设为首页
加入收藏
Nowcasting of Cloud Images Based on Generative Adversarial Network and Satellite Data
-
摘要: 利用风云四号气象卫星A星（FY-4A）红外云图，基于生成对抗网络方法，提出了红外云图外推预报模型，实现了华东区域未来3 h的云图预报，预报的时空分辨率分别为1 h和4 km。结果表明：该外推模型预报的云图可较好描述云团移动、发展和减弱趋势，对研究区域内云团的强度、位置和形态得到较为理想的预报效果。为了验证提出的云图外推模型的有效性，将其与光流法和轨迹门控循环单元模型进行比较。对比试验结果表明：该云图外推模型具有最优的预报效果，说明使用生成对抗网络进行云图外推具有较好的可行性，能有效应用于气象业务中监测云团的生消和移动并提前预警灾害性天气的发生，为天气预报提供重要的参考依据。Abstract: Satellite cloud images contain abundant information, which can reflect daily weather conditions. Nowcasting based on cloud images can strengthen the application of satellite data in the early warning and forecasting of severe weather. At present, the cloud images predicted by most nowcasting methods based on artificial intelligence are not accurate enough and the lead time is limited. Thus, it's necessary to improve the accuracy and period validity of cloud images in nowcasting. Using the infrared cloud image data of Fengyun-4A (FY-4A) and the generative adversarial network (GAN) method, an infrared cloud image extrapolation nowcasting model is proposed. The cloud images in the next 3 hours in East China are predicted by the proposed model, and the spatial resolution of predicted cloud images is 4 km and the temporal resolution is 1 hour. The results show that the evaluation values of SSIM (structural similarity), PSNR (peak signal to noise ratio) and RMSE (root mean square error) predicted by the proposed GAN-based cloud images extrapolation model are 0.75, 20.92 and 10.00 K, respectively. In addition, the MAE (mean absolute error), MSE (mean squared error), and SSIM are chosen as loss function and analyzed, aiming to verify the rationality of the loss function in the generator. Comparative experiments of different loss functions show that it is reasonable and effective to choose SSIM combined with MAE as the loss function. Furthermore, to verify the effectiveness of the proposed GAN-based model, the prediction results are compared with those of the optical flow method and the TrajGRU model with the GAN-based model. The experimental results show that the cloud image extrapolation model based on GAN has the superior prediction performance, with the highest SSIM and PSNR, and the lowest RMSE within 1-3 h of cloud images nowcasting. The observational examples show that the cloud images predicted by the proposed model can well describe the movement, development and dissipation trend of clouds. Meanwhile, the experiments obtain accurate prediction performance on the intensity, location and shape of clouds in the study region. It indicates that the cloud image extrapolation model based on GAN is rational and feasible, which can be effectively applied to the meteorological business to monitor the occurrence and movement of clouds and warn the occurrence of severe weather in advance, and can provide an important reference for weather forecasting.
-
图 2 不同外推模型不同预报时效云图预报能力对比
Fig. 2 Prediction performance of diverse cloud image extrapolation models at different lead times
图 3 2020年6月23日05:00—09:00(输入模型的过去5 h)观测云图
Fig. 3 Observed cloud images in the past 5 hours from 0500 UTC to 0900 UTC on 23 Jun 2020(inputs of models)
图 4 2020年6月23日10:00—12:00未来3 h观测与预报云图对比
Fig. 4 Comparison of the observed and the predicted cloud images in the next 3 hours from 1000 UTC to 1200 UTC on 23 Jun 2020
表 1 生成器中选取不同损失函数训练的云图外推模型的外推效果
Table 1 Performance of the cloud image extrapolation model trained with different loss functions in the generator
