Estimating palaeoclimates and prediction errors from pollen data:

A comparison of techniques

Tibby, J. ¹, DíCosta, D.M. ², Penny, D. ¹ and Kershaw, A.P. ¹

1. Centre for Palynology and Palaeoecology, School of Geography and Environmental Science, Monash University, Vic. 3800, Australia
2. School of Environmental and Marine Sciences, University of Auckland, Private Bag 92019, Auckland, New Zealand

Kershaw and Bulman (1996) and DíCosta and Kershaw (1997), respectively, developed and expanded a pollen database for climate reconstruction in southeastern Australia (now focused on Victorian and Tasmania). The database relates "pre-European" pollen spectra to climatic parameters. We assess the utility of the database by:
 

• evaluating which climate variables appear to most influence pollen assemblages,
• comparing the performance of a number of transfer function techniques for climate reconstruction.

The BIOCLIM derived climate parameters used in the database are shown in Table 1. All temperature and precipitation parameters (with the exception of ranges) are, respectively, highly correlated with mean annual temperature (r²>0.70, p<0.005) and annual precipitation (r²>0.88, p<0.005). Statistical evaluation of the influence of climate variables on the modern pollen assemblages using Canonical Correspondence Analysis (CCA), shows that all precipitation parameters (with the exception of RCVAR) explain a similar amount of pollen variance (>15.5%). Given this, our precipitation based transfer functions focus on annual precipitation. Temperature parameters were less important in explaining taxon distribution, with annual mean temperature accounting for approximately 10% of pollen variance.

Table 1: BIOCLIM parameters generated for pollen database sites
 

TANN	Annual mean temperature
TMNCM	Minimum temperature of the coolest month
TMXWM	Maximum temperature of the warmest month
TSPAN	Annual temperature range (ie: 2 to 3)
TCLQ	Mean temperature of the coolest quarter
TWMQ	Mean temperature of the warmest quarter
TWETQ	Mean temperature of the wettest quarter
TDRYQ	Mean temperature of the driest quarter
RANN	Annual precipitation
RWEM	Precipitation of the wettest month
RDRYM	Precipitation of the driest month
RCVAR	Coefficient of variation of monthly precipitation
RWETQ	Precipitation of the wettest quarter
RDRYQ	Precipitation of the driest quarter
RCLQ	Precipitation of the coolest quarter
RWMQ	Precipitation of the warmest quarter

Much of the influence of individual variables overlaps (or "covaries"). Hence the CCA technique was extended, using variance partitioning (Borcard et al., 1992) which removes the effect of (selected) co-variables. This analysis showed that annual mean temperature explained a significant amount of variation not accounted for by annual precipitation and vice versa. Variance partitioning also showed that, in the context of strong correlations between variables, reconstructing a larger number of climate parameters is unlikely to provide significantly more (reliable) information.

Given that CCA showed that annual precipitation (RANN) and annual mean temperature (TANN) are important in explaining pollen distribution, we assessed the performance of a variety of transfer function techniques in predicting RANN and TANN in the modern data set. A variety of analog matching techniques, along with weighted averaging regression and calibration of pollen data was evaluated. Analog matching operates by attributing to a site (or fossil sample) the climate characteristics of sites with similar pollen assemblages, while weighted averaging utilises taxon optima to derived climate estimates.

There is a significant relationship between "measured" modern climate parameters (derived from BIOCLIM) and those predicted from the pollen data. For annual precipitation the best relationship between measured and predicted values (r²=0.72, p<0.005; Root Mean Squared Error=351 mm) was derived using Squared Chord Distance (Overpeck et al., 1985). For mean annual temperature, the Canberra metric performed best (measured v. predicted values, r²=0.60; p<0.005; Root Mean Squared Error=1.96° C).

Estimates of past climates from a variety of Western Victorian crater lakes have been developed using the above analogue matching methods. Firm interpretations are hampered by the lack of good analogues, a problem which tends to increase with time before present.

References

Borcard, D., Legendre, P. and Drapeau, P. (1992). Partialling out the spatial component of ecological variation. Ecology, 73, 1045-1055.

DíCosta, D.M. and Kershaw, A.P. (1997). An expanded recent pollen database from southeastern Australia and its potential for refinement of palaeoclimatic estimates. Australian Journal of Botany, 45: 583-605.

Kershaw, A.P. and Bulman, D. (1996). A preliminary application of the analogue approach to the interpretation of late Quaternary pollen spectra from southeastern Australia. Quaternary International, 33, 61-71.

Overpeck, J.T., Webb, T., III. and Prentice, I.C. (1985). Quantitative interpretation of fossil pollen spectra: dissimilarity coefficients and the method of modern analogs. Quaternary Research, 23, 87-108.