portal.geobon.org/ebv-detail?id=84
European vertebrates suitability maps
2024-10-23
Sara Si
Moussi
CNRS - Université Grenoble Alpes
France
sara.si-moussi@univ-grenoble-alpes.fr
Predicted habitat suitability of most terrestrial vertebrate species (1,207 species) in Europe at 1km resolution under current conditions (1990-2020). Predictions for each species were obtained from an ensemble of species distribution models (ensemble SDMs). The provided data are the raw weighted mean of habitat suitability through the ensemble SDMs, the rescaled weighted mean of habitat suitability through the ensemble SDMs but spatially constrained by expert range maps, the committee averaging score (% of predicted presence across models), and the coefficient of variation of predicted habitat suitability across models.
N.A
NaturaConnect
https://naturaconnect.eu/
Sara
Si-Moussi
Centre National de la Recherche Scientifique (CNRS)
sara.si-moussi@univ-grenoble-alpes.fr
Wilfried
Thuiller
https://creativecommons.org/licenses/by/4.0
Each vertebrate species was modeled using an ensemble of machine learning algorithms. These algorithms were fitted on curated presence-only data from GBIF with various sets of pseudo-absences, in function of selected environmental data including climate (CHELSA), soil (Soilgrids), terrain (EU-DEM), hydrography (EU-Hydro) and land use (Dou et al 2021) variables. In total, we used 3 algorithms (Random Forest, XGBoost, Neural Networks), 5 pseudo-absence repetitions and 5 spatial cross-validation partitions (i.e., 75 models). We kept only the models that reached a TSS > 0.4 on the cross-validation set. From this, we provide different ensemble predictions. <br />\r\n<br />\r\n1- The weighted mean habitat suitability, i.e., the mean probability of occurrence across the models weighted by their TSS predictive accuracy. This map then shows the potential suitable areas across the whole extent.<br />\r\n<br />\r\n2- Coefficient of variation of suitability scores. This map shows where predictions tend to diverge or converge. <br />\r\n<br />\r\n3- The mean probability of occurrence across the models weighted by their TSS predictive accuracy but rescaled with a spatial constraint layer derived from a probabilistic combination of IUCN / Bird Life expert range maps or known occurrence records. This map is then close to the ‘realized’ distribution of the species since it is spatially constrained by expert range maps. <br />\r\n<br />\r\n4- Committee averaging presence, which represents the percentage of time the species is predicted to be present across the 75 predictions (after having binarized the probability of presence into presence-absence using a threshold maximising the TSS). This map shows the agreement and disagreement of the predictions across the algorithms-pseudoabsence sets-cross-validation sets. Values close to 0 means that all predictions converge to predict the absence of the species, values close to 1 means the way around. <br />\r\n<br />\r\n
Sampled P0000-00-00. The Environment domain covers Terrestrial.
Terrestrial Vertebrates
Europe Union
Species Distributions
Predicted Distributions
Terrestrial
The general content type of the resource (referring ISO codeList) is modelResult.
moussi_spepop_id84_20241023.nc
59848249601
NetCDF
portal.geobon.org/data/upload/84/moussi_spepop_id84_20241023.nc
Continental/Regional - Europe. Spatial resolution/unit: 1000 meter. Coordinate Reference System (CRS): EPSG:3035
723000.0
7700000.0
6615000.0
160000.0
2020-01-01
2020-01-01
N/A
N/A
Species. Scope: Vertebrates species including: amphibians (117), birds (529), mammals (296), reptiles (268)
Weighted Mean habitat suitability
Weighted mean of suitability scores excluding low-performing models of the ensemble (multiplied by a scale factor of 10000).
Probability (x10000)
Uncertainty
Coefficient of variation of suitability scores excluding low-performing models of the ensemble (multiplied by a scale factor of 100).
% (x100)
Occurrence probability
Occurrence probability score obtained by constraining habitat suitability with expert range maps.
Probability
Committee averaging
Average of binary suitability scores excluding low-performing models of the ensemble. Binary scores are obtained from the mean suitability score by applying an optimal TSS threshold for each model (multiplied by a scale factor of 10000).
Probability (x10000)
Species populations
Species distributions