损失函数 评估指标 相对误差绝对值/% 均方根误差/K 结构相似性 峰值信噪比 均方误差 157.80 10.48 0.69 20.50 平均误差绝对值 155.75 10.11 0.74 20.87 结构相似性 158.43 10.36 0.74 20.63 结构相似性结合平均误差绝对值 151.64 10.00 0.75 20.92 
下载: 导出CSV
表 2 使用不同损失函数时，生成对抗网络模型在不同时效上的云图预报能力
Table 2 Predictive performance of the generative adversarial network model for the cloud images at different lead times using different loss functions
损失函数 评估指标 结构相似性 峰值信噪比 1 h 2 h 3 h 1 h 2 h 3 h 平均误差绝对值 0.79 0.74 0.71 23.25 20.44 18.92 均方误差 0.75 0.69 0.61 22.99 19.53 18.98 结构相似性 0.79 0.74 0.71 22.92 20.17 18.82 结构相似性结合平均误差绝对值 0.80 0.75 0.72 23.21 20.51 19.03
下载: 导出CSV
表 3 基于测试集的不同外推模型预报效果评估
Table 3 Prediction performance of different models based on testing set
模型 评估指标 相对误差绝对值% 均方根误差/K 结构相似性 峰值信噪比 生成对抗网络 151.64 10.00 0.75 20.92 轨迹门控循环单元模型 172.84 11.10 0.72 19.99 光流法 186.35 11.97 0.66 19.22
下载: 导出CSV
-
[1] 张弛, 刘钧, 李旭光, 等.基于可见光-红外图像信息融合的云状识别方法.气象与环境学报,2018, 34(1):82-90. doi: 10.3969/j.issn.1673-503X.2018.01.010Zhang C, Liu J, Li X G, et al. A cloud classification method based on information fusion of visible and infrared images. J Meteor Environ, 2018, 34(1): 82-90. doi: 10.3969/j.issn.1673-503X.2018.01.010 [2] 吴晓京, 朱小祥, 毛紫阳, 等. 风云二号气象卫星红外观测在云团降水监测中的应用. 海洋气象学报, 2019, 39(3): 1-10. https://www.cnki.com.cn/Article/CJFDTOTAL-SDQX201903001.htmWu X J, Zhu X X, Mao Z Y, et al. Algorithm design of convective precipitation monitoring and early warning service using FY-2 infrared data. Journal of Marine Meteorology, 2019, 39(3): 1-10. https://www.cnki.com.cn/Article/CJFDTOTAL-SDQX201903001.htm [3] Rojas Y, Minder J R, Campbell L S, et al. Assessment of GPM IMERG satellite precipitation estimation and its dependence on microphysical rain regimes over the mountains of south-central Chile-ScienceDirect. Atmos Res, 2021, 253: 105454. doi: 10.1016/j.atmosres.2021.105454 [4] 覃皓, 郑凤琴, 伍丽泉. 台风威马逊(1409)强度与降水变化的相互作用. 应用气象学报, 2022, 33(4): 477-488. doi: 10.11898/1001-7313.20220408Qin H, Zheng F Q, Wu L Q. The interaction between intensity and rainfall of Typhoon Rammasun(1409). J Appl Meteor Sci, 2022, 33(4): 477-488. doi: 10.11898/1001-7313.20220408 [5] 郑倩, 毛程燕, 丁丽华, 等. 台风利奇马(1909)与台风摩羯(1814)云特征对比. 应用气象学报, 2022, 33(1): 43-55. doi: 10.11898/1001-7313.20220104Zheng Q, Mao C Y, Ding L H, et al. Comparison of cloud characteristics between Typhoon Lekima(1909) and Typhoon Yagi(1814). J Appl Meteor Sci, 2022, 33(1): 43-55. doi: 10.11898/1001-7313.20220104 [6] 高拴柱, 张胜军, 吕心艳, 等. 南海台风生成前48 h环流特征及热力与动力条件. 应用气象学报, 2021, 32(3): 272-288. doi: 10.11898/1001-7313.20210302Gao S Z, Zhang S J, Lü X Y, et al. Circulation characteristics and thermal and dynamic conditions 48 hours before typhoon formation in South China Sea. J Appl Meteor Sci, 2021, 32(3): 272-288. doi: 10.11898/1001-7313.20210302 [7] 郑宗生, 刘敏, 胡晨雨, 等. 基于Seq2Seq和Attention的时序卫星云图台风等级预测. 遥感信息, 2020, 35(4): 16-22. doi: 10.3969/j.issn.1000-3177.2020.04.003Zhang Z S, Liu M, Hu C Y, et al. Prediction of typhoon grade with time series typhoon satellite images based on Seq2Seq and Attention. Remote Sensing Information, 2020, 35(4): 16-22. doi: 10.3969/j.issn.1000-3177.2020.04.003 [8] 邹国良, 侯倩, 郑宗生, 等. 面向卫星云图及深度学习的台风等级分类. 遥感信息, 2019, 34(3): 1-6. https://www.cnki.com.cn/Article/CJFDTOTAL-YGXX201903001.htmZou G L, Hou Q, Zheng Z S, et al. Classification of typhoon grade based on satellite cloud image and deep learning. Remote Sensing Information, 2019, 34(3): 1-6. https://www.cnki.com.cn/Article/CJFDTOTAL-YGXX201903001.htm [9] 任素玲, 方翔, 卢乃锰, 等. 基于气象卫星的青藏高原低涡识别. 