Zhang et al. (2026) Physics-Informed Liquid Neural Network Emulator for CRTM with Atmospheric-Layer Jacobian Capability

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-09-27
• Authors: Feng Zhang, Changyong Cao, Yong Chen, Xi Shao, Tung-Chang Liu
• DOI: 10.3390/rs18193325

Research Groups

[Not explicitly stated in the provided text, but inferred to be groups working in atmospheric science, remote sensing, numerical weather prediction, and machine learning for Earth system applications.]

Short Summary

This study develops a physics-informed Liquid Neural Network (LNN) emulator for the Community Radiative Transfer Model (CRTM) to address increasing computational demands. The CRTM-LNN significantly accelerates forward calculations (up to 18-fold) while maintaining high accuracy and producing more physically consistent Jacobians compared to conventional emulators.

Objective

• To develop and evaluate a physics-informed Liquid Neural Network (LNN) emulator for the Community Radiative Transfer Model (CRTM) that improves computational efficiency, maintains accuracy, and produces physically consistent Jacobians for applications like satellite data assimilation and atmospheric retrievals.

Study Configuration

• Spatial Scale: Global, with regional analysis for biases. Vertical resolution is based on the European Centre for Medium-Range Weather 91-layer (ECMWF91L) grid.
• Temporal Scale: The study evaluates the emulator's performance using instantaneous atmospheric profiles and observations, focusing on the computational efficiency of the forward model rather than long-term temporal dynamics.

Methodology and Data

• Models used:
  • Community Radiative Transfer Model (CRTM): Used as the reference model for emulation.
  • CRTM-LNN: A physics-informed Liquid Neural Network emulator, combining an Ordinary Differential Equation (ODE)-inspired LNN for layer-by-layer optical-depth modeling with an analytic, differentiable radiative-transfer solver.
  • Conventional Multilayer Perceptron (MLP) CRTM emulator: Used for comparative analysis of Jacobian quality.
• Data sources:
  • Infrared Atmospheric Sounding Interferometer (IASI) observations.
  • ECMWF91L forecast profiles (atmospheric temperature, humidity, and trace gas profiles).

Main Results

• CRTM-LNN closely reproduces reference CRTM simulations.
• Global, channel-mean brightness-temperature biases are below 0.1 K in magnitude.
• Regional bias magnitudes reach approximately 0.2–0.4 K for selected channels and latitude bands.
• Correlations with CRTM exceed 0.97 for cumulative optical depth, transmittance, and weighting functions in regimes with cumulative optical depth below five.
• Accelerates forward calculations by up to 18-fold compared to the reference CRTM.
• Efficiently generates Jacobians through automatic differentiation.
• Produces smoother and more physically consistent Jacobians compared to a conventional multilayer perceptron CRTM emulator, with improved vertical localization and fewer spurious oscillations.
• Absolute centroid-pressure errors for temperature Jacobians are reduced across all evaluated channels, ranging from 1.77 to 5.01 hPa for CRTM-LNN versus 5.88–8.17 hPa for CRTM–MLP.

Contributions

• Introduction of a novel physics-informed Liquid Neural Network (LNN) framework for radiative transfer emulation, integrating continuous dynamical-system concepts with an analytic radiative-transfer solver.
• Demonstrated significant computational acceleration (up to 18-fold) for radiative transfer forward calculations, addressing a critical bottleneck in satellite data processing.
• Achieved high accuracy in emulating CRTM, with global brightness-temperature biases below 0.1 K.
• Produced demonstrably more physically consistent and accurate Jacobians (reduced centroid-pressure errors, smoother profiles) compared to conventional neural network emulators, which is vital for robust data assimilation and atmospheric retrievals.
• Established a scalable foundation for future advancements in atmospheric retrievals, satellite data assimilation, and near-real-time radiative-transfer applications.

Funding

[No specific funding projects, programs, or reference codes were mentioned in the provided paper text.]

Citation

@article{Zhang2026PhysicsInformed,
  author = {Zhang, Feng and Cao, Changyong and Chen, Yong and Shao, Xi and Liu, Tung-Chang},
  title = {Physics-Informed Liquid Neural Network Emulator for CRTM with Atmospheric-Layer Jacobian Capability},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18193325},
  url = {https://doi.org/10.3390/rs18193325}
}