Results

Explore how models perform across species, both overall and within groups defined by sampling effort and relative prevalence. We compare a range of single-species SDMs and multi-species DeepSDMs.

Leaderboard

All models are run with five seeds, evaluated on the held-out presence–absence plots. Columns show performance across species groups split at the median of each property.

Metric

Download this table (CSV)

Compare models across species groups

Choose two models to compare their paired performance differences across the four groups defined by sampling effort and relative prevalence.

Model

Reference model

Extreme deciles Show

Scale

Paper comparisons

Evaluate your own model

Here, you can add your own model into the comparison framework above. Generate one probability per plot and species, then upload the resulting CSV. Your model is added to the leaderboard and comparison heatmap.

Write probabilities with shape (42268, 5771). Keep rows in the order of targets/eval_test_targets.h5 and columns in the order of targets/species_names.csv.
Run: python scripts/evaluate_predictions.py --predictions my_model.h5 --data-dir <data_dir> --output-dir out/
Upload out/species_metrics.csv below.

Drop species_metrics.csv here
or click to choose a file

Submit to the public leaderboard

We are happy to add your model or method to this overview! Tell us about your model, training data, paper, and reproducible code, then send the details by email.