Reconstructing land cover from pollen assemblages from small lakes in Denmark

A calibration data set of AD 1800 pollen assemblages and land cover from 30 lake sites (3.5-33 ha) in Denmark is analysed. Canonical ordination (RDA) shows that the distance-weighted land cover of woodland, heathland and open field explain the highest proportion of the variation in the pollen data. Four different distance-weighting functions, 1, 1/d, 1/d2 and an approximation of the Sugita model of pollen dispersal, are applied to the land-cover data, and the 1/d function results in the strongest correlation between distance-weighted land cover and pollen proportions. Partial least squares (PLS) calibration is applied, and land cover around nine test sites (not included in the calibration data set) is reconstructed from pollen assemblages. The Extended R-value (ERV) model is applied to the pollen proportions and distance-weighted plant abundance inferred from the maps. Pollen productivity and background components are estimated from the calibration data set and used to reconstruct plant abundance around the test sites. The reconstructed land cover and plant abundance are compared with historical maps, and the performance of the two models is compared. Root-mean-squared errors for the two models are within the same ranges but tend to be slightly lower for the ERV model. Division of the data set into two regional subsets does not improve the reconstructions.