The last deglaciation as model system thus offers several avenues for studying the responses of species and communities to multiple drivers of global change.[55, 74] Building upon earlier work,[10, 65, 70] much of our recent work has focused on the challenge of understanding why no-analog associations of species emerged during the last deglaciation and what caused them to disappear. Specifically, we have sought to (1) understand the abiotic and biotic factors that shaped the formation, maintenance, and ultimate disappearance of no-analog plant associations in eastern North America during the last deglaciation, and (2) use the no-analog species associations and climates of the last deglaciation as a testing ground for biogeographical models of species distributions and diversity.
Recent investigations have progressed along two primary avenues. First, we collected a series of new fossil pollen records and other ecological proxies from lake sediments in the east-central United States in order to more tightly constrain the timing and duration of no-analog vegetation, as well as test and refine hypotheses of abiotic (climate) and biotic (megafaunal herbivory) controls on the formation of no-analog plant associations.[62, 75-77] Second, we have built a new dataset of fossil pollen distributions in eastern North America and combined it with paleoclimatic simulations from CCSM3[79, 80] and HadUM to estimate pollen–climate response surfaces. These response surfaces have been applied to assess the stability of species’ realized climatic niches during past climates with no modern analog[54, 72, 82, 83] and assess the predictive ability of models of species distribution and diversity.[54, 56, 84-86] Here we summarize key findings from these papers, discuss the implications for contemporary studies into novel ecosystems and climates, and identify ongoing and future research directions.
Site-level studies into biotic and abiotic controls on no-analog vegetation
Several new pollen records have been published from the heart of the region of late-glacial no-analog plant associations in eastern North America (Fig. 3): Crystal Lake, Illinois, Appleman Lake, Indiana, Silver Lake, Ohio, Jackson Pond, Kentucky, and Anderson Pond, Tennessee, with several more records soon to be published (Gill et al. in preparation, Jones et al. in preparation). These new records, based on accelerator mass spectrometry (AMS) radiocarbon dating of terrestrial plant remains and charcoal, are more accurately and precisely dated than previous records in the region, which incorporated inaccuracies of 102–103 years introduced by obtaining radiocarbon dates from bulk lake sediments in a region underlain by radiocarbon-depleted carbonate bedrock. These records clearly show that no-analog plant associations formed first at the south-central sites (Jackson and Anderson Ponds), between 15.9 and 15.4 ka, and roughly 1,700 years later at the upper Midwestern sites (Crystal, Silver, and Appleman), where the establishment of no-analog vegetation dates between 14.2 and 13.7 ka. Additionally, the duration of no-analog vegetation at individual sites is more narrowly constrained than in previously published subcontinental maps. At the northern sites, no-analog vegetation persists for 2,100 to 2,500 years, ending at 11.8 ka at Silver Lake and at 11.7 ka at Crystal Lake (Fig. 3). The duration of no-analog vegetation at the southern sites unfortunately is uncertain because of low sedimentation rates and hiatuses in the upper (end Pleistocene and Holocene) sediment column at Jackson and Anderson Ponds. At all sites, the no-analog plant associations are characterized by rising abundances of deciduous tree taxa (Fraxinus, Ostrya/Carpinus, Quercus, Carya, Ulmus, Salix) combined with moderate to high abundances of conifers (Picea, Larix, Abies) and herbs (Cyperaceae, Poaceae), but the relative abundances of these taxa and the timing of their abundance rises differ among sites (Gill et al., in preparation).
Our working conceptual model is that these plant associations formed in response to a hierarchy of climatic and biotic controls in which (1) climate changes during the last deglaciation set the first-order constraints on plant species distributions as species tracked their climatic niches, and (2) reductions in megafaunal populations and biomass may have altered plant community composition via reduced or altered herbivory by megafauna or a shift from a megafaunal-dominated to fire-dominated disturbance regime[62, 75] in at least some regions. Within this general conceptual model, elucidating the processes responsible for the formation and eventual disappearance of the no-analog plant associations remains a major frontier, and at this point the primary lines of evidence are correlative rather than mechanistic.
