Evaluation of AI Model Predictions for the 2024 Meiyu Season in the Middle and Lower Reaches of the Yangtze River
-
摘要: 目前，人工智能大模型对长江中下游降水的次季节预测效果尚不清楚。采用三个人工智能气象大模型(Pangu-weather、Fuxi和FourCastNet)与欧洲中心次季节-季节模式(EC-S2S)预测资料，以2024年长江中下游梅雨为例，在诊断其降水及其环流演变的基础上，利用相关技巧、功率谱分析等方法，对比评估了气象大模型与欧洲中心S2S模式(EC-S2S)对该年梅雨降水、背景场变量及其低频振荡分量的预测效果，并与传统的EC-S2S模式进行了比较。结果表明：(1) 2024年6月第4候，受西太平洋副热带高压北抬和北侧西风槽发展南伸影响，长江中下游进入梅雨期。此后，梅雨及其相关联的夏季风、冷空气影响和湿度变化均呈现显著的准双周振荡。(2) 三个大模型与EC-S2S模式都能提前10 d较好地预测梅雨相联系的副高和西风槽的演变活动。当预测时效超前11 d时，三个大模型与EC-S2S模式预测对梅雨环流形势的预测不确定性增加，其中仅Pangu模型和EC-S2S模式超前16—20 d的预测能反映出长江中下游南北两侧的冷暖空气活动。(3) FourCastNet、Fuxi两个模型和EC-S2S模式能提前11—15 d给出有显著相关技巧的梅雨降水预测，也能提前11—15 d反映梅雨区降水及其相关联环流的准双周振荡特征。EC-S2S模式对降水量的预测优于大模型，但其准双周振荡功率谱值弱于大模型。Pangu、FourCastNet和EC-S2S模式能提前16—20 d预报长江中下游南侧的夏季风和北侧西风槽活动的准双周振荡影响。尽管大模型在超前半月以上的梅雨环流预测上存在挑战，但对其低频分量的有效超前预测时效更长，部分要素(如经向风和比湿)预测优于EC-S2S模式，表明从低频振荡这一途径出发开展大模型的次季节预测，可为后续应用和改进大模型的次季节预测提供新思路。Abstract: Currently, the performance of AI models in forecasting sub-seasonal precipitation in the middle and lower reaches of the Yangtze River remains unclear. This study used the Meiyu process in the middle and lower Yangtze River region in 2024 as an evaluation case to assess the prediction performance of three AI meteorological models (i.e., Pangu-weather, Fuxi, and FourCastNet) and Sub-seasonal to seasonal (S2S) Prediction data from the ECMWF. Moreover, based on an analysis of precipitation and circulation evolution using methods such as correlation skills and power spectrum analysis, we evaluated Meiyu precipitation, background field variables, and their low-frequency components and compared them with those from the conventional EC-S2S model. The results are as follows: (1) In the 4th pentad of June 2024, the Meiyu season commenced in the middle and lower reaches of the Yangtze River. It was influenced by the northward extension of the Western Pacific Subtropical High and the southward development of the westerly trough. Subsequently, the Meiyu and its associated summer monsoon, cold air influences, and humidity changes exhibited significant quasi-biweekly variations. (2) All three models and EC-S2S successfully captured the evolution of the subtropical anticyclone and the westerly trough within a 10-day lead time. Prediction uncertainties increased for all three models as well as EC-S2S after 11 days of lead time. Only Pangu-weather and EC-S2S continued to provide valuable references for predictions beyond 15 days. (3) FourCastNet, Fuxi, and EC-S2S provided skillful predictions of Meiyu precipitation with significant correlations 11-15 days in advance. They also accurately reflected the quasi-biweekly oscillation characteristics of precipitation and associated circulation in the Meiyu region within the same lead time. The EC-S2S model demonstrated high accuracy in precipitation prediction but had a weaker ability to predict the significance of quasi-biweekly characteristics. Pangu, FourCastNet, and EC-S2S were able to forecast the quasi-biweekly oscillation of the summer monsoon in the south and the westerly trough in the north of the Yangtze River Basin 16-20 days in advance. Although large models faced challenges in forecasting Meiyu circulation beyond a lead time of half a month, their effective lead time for predicting low-frequency components was longer, and their forecasts of some elements (e. g., V-wind and specific were better than those of EC-S2S. The results suggest that exploring sub-seasonal forecasting of large models from the perspective of low-frequency oscillations may provide new insights for subsequent applications and improvements in sub-seasonal forecasting.
