|| Checking for direct PDF access through Ovid
Can the model performance of the landscape reconstruction algorithm (LRA) for small forest hollows be validated through comparison to inventory-based vegetation reconstructions from the last 150 yrs? Does the application of LRA and the comparison to historical data enhance interpretation of the pollen record?Denmark. The Gribskov-Ostrup small forest hollow (56°N, 12°20′ E, 44 m a.s.l.) in the forest of Gribskov, eastern Denmark.Pollen analysis was carried out on a small forest hollow, and LRA used to derive pollen-based quantitative estimates of past vegetation. Historical forest inventory data and maps were used to reconstruct the vegetation within three different circles around the hollow (20, 50 and 200 m ring widths) for five time periods during the last 150 yrs. The results of the two approaches were compared in order to evaluate model performance, and the LRA-based reconstruction used to describe how the model changes interpretation of vegetation development during the last ca. 6500 yrs compared to the use of pollen percentages alone.Distance-weighted inventory-based reconstructions within 200 m of the hollow's edge provide the best match with the LRA-modelled vegetation. Precise validation of the model is not possible due to insufficient historical data, but the comparison indicates that the LRA reconstruction for Gribskov tends to (1) underestimate tree cover and overestimate open areas, (2) give a too high representation of on-site pollen types, (3) give an underestimation of Fagus and (4) a small overestimation of Quercus and Corylus. Despite these uncertainties, application of the LRA model shows a higher degree of openness than would be apparent from the uncorrected pollen diagram, and makes it possible to attempt to distinguish changes at the local scale from regional vegetation changes, thus giving a clearer picture of the vegetation changes at the site.We demonstrate that the estimates of the LRA model applied to pollen data from small forest hollows can be compared with small-scale historical data to evaluate model performance.