应用气象学报, 2019, 30(3): 345-359. doi: 10.11898/1001-7313.20190308Ren S L, Fang X, Lu N M, et al. Recognition method of the Tibetan Plateau vortex based on meteorological satellite data. J Trop Meteor, 2019, 30(3): 345-359. doi: 10.11898/1001-7313.20190308 [10] 王羽飞, 齐彦斌, 李倩, 等. 一次长白山夏季雾的宏微观特征. 应用气象学报, 2022, 33(4): 442-453. doi: 10.11898/1001-7313.20220405Wang Y F, Qi Y B, Li Q, et al. Macro and micro characteristics of a fog process in Changbai Mountain in summer. J Appl Meteor Sci, 2022, 33(4): 442-453. doi: 10.11898/1001-7313.20220405 [11] 马瑞阳, 郑栋, 姚雯, 等. 雷暴云特征数据集及我国雷暴活动特征. 应用气象学报, 2021, 32(3): 358-369. doi: 10.11898/1001-7313.20210308Ma R Y, Zheng D, Yao W, et al. Thunderstorm feature dataset and characteristics of thunderstorm activities in China. J Appl Meteor Sci, 2021, 32(3): 358-369. doi: 10.11898/1001-7313.20210308 [12] 任素玲, 牛宁, 覃丹宇, 等. 2021年2月北美极端低温暴雪的卫星遥感监测. 应用气象学报, 2022, 33(6): 696-710. doi: 10.11898/1001-7313.20220605Ren S L, Niu N, Qin D Y, et al. Extreme cold and snowstorm event in North America in February 2021 based on satellite data. J Appl Meteor Sci, 2022, 33(6): 696-710. doi: 10.11898/1001-7313.20220605 [13] 王磊, 周毓荃, 蔡淼, 等. 华北云特征参数与降水相关性的研究. 气象与环境科学, 2019, 42(3): 9-16. https://www.cnki.com.cn/Article/CJFDTOTAL-HNQX201903002.htmWang L, Zhou Y Q, Cai M, et al. Study on correlation between cloud characteristic parameters and precipitation in North China. Meteor Environ Sci, 2019, 42(3): 9-16. https://www.cnki.com.cn/Article/CJFDTOTAL-HNQX201903002.htm [14] 王帅辉, 韩志刚, 姚志刚, 等. 基于CloudSat资料的中国及周边地区云垂直结构统计分析. 高原气象, 2011, 30(1): 38-52. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201101006.htmWang S H, Han Z G, Yao Z G, et al. Analysis on cloud vertical structure over China and its neighborhood based on CloudSat data. Plateau Meteor, 2011, 30(1): 38-52. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201101006.htm [15] Genkova I S, Pachedjieva B, Ganev G, et al. Cloud motion estimation from METEOSAT images using time mutability method. International Society for Optics and Photonics, 1999, 3571: 297-301. [16] 陈明轩, 王迎春, 俞小鼎. 交叉相关外推算法的改进及其在对流临近预报中的应用. 应用气象学报, 2007, 18(5): 690-701. http://qikan.camscma.cn/article/id/200705105Chen M X, Wang Y C, Yu X D. Improvement and application test of TREC algorithm for convective storm nowcast. J Trop Meteor, 2007, 18(5): 690-701. http://qikan.camscma.cn/article/id/200705105 [17] 张乐坚, 程明虎, 田付友. 人工神经网络及支持向量机在降雨量预报中的应用. 高原气象, 2010, 29(4): 982-991. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201004018.htmZhang L J, Cheng M H, Tian F Y. Application of artificial neural networks and support vector machine in rainfall forecasting. Plateau Meteor, 2010, 29(4): 982-991. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201004018.htm [18] 刘科峰, 张韧, 孙照渤. 基于交叉相关法的卫星云图中云团移动的短时预测. 中国图象图形学报, 2006, 11(4): 586-591. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB200604023.htmLiu K F, Zhang R, Sun Z B. A cloud movement short-time forecast based on crosscorrelation. J Image Graph, 2006, 11(4): 586-591. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB200604023.htm [19] Vukicevic T, Greenwald T, Zupanski M, et al. Mesoscale cloud state estimation from visible and infrared satellite radiances. Mon Wea Rev, 2004, 132(12): 3066-3077. [20] Das S K, Chanda B, Mukherjee D P. Prediction of Cloud for Weather Now-casting Application Using Topology Adaptive Active Membrane//International Conference on Pattern Recognition and Machine Intelligence. Springer, 2009: 303-308. [21] Goswami B, Bhandari G. Automatically Adjusting Cloud Movement Prediction Model from Satellite Infrared Images//2011 Annual IEEE India Conference, 2012: 1-4. [22] 刘延安, 魏鸣, 高炜, 等. FY-2红外云图中强对流云团的短时自动预报算法. 