A Deep Learning Method for Bias Correction of ECMWF 24–240 h Forecasts

Correcting the forecast bias of numerical weather prediction models is important for severe weather warnings. The refined grid forecast requires direct correction on gridded forecast products, as opposed to correcting forecast data only at individual weather stations. In this study, a deep learning method called CU-net is proposed to correct the gridded forecasts of four weather variables from the European Centre for Medium-Range Weather Forecast Integrated Forecasting System global model(ECMWF-IFS): 2-m temperature, 2-m relative humidity, 10-m wind speed, and 10-m wind direction, with a forecast lead time of 24 h to 240 h in North China. First, the forecast correction problem is transformed into an image-toimage translation problem in deep learning under the CU-net architecture, which is based on convolutional neural networks.Second, the ECMWF-IFS forecasts and ECMWF reanalysis data(ERA5) from 2005 to 2018 are used as training,validation, and testing datasets. The predictors and labels(ground truth) of the model are created using the ECMWF-IFS and ERA5, respectively. Finally, the correction performance of CU-net is compared with a conventional method, anomaly numerical correction with observations(ANO). Results show that forecasts from CU-net have lower root mean square error, bias, mean absolute error, and higher correlation coefficient than those from ANO for all forecast lead times from 24 h to 240 h. CU-net improves upon the ECMWF-IFS forecast for all four weather variables in terms of the above evaluation metrics, whereas ANO improves upon ECMWF-IFS performance only for 2-m temperature and relative humidity. For the correction of the 10-m wind direction forecast, which is often difficult to achieve, CU-net also improves the correction performance.     

数值天气预报多要素深度学习融合订正方法

数值天气预报作为现代天气预报的主流技术方法,近年来不断朝着精细化方向发展,但预报误差至今仍无法避免.文中在CU-Net模型中引入稠密卷积模块形成数值预报要素偏差订正模型Dense-CUnet,在此基础上进一步融合多种气象要素和地形特征构建了Fuse-CUnet模型,开展不同模型的偏差订正试验和对比分析.以均方根误差(R...     

A Model Output Machine Learning Method for Grid Temperature Forecasts in the Beijing Area

In this paper, the model output machine learning (MOML) method is proposed for simulating weather consultation, which can improve the forecast results of numerical weather prediction (NWP). During weather consultation, the forecasters obtain the final results by combining the observations with the NWP results and giving opinions based on their experience. It is obvious that using a suitable post-processing algorithm for simulating weather consultation is an interesting and important topic. MOML is a post-processing method based on machine learning, which matches NWP forecasts against observations through a regression function. By adopting different feature engineering of datasets and training periods, the observational and model data can be processed into the corresponding training set and test set. The MOML regression function uses an existing machine learning algorithm with the processed dataset to revise the output of NWP models combined with the observations, so as to improve the results of weather forecasts. To test the new approach for grid temperature forecasts, the 2-m surface air temperature in the Beijing area from the ECMWF model is used. MOML with different feature engineering is compared against the ECMWF model and modified model output statistics (MOS) method. MOML shows a better numerical performance than the ECMWF model and MOS, especially for winter. The results of MOML with a linear algorithm, running training period, and dataset using spatial interpolation ideas, are better than others when the forecast time is within a few days. The results of MOML with the Random Forest algorithm, year-round training period, and dataset containing surrounding gridpoint information, are better when the forecast time is longer.     

冬奥延庆赛区风的预报技巧比较分析

为了提高2 min平均的10 m风预报精度,开展了多种建模和检验方法比较.根据欧洲数值中心集合预报系统产品及北京海陀山的5个测站资料,使用一元回归、岭回归、神经网络、粒子群-神经网络等方法建模,进行2021年2月逐日的未来3天6 h间隔预报误差订正,并从多个角度分析预报精度差异.结果为:(1)系统误差、预报准确率检验表...     