-
Key words:
- AI models /
- Meiyu season /
- subseasonal forecasting /
- quasi-biweekly oscillation
-
图 1 2024年长江中下游区域(沿115~123 °E)梅汛期环流形势时间-纬度剖面实况演变
a.500 hPa高度场(阴影)，红线为纬向风为0等值线；b.2 m温度场(阴影，单位：℃)，红线为22 ℃等值线；c.850 hPa风矢量(单位：m·s-1)，阴影区域为大于2 m·s-1经向风；d. 日降水量(单位：mm)。
图 2 关键区的各气象要素区域平均时间序列实况功率谱分析
a.500 hPa高度场；b.850 hPa经向风；c.相对湿度；d.日降水量。实线表示功率谱，红虚线表示红噪声，蓝虚线表示95%显著性水平。
图 3 三个气象大模型(Pangu、Fourcastnet、Fuxi)与EC-S2S模式超前1—5 d、6—10 d、11—15 d、16—20 d预测的2024年梅汛期长江中下游区域(沿115~123 °E)500 hPa位势高度演变
图 4 两个气象大模型(Fourcastnet、Fuxi)与EC-S2S模式超前1—5 d、6—10 d、11—15 d、16—20 d预测的2024年梅汛期长江中下游区域(沿115~123 °E)降水演变
图 5 FourCastNet、Fuxi大模型与EC-S2S模式超前不同时效预测的长江中下游区域降水量变化的功率谱
图l为降水实况功率谱，实线表示功率谱，红虚线表示红噪声，蓝虚线表示95%显著性水平。
图 6 FourCastNet与Fuxi大模型超前不同时效预测的长江中下游区域相对湿度变化的功率谱
图h为相对湿度实况功率谱，实线表示功率谱，红虚线表示红噪声，蓝虚线表示95%显著性水平。
图 7 Pangu大模型与EC-S2S模式超前不同时效预测的长江中下游区域比湿变化的功率谱
图i为比湿实况功率谱，实线表示功率谱，红虚线表示红噪声，蓝虚线表示95%显著性水平。
图 8 Pangu、FourCastNet、Fuxi大模型与EC-S2S模式超前不同时效预测的长江中下游以北区域上空的500 hPa位势高度变化的功率谱
图p为位势高度实况功率谱实线表示功率谱，红虚线表示红噪声，蓝虚线表示95%显著性水平。
图 9 Pangu、FourCastNet、Fuxi大模型与EC-S2S模式超前不同时效预测的长江中下游区域低层850 hPa经向风变化的功率谱
图l为经向风实况功率谱实线表示功率谱，红虚线表示红噪声，蓝虚线表示95%显著性水平。
图 10 Pangu、FourCastNet、Fuxi大模型与EC-S2S模式超前不同时效预测的长江中下游梅雨北侧关键区500 hPa位势高度
红色实线，蓝色实线为观测)及其低频分量(红色虚线，蓝色虚线为观测，图例中cor为相关系数，p为显著性检验p值。
图 11 Pangu、FourCastNet、Fuxi大模型与EC-S2S模式超前不同时效预测的长江中下游梅雨南侧关键区500 hPa位势高度
红色实线，蓝色实线为观测，图例中cor为相关系数，p为显著性检验p值。
图 12 FourCastNet与Fuxi大模型超前不同时效预测的长江中下游梅雨关键区850 hPa相对湿度(红色实线，蓝色实线为观测)及其低频分量(红色虚线，蓝色虚线为观测，图例中cor为相关系数，p为显著性检验p值)
图 13 Pangu大模型与EC-S2S模式超前不同时效预测的长江中下游梅雨关键区850 hPa比湿(红色实线，蓝色实线为观测) 及其低频分量(红色虚线，蓝色虚线为观测，图例中cor为相关系数，p为显著性检验p值)
图 14 Pangu、FourCastNet、Fuxi大模型与EC-S2S模式超前不同时效预测的长江中下游梅雨区2 m温度
红色实线，蓝色实线为观测，图例中cor为相关系数，p为显著性检验p值。
-
[1] 梁萍, 孙国武, 马悦, 等. 梅雨变率和预测[M]. 北京: 中国环境出版集团, 2023. [2] 梁萍, 杨子凡, 谢潇, 等. 上海梅汛期候降水异常的低频信号及延伸期预报[J]. 气象科技, 2020, 48(5): 685-694. [3] 王遵娅, 丁一汇. 夏季长江中下游旱涝年季节内振荡气候特征[J]. 应用气象学报, 2008, 19(6): 710-715. [4] 曹鑫, 任雪娟, 杨修群, 等. 中国东南部5—8月持续性强降水和环流异常的准双周振荡[J]. 气象学报, 2012, 70(4): 766-778. [5] 陈蔚, 刘梅, 杨华栋, 等. 长江中下游典型涝年准双周振荡特征及其影响分析[J]. 气象学报, 2023, 81(3): 393-415. [6] DING Y H, LIU Y Y, HU Z Z. The record-breaking Meiyu in 2020 and associated atmospheric circulation and tropical SST anomalies[J]. Advances in Atmospheric Sciences, 2021, 38(12): 1 980-1 993. [7] HUANG W R, LIU P Y, CHEN J H, et al. Impact of boreal summer intra-seasonal oscillations on the heavy rainfall events in Taiwan during the 2017 Meiyu season[J]. Atmosphere, 2019, 10(4): 205. [8] DING Y H, LIANG P, LIU Y J, et al. Multiscale variability of Meiyu and its prediction: A new review[J]. Journal of Geophysical Research: Atmospheres, 2020, 125(7): e2019JD031496. [9] 刘芸芸, 丁一汇. 2020年超强梅雨特征及其成因分析[J]. 