遥感学报, 2012, 16(1): 79-82. https://www.cnki.com.cn/Article/CJFDTOTAL-YGXB201201007.htmLiu Y A, Wei M, Gao W, et al. Short-term automatic forecast algorithm of severe convective cloud identification using FY-2 IR images. J Remote Sens, 2012, 16(1): 79-92. https://www.cnki.com.cn/Article/CJFDTOTAL-YGXB201201007.htm [23] Morf H. Sunshine and cloud cover prediction based on Markov processes. Sol Energy, 2014, 110: 615-626. [24] 梁立为, 杨秀洪, 尹洁, 等. 一种云图的短时预测新方法探讨. 高原气象, 2015, 34(4): 1186-1190. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201504030.htmLiang L W, Yang X H, Yin J, et al. The new short-term cloud forecast method. Plateau Meteor, 2015, 34(4): 1186-1190. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201504030.htm [25] Sakaino H. Spatio-temporal image pattern prediction method based on a physical model with time-varying optical flow. IEEE Trans Geosci Remote Sens, 2013, 51(5): 3023-3036. [26] Chow C W, Belongie S, Kleissl J. Cloud motion and stability estimation for intra-hour solar forecasting. Sol Energy, 2015, 115: 645-655. [27] Tian L, Li X, Ye Y, et al. A generative adversarial gated recurrent unit model for precipitation nowcasting. IEEE Geoscience and Remote Sensing Letters, 2019, 17(4): 601-605. [28] 刘娜, 熊安元, 张强, 等. 强对流天气人工智能应用训练基础数据集构建. 应用气象学报, 2021, 32(5): 530-541. doi: 10.11898/1001-7313.20210502Liu N, Xiong A Y, Zhang Q, et al. Development of basic dataset of severe convective weather for artificial intelligence training. J Appl Meteor Sci, 2021, 32(5): 530-541. doi: 10.11898/1001-7313.20210502 [29] 袁凯, 李武阶, 李明, 等. 四种机器深度学习算法对武汉地区雷达回波临近预报的检验和评估. 气象, 2022, 48(4): 428-441. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX202204004.htmYuan K, Li W J, Li M, et al. Examination and avaluation of four machine deep learning algorithms for radar echo nowcasting in Wuhan Region. Meteor Mon, 2022, 48(4): 428-441. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX202204004.htm [30] Shi X, Gao Z, Lausen L, et al. Deep learning for precipitation nowcasting: A benchmark and a new model. Computer Science-Computer Vision and Pattern Recognition, 2017, arXiv. 1706.03458. [31] Pham T Q D, Hoang T V, Van Tran X, et al. Fast and accurate prediction of temperature evolutions in additive manufacturing process using deep learning. J Intell Manuf, 2022: 1-19. DOI: 10.1007/s10845-021-01896-8. [32] Wei J, Huang W, Li Z Q, et al. Estimating 1-km-resolution PM2.5 concentrations across China using the space-time random forest approach. Remote Sens Environ, 2019, 231: 111221. [33] Kristiani E, Lin H, Lin J R, et al. Short-term prediction of PM2.5 using LSTM deep learning methods. Sustainability, 2022, 14(4): 2068. [34] Zhang J, Liu P, Zhang F, et al. CloudNet: Ground-based cloud classification with deep convolutional neural network. Geophys Res Lett, 2018, 45(16): 8665-8672. [35] Hensel S, Rinov M B, Koch M, et al. Evaluation of deep learning-based neural network methods for cloud detection and segmentation. Energies, 2021, 14(19): 6156. [36] 何如, 管兆勇, 金龙. 一种神经网络的云图短时预测方法. 大气科学学报, 2010, 33(6): 725-730. https://www.cnki.com.cn/Article/CJFDTOTAL-NJQX201006010.htmHe R, Guan Z Y, Jin L. A short-term cloud forecast model by neural networks. Trans Atmos Sci, 2010, 33(6): 725-730. https://www.cnki.com.cn/Article/CJFDTOTAL-NJQX201006010.htm [37] Tan C, Feng X, Long J, et al. FORECAST-CLSTM: A New Convolutional LSTM Network for Cloudage Nowcasting. 