A machine learning-based short-term forecasting method for heavy fog in Anhui Province of China
Authors:
Open Access
- 13 January 2026
- journal article
- research article
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Abstract
Based on hourly observations from expressway meteorological stations across Anhui Province and the fifth-generation atmospheric reanalysis (ERA5) data from the European Centre for Medium-Range Weather Forecasts (ECMWF) spanning from January 2011 to December 2023, 340 typical cases of regional heavy fog in Anhui Province were identified. The rotated empirical orthogonal function (REOF) method was then employed to objectively divide the province into seven distinct fog zones. Using five categories of forecasting factors (local water vapor, thermodynamics, stability, temporal variations, and upstream–downstream effects), heavy fog forecasting models were developed based on multiple machine learning algorithms, including random forest (RF), logistic regression, K-nearest neighbors, Gaussian Naive Bayes, and decision tree. With daily meteorological observations at 1600 Beijing Standard Time (BST) and corresponding ECMWF forecast data as inputs, the models generated forecasts for fog occurrence from 20 BST on the same day to 08 BST the following day. The results indicate that Anhui Province can be objectively categorized into seven fog zones using the REOF approach. All five machine learning algorithms demonstrated skill in short-term heavy fog forecasting, with the RF algorithm achieving the highest threat score (TS) and accuracy. Water vapor, thermodynamic conditions, and upstream–downstream effects were identified as the most important predictors for the RF model. Consequently, the RF model was selected as the preferred forecasting method and was further evaluated using operational data from 2023. Comparative analysis against visibility forecasts from ECMWF and the National Meteorological Centre’s gridded forecast guidance product (SCMOC) showed that the RF-based model achieved the highest TS values and the lowest missing alarm rates across all seven fog zones. Although the false alarm rates of the RF model were lower than those of SCMOC, they were slightly higher than those of ECMWF.Keywords
Funding Information
- The National Natural Science Foundation of China (42075188)
This publication has 56 references indexed in Scilit:
- Coupled data assimilation at ECMWF: current status, challenges and future developmentsQuarterly Journal of the Royal Meteorological Society, 2022
- A review on factors influencing fog formation, classification, forecasting, detection and impactsRendiconti Lincei. Scienze Fisiche e Naturali, 2022
- Fog probabilistic forecasting using an ensemble prediction system at six airports in Iran for 10 fog eventsMeteorological Applications, 2021
- Sea Fog Dissipation Prediction in Incheon Port and Haeundae Beach Using Machine Learning and Deep LearningSensors, 2021
- Learning earth system models from observations: machine learning or data assimilation?Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2021
- Physics-informed machine learning: case studies for weather and climate modellingPhilosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2021
- Dutch fog: On the observed spatio‐temporal variability of fog in the NetherlandsQuarterly Journal of the Royal Meteorological Society, 2019
- Integrating precipitation zoning with random forest regression for the spatial downscaling of satellite‐based precipitation: A case study of the Lancang–Mekong River basinInternational Journal of Climatology, 2019
- Algorithm for sea fog monitoring with the use of information technologiesMeteorological Applications, 2012
- A Study of Fog Characteristics Using a Coupled WRF–COBEL Model Over Thessaloniki Airport, GreecePure and Applied Geophysics, 2011