Pan et al. (2026) Mapping daily snow depth with machine learning and airborne lidar across two contrasting snowpacks

Identification

• Journal: ˜The œcryosphere
• Year: 2026
• Date: 2026-07-03
• Authors: Caleb G. Pan, Jeremy Johnston, Jennifer M. Jacobs, Shad O'Neel
• DOI: 10.5194/tc-20-3759-2026

Research Groups

• Geospatial Research Laboratory, Engineer Research and Development Center, US Army Corps of Engineers, Alexandria, VA, 22315, USA
• Earth Systems Research Center, Institute for the Study of Earth, Oceans, and Space, University of New Hampshire, Durham, NH, USA
• Department of Civil and Environmental Engineering, University of New Hampshire, Durham, NH, USA
• Cold Regions Research Engineering Laboratory, Engineer Research and Development Center, US Army Corps of Engineers, Boise, ID, USA

Short Summary

This study presents a portable relative-depth machine-learning framework that converts sparse lidar acquisitions into temporally coherent daily snow depth maps. The approach is evaluated in two contrasting regimes – Mores Creek, Idaho, and Hubbard Brook, New Hampshire – using multi-year lidar records.

Objective

• Investigate the feasibility of extending episodic lidar observations into continuous seasonal snow depth maps using machine learning.
• Evaluate the performance of a random forest model in predicting daily snow depth across different snow-climate regimes and seasons.
• Assess the impact of lidar cadence, acquisition timing, and driver dataset choice on model skill.

Study Configuration

• Spatial Scale: Basin-scale (35 km2 for Mores Creek, 0.42 km2 for Hubbard Brook)
• Temporal Scale: Daily

Methodology and Data

• Models used: Random Forest Regressor
• Data sources:
  • Lidar surveys (airborne helicopter lidar for Mores Creek, UAS lidar for Hubbard Brook)
  • In-situ stations (SNOTEL and SC2) for meteorological data
  • ERA5-Land reanalysis for daily snow depth, air temperature, and wind

Main Results

• The random forest model achieved high performance in predicting relative snow depth across both study areas, with R2 values ranging from 0.89 to 0.97.
• Model skill was influenced by lidar cadence, with diminishing returns after a few acquisitions.
• ERA5-Land reanalysis was found to be a suitable substitute for in-situ forcing in daily prediction.

Contributions

• This study provides a practical route from episodic lidar snow surveys to meter-scale, daily, basin-scale products and actionable guidance on survey timing and frequency.

Funding

• This research was funded by the US Army Corps of Engineers (ERDC) and the National Science Foundation (NSF).

Citation

@article{Pan2026Mapping,
  author = {Pan, Caleb G. and Johnston, Jeremy and Jacobs, Jennifer M. and O'Neel, Shad},
  title = {Mapping daily snow depth with machine learning and airborne lidar across two contrasting snowpacks},
  journal = {˜The œcryosphere},
  year = {2026},
  doi = {10.5194/tc-20-3759-2026},
  url = {https://doi.org/10.5194/tc-20-3759-2026}
}