@article{4834,
  author = {Alla A. Kaybiyaynen},
  title = {A High-Resolution Global METAR Dataset of Aviation Weather Observations from Major Cities (2024-2026): Characterization, Adverse-Event Analysis, and Predictive Modeling},
  journal = {Journal of E-Technology},
  year = {2026},
  volume = {17},
  number = {4},
  doi = {https://doi.org/10.6025/jet/2026/17/4/176-192},
  url = {https://www.dline.info/jet/fulltext/v17n4/jetv17n4_2.pdf},
  abstract = {Aviation operations are highly sensitive to rapidly changing meteorological conditions, yet existing research
often focuses on isolated airports or specific weather hazards. This study addresses these gaps by developing
and analyzing a high resolution, global Meteorological Aerodrome Report (METAR) dataset comprising
958,338 observations from 26 major cities across 15 countries between 2024 and 2026. By integrating
meteorological characterization, adverse-weather identification, and machine learning, the study quantifies
the temporal and geographical variability of aviation weather conditions. Results reveal a severe class
imbalance, with adverse weather events accounting for only 4.29% of observations. While classification
models achieved high overall accuracy (up to 98.1%), minority class precision and recall remained low,
though fog detection demonstrated strong ranking ability (ROC-AUC = 0.860). Conversely, regression
modeling for a continuous severity score proved highly challenging (R² = -1.43), indicating that temperature
alone is insufficient for severity prediction. Spatiotemporal analyses highlighted distinct diurnal peaks in
adverse weather (11:00–14:00 UTC) and significant geographic heterogeneity. Ultimately, this study
provides a comprehensive, multi hazard framework that underscores the need for advanced feature
engineering and cost sensitive learning to support risk based operational decision making in global aviation.},
}