BACKLAND: spatially explicit and high-resolution pollen-based BACKward LAND-cover reconstructions

Publié le 1 février 2024 Mis à jour le 1 février 2024

Dernier article co signé Clara Plancher & Florence Mazier

Studying the interactions between humans, land-cover and biodiversity is necessary for the sustainable management of socio-ecosystems and requires long-term reconstructions of past landscapes, improving the integration of slow processes. The main source of information on past vegetation is fossil pollen, but pollen data are biased by inter-taxonomic differential production and dispersal. The landscape reconstruction algorithm (LRA) approach is the most widely used to correct for these biases.

The LOVE algorithm (LOcal Vegetation estimates), the second step in the LRA approach, also estimates the spatial extent of the local vegetation reconstruction zone (the relevant source area of pollen, RSAP). While LRA estimates have already been integrated into certain past land-cover mapping approaches, none have been designed to allow the diachronic reconstruction of a land-cover mosaic over the long term combining the following points: the direct integration of LOVE estimates as a source of variability in the composition and distribution of pollen taxa, without multiple scenarios, and the integration of spatiotemporal autocorrelation in the taxa distribution between periods.

Here, we propose an innovative approach for BACKward LAND-cover reconstruction (BACKLAND), combining these points and estimating the past land-cover mosaic within a set of RSAPs. Based on three stages using parsimonious assumptions and easy-to-implement probabilistic and statistical tools, this approach requires LOVE estimates of sites close enough to each other for their RSAPs to overlap, botanical data, a digital elevation model and two recent land-cover maps. Developed and tested on a small study area within the mountain landscape of the Bassiès valley (French Pyrenees), BACKLAND achieved the reconstruction of a past land-cover map representing eight land-cover types at a spatial resolution of 20 m with a good level of accuracy.

We show in this study the originality of this approach and discuss its potential for palaeoenvironmental studies, historical ecology and the management of socio-ecosystems.

Lire l'article complet en cliquant ici