Ensemble forecasting of tropical cyclone motion: comparisonbetween regional bred modes and random perturbations

Summary Random perturbations (RPs) and a modified version for breeding of growing modes are used with a regional baroclinic mesoscale model to perform ensemble forecasting of tropical cyclone motion. Based on a sample of six cases, similar conclusions are found as in previous barotropic modeling studies. Even after introducing a larger spatial correlation into the RPs using a multi-quadric analysis scheme, the skill of this ensemble mean track prediction is almost always lower than that of the control forecast in the cases considered. The track prediction performance of the ensemble using regional bred modes (RBMs) as perturbations has a higher average skill. At nearly all forecast intervals except less than 24 h when the initial position error still dominates, the ensemble mean tracks in all six cases are improved over the control forecast. In the 6 h–24 h range, the success rate (ratio of the cases with a forecast improvement to the total number of cases) has a value of 10/24. In the 30 h–48 h range, the success rate increases to 20/24, but drops to 18/24 in the 54 h–72 h range. A relative skill score (RSS) is used to compare the skills of the two perturbation methodologies. It is found that the average RSSs of using RBMs are significantly higher than the corresponding ones of RPs at the 99% confidence level in all three 24-h periods. Note that the above conclusion is only based on ensemble mean forecasts. All of the possibilities from an ensemble-based probabilistic track distribution are not explored in this paper. The ensemble spreads in these RBM ensembles are large enough to include the verifying tracks in all the cases considered. It is also found that the ensemble spread is well correlated with the average error in an ensemble when using RBMs, but not with the ensemble mean forecast error in both methodologies. Received February 7, 2001/Revised April 18, 2001     

Ocean–atmosphere coupling and the boreal winter MJO

The influence of ocean–atmosphere coupling on the simulation and prediction of the boreal winter Madden–Julian Oscillation (MJO) is examined using the Seoul National University coupled general circulation model (CGCM) and atmospheric—only model (AGCM). The AGCM is forced with daily SSTs interpolated from pentad mean CGCM SSTs. Forecast skill is examined using serial extended simulations spanning 26 different winter seasons with 30-day forecasts commencing every 5 days providing a total of 598 30-day simulations. By comparing both sets of experiments, which share the same atmospheric components, the influence of coupled ocean–atmosphere processes on the simulation and prediction of MJO can be studied. The mean MJO intensity possesses more realistic amplitude in the CGCM than in AGCM. In general, the ocean–atmosphere coupling acts to improve the simulation of the spatio-temporal evolution of the eastward propagating MJO and the phase relationship between convection (OLR) and SST over the equatorial Indian Ocean and the western Pacific. Both the CGCM and observations exhibit a near-quadrature relationship between OLR and SST, with the former lagging by about two pentads. However, the AGCM shows a less realistic phase relationship. As the initial conditions are the same in both models, the additional forcing by SST anomalies in the CGCM extends the prediction skill beyond that of the AGCM. To test the applicability of the CGCM to real-time prediction, we compute the Real-time Multivariate MJO (RMM) index and compared it with the index computed from observations. RMM1 (RMM2) falls away rapidly to 0.5 after 17–18 (15–16) days in the AGCM and 18–19 (16–17) days in the CGCM. The prediction skill is phase dependent in both the CGCM and AGCM.     

Skill as a function of time scale in ensembles of seasonal hindcasts

Forecast skill as a function of time lead and time averaging is examined in two 6-member ensembles of seasonal hindcasts. One ensemble is produced with the second generation general circulation model of the Canadian Centre for Climate Modelling and Analysis (GCM2) and the other with a reduced resolution version of the numerical weather prediction model of the Canadian Meteorological Centre (SEF). The integrations are initiated from the NCEP/NCAR reanalyzed data. Monthly sea surface temperature anomalies observed prior to the forecast period are maintained throughout the forecast season. A statistical forecast improvement technique, based on the singular value decomposition of forecast and reanalyzed fields, is discussed and evaluated. A simple analogue of the hindcast integrations is used to examine the behavior of two common skill scores, the correlation skill score and the explained variance skill score. The maximal skill score and the corresponding optimal forecast in this analogue are identified. The total skill of the optimal forecast is a sum of two terms, one associated with the initial conditions and the other with the lower boundary forcing. The two sources of skill operate on different time scales, with initial conditions being more important in the first one-two weeks and the atmospheric response to the boundary forcing becoming more dominant for longer time leads and time averages. This suggests that these sources of skill should be considered separately in forecast optimization. The statistical technique is moderately successful in improving the skill of monthly to seasonal forecasts of 500 hPa height (Z ₅₀₀) and 700 hPa temperature (T ₇₀₀) in the Northern Hemisphere and in the North Pacific/North America sector. The improvement is better when the forecasts for the first week and for the rest of the season are optimized separately. The SEF model produces better Z ₅₀₀ and T ₇₀₀ forecasts than GCM2 in the first one-two weeks whereas GCM2 performs slightly better at longer time leads. The skill of zero time lead forecast decays rapidly with averaging interval for time averages up to about 30–45 days and stabilizes, or even rises, for longer time averages. Excluding the first week from seasonal forecasts results in substantial degradation of predictive skill. Received: 1 November 1999 / Accepted: 24 May 2000     

