Simulation of Ground Visibility Based on Atmospheric Boundary Layer Data Using K-Nearest Neighbors and Ensemble Model Algorithms
Abstract Low visibility will seriously affect traffic safety, and accurate prediction of low visibility can effectively reduce safety risks. This study introduces a machine learning approach for simulating visibility, utilizing the K-Nearest Neighbors algorithm and an ensemble model, which incorpora...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2024-11-01
|
| Series: | Aerosol and Air Quality Research |
| Subjects: | |
| Online Access: | https://doi.org/10.4209/aaqr.240145 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Low visibility will seriously affect traffic safety, and accurate prediction of low visibility can effectively reduce safety risks. This study introduces a machine learning approach for simulating visibility, utilizing the K-Nearest Neighbors algorithm and an ensemble model, which incorporate data from atmospheric boundary layer detection and conventional ground meteorological observations as simulation inputs. We developed three distinct visibility simulation schemes to identify the most effective algorithm and to assess the influence of the atmospheric boundary layer on the simulation outcomes. Our results revealed that during two separate fog events, the ensemble model consistently outperformed the KNN algorithm. In the first fog event, the ensemble model achieved a more significant reduction in RMSE compared to the MAE within the same range of visibility (for VIS < 200 m, Scheme 2 reduced MAE by 33% and RMSE by 24%). Moreover, the integration of atmospheric boundary layer data notably enhanced model accuracy in both fog events, with the enhancement being particularly marked in the first event (ensemble model: for VIS < 200 m, Schemes 2 and 3 had MAEs of 20.5 m, corresponding to a relative error of less than 10.3%, and 22.9 m, corresponding to a relative error of less than 11.5%, respectively). In the second fog event, the addition of atmospheric pollutant concentration data from the boundary layer further improved results (ensemble model: for VIS < 200 m, Schemes 2 and 3 had MAEs of 20.1 m, corresponding to a relative error of less than 10.1%, and 11.4 m, corresponding to a relative error of less than 5.7%, respectively). These findings underscore the importance of incorporating atmospheric boundary layer observations in enhancing the fidelity of visibility simulations based on KNN and ensemble model algorithms and their potential to significantly improve transportation safety and reduce economic losses. |
|---|---|
| ISSN: | 1680-8584 2071-1409 |