Virtual Species Data for Evaluating Variable Importance in Species Distribution Models

A 3.3 GB dataset by He Wu, last updated in April 2026, containing virtual species data and intermediate results. The project developed a sample perturbations diagnostic framework to assess the accuracy and stability of variable importance inference in species distribution models (SDMs). It includes code for drawing and calculation, and establishes different scenarios with metrics like Driver Rank Accuracy (DRA) and Change Variable Importance (|∆VI|).

Use Cases

• Benchmarking the stability of variable importance rankings across different SDM algorithms based on the described diagnostic framework.
• Analyzing the impact of individual occurrence records on model inference using the sample perturbation methodology.
• Investigating underlying mechanisms for instability in variable importance inference as described in the project.
• Testing new SDM algorithms against established metrics like Driver Rank Accuracy (DRA) and Change Variable Importance (|∆VI|).

Strengths

• Dataset size is 3.3 GB, indicating substantial computational experiments and results.
• Includes both raw virtual species data and intermediate results for reproducibility.
• Project framework establishes specific evaluation metrics like DRA and |∆VI| for systematic assessment.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is based on virtual species, which may not fully capture complexities of real-world ecological data.

Provenance

Source

figshare

Collection Method

Generated via computational experiments with virtual species and a sample perturbations diagnostic framework.

Freshness

Last updated 2026-04-13 15:12:58