气象, 2020, 46(11): 1 393-1 404. [10] 丁一汇, 梁萍. 基于MJO的延伸预报[J]. 气象, 2010, 36(7): 111-122. [11] 金荣花, 马杰, 任宏昌, 等. 我国10~30天延伸期预报技术进展与发展对策[J]. 地球科学进展, 2019, 34(8): 814-825. [12] WCRP JOINT SCIENTIFIC COMMITTEE (JSC). World Climate Research Programme Strategic Plan 2019-2028[R]. Geneva: WCRP Publication, 2019. [13] ZHANG J P, SUN L Y, JIN C, et al. Recent advances in artificial intelligence generated content[J]. Frontiers of Information Technology & Electronic Engineering, 2024, 25(1): 1-6. [14] LYU Y, ZHU S P, ZHI X F, et al. Significant advancement in subseasonal-to-seasonal summer precipitation ensemble forecast skills in China mainland through an innovative hybrid CSG-UNET method[J]. Environmental Research Letters, 2024, 19(7): 074055. [15] LYU Y, ZHU S P, ZHI X F, et al. Improving subseasonal-to-seasonal prediction of summer extreme precipitation over southern China based on a deep learning method[J]. Geophysical Research Letters, 2023, 50(24): e2023GL106245. [16] 杨淑贤, 零丰华, 应武杉, 等. 人工智能技术气候预测应用简介[J]. 大气科学学报, 2022, 45(5): 641-659. [17] DUEBEN P D, BAUER P. Challenges and design choices for global weather and climate models based on machine learning[J]. Geoscientific Model Development, 2018, 11(10): 3 999-4 009. [18] WEYN J A, DURRAN D R, CARUANA R. Can machines learn to predict weather? Using deep learning to predict gridded 500-hPa geopotential height from historical weather data[J]. Journal of Advances in Modeling Earth Systems, 2019, 11(8): 2 680-2 693. [19] ZHANG Y C, LONG M S, CHEN K Y, et al. Skillful nowcasting of extreme precipitation with NowcastNet[J]. Nature, 2023, 619(7970): 526-532. [20] SCHULTZ M G, BETANCOURT C, GONG B, et al. Can deep learning beat numerical weather prediction?[J]. Philosophical Transactions of the Royal Society A, 2021, 379(2194): 20200097. [21] KURTH T, SUBRAMANIAN S, HARRINGTON P, et al. Fourcastnet: Accelerating global high-resolution weather forecasting using adaptive Fourier neural operators[C]//Proceedings of the platform for advanced scientific computing conference. 