Quantitative Paleoclimatic Reconstructions Based On Pollen Analysis

Paleoclimatic reconstruction from fossil pollen spectra is based on the notion that, as vegetation distribution is largely determined by climate, it should be possible to use that distribution (as represented in the fossil pollen spectra) to reconstruct past climate. Large databases of surface pollen samples (generally from the surface sediments of lakes) are now available for many parts of the world (see Appendix B) and these have enabled pollen assemblages to be calibrated directly in terms of climate. Although in many regions natural vegetation has been dramatically reduced, the broad-scale relationships between pollen rain and climate are sufficiently robust that they can be used to make reliable paleoclimatic reconstructions. This has been demonstrated several times (Bartlein et al., 1984; Huntley, 1990b; Huntley and Prentice,

1993), but Guiot (1990) believes this factor (the reduction in natural vegetation) creates a great deal of noise in paleoclimate estimates (see discussion that follows).

The simplest approach to determining a relationship between modern pollen rain assemblages and contemporary climate is through multiple linear regression, such that:

where Cm is the modern climatic data, Pm is the modern pollen rain, and Tm is a functional coefficient or set of coefficients ("transfer functions") derived from the relationship between modern climate and pollen data. Former climatic conditions (C^) are then derived by using the fossil pollen assemblage (P^ and the modern transfer function (Tm). In studies of pollen-climate relationships in eastern North America the equations were derived for different regions, defined in such a way that a "clear and monotonic" relationship could be recognized between climate and pollen percentages (generally using a pollen sum of the major forest taxa, plus Cyperaceae and prairie forbs [herbs]) (Bartlein and Webb, 1985). Those species associated with human settlement (e.g., Ambrosia) were eliminated to minimize anthropogenic influences on the pollen sum.

Scatter plots of pollen percentages and climate variables (e.g., % oak and July mean temperature across a region) commonly show nonlinear relationships between the pollen percentage and the climate variable. Such non-linearities can often be resolved by transforming the pollen data by some power function (Fig. 9.11). The transformed data are then used in developing an equation in which climate is the dependent variable and pollen percentages are the independent (predictor) variables. Thus, for the New England region of the United States, the following equation was constructed (R2= 0.77) (Bartlein and Webb, 1985):

July Tmean (°C) = 17.76 + 1.73(Quercus)0-25 + 0.09(Juniperus) + 0.51(Tsuga)0-25 - 0.41(Pinus)°-5 - 0.12(Acer) - 0.04 (Fagus)

Using this approach, Bartlein and Webb (1985) estimated that at 6 ka B.P. July temperatures over the north-central and eastern U.S. and southern Canada were 1-2 °C warmer than modern temperatures. Using a very similar approach, Huntley and Prentice (1988) estimated that July temperatures in central and southern Europe were as much as 4 °C warmer at 6 ka B.P. as compared to modern temperatures. The assumptions underlying this method (both ecological and statistical) are discussed at length in Howe and Webb (1983). Most important is the key uniformitar-ian assumption that any changes seen in pollen percentages of the past can be interpreted in terms of modern climate-pollen relationships. Indeed, this assumption underlies almost all paleoclimatic research involving calibration of paleo-records by means of modern data. One must also accept that paleoclimatic reconstructions based on transfer functions are only reliable if the modern calibration data set is extensive enough to be representative of all (or nearly all) conditions occurring in the past, otherwise unwarranted extrapolations may be made. Hence, transfer functions will not help to explain in climatic terms fossil pollen assemblages in the past that bear no relation to modern experience (the no-analog situations).

Pollen Transfer Function

20.0 40.0 60.0 80.0

% Quercus

20.0 40.0 60.0 80.0

% Quercus

Pollen Paleoclimatology

> Prairie Forbs

FIGURE 9.1 I Scatter diagrams for (A) July mean temperature vs the percentages of Quercus (oak) pollen; (B) July mean temperature vs the percentages of Quercus pollen raised to the 0.25 power; (C) annual precipitation vs the percent of prairie-forb pollen (excluding Ambrosia) and (D) annual precipitation vs the percent of prairie-forb pollen raised to the 0.5 power (Bartlein et al„ 1984).

FIGURE 9.1 I Scatter diagrams for (A) July mean temperature vs the percentages of Quercus (oak) pollen; (B) July mean temperature vs the percentages of Quercus pollen raised to the 0.25 power; (C) annual precipitation vs the percent of prairie-forb pollen (excluding Ambrosia) and (D) annual precipitation vs the percent of prairie-forb pollen raised to the 0.5 power (Bartlein et al„ 1984).

