ISSN 1000-0526
CN 11-2282/P
CN 11-2282/P
Research on the Identification of Varying Grades of Severe Convective wind with Convolutional Neural Networks Optimized by Immune Evolutionary Algorithm and its Applications
- Article
- Figures
- Metrics
- Preview PDF
- Reference
- Related
- Cited by
- Materials
Abstract:
The severe imbalance in sample distribution,characterized by a sharp decrease in the occurrence frequency of severe convective wind with increasing wind speed，is identified as the predominant factor hindering the accurate intensity-based classification of severe convective wind by various existing algorithms.To address this issue，this paper proposes using a non-differentiable Probability of Detection (POD) as the loss function for a Convolutional Neural Network (CNN)，with Bias as its constraint condition.Subsequently，the Multi-objective Optimization Immune Evolution Algorithm (MOIEA) was employed to optimize all the model parameters of the CNN.This led to the development of an identification algorithm for identifying severe convective wind of 17.2、20.8、24.5 m?s-1 and above，named Severe Convective Wind Identification Network (SCWINet). SCWINet leverages data from 2022 to 2024，including radar vertical liquid water content，three-dimensional radar reflectivity，lightning location data，and minute-level ground-based automatic observation station data from Zhejiang Province.It achieves different levels of severe convective wind identification with a temporal resolution of 6 minutes and a spatial resolution of 0.01°.The performance of SCWINet was compared to approaches using the same CNN structure but with the differentiable loss functions of Weighted Mean Squared Error (WMSE)and Balanced Mean Squared Error (BMSE).The applicability of SCWINet was then evaluated based on Threat Score (TS)，Bias，POD，False Alarm Ratio (FAR) using the neighborhood method (with a scanning radius of 5 km)，and the planar distribution characteristics of severe convective wind.The main results are as follows: SCWINet effectively identifies severe convective wind of 17.2、20.8、24.5 m?s-1 and above, associated with both systematic and scattered severe convective systems，with better performance observed in identifying severe convective wind triggered by systematic convection compared to scattered convection.Furthermore，the identification effectiveness generally decreases as wind speed increases，with increased false alarms and missed detections being the primary causes of this pattern.The commonly used WMSE and BMSE approaches，however, showed no ability to identify severe convective wind，with all identified severe convective wind being below 17.2 m?s-1.Nonetheless，the data used in this study are still somewhat limited in terms of feature completeness and volume.Future enhancements in identification accuracy could be achieved by incorporating additional features and data，such as radar radial velocity，Specific Differential Phase (KDP), Differential Reflectivity (ZDR)，and satellite data，which could extend the applicability to even higher wind speeds.
Keywords:
Project Supported:
Clc Number:
Mobile website
