Habitat Inference from Biodiversity Field Notes with CORINE Land Cover Labels

A 226.5 MB dataset compiled from biodiversity occurrence records with textual field notes and spatial information. It was created by Mahsa Hadikhah Mozhdehi for the paper "Habitat Inference from Biodiversity Field Notes" and last updated in June 2026. The main text column, `aggregated_text`, contains aggregated habitat descriptions for each spatial grid cell.

Use Cases

• Train machine learning models for land-cover classification based on aggregated textual habitat descriptions.
• Analyze the relationship between environmental variables and habitat types as indicated by CORINE labels.
• Develop spatial or graph-based models for habitat inference using the provided coordinate and environmental data.

Strengths

• The dataset is 226.5 MB, indicating a substantial volume of compiled information.
• It integrates multiple data sources, including textual field notes, spatial coordinates, and environmental variables.
• The target label (`Label3`) is a standardized Level-3 CORINE land-cover class.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset excludes Class 44 (sea and ocean), focusing only on terrestrial habitats.

Provenance

Source

figshare, associated with a paper submitted to KDD AI for Science track.

Collection Method

Created from biodiversity occurrence records with associated textual field notes and spatial information.

Freshness

Last updated 2026-06-03 13:11:23; freshness should be verified.

Geography

Spatial coverage is implied by coordinate columns and CORINE labels, but specific geography is not stated.