The anomalies documented here are not trivial. The biogeographic patterns expected for rapidly migrating species are observed in the glaciated zone, but rarely in the unglaciated zone. Boreal zone species exhibit low genetic variation or diversity gradients, but species in the unglaciated zone show high genetic richness, no gradients, and deep (ancient) DNA divergence between populations. Endemics are very rare in the glaciated zone, and endemics in this zone appear to be mostly recent (due to hybridization, for example). Sister species, subspecies, and races are common across the unglaciated East but not in the glaciated zone. To keep all of these groups separate during the Pleistocene so that they did not hybridize/homogenize out of existence would require multiple LGM refuges, but no such refuges have been found in the far South. If the pollen data say one thing and the biogeographic data flatly contradict this inference, what can we conclude?
There is a possible resolution to this paradox based on climate and CO2 factors. First, glacial climates are not simply a general cooling of the climate. Glacial epoch climates were cooler, but more so in summer (Pielou, 1991). Thus, the existence of “boreal” species farther south than today does not mean that climates were bitterly cold. Second, a factor that also varied during the glacial period is CO2, which was so low that it caused CO2 starvation. CO2 starvation does not affect all species equally.
At the LGM, about 18 000 years ago, CO2 levels were very low, less than 200 ppm. This low level of CO2 constitutes a severe deficiency for growth, and may have shifted competitive dominance between different plant types, as well as affecting overall vegetation biomass (Levis and Foley, 1999). Altered CO2 levels could also affect pollen production, which would bias biome reconstruction. C4 grasses have a strong advantage over other plants under low CO2 (Cole and Monger, 1994; Farquhar, 1997; Polley et al., 1993, 1995), though not under low temperatures. Robinson (1994) shows that the extent of stomatal regulation plants exhibit in response to CO2 level varies by taxa in a manner that suggests that this trait is a modern adaptation. Ancient taxa such as conifers exhibit little stomatal responsiveness compared to angiosperms. The benefit of stomatal responsiveness trades off against greater water loss at low CO2 levels resulting from stomates being open longer (Drake et al., 1997; Saxe et al., 1998). Where water is not limiting, a cool low CO2 climate should favor C3 grasses, forbs and angiosperm trees over conifers (Beerling, 1996; Beerling and Woodward, 1993; Jolly and Haxeltine, 1997; Robinson, 1994; Saxe et al., 1998). This suggests that mesic broadleaf trees should have remained competitive in mesic microsites during the Pleistocene. On most upland sites, however, a low CO2 climate would favor xeric species such as conifers, which have consistently low stomatal conductance, and herbaceous species, including grasses and sedges. Oaks, as trees with intermediate responsiveness, would have persisted as well on upland sites. We in fact observe extensive open conifer forest replacement of broadleaf forest at the LGM in eastern North America, with a persistent oak component. Simulation studies also show that the reduced water use efficiency expected at the LGM would produce a xeric, open forest (with low leaf area index) south of the ice dominated by conifers (Cowling, 1999; Harrison and Prentice, 2003; Jolly and Haxeltine, 1997; Levis and Foley, 1999). This explains the “no-analog” open forest (or parkland) often remarked upon. This discussion shows that species that currently occur together would diverge in their responses under the LGM climate, and that climate predictions based on their occurrence would not be correct. Exact simulations of vegetation would need to be based on more detailed studies of the physiology of each species, few of which have been done.
Simulations of the effect of low CO2 levels show that it could cause a major lowering of alpine treelines (Bennett and Willis, 2000; Cowling and Sykes, 1999; Street-Perrott et al., 1997). Jolly and Haxeltine (1997) simulated the montane ecotone for African mountains and showed that the entire LGM lowering of treeline at their study site is consistent with the effect of low CO2 without assuming any drop in temperature. Lower treelines at the LGM have typically been taken as evidence for colder temperatures, but could really be the result of changes in CO2 levels. They would thus not correspond to snowline depressions (e.g., Greene et al., 2002) or other indicators of temperature.