Likely climatic factors contributing to the formation of no-analog plant associations include rising temperatures at the end of the Pleistocene, higher-than-present insolation and temperature seasonality, and, possibly, moisture availability. The importance of rising temperatures is suggested by the south-to-north sequence of no-analog formation and the rising abundances of temperate deciduous trees. Rising temperatures alone, however, cannot explain why some temperate deciduous taxa were relatively abundant in the late-glacial no-analog associations (e.g., Fraxinus, Salix, Ostrya/Carpinus) and relatively uncommon in mid- to late-Holocene forests. Paleoclimatic simulations from general circulation models consistently indicate that the no-analog associations were associated with higher-than-present seasonality of insolation and temperature,[10, 65] which may explain why some temperate deciduous taxa were more abundant during the late Pleistocene. Paleoclimatic simulations and pollen-based paleoclimatic reconstructions disagree on whether regional moisture availability was higher or lower than present during the period of no-analog vegetation. One possibility is that moisture availability varied across these recently deglaciated landscapes, with wetter-than-present conditions in low-lying poorly drained soils and dryer conditions in the uplands. Note that the timing of these vegetation changes and their asynchronous nature rule out any hypothesis that invokes a single instantaneous event, such as a putative extraterrestrial impact event at 12.9 ka,[90, 91] as a cause of either the formation of the no-analog associations or their disaggregation.
At several of the northern sites (Appleman, Silver, and also Binnewater Pond, New York), the formation of no-analog vegetation coincides with or is slightly preceded by a decline in abundances of spores of Sporormiella (Fig. 3), a coprophilous fungus[93, 94] that has been used to track the worldwide megafaunal population declines and extinctions during the late Pleistocene and Holocene.[72, 95-99] Sporormiella is useful because its spores are readily identifiable during standard pollen counts and Sporormiella lives exclusively on dung for part of its life cycle. The correspondence of Sporormiella declines with the formation of the no-analog plant associations at Appleman and Silver Lakes in the upper Midwest (Fig. 3), which may suggest that the rapid increases in hardwood tree abundances and development of no-analog vegetation in the upper Midwest was amplified by the end-Pleistocene megafaunal population declines and extinctions. Possible mechanisms include the release of relatively palatable taxa from megaherbivory, increased fuel load and altered fire regime, or changed nutrient cycling.[62, 75, 92]
However, Sporormiella was scarce throughout the records at the southern sites (Anderson and Jackson Ponds), suggesting that either the population densities of megafaunal species were always low near these southern sites or that the coring locations within the lake basins were too far from Sporormiella sources to detect Sporormiella spores, which have a dispersal distance on the order of tens to hundreds of meters.[100-103] More data are needed to understand the representation of Sporormiella within late-glacial sediments, and more process-oriented modeling is needed to test mechanisms linking megaherbivore declines with vegetation change in eastern North America.
Data synthesis and chronology revision for vegetation mapping and modeling
Most of these new fossil pollen records, combined with other recently published and well-dated records (e.g., Ref. ), have formed the basis for a new dataset of plant distributions in eastern North America. The challenge tackled here is how best to make use of the ecological information from the hundreds of late-Quaternary fossil pollen records in eastern North America, given variations among sites in dating precision and accuracy. Blois et al. established a systematic scheme for ranking the accuracy and precision of radiocarbon dates and the age models on the basis of those dates, then applied the ranking scheme to identify benchmark fossil pollen records in eastern North America with well-constrained age models. Widespread ecological events (e.g., the end-Pleistocene Picea decline and Quercus rise) were identified within individual records using Bayesian change-point analysis, and the ages of these events were spatially interpolated from the benchmark records to refine the age models for other records, with a cross-validation interpolation uncertainty of about 500 years. Age models were subsequently fitted using the R software package clam, and pollen abundances linearly interpolated between samples to build datasets of fossil pollen abundances spaced at 1000-year intervals. This dataset, combined with paleoclimatic simulations from the CCSM3 and HadUM earth system models,[79, 81] is the foundation of the subsequent analyses described next.
