Combining remote sensing and ancillary data to improve species
distribution models
Jessica Delangre*, Julien Radoux**, Floriane Jacquemin*, Marc Dufrêne*
*University of Liège, Gembloux Agro-Bio Tech, Biodiversity and Landscape Unit **Université Catholique de Louvain, Earth and Life Institute, Environmental Sciences
In the Lifewatch-WB project, a database combining segmentation in homogeneous landscape units (called « ecotopes ») and environmental attributes derived from regularly updated remote sensing data (quantitative land cover attributes, solar energy,…) and other data sources (climate, topography,…) has been designed.
Species distribution modelling: a data hungry but useful tool for biological conservation
Species distribution model Species occurrences Environmental attributes
Climate Topography Land cover Soil
properties Hydrology … Invasive species management Habitat restoration Network design Wild fauna monitoring Natural reserve selection Identification of biodiversity hotspots
Objectives Study species
Study area: Wallonia
Three different databases were tested: the ecotopes, a grid with the same environmental variables and a categorical land cover database (COSW).
Lycaena dispar Lycaena helle Boloria dia Podarcis muralis
Turdus pilaris Anthus pratensis
Euphydryas aurinia Barbier Y. Hernàndez V. Bohdal J. Ciconia nigra Daly S. Fichefet V. Imbaud C. Dufrêne M. Smits Q. Fichefet V. Meles meles Smits Q. Zootoca vivipara Methods ≠ ≠
For each species, 5 modelling algorithms were tested. Model performance was assessed by 5-fold cross-validation using the Area Under the ROC Curve (AUC), and the best models obtained with each of the 3 data sources were compared.
New variables derived from ancillary data were added when they significantly improved the AUC.
Ecotopes Grille COSW
Results
No significant differences (p>0.05) between ecotopes, grid and COSW
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 A UC COSW Ecotopes Grid
However, ecotopes are better than COSW for species with relatively small sample sizes (<200 occurrences)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 A UC
Lifewatch attributes only
Lifewatch attributes + ancillary data
New variables improve model quality for 6 of 10 species
Distance to hydrological network Terrain roughness (MNT ERRUISSOL) Soil texture (CNSW) Soil drainage (CNSW) Soil marginality
• The ecotopes database provided acceptable to good model quality (AUC>0.7) for all species
• The quantitative land cover attributes of the ecotopes allow species distribution modelling with
relatively small sample sizes (<200 occurrences) while a land cover classification fails
• Attributes related to soil, hydrology and roughness have been integrated in the ecotope database
• To assess the usefulness of « ecotope »
delineation and descriptors for species distribution modelling
• To propose further improvements of the
« ecotope » database