Braz et al. (2025) Development of Automatic Labels for Cold Front Detection in South America: A 2009 Case Study for Deep Learning Applications

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

Identification

• Journal: Climate
• Year: 2025
• Date: 2025-10-08
• Authors: Dejanira Ferreira Braz, Luana Albertani Pampuch, Michelle Simões Reboita, Tércio Ambrizzi, Tristan Pryer
• DOI: 10.3390/cli13100211

Research Groups

Not explicitly provided in the text.

Short Summary

This study introduces an automatic cold front detection method using ERA5 reanalysis for South America, generating spatially consistent labels for machine learning applications, and demonstrating high spatial concordance with manual charts despite pixel-level differences.

Objective

• To develop and validate an automatic cold front detection method using ERA5 reanalysis data to generate physically and spatially consistent training labels for deep learning applications in atmospheric pattern recognition, specifically calibrated for South America.

Study Configuration

• Spatial Scale: South America, at the 850 hPa atmospheric level.
• Temporal Scale: Validation against 1426 manual charts from 2009; analysis of seasonal variability.

Methodology and Data

• Models used: An automatic cold front detection method combining the Thermal Front Parameter (TFP) and temperature advection (AdvT), applying optimized thresholds (TFP < 5 × 10−11 K m−2; AdvT < −1 × 10−4 K s−1), morphological filtering, and polynomial smoothing.
• Data sources: ERA5 reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF) at 850 hPa; 1426 manual charts from 2009 for comparison.

Main Results

• The automatic method utilizes optimized thresholds of TFP < 5 × 10−11 K m−2 and AdvT < −1 × 10−4 K s−1.
• Comparison with manual charts revealed systematic spatial displacement, with mean offsets of approximately 502 km.
• Despite low pixel-level overlap (Intersection over Union (IoU) = 0.013; Dice coefficient (Dice) = 0.034), spatial concordance exceeded 99%, confirming both methods identify the same synoptic systems.
• The automatic method detects 58% more fronts over the South Atlantic and 44% fewer over the Andes compared to manual charts.
• Seasonal variability shows maximum front activity in austral winter (31.3%) and minimum in summer (20.1%).

Contributions

• Introduces the first automatic cold front detection system specifically calibrated for South America.
• Addresses the critical problem of spatial misalignment between training labels and reanalysis input fields, which limits deep learning applications in atmospheric sciences.
• Provides a method for generating physically and spatially consistent training labels directly from reanalysis data for machine learning applications.

Funding

Not explicitly provided in the text.

Citation

@article{Braz2025Development,
  author = {Braz, Dejanira Ferreira and Pampuch, Luana Albertani and Reboita, Michelle Simões and Ambrizzi, Tércio and Pryer, Tristan},
  title = {Development of Automatic Labels for Cold Front Detection in South America: A 2009 Case Study for Deep Learning Applications},
  journal = {Climate},
  year = {2025},
  doi = {10.3390/cli13100211},
  url = {https://doi.org/10.3390/cli13100211}
}