If lowered CO2 affected vegetation in eastern North America, what changes would be expected? Grasses and other grassland species would be expected to expand their range into forest, perhaps creating parkland. Forest communities would shift to drought tolerant species such as pines and oaks. Significantly, mesic microsites such as coves, north slopes, and stream valleys would provide refuges for mesic species across their entire original range, especially in the highly dissected Appalachian region, because the effect of CO2 would be similar to an overall drying. High precipitation, midelevation regions such as the Cumberland Plateau and the Southern Blue Ridge Escarpment would provide prominent refuges. In contrast, topography provides much less protection against extreme cold. An LGM African montane site (Jolly and Haxeltine, 1997) shows this pattern, with low levels of tropical montane forest pollen persisting in LGM profiles dominated by ericaceous shrubs. While drought-tolerant boreal elements such as white spruce (or the extinct temperate Picea crutchfieldii) could have moved south under a cooler, low CO2 climate, most eastern boreal trees are not drought tolerant and would not have been able to move far south. We observe exactly this pattern of change in the pollen record, with most boreal trees being absent from the supposed boreal parkland south of the ice. Several authors have proposed that the southern Appalachians were a refuge (e.g., Braun, 1950; 1951; Church et al., 2003; Harvill, 1973; Hewitt, 2004). This view now has increased plausibility. This analysis provides independent biogeographic support for model-based projections of the effects of CO2 on LGM vegetation. Other factors such as altered fire regimes and herbivores could also impact the details of reconstructions. The above discussion has implications for the locations of refugia in Europe. The traditional reconstruction of LGM Western Europe has been a treeless steppe-tundra as far south as southern France (Stewart and Lister, 2001), with presumed refugia in the Mediterranean areas such as the Iberian and Italian peninsulas (Stewart and Lister, 2001). However, fossils of thermophilous trees and mammals have been dated to the LGM in Belgium, England, Hungary, Slovakia, and Germany, among other places (Stewart and Lister, 2001; Willis and van Andel, 2004). In addition, populations of numerous trees and animals such as Scots pine in Scotland (Stewart and Lister, 2001) and hedgehog in Germany (Willis and Whittaker, 2000) are either genetically distinct from Mediterranean populations (Scots pine) or have no southern relatives (hedgehog). For these species, it appears that they persisted throughout the LGM in refugia scattered across Europe in what is assumed to be a climate too cold for them to persist. Stewart and Lister (2001) interpret these refugia, most of those known being fossils from cave sites in steep valleys, as thermal refuges from the ice age cold. However, steep valleys are not particularly known for providing thermal refuges in present-day northern habitats. In fact, in hilly terrain, it is south and west facing midslopes and ridges that are warmest, not valleys, which receive less sun and are subject to cold air drainage. It is more likely that these valleys provided mesic refuges from the combined effects of a drier climate and the reduced water use efficiency caused by the CO2 effect. These mesic plant communities would provide a home for the animals found there. Just such steep valley and stream-margin refuges are found today in dry regions across the world. It is noteworthy that tree species that went extinct in Europe were less drought tolerant than surviving species (Svenning, 2004), as predicted by the CO2 effect model. To an even greater extent than in North America, the reduction of tree cover due to the CO2 effect may have given a false impression of extreme dryness in Europe. It is not asserted that all aspects of the situation are analogous. There are a number of implications of these results for the practice of climate reconstruction and for testing climate models against historical proxy data. Regression or response surface approaches (e.g., Farrera et al., 1999; Nakagawa et al., 2002; Peyron et al., 1998; Tarasov et al., 1999; Web et al., 1993, 1997) implicitly assume that species’ climate responses are stable over time, but the CO2 effect model suggests that this assumption is violated. The use of plant functional types to reconstruct either biomes (e.g., Elenga et al., 2000) or climate (e.g., Peyron et al., 1998; Tarasov et al., 1999) has the same issues as response surface models and also assumes that similar species will respond to a changing climate as a group. It was shown above that normally co-occurring species will show divergent responses to CO2 change. Very few studies can be found that take explicit account of CO2 effects when reconstructing vegetation and climate (e.g., Guiot et al., 2000; Levis et al., 1999; Jolly and Haxeltine, 1997). Regression and response surface models, including models of plant functional types, also are unable to account for other changes in the conditions at the LGM. For instance, seasonal distributions of temperature were not strictly analogous to latitudinal shifts. That is, the colder temperatures at the LGM did not produce seasonal shifts that correspond to a simple northward movement of place. This throws off calibration of models of plant response to climate. Annual variability may have been altered, fire regimes may have been different, and in North America the presence of large herbivores that later went extinct really should not be ignored. All of these factors could have shifted the community composition in ways that would give the impression of some climate effect.
Testing of general circulation models is likewise affected by the CO2 response of vegetation. While there is some recognition that CO2 could affect pollen interpretations, the range of weight given to this issue ranges from none (Nakagawa et al., 2002; Pinot et al., 1999) to a consideration of only the C3 vs. C4 effect (Crowley, 2000; Farrera et al., 1999). Few climate model comparisons to paleotemperature proxies have fully factored in the CO2 effects on leaf area, plant types, and water use efficiency in the temperature reconstruction. The treatment of vegetative cover and transpiration rate within climate simulation models is probably also deficient in this regard, though published descriptions of this model component are not usually adequate to determine exactly how vegetation is modeled.
In summary, the entire interpretation of LGM vegetation and climate in eastern North America may have been biased by the CO2 effect. It was neither as cold nor as dry as has been assumed. Lower treelines were probably caused by the CO2 effect. The “no-analog” conifer woodlands are the direct result of changes in water use efficiency between taxa and the lumping of spruce species and sedge species into generic-level categories. The presence of spruce in eastern North America was not an indicator of a boreal climate. Massive vegetation dislocations and migrations did not occur prior to ice melt, and the entire LGM unglaciated region acted as a refuge for species of the eastern deciduous forest. This explains the almost complete lack of tree species extinctions in this region. It also suggests that the use of plant remains to predict climate for any period of the past in which CO2 level was appreciably different from today may lead to incorrect conclusions unless the effect of CO2 on relative growth rates is accounted for.