2023: 1-11. [22] CHEN L, ZHONG X, ZHANG F, et al. FuXi: A cascade machine learning forecasting system for 15-day global weather forecast[J]. npj Climate and Atmospheric Science, 2023, 6(1): 190. [23] LAM R, SANCHEZ-GONZALEZ A, WILLSON M, et al. Learning skillful medium-range global weather forecasting[J]. Science, 2023, 382 (6677): 1 416-1 421. [24] BI K F, XIE L X, ZHANG H H, et al. Accurate medium-range global weather forecasting with 3D neural networks[J]. Nature, 2023, 619 (7970): 533-538. [25] CHEN K, HAN T, GONG J, et al. Fengwu: Pushing the skillful global medium-range weather forecast beyond 10 days lead[J]. arXiv, 2023. [26] 黄小猛, 林岩銮, 熊巍, 等. 数值预报AI气象大模型国际发展动态研究[J]. 大气科学学报, 2024, 47(1): 46-54. [27] OLIVETTI L, MESSORI G. Do data-driven models beat numerical models in forecasting weather extremes? A comparison of IFS HRES, Pangu-Weather and GraphCast[J]. EGUsphere, 2024, 2024: 1-35. [28] XU H X, ZHAO Y, ZHAO D J, et al. Improvement of disastrous extreme precipitation forecasting in North China by Pangu-weather AI-driven regional WRF model[J]. Environmental Research Letters, 2024, 19(5): 054051. [29] CHARLTON-PEREZ A J, DACRE H F, DRISCOLL S, et al. Do AI models produce better weather forecasts than physics-based models? A quantitative evaluation case study of Storm Ciarán[J]. npj Climate and Atmospheric Science, 2024, 7(1): 93. [30] 王会军, 戴永久, 杨崧, 等. 气候系统预测: 基础创新和集成应用[J]. 大气科学学报, 2024, 47(2): 161-172. [31] HERSBACH H, BELL B, BERRISFORD P, et al. The ERA5 global reanalysis[J]. Quarterly Journal of the Royal Meteorological Society, 2020, 146(730): 1 999-2 049. [32] HUANG G, WANG Y, HAM Y G, et al. Toward a learnable climate model in the artificial intelligence era[J]. Advances in Atmospheric Sciences, 2024, 41: 1 281-1 288. [33] GUIBAS J, MARDANI M, LI Z Y, et al. Efficient token mixing for transformers via adaptive Fourier neural operators[C]//International Conference on Learning Representations. 2022. [34] LI Y, SUN Y, WANG W W, et al. U-shaped transformer with frequency-band aware attention for speech enhancement[J]. IEEE / ACM Transactions on Audio, Speech, and Language Processing, 2023, 31: 1 511-1 521. [35] 姚菊香, 王盘兴, 李丽平. 季节内振荡研究中两种数字滤波器的性能对比[J]. 南京气象学院学报, 2005, 28(2): 248-253. [36] 王文, 孙畅, 蔡晓军, 等. 南亚高压低频振荡与长江中下游地区旱涝的关系[J]. 地球科学进展, 2016, 31(5): 529-541. [37] CHEN L, ZHONG X H, LI H, et al. A machine learning model that outperforms conventional global subseasonal forecast models[J]. Nature Communications, 2024, 15(1): 6 425. -
下载:
点击查看大图
计量
- 文章访问数: 8
- HTML全文浏览量: 2
- PDF下载量: 0
- 被引次数: 0
粤公网安备 4401069904700003号