Abstract This study presents FogCast, a Bayesian neural network (BNN) ‐based model that forecasts fog using the outputs of GraphCast, a machine‐learning global weather prediction system. Compared to existing operational forecasts limited to 3–5 days lead time, FogCast extends fog forecasting capabilities to 10 days, making it a valuable tool for medium‐range forecasts. FogCast provides probabilistic fog/no‐fog forecasts while accounting for aleatoric and epistemic uncertainties, thereby improving forecast reliability. FogCast is trained and validated using visibility observations for seven cities in the Indo‐Gangetic Plains from 2017 to 2023. The model achieves Critical Success Index (CSI) values between 0. 44 (for six‐hour lead time) and 0. 36 (for 10‐day lead time). Moreover, it correctly detects over 90% of very dense fog observations (visibility <50 m). The uncertainty analysis shows that aleatoric uncertainty is typically higher than epistemic uncertainty, suggesting potential avenues for forecast improvements. The real‐time fog forecasts generated by FogCast are accessible at https: //fog. iitk. ac. in/fog‐prediction/fogcastₗiteforecast. php.
Building similarity graph...
Analyzing shared references across papers
Loading...
Prasad Deshpande
Shubhangi Agarwal
Kritika Bansal
Quarterly Journal of the Royal Meteorological Society
Kansas State University
Indian Institute of Technology Kanpur
Building similarity graph...
Analyzing shared references across papers
Loading...
Deshpande et al. (Fri,) studied this question.
www.synapsesocial.com/papers/69af963170916d39fea4e1eb — DOI: https://doi.org/10.1002/qj.70150
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: