EUNIS habitat maps: enhancing thematic and spatial resolution for Europe through machine learning

Area of Science:

• Ecology
• Environmental Science
• Geospatial Analysis

Background:

• The EUNIS habitat classification is vital for European nature conservation policies, including the Nature Restoration Law.
• Accurate and detailed habitat data is increasingly needed for effective conservation and restoration planning.
• Existing habitat maps may lack the resolution or detail required for current policy needs.

Purpose of the Study:

• To generate high-resolution spatial predictions for 260 EUNIS habitat types across the EEA39 territory.
• To provide independent validation and uncertainty analyses for the generated habitat maps.
• To create a valuable resource for European conservation and restoration initiatives.

Main Methods:

• Utilized ensemble machine learning models trained on high-resolution satellite imagery.
• Incorporated ecologically relevant climatic, topographic, and edaphic variables.
• Performed spatial block cross-validation and external validation using independent datasets.

Main Results:

• Produced a 100-m resolution European habitat map identifying the most probable habitat type.
• Quantified prediction uncertainty and identified probable habitats within EUNIS level 1 formations.
• Achieved strong predictive performance, with F1-scores up to 0.95 in external validation.
• Demonstrated improved accuracy for rare or localized habitats when considering top predictions.

Conclusions:

• The developed European habitat map offers unprecedented detail and accuracy for conservation and restoration.
• The spatial predictions and uncertainty analyses are crucial tools for implementing European environmental policies.
• This resource supports evidence-based decision-making for biodiversity protection and habitat management.