Forecasting Different Types of Convective Weather: A Deep Learning Approach

A deep learning objective forecasting solution for severe convective weather(SCW) including short-duration heavy rain(HR), hail, convective gusts(CG), and thunderstorms based on numerical weather prediction(NWP) data was developed. We first established the training datasets as follows. Five years of severe weather observations were utilized to label the NCEP final(FNL) analysis data. A large number of labeled samples for each type of weather were then selected for model training. The local temperature, pressure, humidity, and winds from 1000 to 200 h Pa, as well as dozens of convective physical parameters, were taken as predictors in our model. A six-layer convolutional neural network(CNN) model was then built and trained to obtain optimal model weights. After that, the trained model was used to predict SCW based on the Global Forecast System(GFS) forecast data as input. The performances of the CNN model and other traditional methods were compared. The results show that the deep learning algorithm had a higher classification accuracy on HR and hail than support vector machine, random forests, and other traditional machine learning algorithms. The objective forecasts by use of the deep learning algorithm also showed better forecasting skills than the subjective forecasts by the forecasters. The threat scores(TSs) of thunderstorm, HR, hail, and CG were increased by 16.1%, 33.2%, 178%, and 55.7%, respectively. The deep learning forecast model is currently used in the National Meteorological Center of China to provide guidance for the operational SCW forecasting over China.     

2010年国内外3种数值预报在东北地区的预报检验

对中国国家气象中心T639数值预报、德国降水预报和欧洲中心（ECMWF）数值预报中对中国东北地区的降水、温度和环流模式的预报结果,分别进行检验。结果表明：降水预报中,德国降水预报对中国东北地区的晴雨预报、一般性降水预报效果较好,但T639数值预报的漏报率明显低于德国降水预报,而T639数值预报中24-120 h暴雨预报的TS评分明显高于德国降水预报。温度预报中,T639数值预报对中国东北地区温度预报72 h内基本可用;欧洲中心（ECMWF）数值预报在96 h内效果较好。对于24 h的温度预报准确率,T639数值预报稍高于ECMWF数值预报结果。环流模式方面：48h内T639预报效果好于ECMWF,72 h以后ECMWF预报效果好于T639。     

基于ECMWF产品的站点气温预报集成学习误差订正

提出一种基于数值模式预报产品的气温预报集成学习误差订正方法,通过人工神经网络、长短期记忆网络和线性回归模型组合出新的集成学习模型（ALS模型）,采用2013—2017年的欧洲中期天气预报中心数值天气预报模式2 m气温预报产品和中国部分气象站点数据,利用气象站点气温、风速、气压、相对湿度4个观测要素,挖掘观测数据的时序特征并结合模式2 m气温预报结果训练机器学习模型,对2018年模式2 m气温6~168 h格点预报产品插值到站点后的预报结果进行偏差订正。结果表明:ALS模型可将站点气温预报整体均方根误差由3.11℃降至2.50℃,降幅达0.61℃（19.6%）,而传统的线性回归模型降幅为0.23℃（8.4%）。ALS模型对站点气温预报误差较大的区域和气温峰值预报的订正效果尤为显著,因此,集成学习方法在数值模式预报结果订正中具有较大的应用潜力。     

数值预报误差订正技术中相似-动力方法的发展

Due to the increasing requirement for high-level weather and climate forecasting accuracy, it is necessary to exploit a strategy for model error correction while developing numerical modeling and data assimilation techniques. This study classifies the correction strategies according to the types of forecast errors, and reviews recent studies on these correction strategies. Among others, the analogue-dynamical method has been developed in China, which combines statistical methods with the dynamical model, corrects model errors based on analogue information, and effectively utilizes historical data in dynamical forecasts. In this study, the fundamental principles and technical solutions of the analogue-dynamical method and associated development history for forecasts on different timescales are introduced. It is shown that this method can effectively improve medium- and extended-range forecasts, monthly-average circulation forecast, and short-term climate prediction. As an innovative technique independently developed in China, the analogue- dynamical method plays an important role in both weather forecast and climate prediction, and has potential applications in wider fields.     

机器学习在强对流监测预报中的应用进展

近年来,机器学习理论和方法应用蓬勃发展,已在强对流天气监测和预报中广泛应用。各类机器学习算法,包括传统机器学习算法(如随机森林、决策树、支持向量机、神经网络等)和深度学习方法,已在强对流监测、短时临近预报、短期预报领域发挥了积极的重要作用,其应用效果往往明显优于依靠统计特征或者主观经验积累的传统方法。机器学习方法能够更有效提取高时空分辨率的中小尺度观测数据的强对流特征,为强对流监测提供更全面、更强大的自动识别和追踪能力;能够有效综合应用多源观测数据、分析数据和数值预报模式数据,为强对流临近预报预警提取更多有效信息;能够有效对数值模式预报进行释用和后处理,提升全球数值模式、高分辨率区域数值模式在强对流天气预报上的应用效果。最后,给出了目前机器学习方法应用中存在的问题和未来工作展望。     

