RESULTS

Figure 4 illustrates that in the Miocene and present-day simulations, the global temperature increases and the global sea ice cover decreases with an increased atmospheric carbon dioxide. The global temperature of the Tortonian runs is generally higher than in the present-day simulations (if CO₂ is the same in CTRL and TORT). Accordingly, the sea ice cover of TORT-200 to TORT-INC is also lower than in CTRL-280 to CTRL-700. With increasing CO₂, the global temperature and sea ice cover of TORT-INC follows quite well the distribution of the other runs TORT-200 to TORT-700. Hence, the CO₂ increase of +1 ppm is small enough to keep TORT-INC close to the equilibriums of TORT-200 to TORT-700.

Comparing CTRL-360 and CTRL-700 vs. CTRL-280 and the Tortonian runs, the response on an increased pCO₂ is more pronounced in the present-day simulations. The temperature difference between CTRL-700 and CTRL-360 is +2.5°C, whereas it is +1.9°C between TORT-700 and TORT-360. Sea ice cover is reduced by –2.9% (CTRL-700 minus CTRL-360) and by –2.1% (TORT-700 minus TORT-360), respectively. Hence, the weaker response to a CO₂ increase is explained by the generally lower amount of sea ice in the Miocene experiments, i.e., the ice-albedo feedback is weaker.

Figure 5 illustrates the zonal average sea ice cover of TORT-INC with respect to CO₂. A critical threshold for the Arctic ice cover is around 1,250 ppm. At this level, the northern sea ice entirely vanishes for the first time, but it sensitively responds on climate fluctuations. If there is a small deviation (climate variability), ice-free conditions cannot be maintained. With an atmospheric carbon dioxide concentration of about 1,400 ppm, the Northern Hemisphere is permanently ice-free in TORT-INC. The sea ice cover on the Southern Hemisphere is generally maintained. However, only a few small fractions of sea ice remain if pCO₂ is higher than about 1,500 ppm.

Figure 6 shows the zonal average temperatures of the simulations wherein the Miocene experiments are represented against TORT-280 and the present-day runs and TORT-280 against CTRL-280, respectively. TORT-280 is much warmer than CTRL-280 at around 30°N (+4°C) and farther to the northern high latitudes (+5°C). Close to the equator, CTRL-280 and TORT-280 do not differ much (less than +1°C). Thus, TORT-280 represents a weaker-than-present meridional temperature gradient of –4°C. The meridional gradient in TORT-200 is less pronounced than in TORT-280, but it is still weaker than in CTRL-280. With increasing carbon dioxide in the atmosphere, the low to high latitudes get successively warmer, but polar warming is much more intense. In TORT-INC at 2,000 ppm, the high latitudes heat up by +9°C as compared to TORT-280, and tropical latitudes are +3.5°C warmer. Thus, the temperature difference between pole and equator is –5.5°C lower than in TORT-280. The successive reduction of the meridional temperature gradient is a result of the sea ice-albedo feedback (cf. Figure 4 and Figure 5). The temperature difference between TORT-2000 (i.e., TORT-INC at 2000 ppm) and TORT-1500 is generally less than for TORT-1500 vs. TORT-1000. The weaker response is due to the fact that northern sea ice vanishes at around 1,400 ppm (Figure 5).

Between CTRL-360 and CTRL-280, there are just minor differences of less than +1°C, which is similarly seen from TORT-360 as compared to TORT-280. However, CTRL-700 vs. CTRL-280 as compared to CTRL-360 vs. CTRL-280 demonstrates a more amplified polar warming than TORT-700 and TORT-360 vs. TORT-280, respectively. The CO₂ doubling from 360 ppm to 700 ppm leads to a polar warming of +4°C under present-day conditions, whereas it is only +3°C in the Miocene. In lower latitudes, the response to the CO₂ increase is about the same. Thus, the sea ice-albedo feedback tends to be weaker under Miocene boundary conditions than using present-day conditions (cf. Figure 4).

