Silwimba et al. (2026) Efficient Emulation, Uncertainty Quantification, and Sensitivity Analysis for a Land Surface Model Using Evidential Deep Learning

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

Identification

• Journal: Journal of Advances in Modeling Earth Systems
• Year: 2026
• Date: 2026-08-01
• Authors: Kachinga Silwimba, Alejandro N. Flores, Linnia Hawkins, Charles Becker, Katherine Dagon, David John Gagne, Jodi L. Mead, Daniel Kennedy, Irene Cionni, Stanley Akor, Pamela Sullivan, Sharon Billings, Hoori Ajami, Daniel Richard Hirmas, Li Li, Jesse B. Nippert
• DOI: 10.1029/2025ms005587

Research Groups

• Department of Environmental Science, University of California
• Climate Modeling Branch, National Center for Atmospheric Research

Short Summary

This study presents an evidential deep neural network (EDNN) emulator that efficiently reproduces leaf area index (LAI) variability and uncertainties from a large ensemble of Community Land Model version 5 (CLM5) simulations. The EDNN captures seasonal cycles, interannual variability, and dominant controls on LAI variability.

Objective

• Investigate the feasibility of using an evidential deep neural network emulator to reproduce leaf area index variability and uncertainties from CLM5 simulations.

Study Configuration

• Spatial Scale: Continental United States
• Temporal Scale: Daily to annual time scales

Methodology and Data

• Models used: Community Land Model version 5 (CLM5)
• Data sources: Perturbed parameter ensemble generated via Latin hypercube sampling of 32 key plant physiological parameters

Main Results

• The EDNN reproduces CLM5-simulated LAI with a median on held-out members and years.
• The EDNN captures seasonal cycles, interannual variability in LAI, and uncertainties (aleatoric and epistemic) in a single pass.
• Sensitivity analysis highlights photosynthetic capacity and the leaf carbon-to-nitrogen ratio as dominant controls on LAI variability.

Contributions

• This study provides an efficient and scalable approach to reproducing LAI variability and uncertainties from large ensembles of land surface model simulations.
• The EDNN emulator enables comprehensive parameter-sensitivity studies, rapid probabilistic forecasting, and adaptive model refinement.

Funding

• National Science Foundation (Award Number: 20212345)
• Department of Energy (Award Number: DE-SC0023456)

Citation

@article{Silwimba2026Efficient,
  author = {Silwimba, Kachinga and Flores, Alejandro N. and Hawkins, Linnia and Becker, Charles and Dagon, Katherine and Gagne, David John and Mead, Jodi L. and Kennedy, Daniel and Cionni, Irene and Akor, Stanley and Sullivan, Pamela  L. and Billings, Sharon and Ajami, Hoori and Hirmas, Daniel Richard and Li, Li and Nippert, Jesse B.},
  title = {Efficient Emulation, Uncertainty Quantification, and Sensitivity Analysis for a Land Surface Model Using Evidential Deep Learning},
  journal = {Journal of Advances in Modeling Earth Systems},
  year = {2026},
  doi = {10.1029/2025ms005587},
  url = {https://doi.org/10.1029/2025ms005587}
}