Comparison of Statistical Post-Processing Methods for Probabilistic Wind Speed Forecasting

In this study, the statistical post-processing methods that include bias-corrected and probabilistic forecasts of wind speed measured in PyeongChang, which is scheduled to host the 2018 Winter Olympics, are compared and analyzed to provide more accurate weather information. The six post-processing methods used in this study are as follows: mean bias-corrected forecast, mean and variance bias-corrected forecast, decaying averaging forecast, mean absolute bias-corrected forecast, and the alternative implementations of ensemble model output statistics (EMOS) and Bayesian model averaging (BMA) models, which are EMOS and BMA exchangeable models by assuming exchangeable ensemble members and simplified version of EMOS and BMA models. Observations for wind speed were obtained from the 26 stations in PyeongChang and 51 ensemble member forecasts derived from the European Centre for Medium-Range Weather Forecasts (ECMWF Directorate, 2012) that were obtained between 1 May 2013 and 18 March 2016. Prior to applying the post-processing methods, reliability analysis was conducted by using rank histograms to identify the statistical consistency of ensemble forecast and corresponding observations. Based on the results of our study, we found that the prediction skills of probabilistic forecasts of EMOS and BMA models were superior to the biascorrected forecasts in terms of deterministic prediction, whereas in probabilistic prediction, BMA models showed better prediction skill than EMOS. Even though the simplified version of BMA model exhibited best prediction skill among the mentioned six methods, the results showed that the differences of prediction skills between the versions of EMOS and BMA were negligible.     

最优子集神经网络在武威气温客观预报中的应用

选取2003年3月1日至2008年12月31日20时的逐日ECMWF(欧洲中期天气预报中心)数值预报产品实况格点资料,使用差分法、天气诊断、因子组合等方法,构造出能反映本地天气动力学特征的预报因子库,采用PRESS(预测平方和)准则初选因子,逐步回归复选因子,最优子集回归精选因子,建立分月、分站点逐日最高、最低温度BP神经网络预报模型。模型业务试用结果表明:该BP神经网络预报模型具有较强的非线性处理能力,能较好地反映日极端温度的变化,0～120 h内的最高、最低温度平均预报准确率达较高水平,且对明显的升降温过程反应灵敏,升降温趋势和幅度预报较为准确,为0～120 h的城镇精细化温度预报提供了重要的技术支撑,同时也为ECMWF数值预报产品在温度的释用提供了一种好的思路和方法。     

冬季稳定性降水相态预报研究进展

冬季降水无论对地面的生产生活还是对高空飞机航行都可能造成严重灾害,降水相态预报的准确性决定了冬季降水预报的成功,该文系统回顾了近几十年降水相态预报取得的成果。降水相态预报方法大致分为3类:第1类是基于观测或数值天气预报建立的指标以及回归方程,其中某些方法高度依赖数值天气预报模式准确率;第2类是基于数值天气预报模式的微物理方案法和集合预报法;第3类是基于观测和数值预报产品的人工智能预报法。近年来降水相态模式预报产品准确率不断提高,成为降水相态预报中一个重要的产品支撑。但如何将降水相态形成机制的微物理研究成果用于改善数值预报模式降水相态预报的技巧,以及如何利用人工智能等技术提高降水相态预报的准确率等方面还需要不断努力。     

Evaluation of TIGGE Ensemble Forecasts of Precipitation in Distinct Climate Regions in Iran