集合方法在月动力预报信息提取中的应用

本工作将集合方法应用于提取月动力预报有用信息。利用中国气象局国家气候中心Ｔ６３Ｌ１６全球谱模式的５００百帕高度场月集合预报产品（集合成员数为８个，初始场的选取采用滞后方法（ＬＡＦ），即相邻两天的００００，０６００，１２００和１８００ＧＭＴ的初始化资料），就１９９７年１月至５月共１５次预报，分析了集合预报成员间的离散度与预报评分（距平相关系数和均方根误差）的关系，研究了用集合各成员预报离散度作为各个成员逐日预报的权重对月预报效果的影响。结果表明集合预报成员的离散度与预报评分有显著的相关，是有效预报长度Ｎ的一个很好估计；用离散度作为权重平均的月预报高度距平相关系数明显高于算术平均和线性权重，此外个例分析表明月平均环流及其异常的预报得到明显的提高。     

Application of a serial extended forecast experiment using the ECMWF model to interpret the predictive skill of tropical intraseasonal variability

The extended-range forecast skill of the ECMWF operational forecast model is evaluated during tropical intraseasonal oscillation (ISO) events in the Indo-West Pacific warm pool. The experiment consists of ensemble extended serial forecasts including winter and summer ISO cases. The forecasts are compared with the ERA-40 analyses. The analysis focuses on understanding the origin of forecast errors by studying the vertical structure of relevant dynamical and moist convective features associated with the ISO. The useful forecast time scale for circulation anomalies is in average 13 days during winter compared to 7–8 days during summer. The forecast skill is not stationary and presents evidence of a flow-dependent nature, with states of the coupled system corresponding to long-lived convective envelopes associated with the ISO for which the skill is always low regardless of the starting date of the forecast. The model is not able to forecast skillfully the generation of specific humidity anomalies and results indicate that the convective processes in the model are associated with the erosion of the ISO forecast skill in the model. Circulation-associated anomalies are forecast better than moist convective associated anomalies. The model tends to generate a more stable atmosphere, limiting the model’s capability to reproduce deep convective events, resulting in smaller humidity and circulation anomalies in the forecasts compared to those in ERA-40.     

基于背景场奇异向量的CMA全球集合预报试验

目前中国气象局全球集合预报系统（China Meteorological Administration Global Ensemble Prediction System,CMA-GEPS）利用CMA全球数值预报系统分析场计算奇异向量（ANSV）,欧洲中期天气预报中心采用同化背景场计算奇异向量（FCSV）,在业务流程上先于计算ANSV,可优化集合预报系统运行时间。为此,在CMA-GEPS中探索采用FCSV进行集合预报的可行性,分析ANSV和FCSV的空间分布及相似指数,进而针对夏秋季节10个个例开展采用ANSV和FCSV的全球集合预报试验,从等压面要素集合预报技巧、中国地区24 h累积降水概率预报技巧、台风路径集合预报技巧、台风中心最低海平面气压预报技巧等方面对比二者结果。结果表明:ANSV和FCSV的主要结构特征相似,两组集合预报结果相当,表明在CMA-GEPS中使用FCSV可行,可作为未来高分辨率CMA-GEPS业务系统建设的选项。     

Improvement of model forecast on the Asian summer rainfall anomaly with the application of a spatial filtering scheme

Summary A revised 25-point Shuman-Shapiro Spatial Filter (RSSSF) has been applied to six atmospheric circulation models and multi-model ensemble (MME) predictions, and its effect on the improvement of model forecast skill scores of the Asian summer precipitation anomaly is discussed in this paper. On the basis of 21-yr model ensemble predictions, the RSSSF can remove the unpredictable ‘noise’ with respect to the 2-grid wavelength in the model precipitation anomaly fields and maintain the large-scale counterpart, which is related to the response of the model to large-scale boundary forcing. Therefore, this could possibly enhance the forecast skill of the Asian summer rainfall anomaly in the models and the MME. The potential improvement of model forecasting skill is found in the Asian summer monsoon region, where the anomaly correlation coefficient (ACC) has been improved by 7–40%, corresponding to the decreased root mean square error (RMSE) in the model and the MME precipitation anomaly forecasts.     

