基于TCP-GAN的热带气旋路径预测
TC track forecasting using a TCP-GAN network
张芮 1杭仁龙 2刘英杰2
作者信息
- 1. 江苏科技大学 计算机学院,镇江 212100;南京信息工程大学 计算机与软件学院,南京 210044
- 2. 南京信息工程大学 计算机与软件学院,南京 210044
- 折叠
摘要
准确预测热带气旋路径对于中国沿海地区的防灾减灾具有重要作用.卫星数据是预测热带气旋的重要手段,针对现有方法生成的卫星图像不够清晰,很难准确判断热带气旋云系的轮廓,提出了一种基于TCP-GAN的热带气旋路径预测方法.采用了生成对抗网络,并加入了感知损失,使得生成的图像更加细致.模型在不同输入图像序列长度(2、4 和 6)下进行试验,结果表明,当长度为 4 时,此时的路径误差是最小的,约为 45.36 km.此外,进行了滚动预测,以验证模型在 12、18和 24h的预测性能,提出的热带气旋路径预测模型生成的图像能够很好地描述云系的细致纹理结构,预测的路径误差相比于同类方法也更小.
Abstract
Accurate prediction of TC track plays an important role in disaster prevention and mitigation in coastal areas of China.Satellite data are an important tool for forecasting tropical cyclones.However,the satellite ima-ges generated by the existing methods are not clear enough,so that it is difficult to accurately determine the con-tour of the TC cloud system.To address this issue,we propose a TC track prediction model,named TCP-GAN.This model uses a generative adversarial network,and adds a new loss function to make generated images more detailed.The model is tested using different input image sequence lengths(N=2,4,and 6).The results show that the mean error is the smallest with N being 4,about 45.36 km.In addition,this paper carries out rolling forecasts to verify the forecasting performance of the model at 12,18,and 24 hours.In the study,the generated images of the proposed model can describe in detail the textures and shapes of cloud systems,and the TC predic-tion errors are also smaller than those of similar methods.
关键词
生成对抗网络/卫星云图外推/热带气旋路径预测/葵花-8Key words
generative adversarial network/satellite image extrapolation/TC track forecasting/HIMAWARI-8引用本文复制引用
基金项目
国家自然科学基金青年基金项目(62206115)
江苏省自然科学基金青年基金项目(BK20220646)
出版年
2024
NETL
NSTL
万方数据