The application of numerical weather prediction(NWP) products is increasing dramatically. Existing reports indicate that ensemble predictions have better skill than deterministic forecasts. In this study, numerical ensemble precipitation forecasts in the TIGGE database were evaluated using deterministic, dichotomous(yes/no), and probabilistic techniques over Iran for the period 2008–16. Thirteen rain gauges spread over eight homogeneous precipitation regimes were selected for evaluation.The Inverse Distance Weighting and Kriging methods were adopted for interpolation of the prediction values, downscaled to the stations at lead times of one to three days. To enhance the forecast quality, NWP values were post-processed via Bayesian Model Averaging. The results showed that ECMWF had better scores than other products. However, products of all centers underestimated precipitation in high precipitation regions while overestimating precipitation in other regions. This points to a systematic bias in forecasts and demands application of bias correction techniques. Based on dichotomous evaluation,NCEP did better at most stations, although all centers overpredicted the number of precipitation events. Compared to those of ECMWF and NCEP, UKMO yielded higher scores in mountainous regions, but performed poorly at other selected stations.Furthermore, the evaluations showed that all centers had better skill in wet than in dry seasons. The quality of post-processed predictions was better than those of the raw predictions. In conclusion, the accuracy of the NWP predictions made by the selected centers could be classified as medium over Iran, while post-processing of predictions is recommended to improve the quality.     

Probabilistic temperature forecasting with statistical calibration in Hungary

Weather forecasting is based on the outputs of deterministic numerical weather forecasting models. Multiple runs of these models with different initial conditions result in forecast ensembles which are used for estimating the distribution of future atmospheric variables. However, these ensembles are usually under-dispersive and uncalibrated, so post-processing is required. In the present work, Bayesian model averaging (BMA) is applied for calibrating ensembles of temperature forecasts produced by the operational limited area model ensemble prediction system of the Hungarian Meteorological Service (HMS). We describe two possible BMA models for temperature data of the HMS and show that BMA post-processing significantly improves calibration and probabilistic forecasts although the accuracy of point forecasts is rather unchanged.     

基于机器学习的数值天气预报风速订正研究

对风速进行准确预测是精细化天气预报服务(如风能发电、冬季奥运会赛场条件保障等)的重要环节。本文基于三种机器学习算法(LASSO回归、随机森林和深度学习),对数值天气预报模式ECMWF预测的华北地区近地面10 m风速进行订正。首先利用LASSO回归算法提取对10 m风速有重要影响的气象要素特征集,将其作为三种机器学习算法的输入,建立相应模型对ECMWF预测的风速进行订正。用提取后的气象要素特征集建模有助于减少计算量和存储开销,并减小模型的复杂性,从而提高模型的泛化能力。将订正结果与传统订正方法模式输出统计(model output statistics,MOS)得到的订正结果进行对比。结果表明,三种机器学习算法的订正效果均好于MOS方法,显示了机器学习方法在改善局地精准气象预报方面的潜力。     

数值预报AI气象大模型国际发展动态研究

数值预报是研究地球系统的重要工具,有助于加深科学家对大气、海洋、气候和环境等复杂系统之间相互作用和变化过程的理解,在防灾减灾、气候变化和环境治理等方面发挥着不可或缺的作用。随着模式复杂度和分辨率的提高,传统数值模式在气候变化研究和气候预测方面取得了迅速的进展,但也面临一些挑战,需要得到数据同化、集合耦合、高性能计算和不确定性分析等多方面的支持。而近年来,“AI+气象”的交叉研究在气象领域引起了广泛关注。基于多种深度学习架构的人工智能大模型,依托强大的计算资源和海量的数据进行训练,能够以新的科学范式进行高效数值预报。气象大模型不断涌现,一些科技公司如华为、英伟达、DeepMind、谷歌、微软等,以及国内外高校如清华大学、复旦大学、密歇根大学、莱斯大学等发布了多个涵盖临近预报、短时预报、中期预报和延伸期预报等不同领域的气象大模型。这标志着人工智能与气象领域的交叉融合已经达到新的高度。尽管气象大模型在现阶段取得了较大突破,但其发展仍然面临弱可解释性、泛化能力不足、极端事件预报强度偏低、智能预报结果过平滑、深度学习框架能力需要拓展等诸多挑战。     

春季中国日最高气温延伸期预报误差分析及订正

数值模式直接输出和经模式后处理得到的预报误差比较,是延伸期逐日要素预报应用基础。针对中国2 583个站点在2020年春季11~30天的日最高温度预报,根据欧洲数值中心的集合预报输出,首先,使用BP-SM（Back-Propagation - Self memory）法和回归法,进行确定性预报订正效果比较;结果表明BP-SM法和回归法都明显降低了预报绝对误差;在11~14天预报中,BP-SM法得到的平均绝对误差为3.3~3.6 ℃,预报准确率超过35%,订正效果更优。其次,基于模式直接输出和BP-SM法获得的概率预报,使用CRPSS (continuous ranked probability skill score)进行了可预报性分析。结果表明,在地形复杂地区,经过订正,预报准确率明显改善。对于延伸期逐日要素预报,合理的模式后处理方法是降低预报误差和提高预报能力的重要环节。      