Skill of seasonal hindcasts as a function of the ensemble size

 Forecast skill as a function of the ensemble size is examined in a 24-member ensemble of northern winter (DJF) hindcasts produced with the second generation general circulation model of the Canadian Centre for Climate Modelling and Analysis. These integrations are initialized from the NCEP reanalyses at 6 h intervals prior to the forecast season. The sea surface temperatures that are applied as lower boundary conditions are predicted by persisting the monthly mean anomaly observed prior to the forecast period. The potential predictability that is attributed to lower boundary forced variability is estimated. In lagged-average forecasting, the forecast skill in the first two weeks, which originates predominately from the initial conditions, is greatest for relatively small ensemble sizes. The forecast skill increases monotonically with the ensemble size in the rest of the season. The skill of DJF 500 hPa geopotential height hindcasts in the Northern Hemisphere and in the Pacific/North America sector improves substantially when the ensemble size increases from 6 to 24. A statistical skill improvement technique based on the singular value decomposition method is also more successful for larger ensembles. Received: 22 February 2000 / Accepted: 6 December 2000     

STUDY OF THE MODIFICATION OF MULTI-MODEL ENSEMBLE SCHEMES FOR TROPICAL CYCLONE FORECASTS

This study investigates multi-model ensemble forecasts of track and intensity of tropical cyclones over the western Pacific, based on forecast outputs from the China Meteorological Administration, European Centre for Medium-Range Weather Forecasts, Japan Meteorological Agency and National Centers for Environmental Prediction in the THORPEX Interactive Grand Global Ensemble (TIGGE) datasets. The multi-model ensemble schemes, namely the bias-removed ensemble mean (BREM) and superensemble (SUP), are compared with the ensemble mean (EMN) and single-model forecasts. Moreover, a new model bias estimation scheme is investigated and applied to the BREM and SUP schemes. The results showed that, compared with single-model forecasts and EMN, the multi-model ensembles of the BREM and SUP schemes can have smaller errors in most cases. However, there were also circumstances where BREM was less skillful than EMN, indicating that using a time-averaged error as model bias is not optimal. A new model bias estimation scheme of the biweight mean is introduced. Through minimizing the negative influence of singular errors, this scheme can obtain a more accurate model bias estimation and improve the BREM forecast skill. The application of the biweight mean in the bias calculation of SUP also resulted in improved skill. The results indicate that the modification of multi-model ensemble schemes through this bias estimation method is feasible.     

Assessment of probability multimodel seasonal forecast based on the APCC model data

The probability multimodel forecast system based on the Asia-Pacific Economic Cooperation Climate Center (APCC) model data is verified. The winter and summer seasonal mean fields T ₈₅₀ and precipitation seasonal totals are estimated. To combine the models into a multimodel ensemble, the probability forecast is calculated for each of single models first, and then these forecasts are combined using the total probability formula. It is shown that the multimodel forecast is considerably more skilful than the single-model forecasts. The forecast quality is higher in the tropics compared to the mid- and high latitudes. The multimodel ensemble temperature forecasts outperform the random and climate forecasts for Northern Eurasia in the above- and below-normal categories. Precipitation forecast is less successful. For winter, the combination of single-model ensembles provides the precipitation forecast skill exceeding that of the random forecast for both Northern Eurasia and European Russia.     

Prediction and monitoring of monsoon intraseasonal oscillations over Indian monsoon region in an ensemble prediction system using CFSv2

An ensemble prediction system (EPS) is devised for the extended range prediction (ERP) of monsoon intraseasonal oscillations (MISO) of Indian summer monsoon (ISM) using National Centers for Environmental Prediction Climate Forecast System model version 2 at T126 horizontal resolution. The EPS is formulated by generating 11 member ensembles through the perturbation of atmospheric initial conditions. The hindcast experiments were conducted at every 5-day interval for 45 days lead time starting from 16th May to 28th September during 2001–2012. The general simulation of ISM characteristics and the ERP skill of the proposed EPS at pentad mean scale are evaluated in the present study. Though the EPS underestimates both the mean and variability of ISM rainfall, it simulates the northward propagation of MISO reasonably well. It is found that the signal-to-noise ratio of the forecasted rainfall becomes unity by about 18 days. The potential predictability error of the forecasted rainfall saturates by about 25 days. Though useful deterministic forecasts could be generated up to 2nd pentad lead, significant correlations are found even up to 4th pentad lead. The skill in predicting large-scale MISO, which is assessed by comparing the predicted and observed MISO indices, is found to be ~17 days. It is noted that the prediction skill of actual rainfall is closely related to the prediction of large-scale MISO amplitude as well as the initial conditions related to the different phases of MISO. An analysis of categorical prediction skills reveals that break is more skillfully predicted, followed by active and then normal. The categorical probability skill scores suggest that useful probabilistic forecasts could be generated even up to 4th pentad lead.     

