Refining data–data and data–model vegetation comparisons using the Earth mover's distance (EMD)

Abstract. Comparing temporal and spatial vegetation changes between reconstructions or model simulations requires carefully selecting an appropriate evaluation metric. A common way of comparing reconstructed simulated involves measuring the agreement pollen- model-derived unary estimates, such as biome plant functional type (PFT) with highest affinity scores. While this approach based on summarising signal into estimates performs well in general, it overlooks details underlying structure. However, data structure can influence conclusions since minor variations pollen percentages modify which PFT has score (i.e. estimate). To overcome limitation, we propose using Earth mover's distance (EMD) to quantify mismatch distributions The EMD circumvents issue by considering entire range scores calculate a compared entities. In addition, each be given specific weight account for case-specific ecological distances or, said differently, fact that reconstructing temperate forest instead boreal is ecologically more coherent than desert. We also introduce two EMD-based statistical tests determine (1) if similarity samples significantly better random association particular context (2) pairing datasets might expected chance. illustrate potential advantages comparison studies, reproduce different case studies previously published Europe capitalise refine interpretations past highlighting flickering give impression high instability, correspond gradual low values contiguous (case study 1). data–model comparisons five time slices identify those are statistically robust while accounting sample 2). included paleotools R package (https://github.com/mchevalier2/paleotools, last access: 3 May 2023).