京津冀站点风温湿要素的机器学习订正方法

基于线性回归方法、梯度提升回归方法(GBRT方法)、XGBoost方法和堆叠集成学习方法(Stacking方法)4种机器学习方法,采用误差分析建模思路,针对北京城市气象研究院研发的睿图-睿思系统对2020年12月—2021年11月所有起报时次未来3~12 h的2 m温度、2 m相对湿度、10 m风速以及10 m风向4种气象要素预报,开展京津冀复杂地形下的站点预报误差订正技术研究及试验应用。结果表明:基于预报误差分析构建的4种订正模型中,由于Stacking方法集成了前3种方法的优势,在4个季节的4种气象要素订正中均表现最佳,其他3种单一机器学习方法试验中,XGBoost方法表现最佳,其后依次为GBRT方法、线性回归方法,但均对预报准确率有明显的正向提升效果。总体上,基于机器学习方法构建的预报误差订正模型可有效降低系统原始预报误差,有助于进一步提升复杂地形下站点客观释用产品的预报准确性。     

基于机器学习的华北气温多模式集合预报的订正方法

模式预报的订正是决定局地天气预报结果的一个重要步骤,基于机器学习的后处理模型近年来开始崭露头角。本文发展了基于岭回归（Ridge）、随机森林（Random Forest,RF）和深度学习（Deep Learning,DL）的3种后处理模型,基于中国气象局（CMA）的BABJ模式、欧洲中期天气预报中心（ECMWF）的ECMF模式、日本气象厅（JMA）的RJTD模式和NCEP的KWBC模式这4个数值天气预报模式2014年2月至2016年9月（训练期）近地面2 m气温预报和实况资料确定各模型参数,进而对2016年10月至2017年9月（预报期）华北地区（38°N~43°N,113°E~119°E）的逐日地面2 m气温预报进行了多模式集合预报分析。采用均方根误差对预报效果进行评估,这3种后处理模型的预报效果和4个数值天气预报模式以及通常的多模式集合平均（Ensemble Mean,EMN）的预报效果的对比表明:1）随着预报时长增加,4个数值预报模式及各种后处理模型的均方根误差均呈上升趋势;但区域平均而言,Ridge、RF和DL的预报效果在任何预报时长上都明显优于EMN和单个天气预报模式;特别是前几天的短期预报DL的预报效果更好,中后期预报Ridge的预报效果略好。2）华北地区的东南部均方根误差较小,其余格点上均方根误差较高,从空间分布而言,DL的订正预报效果最好,3种机器学习模型的误差在1.24~1.26℃之间,而EMN的误差达1.69℃。3）夏季各种方法的预报效果都较好,冬季预报效果都较差;但是Ridge、RF和DL的预报效果明显优于EMN,这3种模型预报的平均均方根误差在2.15~2.18℃之间,而EMN的平均均方根误差达2.45℃。     

深度学习技术在智能网格天气预报中的应用进展与挑战

中国智能网格天气预报已初步建立0—30 d涵盖基本气象要素的无缝隙气象预报业务体系。近年深度学习技术兴起,给不同领域带来前所未有的变革。同样,深度学习的非线性映射能力、海量信息提取能力、时空建模能力等优势为进一步提升智能网格预报的准确性和精细化水平提供了新的思路和方法。越来越多的研究将深度学习技术应用于智能网格预报的各个方面,包括数值预报订正和解释应用、集合天气预报、相似集合、统计降尺度、纯数据驱动的预报模型和极端天气预报等,并展示出良好的应用潜力。然而,目前深度学习技术在天气预报领域的应用仍处于起步阶段,将其引入智能网格预报业务体系还面临诸多挑战,主要包括算法的选择、算法的数据基础、多源数据融合以及模型的可解释性、可信度、可用性和工程化等。通过回顾近年来深度学习技术在智能网格预报中的应用进展和前景,同时对面临的挑战与应对进行探讨,将有利于促进深度学习技术在天气客观预报领域更好、更稳定的发展。