本文已被:浏览 50次 下载 0次 |
 码上扫一扫! |
|
基于深度学习模型的雷电落区预报 |
任照环1,2, 林锐3, 周浩3, 覃彬全4, 李卫平1,2, 许伟1,2
|
1. 中国气象局气候资源经济转化重点开放实验室,重庆市防雷中心,重庆 401147;2. 中国气象局金佛山国家综合气象观测专项试验外场,重庆 401147;3. 重庆莱霆防雷技术有限责任公司,重庆 401147;4. 重庆市气象安全技术中心,重庆 401147
|
|
摘要: |
以 PredRNN 时空预测模型作为骨干网络,ASPP 模块为分类器,构建了雷电落区预报的深度学习模型(Lightning-Net)。模型以前 1 h的雷达组合反射率和雷电定位数据为预报因子,输出未来 1 h的雷电落区。利用 2020—2021 年重庆市雷达、雷电定位数据对模型进行训练,在 2022 年的数据集上测试。结果表明:构建的Lightning-Net 模型 TS(Threat score,威胁分数)评分为 0.53、命中率(POD)为 0.82,相比传统的光流法和 UNet 模型具有一定的优势;个例检验发现,模型对强雷暴的预报效果优于弱雷暴,模型对雷电落区变化的整体趋势能很好的把握,但对雷暴主体周围的零星雷电预报能力不足。 |
关键词: 深度学习 雷电落区预报 雷达 雷电定位数据 |
DOI:10.16032/j.issn.1004-4965.2024.068 |
分类号: |
基金项目: |
|
Lightning Strike Location Prediction Using a Deep Learning Model |
REN Zhaohuan1,2, LIN Rui3, ZHOU Hao3, QIN Binquan4, LI Weiping1,2, XU Wei1,2
|
1. China Meteorological Administration Economic Transformation of Climate Resources Key Laboratory/Chongqing Lightning Protection Center, Chongqing 401147, China;2. Jinfoshan National Comprehensive Meteorological Observation Special Test Field of China Meteorological Administration, Chongqing 401147, China;3. Chongqing Leting Lightning Protection Technology Co., Ltd, Chongqing 401147, China;4. Chongqing Meteorological Safety Technology Center, Chongqing 401147, China
|
Abstract: |
In this study, Lightning-Net, a deep learning model for lightning strike location prediction, was developed using a PredRNN spatio-temporal prediction model as the backbone network, and an Astrus spatial pyramid pooling module as the classifier. The model employed the radar composite reflectivity and the lightning location data of the preceding hour as predictive factors to predict the lightning strike location for the subsequent hour. The model was trained with the radar composite reflectivity and the lightning location data of Chongqing during 2020—2021 and tested with the dataset of Chongqing in 2022. The results show that the Lightning-Net model, with a threat score of 0.53 and a probability of detection of 0.82, demonstrated advantages over traditional optical flow methods and U-Net models. Case studies show that the model’s forecasting performance for severe thunderstorms was better than that for weak thunderstorms. While the model adeptly captured the overall trend of lightning strike location changes, it exhibited limitations in predicting sporadic lightning around the main body of thunderstorms. |
Key words: deep learning lightning strike location prediction radar lightning location data |