2018 IEEE Visual Communications and Image Processing(VCIP), 2018. [38] Su X, Li T, An C, et al. Prediction of short-time cloud motion using a deep-learning model. Atmosphere, 2020, 11(11): 1151. [39] Lee J H, Lee S S, Kim H G, et al. MSCIP Net: Multichannel satellite image prediction via deep neural network. IEEE Trans Geosci Remote Sens, 2019, 58(3): 2212-2224. [40] 陶润喆. 基于风云4号卫星图像的西藏地区云检测和降水外推预报研究. 南京: 南京信息工程大学, 2021.Tao R Z. Cloud Detection and Precipitation Extrapolation Forecast in Tibet Based on Fengyun-4 Satellite Images. Nanjing: Nanjing University of Information Science & Technology, 2021. [41] 瞿建华, 张烺, 陆其峰, 等. 基于ERA5的快速辐射传输模式与FY-4A成像仪观测结果的偏差分析. 气象学报, 2019, 77(5): 911-922. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201905009.htmQu J H, Zhang L, Lu Q F, et al. Characterization of bias in FY-4A advanced geostationary radiation imager observations from ERA5 background simulations using RTTOV. Acta Meteor Sinica, 2019, 77(5): 911-922. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201905009.htm [42] Tang F, Zhuge X, Zeng M, et al. Applications of the advanced radiative transfer modeling system(ARMS) to characterize the performance of Fengyun-4A/AGRI. Remote Sens, 2021, 13(16): 3120. [43] 高洋, 蔡淼, 曹治强, 等. "21·7"河南暴雨环境场及云的宏微观特征. 应用气象学报, 2022, 33(6): 682-695. doi: 10.11898/1001-7313.20220604Gao Y, Cai M, Cao Z Q, et al. Environmental conditions and cloud macro and micro features of "21·7" extreme heavy rainfall in Henan Province. J Appl Meteor Sci, 2022, 33(6): 682-695. doi: 10.11898/1001-7313.20220604 [44] 齐道日娜, 何立富, 王秀明, 等. "7·20"河南极端暴雨精细观测及热动力成因. 应用气象学报, 2022, 33(1): 1-15. doi: 10.11898/1001-7313.20220101Chyi D, He L F, Wang X M, et al. Fine observation characteristics and thermodynamic mechanisms of extreme heavy rainfall in Henan on 20 July 2021. J Appl Meteor Sci, 2022, 33(1): 1-15. doi: 10.11898/1001-7313.20220101 [45] 陈元昭, 林良勋, 王蕊, 等. 基于生成对抗网络GAN的人工智能临近预报方法研究. 大气科学学报, 2019, 42(2): 311-320. https://www.cnki.com.cn/Article/CJFDTOTAL-NJQX201902014.htmChen Y Z, Lin L X, Wang R, et al. A study on the artificial intelligence nowcasting based on generative adversarial networks. Trans Atmos Sci, 2019, 42(2): 311-320. https://www.cnki.com.cn/Article/CJFDTOTAL-NJQX201902014.htm [46] Shi X, Chen Z, Wang H, et al. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. Computer Science-Computer Vision and Pattern Recognition, 2015, 28: 802-810. [47] Jing J, Li Q, Peng X. MLC-LSTM: Exploiting the spatiotemporal correlation between multi-level weather radar echoes for echo sequence extrapolation. Sensors, 2019, 19(18): 3988. [48] Kingma D P, Ba J. Adam: A method for stochastic optimization. Computer Science-Machine Learning, 2014, arXiv: 1412.6980. [49] Ruder S. An overview of gradient descent optimization. Comput Sci-Machine Learning, 2016, arXiv: 1609.04747. [50] Zhou W, Bovik A C, Sheikh H R, et al. Image quality assessment: From error visibility to structural similarity. IEEE Trans Image Process, 2004, 13: 600-612. [51] Zhao H, Gallo O, Frosio I, et al. Loss functions for image restoration with neural networks. IEEE Trans Comput Imaging, 2016, 3(1): 47-57. [52] Ayzel G, Heistermann M, Winterrath T. Optical flow models as an open benchmark for radar-based precipitation nowcasting(rainymotion v0.1). Geosci Model Dev, 2019, 12(4): 1387-1402. -
图(8) / 表(3)
计量
- 摘要浏览量: 1583
- HTML全文浏览量: 551
- PDF下载量: 175
- 被引次数: 0