Modeling taxon-level dynamics under no-analog climates and truncated niches
A general problem posed by no-analog climates is how to infer species–climate relationships for truncated portions of species’ fundamental niches (Fig. 4), that is, portions of fundamental niches that lie outside the bounds of the currently available and observable climate space.[10, 12, 82] No-analog climates and truncated niches are a fundamental challenge for both paleoclimatologists (who use contemporary species–climate relationships and space-for-time substitution to make inferences about past climates given information about past species distributions and abundances)[106, 107] and contemporary biodiversity modelers (who use species–climate relationships to make inferences about future and past species distributions given information about future or past climate). Inferences about 21st-century climate change effects on biodiversity are highly sensitive to assumptions about whether species can tolerate temperatures outside the 20th- and 21st-century observational domain.[12, 109-111]
Figure 4. Diagram of the expanded response surface (ERS) method for building response surfaces for truncated portions of a fundamental niche. In this approach, the distribution of pollen abundances along an environmental gradient are assumed to be stable and symmetric around a central mode, allowing information from the complete limb of the response surface to be used to fill in the incomplete limb.
Download figure to PowerPoint
Gonzales et al. tackled the problem of truncated niches and no-analog plant associations using a paleoclimatic approach. They developed the extended response surface (ERS) method, which assumed that (1) fossil pollen abundances were characterized by symmetric and unimodal distributions along environmental gradients; (2) some species have fundamental niches that could extend beyond the current set of realized climates; and (3) species–climate response surfaces were stable over time, except where covered or uncovered by shifts in the realized climate envelope (Fig. 4). These assumptions allowed a nonparametric approach to estimating truncated portions of species’ fundamental niches, in which fossil pollen abundances for a taxon along an environmental gradient were mirror-imaged around a central mode, from nontruncated to truncated sections of a species-response surface. The ERS method successfully made inferences about past climates from the fossil pollen record at Crystal Lake, Illinois, where standard analog-based methods failed. Reconstructed late-glacial climates during the height of no-analog vegetation at Crystal Lake (14.2–12.4 ka) were characterized by summer and winter temperatures that were slightly cooler than present (but warmer than the preceding and following intervals) and by higher-than-present winter precipitation. However, the ERS reconstructions suggested that temperature seasonality was similar to the present, contrary to earlier paleoclimatic simulations by general circulation models.[10, 65] This discrepancy implies that either the climate models overestimate the effects of insolation on temperature seasonality or the ERS-based reconstructions could not disentangle the strong present-day correlation between summer and winter temperatures and accommodate past shifts in the joint distribution between summer and winter temperature.[82, 106, 112] We favor the latter interpretation.
Veloz et al. took an alternate approach to the problem of no-analog climates and truncated niches. After intersecting the fossil pollen dataset from Blois et al. with paleoclimatic simulations from CCSM3[79, 80] and HadUM, Veloz et al. then assessed the stability of the resultant pollen–climate relationships as a measure of niche stability and the predictive ability of species-distribution models under conditions of no-analog climates and shifts of realized niches for some taxa. This approach fundamentally differs from the ERS method, in that it does not assume that species-response surfaces are stable through time. Rather, Veloz et al. assumed that the CCSM3 and HadUM simulations were accurate depictions of past climates, based on consistency in results between these two simulations and earlier assessments of the CCSM3 simulations against paleoclimatic proxy data.[79, 114] Neither of the assumptions made by Gonzales et al. or Veloz et al. are likely to be fully true; it is an open question as to which is a better working set of assumptions when modeling species–climate relationships under no-analog climates.
