DISCUSSION

For the Late Miocene, we defined some CO₂ climate modelling sensitivity scenarios 1) to test how much CO₂ is necessary to produce ice-free conditions on the Northern Hemisphere in the Miocene, 2) to analyse the Miocene climate sensitivity with respect to variations of CO₂, and 3) to validate the consistency of the model results with proxy data in order to estimate how high CO₂ might have been in the Late Miocene.

The first question can be answered now "easily". Based on our model results, a pCO₂ of at least 1500 ppm is necessary to produce an ice-free Arctic Ocean. Even the most "optimistic" estimations for the Miocene carbon dioxide concentration, however, give values of about 1000 ppm or less (e.g., Cerling 1991; Retallack 2001). Thus, our model results suggest that an ice-free Arctic Ocean in the Miocene is unlikely in the absence of other processes, which might have caused additional warming. Recently found evidences from the Miocene fossil record also suggest some significant cooling events and support sea ice cover on the Northern Hemisphere (e.g., Moran et al. 2006; Kamikuri et al. 2007; Jakobsson et al. 2007).

What is the Late Miocene climate sensitivity with respect to different concentrations of CO₂? For the present-day simulations TORT-700 vs. TORT-360, the global temperature increase is +2.5°C. McGuffie et al. (1999) demonstrated that a doubling of CO₂ under present-day conditions leads to a warming of +2.5°C to +4.5°C. The most recent future climate change projections of the IPCC (Meehl et al. 2007), which consider a doubling of CO₂, give a similar temperature increase between +2°C to +4.5°C. Our present-day sensitivity experiments are at the lower end but within the range of these climate change scenarios. Our Late Miocene simulations demonstrate a global temperature increase of +1.9°C (TORT-560 vs. TORT-280, TORT-700 vs. TORT-360) due to a doubling of CO₂. The Late Miocene climate sensitivity on changed pCO₂ is weaker than today. Steppuhn et al. (2007) observed that a doubling of CO₂ under Miocene boundary conditions leads to a global warming of +3°C. This is well within future climate change predictions, but stronger than in our simulations. The sea ice cover in Steppuhn et al.'s (2007) Tortonian reference run is close to the modern situation. In our simulations, even TORT-200 has a lower ice volume than CTRL-360 (cf. Figure 4). Thus, the lower sea ice-albedo feedback dampens the Late Miocene climate sensitivity on a CO₂ increase, although the Late Miocene is still comparable to the modern situation. The amount of sea ice is lower in our Miocene simulations than in previous Tortonian experiments (Micheels et al. 2007; Steppuhn et al. 2006, 2007) because we specified the modern ocean flux correction (cf. Figure 3), whereas the previous studies considered a weaker-than-present ocean heat transport. The reduced sea ice cover in our Tortonian runs as compared to the present-day simulations is explained by the warming effect of Greenland (no glaciers and lower elevation, cf. Figure 1) and the larger forest cover (cf. Figure 2) in the palaeoclimate experiments. These results are consistent to the previous studies.

Future climate change projections also demonstrated that continents are much more affected by the global warming (e.g., Meehl et al. 2007). This is consistent to our simulations. However, the climate change predictions represent a pronounced warming of higher latitudes (e.g., Meehl et al. 2007). This is in agreement to TORT-700 vs. TORT-360. In contrast, the Late Miocene simulations do not show as much pronounced high-latitude warming (cf. Figure 4). The reduced ice-albedo feedback in the Tortonian runs as compared to the present-day situation reduces the response to a CO₂ increase in higher latitudes. In the lower latitudes, future climate change projections and our Miocene runs are quite comparable. In contrast to Steppuhn et al. (2007), in our simulations the different settings of boundary conditions (Miocene vs. modern) make a difference in the climate response to enhanced CO₂-scenarios. In general, our Miocene experiments demonstrate a weaker sensitivity on higher CO₂ than future climate change scenarios because sea ice is already reduced in the Miocene reference run.