The spatial distribution of mean annual temperature differences and the sea ice margin between our simulations are shown in Figure 7. The increase of CO₂ leads to a generally more pronounced warming in the present-day experiments as compared to the Miocene runs (_blank). For CTRL-280 to CTRL-700, the ice volume is greater than in TORT-200 to TORT-2000 (cf. sec. 3.1 and 3.2). Therefore, the ice-albedo feedback is more intense under present-day conditions. Moreover, the Paratethys dampens the general warming trend due to enhanced CO₂ in the Miocene simulations. In TORT-200 as compared to TORT-280, the cooling because of a decreased CO₂ occurs primarily in higher latitudes. This is a contrast to the other Tortonian runs. Generally, interior parts of the continents become warmer when CO₂ increases. Not until a high concentration of CO₂ and ice-free conditions are reached, polar warming is in the same order of magnitude as over the continental areas. In Central Africa, temperatures in the Tortonian runs remain more or less the same when increasing CO₂. An intensified evapotranspiration (evaporative cooling) dampens the temperature increase due to the greenhouse effect.

Steppuhn et al. (2007) established a method to compare Late Miocene model experiments with quantitative terrestrial proxy data. We now use basically the same method to test how consistent the mean annual temperatures (MAT) of the different Late Miocene CO₂ scenarios are as compared to the fossil record. The main data set of terrestrial proxy data is given in Steppuhn et al. (2007). All data for mean annual temperature (MAT) are based on quantitative climate analyses of fossil plant remains from the Tortonian stage (early Late Miocene, ~ 11 to 7 Ma). For most of the data, the Coexistence Approach (Mosbrugger and Utescher 1997) was applied on micro- (pollen and spores) and macro-botanical (leaves, fruits, and seeds) fossils. The results of this method are 'coexistence intervals,' which express the minimum-maximum range of temperature at which a maximum number of taxa of a given flora can exist. Relying mainly on one reconstruction method reduces the impact of methodological inconsistencies. However, such data are not yet available for North America. Therefore, we also included some published quantitative climate data based on the CLAMP technique (Wolfe 1993), which has proven to be a reliable method especially for climate quantification on the American continent (cf. Wolfe 1995, 1999).

Because such data usually do not include a minimum-maximum range of temperature, a standard range of uncertainty of ±1 °C was assumed for the data-model-comparison. We completed the Steppuhn et al. (2007) data set with additional climate information from Wolfe et al. (1997) and new data from the NECLIME program (see http://www.neclime.de) published in Akgün et al. (2007), Bruch et al. (2007) and Utescher et al. (2007). The actual data set used in this study now comprises 78 localities. Because most of the proxy data represent a climatic range from a minimum to a maximum mean annual temperature at a locality, Steppuhn et al. (2007) constructed a similar MAT range from the minimum and maximum mean annual temperature of their 10-year-model integrations at the specific grid points. In case of an overlap of both climate intervals, model results and proxy data are defined to be consistent (i.e., the temperature difference is equal zero); in case both intervals do not overlap, the smallest distance between them is defined to be a measure for the inconsistency. We slightly modify the validation method of Steppuhn et al. (2007) because results of TORT-1000, TORT-1500, and TORT-2000 do not represent a time series over 10 years. Instead of creating temperature intervals from our simulations, we use "point" data. For TORT-1000 to TORT-2000, the point data are simply the mean annual temperatures of the years 900, 1400, and 1900, respectively. For TORT-200 to TORT-700, the point data are the mean annual temperatures averaged over the last 10 years of the model integrations. Except of this difference, the validation method follows the same principle as before (Steppuhn et al. 2007). Table 2 summarises the overall agreement of the Tortonian simulations with proxy data. In Figure 8, the temperature differences between the simulations and proxy data are mapped for all localities.

On the global scale (Table 2), the experiment TORT-280 fits best with terrestrial proxy data, but also TORT-200, TORT-360, and TORT-460 give more or less consistent results. Discrepancies of TORT-200 to TORT-460 are within ±1°C. These deviations to the fossil record are quite acceptable. As one might expect, TORT-2000 demonstrates the worst overall consistency to proxy data and is globally much too warm.

Figure 8 shows details of the comparison of model results and proxy data. TORT-200 and TORT-280 globally demonstrate the best agreement to proxy data, but they are systematically too cool in higher latitudes. TORT-360 and TORT-460 indicate a better agreement in higher latitudes. However, both runs tend to be slightly too warm in the mid-latitudes, particularly in Europe. At the expense of heating up lower and mid-latitudes, TORT-560 to TORT-2000 are continuously more consistent to high-latitude proxy data. Figure 8 illustrates that TORT-360 to TORT-560 basically agree best with proxy data. This agreement seems to be in contrast to Table 2, but one has to keep in mind that the spatial distribution of proxy data is highly concentrated on Europe. Consequently, discrepancies in this region are over-weighted as compared to others. Our results support that pCO₂ in the Late Miocene is within the range of 360 to 560 ppm.