The observed changes in pollen–climate relationships inferred from the paleoclimatic simulations suggested that realized niches for some plant taxa substantially shifted during the late-glacial no-analog climates but were relatively stable during the Holocene.[54, 83] For example, the distribution of Fraxinus pollen abundances shifted in both the mode and the maximum value (Fig. 5). These shifts in realized niches are interpreted to have been caused by plant taxa exploiting past climates that are available no longer available today. An alternate hypotheses—that the observed shifts in pollen–climate relationships resulted from migrational disequilibrium—appears unlikely given a good correspondence between spatial rates of climate change (i.e., climate velocity) and the spatial rates of distribution shifts for plant taxa during the last deglaciation. However, alternate hypotheses for the realized niche shifts, such as evolutionary shifts, or errors in the pollen data or GCMs, cannot be fully ruled out. Regardless of the cause, the observed shifts in species realized niches had a major and negative effect on the predictive accuracy of species distribution models (SDMs), and taxa with larger niches shifts were more poorly predicted by SDMs. The predictive accuracy of SDMs across time was consistently lower than their predictive accuracy across space, but there was a correlation between the two forms of cross-validation.[54, 115] Under a scenario of a hypothetical Ice Age Ecologist charged with conserving species diversity during the last deglaciation, we showed that the variable predictive ability of SDMs during the last deglaciation negatively affected attempts to select reserves to conserve biological diversity, although SDM-based approaches outperformed simple strategies on the basis of dispersing reserves across environmental space.
Figure 5. Inferred response surface for Fraxinus for 15,000, 14,000, 13,000, and 8,000 years BP based on cross-referencing fossil pollen relative abundances for Fraxinus in eastern North America with CCSM3 paleoclimatic simulations redrawn from Ref. . Over this time period, the shape of the Fraxinus response surface changed substantially, including a decrease in the height of the mode and a shift in the mode toward wetter and warmer conditions.
Download figure to PowerPoint
The above work focused on modeling climatic controls on vegetation distributions at the level of individual taxa. An intriguing alternative solution to the no-analog challenge is to focus instead on modeling community-level phenomena, such as species interactions inferred from co-occurrence data [116, 117] or emergent community-level phenomena such as α-diversity and β-diversity.[118-120] These emergent patterns of biodiversity might be more predictable under no-analog climates than the distributions of individual species, the modeling of which may be more susceptible to truncations of species fundamental niches. If so, community-based approaches might be less sensitive to no-analog climates than species-level models. So far, comparisons of community-based versus species-level approaches using contemporary distributions of taxa have yielded mixed results.
Blois et al. used one community model, the generalized dissimilarity model (GDM), to test whether the spatial relationships between climate and plant compositional dissimilarity (as inferred from pollen data) were constant during the Pleistocene–Holocene transition. GDM is attractive for paleoecological modeling because it directly models community dissimilarity (a standard metric in paleoecological analysis) as a function of environmental distance, and incorporates the assumption that community dissimilarity accumulates monotonically but nonlinearly along environmental gradients. GDM does not assume fixed community types and hence can accommodate individualistic species-level responses to climate change. GDM has been shown to have a strong predictive capacity when compared to species-distribution models.
The analyses indicated that a key feature of late-glacial vegetation was high spatial heterogeneity of community composition (14–12 ka, Fig. 6). The full-glacial periods (21–18 ka) and Holocene (11 ka to present) each have large regions of spatial homogeneity, indicating that pollen assemblages (and presumably the source plant communities) were broadly uniform. The predictive ability of GDM was reasonably strong, explaining 32–51% of community dissimilarity in space. However, as with SDMs, predictive ability decreased as the temporal distance increased between the calibration and testing time periods. In summary, GDM suggests that community-level models offer a promising alternative to species-level models, but more work is needed to further develop community-level models and systematically compare the behavior and predictive accuracy of species- and community-level models[118, 122] under no-analog climates. Such work is critical because community-level models, like species-level models, have applications in conservation biology (e.g., filling gaps in surveys, identifying priority areas for conservation, and modeling climate effects) and because of the fundamental need to understand and accurately model species interactions under changing climates.[18, 123]
Figure 6. Maps of spatial compositional dissimilarity among fossil pollen assemblages in eastern North America during the last deglaciation, as modeled by GDM (background color). GDM simulations are based on gridded pollen abundances that were interpolated from networks of fossil pollen records (dots). Color shading indicates the mean compositional dissimilarity of each grid cell to all other grid cells within a 500-km radius, with high values indicating high spatial heterogeneity. Numbers in lower right of each map panel indicate time before present, in thousands of years (ka). Redrawn from Ref. .
Download figure to PowerPoint