How consistent are the different Late Miocene CO₂-scenarios as compared to the fossil record and can we estimate how high CO₂ was in the Late Miocene? Considering the fossil record, our model fits quite well with proxy data for a pCO₂ of 360 to 560 ppm (cf. Figure 8). TORT-200 to TORT-460 represent a global temperature difference of less than ±1°C as compared to proxy data. Micheels et al. (2007) find a global discrepancy to proxy data of –2.4°C from a Tortonian simulation with the AGCM ECHAM4/ML using almost the same boundary conditions and proxy data base. If this difference is acceptable, then even TORT-700 could still be a realistic scenario for the Late Miocene (Table 2).

There are, however, some crucial points, which limit our interpretations. Most Late Miocene proxy data cover the European realm, whereas most other (key) regions such as the high latitudes, Africa and Asia are poorly covered. Our sensitivity runs demonstrate that lower to mid-latitudes are warmer than suggested by proxy data, but this statement is based on only a few localities. With regard to the quality of proxy data, there are some shortcomings that may influence detailed data intercomparison. This includes the time span covered by the proxy data that may hide temporal climatic variability, palaeogeographical changes that may change the exact position of data points with respect to the grid cell compared with (e.g. in the Pannonian realm, cf. Erdei et al. 2007) and possible taphonomic biases that may influence the results of different reconstruction methods (e.g., Liang et al. 2003; Uhl et al. 2003, 2006). Despite some possible shortcomings, quantitative climate proxy data based on plant fossils are consistent on a larger scale (Bruch et al. 2004, 2006, 2007), and they provide reliable information to validate global climate model results (Micheels et al. 2007; Steppuhn et al. 2007).

For our purposes, we consider that proxy data are more reliable than the model results. The model experiments can be unrealistic because of weak points in the model and its boundary conditions. The EMIC Planet Simulator with its simplified parameterisation schemes and the boundary conditions explain rather more inconsistencies in our experiments. For some localities (e.g., in Africa, North America), the orography in the model might be unrealistic because of the coarse model resolution, the spectral transformation method or simply because the reconstruction of the orography is not fully correct.

The coarse model resolution also does not allow for a representation of details of the land-sea distribution. For example, the model cannot resolve Iceland (cf. Figure 8). This can explain why TORT-200 to TORT-630 represent (more or less) conditions in the North Atlantic region that are too cool. In addition, the settings for the ocean include some uncertainties. As mentioned in the model description, the sea ice model tends to overestimate the amount of ice and the performance of the ice model for seasonal ice is better than for multiyear ice. However, with increasing CO₂ concentrations in our runs, the amount of multiyear ice successively reduces, and the model performance should increase.

The absence of ocean and sea-ice dynamics is a potential source of errors in our study, but it is computationally expensive to run fully-coupled atmosphere-ocean general circulation models in particular for a series of sensitivity experiments. For this reason, AGCM experiments for the Miocene used either prescribed sea surface temperatures and sea ice (e.g., Ramstein et al. 1997; Lunt et al. 2008) or slab ocean models (e.g., Dutton and Barron 1997; Micheels et al. 2007). For past climates, ocean dynamics plays an important role (e.g., Bice et al. 2000). Due to an open Panama Isthmus, the northward heat transport in the Atlantic Ocean was weaker than today in the Miocene (e.g., Bice et al. 2000; Steppuhn et al. 2006). Because of using the present-day flux correction, the climate in Europe should be too warm in our sensitivity experiments.

If we consider that temperatures in Europe might need to be corrected to slightly cooler conditions, the scenarios TORT-460 to TORT-630 would be more realistic. Taking all weak points into account, the scenarios TORT-360 and TORT460 fit best with the fossil record. Hence with some limitations, the simulations support a slightly higher-than-present ('intermediate') pCO₂ of higher than 360 ppm and less than 460 ppm in the Late Miocene. Consistent to this result, MacFadden (2005) proposed a pCO₂ of 500 ppm in the Miocene and for Middle Pliocene model simulations an atmospheric carbon dioxide of 400 ppm is a reasonable value (e.g., Haywood and Valdes 2004, 2006).