NUMERICAL ANALYSIS OF POLLEN SURFACE SAMPLE DATA FROM WESTERN NORTH AMERICA

The reconstruction of past environments from palynological assemblages traditionally follows a sequence of the interpretation of past vegetation followed by the estimation of climate change - typically annual temperature or precipitation. Numerical techniques may follow the same procedure (the "biome approach"), or they may directly estimate past climate from palynological data (the "analog approach"). We have analyzed a dataset of 1367 contemporary pollen samples from western North America (Davis, 1995) using both methods. This database includes metadata of site, latitude, longitude, elevation, and vegetation type, in addition to mean precipitation (mm) and mean annual temperature (°C). These data were analyzed using multiple regression techniques. In this dataset, pollen values (analog approach) explain more variance in the precipitation and temperature data than does vegetation type (biome approach). The analog approach accounts for 60% of the variance for temperature and 57% of the variance for precipitation, compared to 39% and 35% of the variance for the biome approach. Both estimates are much improved when study site is included as a predictor. When site is included, the analog approach accounts for 69% of the variance in temperature and 87% of the variance in precipitation; the biome approach accounts for 70% of the variance in temperature and 91% of the variance in precipitation. We attribute this dramatic increase to two factors - first the site-specific biotic and physical environmental factors that are unique to each study region; and second, the unique pollen analytic techniques used by each author. Standardization of the pollen nomenclature for eurypalynous taxa, as done in this study, likely combines important regional differences in the pollen flora, as well as differences in the identified taxa by different investigators.