Gorooh et al. (2026) Enhancing Deterministic Freezing-Level Predictions in the Northern Sierra Nevada through Deep Neural Networks

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

Identification

• Journal: Journal of Hydrometeorology
• Year: 2026
• Date: 2026-01-16
• Authors: Vesta Afzali Gorooh, Agniv Sengupta, Shawn Roj, Rachel Weihs, Brian Kawzenuk, Luca Delle Monache, F. Martin Ralph
• DOI: 10.1175/jhm-d-25-0052.1

Research Groups

The abstract does not explicitly list specific academic research groups, labs, or departments. However, the work involves: - West Weather Research and Forecasting (West-WRF) Model development and application. - California Nevada River Forecast Center (CNRFC) for FZL estimates. - Research in deep learning and hydrometeorological forecasting, likely involving institutions focused on atmospheric science, hydrology, and artificial intelligence.

Short Summary

This study develops a deep learning-based postprocessing framework using U-Net convolutional neural networks to refine freezing level (FZL) forecasts from the West Weather Research and Forecasting (West-WRF) Model. The proposed U-Net models significantly reduce forecast errors by up to 20% and increase forecast-observation correlation by approximately 10% compared to raw West-WRF outputs, improving FZL prediction for hydrometeorological applications.

Objective

• To develop a deep learning-based postprocessing framework using the U-Net convolutional neural network (CNN) architecture to refine freezing level (FZL) forecasts from the West Weather Research and Forecasting (West-WRF) Model.

Study Configuration

• Spatial Scale: Northern Sierra Nevada watersheds, specifically the Yuba–Feather watershed.
• Temporal Scale: Across various forecast lead times, considering storm variability and diurnal forcing.

Methodology and Data

• Models used:
  • U-Net convolutional neural network (CNN) architecture.
  • U-Net-log (utilizing log-cosh cosine of error loss function).
  • U-Net Gaussian mixture model (U-Net-GMM) (utilizing Gaussian mixture model loss functions).
  • West Weather Research and Forecasting (West-WRF) Model (as the baseline and source of raw forecasts).
• Data sources:
  • Reforecast data from West-WRF.
  • Freezing level (FZL) estimates from the California Nevada River Forecast Center (CNRFC).

Main Results

• U-Net-based postprocessing reduces centered root-mean-square errors by up to 20% compared to raw West-WRF forecasts.
• Forecast-observation correlation is increased by approximately 10% compared to raw West-WRF forecasts.
• The U-Net-GMM model demonstrates consistent improvements across different lead times when evaluated using the continuous ranked probability score (CRPS).
• Both U-Net-GMM and U-Net-log consistently outperform the baseline West-WRF model, despite performance fluctuations with forecast horizon, storm variability, and diurnal forcing.
• The models effectively capture the spatiotemporal variability of the FZL across various elevations, successfully mitigating biases present in the West-WRF Model.

Contributions

• Introduction of a novel deep learning-based postprocessing approach using U-Net CNNs for refining FZL forecasts.
• Development and evaluation of two specialized U-Net model variants, U-Net-log and U-Net-GMM, which employ advanced loss functions (log-cosh cosine of error and Gaussian mixture model loss, respectively) to enhance forecast skill beyond standard benchmarks.
• Demonstration of a promising pathway for integrating machine learning into operational hydrometeorological forecasting and decision support systems, particularly within the Forecast-Informed Reservoir Operations (FIRO) framework.

Funding

The abstract does not contain information regarding specific funding projects, programs, or reference codes.

Citation

@article{Gorooh2026Enhancing,
  author = {Gorooh, Vesta Afzali and Sengupta, Agniv and Roj, Shawn and Weihs, Rachel and Kawzenuk, Brian and Monache, Luca Delle and Ralph, F. Martin},
  title = {Enhancing Deterministic Freezing-Level Predictions in the Northern Sierra Nevada through Deep Neural Networks},
  journal = {Journal of Hydrometeorology},
  year = {2026},
  doi = {10.1175/jhm-d-25-0052.1},
  url = {https://doi.org/10.1175/jhm-d-25-0052.1}
}