The no-analog problem was encountered by Overpeck et al. (1985), who quantified the relationship between modern pollen rain and fossil pollen spectra by calculating "dissimilarity coefficients" for each level in a pollen diagram (Prell, 1985; Bartlein and Whitlock, 1993). By matching the locations that were most similar (or least dissimilar) to a given pollen spectra in the past, they were able to characterize Holocene vegetation change in the eastern U.S. in terms of modern analogs, thereby making deductions about former climatic conditions. This approach worked well for most Holocene samples, but for the period from -11-9 ka in the Upper Midwest, no close analogs could be found in the modern pollen rain. This was probably because climate at that time was rapidly changing and individual species responded to those changes at different rates, creating transient ecosystems not observed today.

Another mutivariate approach to paleoclimatic reconstruction involves the computation of "response surfaces." Pollen distribution is considered as occupying "climate space" defined by a three-dimensional (3D) array of climate variables.

Figure 9.12a-c illustrates the concept, beginning with simple bivariate graphs, relating spruce (Picea) pollen percentages to mean January and July temperatures, and to annual precipitation. The graphs are based on over 1000 modern pollen samples from across northern North America and Greenland and climatic data from the nearest weather stations (Anderson et al., 1991). Clearly, it would be difficult to use any one of these graphs alone in interpreting pollen data. By plotting spruce pollen percentages in relation to both January and July mean temperatures (Fig. 9.12d) a more coherent picture emerges with the highest pollen percentages associated with mean July temperatures of 12-15 °C and mean January temperatures of -17 °C to -20 °C. Similarly, maximum spruce pollen percentages are found where annual precipitation is around 900 mm and July temperatures are 11-13 °C. Spruce pollen percentages thus vary within the 3D "climate space" defined by the two monthly temperatures and annual precipitation. This can be envisaged by considering how pollen percentages vary

100 90

Mean January Temperature (°C)
















0 320 640 980 1280 1600

Total Annual Precipitation (mm)

Disney Character Silhouettes Mickey Head
Mean July Temperature (oC)

Mean January Temperature (°C)

Mean January Temperature (°C)

FIGURE 9.12 Scatter diagrams and response surfaces of Picea (spruce) pollen percentages and (a) mean January temperature, (b) mean July temperature, (c) total annual precipitation, and (d) mean July temperature and mean January temperature (Anderson et al., 1991).

in relation to both January and July temperatures, as annual precipitation changes along a third axis — represented in Fig. 9.13 as a series of slices at selected precipitation intervals. Spruce pollen is highest in areas with July temperatures of 10-13 °C, January temperatures of-12 to -18 °C and annual precipitation of 880-1600 mm. These figures show schematically that the pollen percentage data define a body of variable density (i.e., the % data) suspended in climate space (Prentice et al., 1991).

Response surfaces are mathematical expressions of these relationships, obtained by locally weighted regression techniques. Each pollen taxa is described in this way so that with a set of several equations it is possible to define the combination of climatic conditions to which a given pollen spectra (made up of many different pollen types) corresponds. This is made clear by considering what climatic conditions might be represented (in a fossil pollen spectra) by, say, 20% spruce. Clearly, there are many possible combinations of climate variables where 20% spruce pollen would be

Pollen Analysis Paleoclimate Proxies Pollen

Mean January Temperature (°C)

FIGURE 9.13 Scatter diagrams and response surfaces of P/cea (spruce) pollen percentages as a function of mean July and January temperatures for four ranges of annual precipitation: 0-400 mm (top left); 400-640 mm (top right); 640-880 mm (bottom left); and 880-1600 mm (bottom right) (Anderson et al., 1991).

expected (Fig. 9.13d). But those options would be more limited if the spectra had 20% spruce and 10% pine (which has its own constraints in climate space), and as more pollen types are added, each one uniquely delimited in climate space, the number of possible options becomes more and more limited. Eventually, the appropriate climatic conditions corresponding to the combination of pollen types is isolated, or at least the range of options is minimized, and the centroid of the final climate space, so defined, could be taken as the best estimate of the climate.

