Atmospheric Science
A machine learning-based short-term forecasting method for heavy fog in Anhui Province of China
This paper presents a machine learning (ML) based short-term heavy fog forecasting method for Anhui Province, China. Utilizing hourly observations and ERA5 reanalysis data, 340 heavy fog cases were identified. The province was objectively divided into seven distinct fog zones using Rotated Empirical Orthogonal Function (REOF). Five categories of forecasting factors were used to develop models with multiple ML algorithms (Random Forest, Logistic Regression, K-Nearest Neighbors, Gaussian Naive Bayes, Decision Tree). Random Forest (RF) achieved the highest threat score and accuracy, identifying water vapor, thermodynamic conditions, and upstream-downstream effects as key predictors. Comparative analysis against ECMWF and SCMOC data confirmed the RF model's superior performance in terms of threat score and lower missing alarm rates.
Executive Impact at a Glance
Key metrics demonstrating the potential for enhanced operational efficiency and safety through advanced fog forecasting.
0
Forecast Accuracy
0
Threat Score (Peak)
0
MAR Reduction
0
Optimized Fog Zones
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Precision Regional Fog Zoning
The study effectively categorizes Anhui Province into seven distinct fog-prone areas using Rotated Empirical Orthogonal Function (REOF) analysis. This foundational step is crucial for developing region-specific forecasting models, addressing the spatial variability of fog formation in complex terrains.
Enterprise Process Flow
Historical Visibility Data Collection
→
Typical Heavy Fog Case Identification
→
REOF Decomposition & Significance Test
→
Objective Division into Seven Fog Zones
→
Region-Specific Model Development
Advanced ML for Predictive Accuracy
Among the evaluated machine learning algorithms (Random Forest, Logistic Regression, K-Nearest Neighbors, Gaussian Naive Bayes, Decision Tree), the Random Forest (RF) model demonstrated the highest threat score (TS) and accuracy for short-term heavy fog forecasting. This highlights RF's robustness and capability in handling complex meteorological datasets.
RF
Top-Performing Algorithm for Fog Forecasting
Key Meteorological Predictors Unveiled
Feature importance analysis using the RF model identified water vapor, thermodynamic conditions (temperature-dew point difference, temperature inversion), and upstream-downstream effects (advective transport) as the most crucial factors for heavy fog formation. This provides critical insights into the physical mechanisms driving fog events in the region.
| Factor Category | Key Predictors |
|---|---|
| Water Vapor |
|
| Thermodynamics |
|
| Advective Transport |
|
Robust Performance in Real-World Scenarios
The RF-based model was rigorously evaluated using operational data from 2023 and compared against ECMWF and SCMOC forecasts. The results showed that the RF model achieved the highest Threat Score and significantly lower Missing Alarm Rates across all seven fog zones, demonstrating its superior real-world applicability for continuous heavy fog events.
2023 Anhui Province Heavy Fog Event
Scenario: A prolonged and extensive heavy fog event from December 28-31, 2023, across Anhui Province was used to validate the model's performance against existing operational products.
Challenge: Current mainstream numerical prediction products (ECMWF, SCMOC) struggled with spatial continuity and accurate delineation of heavy fog areas, leading to underreporting or dispersed forecasts.
Solution: The STFM successfully forecasted heavy fog across most parts of the province for three consecutive days, accurately capturing both the continuity and spatial extent of the event. It achieved a provincial-average TS of 0.61, significantly outperforming ECMWF (0.26) and SCMOC (0.30).
Outcome: The RF-based STFM model proved to be a superior tool for short-term heavy fog prediction in complex regions, effectively addressing the deficiencies of current operational forecasts.
Calculate Your Potential ROI
See how leveraging advanced AI for forecasting can translate into tangible savings and reclaimed productivity for your enterprise.
Annual Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A structured approach to integrating advanced fog forecasting into your operations, ensuring seamless adoption and measurable results.
Data Acquisition & Preprocessing
Establish automated pipelines for collecting hourly expressway observations and ERA5 reanalysis data. Develop robust data cleaning, quality control, and feature engineering modules.
Regional Zoning & Model Training
Implement the REOF method for dynamic fog zone identification. Train and validate Random Forest models for each identified zone, optimizing hyperparameters for peak performance.
Operational Integration & Real-time Deployment
Integrate the RF forecasting system into existing meteorological infrastructure. Develop user-friendly interfaces for real-time visualization of fog forecasts and alert generation.
Continuous Improvement & AI Governance
Establish feedback loops for model retraining with new data. Implement MLOps practices for model monitoring, bias detection, and ensuring ethical AI governance.
Ready to Transform Your Operations?
Schedule a personalized consultation to discuss how our AI solutions can address your specific forecasting challenges and drive significant value.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat