A new methodology for reconstructing climate and vegetation from modern pollen assemblages: an example from British Columbia

Abstract

Aim We used modern pollen assemblages to develop a method for climate reconstruction that reduces the spatial autocorrelation of residuals and accounts for the strong topographic and climatic variation that occurs in British Columbia, Canada.Location British Columbia, Canada, including sites both on the mainland and on adjacent islands (Queen Charlotte Islands and Vancouver Island).Methods New pollen assemblages from surface-sediment samples collected in British Columbia were combined with other published and unpublished samples (n = 284). Multivariate rank-distances between sample sites and a randomized set of sites within the province were calculated for climate parameters to determine whether gaps in the current network of present-day pollen sample sites exist. Lacustrine surface-sediment pollen assemblages (n = 145) were ordinated using non-metric multidimensional scaling (NMDS), and a generalized additive model (GAM) was used to reconstruct modern mean warmest month temperature (MWMT) and mean annual precipitation (MAP) from the NMDS ordination. The results were compared with standard climate reconstruction techniques, including the modern analogue technique, partial least squares, weighted averaging, weighted averaging-partial least squares and factor analysis. ResultsReconstructions of MWMT and MAP using NMDS and GAM were comparable to those of existing models. When reconstructing MWMT, the NMDS/GAM method had a lower root-mean-squared error of prediction (RMSEP), lower spatial autocorrelation and higher correlation with observed temperature values than the other methods tested. When reconstructing MAP, the partial least squares method performed better than the NMDS/GAM method for RMSEP and correlation with observed values; however, the NMDS/GAM method had a lower spatial autocorrelation of residuals.Main conclusions NMDS reveals strong relationships among modern pollen assemblages, vegetation and climate parameters. Climate models using NMDS and GAM are comparable to other palaeoecological reconstruction models, but provide lower spatial autocorrelation of residuals for both parameters tested. An inverse distance-weighted surface of multivariate rank-climate distances generated from the network of pollen sample sites indicates that greater sampling intensity in north-western and central-interior British Columbia is required in order to obtain an accurate representation of climatic and vegetation diversity in the province.

Novel and Lost Forests in the Upper Midwestern United States, from New Estimates of Settlement-Era Composition, Stem Density, and Biomass

“…Many papers have used calibration datasets comprised of modern pollen samples to build transfer functions for inferring past climates and vegetation from fossil pollen records [116–119]. However, modern pollen datasets are potentially confounded by recent land-use, which can alter paleoclimatic reconstructions using pollen data [118].…”

Strengths and Weaknesses of Quantitative Climate Reconstructions Based on Late-Quaternary Biological Proxies

“…Vasko et al 2000;Toivonen et al 2001;Korhola et al 2002;Gersonde et al 2005;Haslett et al 2006;Holden et al 2008;Hübener et al 2008;Goring et al 2009;Velle et al 2011), we are probably close to the resolution of current data and methods. Decomposition of the variance in model RMSEP shows that the largest part (75% or more) is due to the fact that at a given environmental value today, there is still siteto-site variation in modern assemblages (Birks et al 1990).…”

Pollen assemblage richness does not reflect regional plant species richness: a cautionary tale

“…Pollen data were obtained from the British Columbia Modern Pollen Database (BCMPD, Goring et al 2009Goring et al , 2010. This data set includes 546 lacustrine pollen sample sites from British Columbia, Canada, and Washington, Oregon, Idaho and Montana in the USA, of which 167 points are located within Canada.…”

“…1), explained largely by differential investigator effort and purpose. Limitations of the distribution of pollen data are discussed further in Goring et al (2009). Taxonomy for each data set was standardized to be internally consistent to the species level, excluding hybrid species in the BC Plant data set, to the pollen taxonomy of Whitmore et al (2005).…”