Testing the Landscape Reconstruction Algorithm for spatially explicit reconstruction of vegetation in northern Michigan and Wisconsin

Abstract

AbstractThe Landscape Reconstruction Algorithm (LRA) overcomes some of the fundamental problems in pollen analysis for quantitative reconstruction of vegetation. LRA first uses the REVEALS model to estimate regional vegetation using pollen data from large sites and then the LOVE model to estimate vegetation composition within the relevant source area of pollen (RSAP) at small sites by subtracting the background pollen estimated from the regional vegetation composition. This study tests LRA using training data from forest hollows in northern Michigan (35 sites) and northwestern Wisconsin (43 sites). In northern Michigan, surface pollen from 152-ha and 332-ha lakes is used for REVEALS. Because of the lack of pollen data from large lakes in northwestern Wisconsin, we use pollen from 21 hollows randomly selected from the 43 sites for REVEALS. RSAP indirectly estimated by LRA is comparable to the expected value in each region. A regression analysis and permutation test validate that the LRA-based vegetation reconstruction is significantly more accurate than pollen percentages alone in both regions. Even though the site selection in northwestern Wisconsin is not ideal, the results are robust. The LRA is a significant step forward in quantitative reconstruction of vegetation.