1. An Optimal Weighted Ensemble Machine Learning Approach to Accurate Estimate the Coastal Boundary Layer Height Using ERA5 Multi‐Variables.
- Author
-
Peng, Kecheng, Xin, Jinyuan, Zhu, Xiaoqian, Xu, Qiang, Wang, Xiaoyuan, Wang, Weifeng, Tan, Yulong, Zhao, Dandan, Jia, Danjie, Cao, Xiaoqun, Ren, Xinbing, Ma, Yongjing, Wang, Guangjie, and Wang, Zifa
- Subjects
MACHINE learning ,MARINE meteorology ,HUMIDITY ,WIND speed ,WEATHER - Abstract
The coastal boundary layer height (CBLH/Coastal‐BLH) is critical in determining the exchange of heat, momentum, and materials between the land and ocean, thereby regulating the local climate and weather change. However, due to the complexity of geographical characteristics and meteorological conditions, accurate estimation of the CBLH remains challenging. Herein, based on continuous high‐resolution measurements of CBL performed from November 2019 to April 2020 in coastal Ningbo city in eastern China, an optimal weighted ensemble model (OWEM) integrating multi‐meteorological variables of the ERA5 reanalysis data sets is constructed and validated to estimate the CBLH. The mean absolute percentage error of the derived CBLH by OWEM is as low as 3%–5%, significantly lower than that of 36%–65% of the ERA5 CBLH products. Furthermore, three categories of different weather scenarios, that is, sunny, cloudy, and rainy, are separately discussed, and OWEM shows greater performance and higher accuracies in comparison with the traditional Least Absolute Shrinkage and Selection Operator, Random Forest, Adaboost, LightGBM, and ensemble model, among which, OWEM under fair weather days behave best, with a robust R2 of 0.97 and a minimum mean absolute error (MAE) of 23 m. Further training results based on wind flow classification, that is, land breeze, sea breeze, and parallel wind, also indicate the outperformance of OWEM than other models, with a relatively large error in parallel wind of 50 m. Subsequent analysis of the Shapley Additive Explanations method strongly correlated with model feature importance, both reveal that thermodynamic factors such as temperature (T2m) and wind velocity (10 m U) are the major factors positively related to estimation accuracy during sunny days. Nevertheless, Relative Humidity dominates on rainy and cloudy days, TP on land breeze days, and dynamic variables like 10 m U and 10 m V on entire types of wind flow weather. In conclusion, the accurate estimation of CBLH from OWEM serves as a feasible and innovative approach, providing technical support for marine meteorology and related engineering applications, for example, onshore wind power, coastal ecological protection, etc. Plain Language Summary: The coastal boundary layer height (CBLH) is fast becoming a key parameter in atmosphere‐ocean interaction studies, and machine learning has been thought of as an effective way to accurate estimate it. In this work, we present a new model called optimal weighted ensemble model (OWEM), which integrates several meteorological variables from reanalysis data sets to estimate the CBLH and make a comparison with the ERA5 BLH under different weather conditions. We also analyze the feature importance that affects estimation accuracy using the Shapley Additive Explanations method and identify several variables as essential factors, to better understand the applicability and characterization of the OWEM for sunny, cloudy, rainy, and different wind flow days. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF