OpenTIR-PINN Benchmark v1.1.0: Multi-Atmospheric Radiative Transfer and Double-Quarantine Golden Target Dataset for Ill-Posed Temperature and Emissivity Separation
Authors/Creators
Description
# OpenTIR-PINN Benchmark (Version 1.1.0)
Comprehensive open-science benchmark dataset for Physics-Informed Neural Networks (PINN), radiative transfer inversion, and operational Temperature and Emissivity Separation (TES) algorithms across multispectral (ASTER, 5 bands) and hyperspectral (HyTES, 256 bands) thermal infrared sensors.
### Key Upgrades & Erratum in Version 1.1.0:
1. Complete Radiative Transfer Coupling: All simulated radiances strictly satisfy the full surface-leaving boundary condition L_surf = epsilon*B(T) + (1-epsilon)*L_down/pi and Top-of-Atmosphere coupling L_TOA = tau*L_surf + L_up.
2. 30 Distinct MODTRAN 5.2 Profiles: Full parameter sweep across 6 standard atmospheric profiles (Tropical, Mid-Latitude Summer, Mid-Latitude Winter, Sub-Arctic Summer, Sub-Arctic Winter, 1976 US Standard) and 5 water vapor scales (0.50x to 2.00x, PWV from 0.21 to 8.23 g/cm²).
3. Dual-Input Architectural Telemetry: Datasets now supply both [L_surf, L_down] enabling real-world atmospheric ingestion.
4. Rigorous Double-Quarantine Suite: Immutable benchmark featuring 7 Golden Targets tested across In-Distribution Atmospheres (Quadrant II, N=308) and Severe Winter Out-of-Distribution Atmospheres (Quadrant IV, N=770), with Target #7 properly documented as Non-Photosynthetic Vegetation (Abies concolor White-Fir Bark).
5. Open Reproducibility: Includes 5-Fold stratified cross-validation splits, full laboratory FTIR continuous inventories (1,205 spectra), and a standalone 7-gate Python verification test suite (quickstart_demo.py).
Files
Additional details
Related works
- Is supplement to
- Software: https://github.com/atilioebg/TIR_PINN (URL)
- Dataset: 10.5281/zenodo.23571365 (DOI)
Software
- Repository URL
- https://github.com/atilioebg/TIR_PINN
- Programming language
- Python
- Development Status
- Active