By Frederik Van Daele, Thierry Onkelinx, Kris Verheyen, Hans Van Calster, Maud Raman, Jasper Van Ruijven & Luc De Keersmaeker
Species-rich Nardus grassland is a priority habitat for conservation and restoration in the European habitats directive, as it has seriously declined in the past century. The causes of this decline are eutrophication, acidification, fragmentation, and changes in management. Reinstating the original management practices, i.e. mowing or grazing, is often insufficient and measures to restore the abiotic and biotic conditions can be necessary. As ecological restoration can be very costly or can have negative side effects, careful selection of sites with a high restoration potential is essential.
Previous research mostly focussed on two types of variables to explain the quality of Nardus grassland: chemical soil variables and historical landscape or management characteristics. Both are sampled in a very distinct way: chemical soil variables always require a field survey, whereas land-use or management characteristics are quantified using old maps or archive data, that can be disclosed in a GIS environment. We compared the statistical complementarity of three statistical models, using either chemical soil variables, historical land-use metrics, or both, to predict the potential number of Nardus indicator species in Flanders (northern Belgium).
Our results indicated that historical land use variables can generate a restoration potential map that discerns a high-priority zone with high accuracy but with relatively low sensitivity. To further increase the overall accuracy of the high-priority zone and detect additional areas with high restoration potential, soil characteristics are necessary. Consulting the habitat suitability map as a first step can considerably reduce the search area for a field survey aiming to collect soil samples.
To illustrate this approach, we manually digitized the land-use history of a Natura 2000 area and created a Nardus grassland habitat suitability map for this test site. Since manual digitization of land use is time-consuming, we are now exploring machine learning techniques to do the same for the whole territory, based on historical maps shared by the Cartesius platform (https://www.cartesius.be/CartesiusPortal/#). The outcome of this digitization will further increase the potential of desktop analyses of historical land-use patterns, not only to create a suitability map for Nardus grassland, but also to support other desktop analyses for target habitat or species.
This is a plain language summary of the paper by Van Daele et al. published in Applied Vegetation Science (https://doi.org/10.1111/avsc.12681). This post was prepared by Luc De Keersmaeker.