Prentice et al. (1991) demonstrate that, in eastern North America, at least 6 major pollen taxa (spruce, birch, northern pines, southern pines, oak, and prairie forbs) are needed to avoid non-unique or indeterminate solutions. These six types are sufficient to define past climatic conditions at fossil pollen sites across eastern North America (Fig. 9.14) although it must be recognized that not all values on the map are as reliable as others; this depends on the goodness of fit of the response surface for the modern climate-pollen relationship and the "size" of the area in climate space to which the fossil pollen spectra correspond. The maps would be improved if some indication of relative error or reliability was represented (in time and space). Using more taxa should decrease the uncertainty in the paleoclimatic estimates (T. Webb et al., 1993b, who used 14 taxa). In the example given, pollen rain is assumed to reflect January and July mean temperatures and annual precipitation. However, other parameters such as soil moisture or an index of continentality might be more discriminating variables. For example, R. Webb et al. (1993) used a soil moisture index to define response surfaces, then reconstructed soil moisture changes over the northeastern U.S. since 12 ka B.P. Changes in precipitation were obtained from another set of response functions. This analysis revealed that although precipitation was lowest at 12 ka B.P., effective soil moisture was lowest at 9 ka B.P., a time when pine became most abundant in this region. The reconstruction is supported by lake-level data that point to the early Holocene as a time of significantly drier conditions.

Guiot (1987) argues that in many areas (such as in Europe) the variable topography and a long history of human impact make it very difficult to capture, in the contemporary pollen rain, all the information necessary to interpret fossil pollen spectra. These problems introduce considerable noise into paleoclimate estimates. Furthermore, the response of vegetation to a given change in climate will not always be represented in the pollen spectra in the same way; it will depend on the preceding vegetation state. There will be a certain autocorrelation in the pollen series that is not accommodated by the transfer or response function approaches (though Webb et al., 1987 and Prentice et al., 1991, implicitly take this into account by reconstructing climate only at 3 ka intervals).

Guiot proposes a method to reduce the variability (noise) in a paleoclimate reconstruction based on transfer or response functions (which he terms "analog climates"). First, the changes in fossil pollen spectra from a limited region are analyzed to extract those variations that are common to the different locations. This is achieved by a method akin to principal components analysis. The main component might account for, say, 80% of the common variance in two or more records and it is assumed that this common signal (termed the "paleobioclimate") represents the overriding influence of climatic variations on vegetation changes at all the

9,000 6,000 3,000 0 yr BP



FIGURE 9.14 Climatic conditions at 3 ka intervals from 18 ka B.P. to the present inferred from fossil pollen and the modern abundances of six pollen types (spruce, birch, northern pines, southern pines, oak, and prairie forbs) using response surfaces.The blank area is the Laurentide ice sheet (Prentice et al., 1991).

Paleoclimatology Pollen

9,000 6,000 3,000 0 yr BP




FIGURE 9.14 Climatic conditions at 3 ka intervals from 18 ka B.P. to the present inferred from fossil pollen and the modern abundances of six pollen types (spruce, birch, northern pines, southern pines, oak, and prairie forbs) using response surfaces.The blank area is the Laurentide ice sheet (Prentice et al., 1991).

sites. This series is then used in identifying the most appropriate modern analog climate, based on a measure of similarity between fossil and modern pollen spectra (Guiot et al., 1989). By reducing what is assumed to be non-climatic noise in the original series, the final paleoclimate estimates have a larger climate signal-to-noise ratio than would otherwise be obtained. Using this approach, Guiot et al. (1989) were able to estimate mean annual temperature and precipitation variations over the last 140,000 yr at two sites in France (Fig. 9.15). These show temperatures similar to or slightly higher than Holocene levels at the height of the last interglacial in Europe (the Eemian); two subsequent interstadials (St. Germain I and II) were almost as warm. Extremely cold and dry conditions were first experienced around -65 ka B.P. Three or four other cold, dry intervals occurred in the main (Wiirm/ Weichselian) glacial period, followed by a change to much warmer and wetter conditions in the Holocene. These reconstructions, and their limitations, are further discussed in Section 9.7.1.

One way of examining the coherency of paleoclimatic estimates quantitatively was suggested by Webb et al. (1987). They used the NCAR community climate model (CCM1) simulations of past climate (at 3 ka intervals, from 18 ka to 3 ka B.P.) to "predict" the pollen rain expected at those times in the past, using simulated climatic conditions applied to the response surface equations derived from modern calibration studies. These predictions of the expected pollen rain can then be






Les Echets Pollen

Continue reading here: Lower Pleni Glacial

Was this article helpful?

0 0


  • Ky
    How is paleoclimate reconstruction data represented?
    3 years ago
  • amber
    How can pollen analysis be used in the reconstruction of climatic conditions?
    3 years ago