Maxwell (2026) Land surface model snow formulation sensitivity and a solar-zenith-angle snow covered area scheme evaluated across the western United States

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-10-09
• Authors: R. M. Maxwell
• DOI: 10.1016/j.jhydrol.2026.136567

Research Groups

• Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ, United States

Short Summary

This study evaluates 58 snow parameterizations in the ParFlow-CLM model across 220 SNOTEL site-years in the western United States, identifying fractional snow-covered area (fSCA) as the dominant control on snow water equivalent (SWE) accuracy. It introduces a novel solar-zenith-angle (SZA) modulated fSCA scheme that effectively balances peak SWE accuracy and melt timing, achieving performance comparable to machine learning models.

Objective

• To expand the ParFlow-CLM snow model with new parameterizations, particularly a solar-zenith-angle (SZA) modulated fractional snow-covered area (fSCA) scheme, and systematically evaluate their impact on snow water equivalent (SWE) prediction accuracy and melt timing across the western United States.

Study Configuration

• Spatial Scale: Western United States, specifically 220 SNOTEL site-years across 100 unique SNOTEL stations in 11 states. Simulations were conducted in a single-column configuration, with potential application to the 1 km CONUS2 domain and finer resolution.
• Temporal Scale: 220 site-years, spanning water years 1985 to 2024. Simulations used hourly atmospheric forcing with a 1-hour timestep, aggregated to daily means for comparison.

Methodology and Data

• Models used:
  • ParFlow-CLM (PF-CLM): Integrated hydrologic model coupling 3D subsurface flow to a land surface scheme.
  • CLM (Community Land Model) version 3.0: Baseline land surface model physics.
  • BATS (Biosphere-Atmosphere Transfer Scheme): Default snow albedo aging model.
  • VIC (Variable Infiltration Capacity) albedo scheme: Alternative albedo formulation.
  • Tarboton albedo scheme: Alternative albedo formulation.
  • SNTHERM.89: Model from which snow compaction processes are derived.
  • Random Forest (RF): Machine learning model used for atmospheric forcing bias correction.
  • LSTM (Long Short-Term Memory): Machine learning model used as a benchmark for SWE prediction.
• Data sources:
  • SNOTEL (Snow Telemetry) network: Daily snow water equivalent (SWE), precipitation, and temperature observations from approximately 900 automated stations across the western United States.
  • CW3E (Center for Western Weather and Water Extremes) atmospheric forcing: Hourly atmospheric forcing data at 1 km resolution (downward shortwave and longwave radiation, precipitation, air temperature, wind speed, surface pressure, specific humidity).
  • HydroData API: Python package used for accessing SNOTEL observations and CW3E forcing data.

Main Results

• Fractional snow-covered area (fSCA) is the dominant control on snow water equivalent (SWE) prediction accuracy in ParFlow-CLM, producing larger improvements than any tested changes to albedo schemes, rain-snow partitioning, or snow aging.
• Static fSCA corrections improve peak SWE but introduce a trade-off with melt timing (e.g., delaying melt-out by two to three weeks).
• The new solar-zenith-angle (SZA) modulated fSCA scheme (SZ3 configuration) resolves this trade-off, achieving a median Nash-Sutcliffe Efficiency (NSE) of 0.79 and a melt-out bias of -1 day across 100 low-bias validation site-years.
• When combined with random forest-based forcing bias correction, the SZA-fSCA scheme (SZ3) achieves a median NSE of 0.82 and an RMSE of 0.0489 m across 100 high-bias site-years, closely approaching the performance of a Long Short-Term Memory (LSTM) model (NSE 0.86, RMSE 0.0485 m).
• Compared to the baseline ParFlow-CLM, the SZA-fSCA scheme (SZ3 with RF correction) approximately halves the absolute bias in peak SWE (median 38% to 19%) and centers the median melt-out bias around zero (improving by about one month).
• The physics improvements from the SZA-fSCA scheme had a greater impact on model performance than atmospheric forcing bias correction at high-bias sites.
• The number of sites classified as "very good" (RSR < 0.5) more than doubled from 60 (baseline) to 131 (SZ3), while the number of "failure" sites (NSE < 0) dropped from 53 to 15 across the 220 site-years.

Contributions

• Development and evaluation of a novel solar-zenith-angle (SZA) modulated fractional snow-covered area (fSCA) scheme for the ParFlow-CLM model, which dynamically adjusts effective roughness length based on seasonal solar geometry.
• Demonstration through systematic sensitivity analysis that fSCA is the dominant control on SWE prediction accuracy in ParFlow-CLM, resolving a critical trade-off between peak SWE magnitude and melt timing.
• Provision of physics-based snow options for ParFlow-CLM that achieve performance comparable to state-of-the-art machine learning models (LSTM) and surpass reanalysis products and default process models.
• Comprehensive evaluation of 58 snow parameterizations across 220 SNOTEL site-years, establishing a robust benchmark for snow physics in integrated hydrologic models.
• Enhancement of ParFlow-CLM's capability for more accurate integrated continental hydrologic simulation over the United States, particularly for snowmelt-dominated regions.

Funding

• U.S. National Science Foundation Convergence Accelerator Program
• U.S. National Science Foundation National Artificial Intelligence Research Resource (NAIRR) (grant no. CA-2040542)
• U.S. Department of Agriculture (Innovations at the Food, Energy, and Water Systems NSF Grant No. 67020-30130)
• U.S. Department of Defense Strategic Environmental Research and Development Program (SERDP Grant RC25-4949)

Citation

@article{Maxwell2026Land,
  author = {Maxwell, R. M.},
  title = {Land surface model snow formulation sensitivity and a solar-zenith-angle snow covered area scheme evaluated across the western United States},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136567},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136567}
}