the database. More information is available online (http://pmip2.lsce.ipsl.fr/ and http://motif.lsce.ipsl.fr/).