Research map / Probabilistic and scientific modeling / Molecules and proteins
SatBird: a Dataset for Bird Species Distribution Modeling using Remote Sensing and Citizen Science Data
Mélisande Teng, Amna Elmustafa, Benjamin Akera, Yoshua Bengio, Hager Radi, Hugo Larochelle, David Rolnick
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add environmental rasters to RGB image input (ablation) | ResNet-18 · SatBird-USA-summer | Top-k accuracy: 54.5 -> 66.9 | Improved |
| add environmental rasters to RGB image input (ablation) | ResNet-18 · SatBird-USA-winter | Top-k accuracy: 56.9 -> 70.9 | Improved |
| use RGBNIR reflectance instead of RGB image (no env data) (ablation) | ResNet-18 · SatBird-USA-summer | Top-10 accuracy: 35.5 -> 36.1 | Improved |
| image-only ResNet-18 compared with environmental baseline | ResNet-18 (img) · SatBird-USA-summer | Top-k accuracy: 63.8 -> 54.5 | Hurt |
| add range maps (ablation) | ResNet-18 (refl+env) · SatBird-USA-summer | Top-k accuracy: 66.6 -> 66.7 | No clear effect |
| add range maps (ablation) | ResNet-18 (refl+env) · SatBird-US-summer subset | Top-k accuracy: 72.9 -> 73.4 | Improved |
| add location encoder instead of range maps (ablation) | ResNet-18 (refl+env) · SatBird-US-summer subset | Top-k accuracy: 72.9 -> 72.0 | Hurt |
| replace ImageNet initialization with Seasonal Contrast weights (ablation) | ResNet-18 (refl+env+RM) · SatBird-USA-summer | Top-k accuracy: 66.8 -> 66.6 | No clear effect |
| replace cross entropy loss with L1 loss (ablation) | ResNet-18 (refl+env+RM) · SatBird-USA-summer | Top-k accuracy: 66.78 -> 61.13 | Hurt |
| replace cross entropy loss with focal loss (ablation) | ResNet-18 (refl+env+RM) · SatBird-USA-summer | Top-k accuracy: 66.78 -> 60.68 | Hurt |
| replace cross entropy loss with L1 loss (ablation) | ResNet-18 (refl+env+RM) · SatBird-USA-summer | MAE [1e-2]: 2.16 -> 1.86 | Improved |
| use frozen pretrained Satlas instead of ResNet-18 on image | Satlas · SatBird-USA-summer | Top-10 accuracy: 35.5 -> 30.5 | Hurt |
Closest papers by meaning
- Combining Observational Data and Language for Species Range Estimation (NeurIPS 2024 · similarity 0.88)
- Feedforward Few-shot Species Range Estimation (ICML 2025 · similarity 0.87)
- GeoPlant: Spatial Plant Species Prediction Dataset (NeurIPS 2024 · similarity 0.87)
- Spatial Clustering of Citizen Science Data Improves Downstream Species Distribution Models (AAAI 2025 · similarity 0.87)
- WildlifeMapper: Aerial Image Analysis for Multi-Species Detection and Identification (CVPR 2024 · similarity 0.85)
- BirdCollect: A Comprehensive Benchmark for Analyzing Dense Bird Flock Attributes (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