A Comparison of Three Kinds of Multimodel Ensemble Forecast Techniques Based on the TIGGE Data

Based on the ensemble mean outputs of the ensemble forecasts from the ECMWF (European Centre for Medium-Range Weather Forecasts),JMA (Japan Meteorological Agency),NCEP (National Centers for Environment...     

Improved seasonal climate forecasts for the Caribbean region using the Florida State University Synthetic Superensemble

Summary Climate variations in the Caribbean, largely manifest in rainfall activity, have important consequences for the large-scale water budget, natural vegetation, and land use in the region. The wet and dry seasons will be defined, and the important roles played by the El Ni?o-Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO) in modulating the rainfall during these seasons will be discussed. The seasonal climate forecasts in this paper are made by 13 state of the art coupled atmosphere-ocean general circulation models (CGCMs) and by the Florida State University Synthetic Superensemble (FSUSSE), whose forecasts are obtained by a weighted combination of the individual CGCM forecasts based on a training period. The success of the models in simulating the observed 1989–2001 climatology of the various forecast parameters will be examined and linked to the models’ success in predicting the seasonal climate for individual years. Seasonal forecasts will be examined for precipitation, sea-surface temperature (SST), 2-meter air temperature, and 850 hPa u- and v-wind components during the period 1989–2001. Evaluation metrics include root mean square (RMS) error and Brier skill score. It will be shown that the FSUSSE is superior to the individual CGCMs and their ensemble mean both in simulating the 1989–2001 climatology for the various parameters and in predicting the seasonal climate of the various parameters for individual years. The seasonal climate forecasts of the FSUSSE and of the ensemble mean of the 13 state of the art CGCMs will be evaluated for years (during the period 1989–2001) that have particular ENSO and NAO signals that are known to influence Caribbean weather, particularly the rainfall. It will be shown that the FSUSSE provides superior forecasts of rainfall, SST, 2-meter air temperature, and 850 hPa u- and v-wind components during dry summers that are modulated by negative SOI and/or positive NAO indices. Such summers have become a feature of a twenty-year pattern of drought in the Caribbean region. The results presented in this paper will show that the FSUSSE is a valuable tool for forecasting rainfall and other atmospheric and oceanic variables during such periods of drought.     

An Ensemble Seasonal Forecast of Human Cases of St. Louis Encephalitis in Florida Based on Seasonal Hydrologic Forecasts

We present a method for the ensemble seasonal prediction of human St. Louis encephalitis (SLE) incidence and SLE virus transmission in Florida. We combine empirical relationships between modeled land surface wetness and the incidence of human clinical cases of SLE and modeled land surface wetness and the occurrence of SLE virus transmission throughout south Florida with a previously developed method for generating ensemble, seasonal hydrologic forecasts. Retrospective seasonal forecasts of human SLE incidence are made for Indian River County, Florida, and forecast skill is demonstrated for 2–4 months. A sample seasonal forecast of human SLE incidence is presented. This study establishes the skill of a potential component of an operational SLE forecast system in south Florida, one that provides information well in advance of transmission and may enable early interventions that reduce transmission. Future development of this method and operational application of these forecasts are discussed. The methodology also will be applied to West Nile virus monitoring and forecasting.     

New Models for Long Range Forecasts of Summer Monsoon Rainfall over North West and Peninsular India

Summary New models based on (a) Multivariate Principal Component Regression (PCR) (b) Neural Network (NN) and (c) Linear Discriminant Analysis (LDA) techniques were developed for long-range forecasts of summer monsoon (June–September) rainfall over two homogeneous regions of India, viz., North West India and Peninsular India. The PCR and NN models were developed with two different data sets. One set consisted 42 years (1958–1999) of data with 8 predictors and the other, 49 years (1951–1999) of data with 6 predictors. The predictors were subjected to the Principal Component Analysis (PCA) before model development. Two different neural networks were designed with 2 and 3 hidden neurons. To avoid the nonlinear instability, 20 ensemble runs were made while training the network and the ensemble mean results are discussed. The LDA model was developed with 42 years of data (1958–1999) for classifying three rainfall intervals with equal prior probability of 0.33. Both the PCR and NN models showed useful forecast skill for NW India and Peninsular India. Models with 8 predictors performed better than the models with only 6 predictors. The NN model with 3 hidden neurons performed better than model with 2 hidden neurons. For NW India, the NN model performed better than the PCR model. The RMSE of the NN model and PCR model with 8 predictors for NW India (Peninsular India) during the independent period 1984–99 was 12.5% (12.2%) and 12.6% (11.5%), respectively. Corresponding figures for the models with 6 predictors are 15.0% (13.0%) and 13.9% (11.4%) respectively. During the independent period, model errors were large in 1991, 1994, 1997 and 1999. However all the models showed deteriorating predictive skill after 1988, both for NW India and Peninsular India. The LDA model correctly classified 62% of grouped cases for NW India and Peninsular India. The LDA model showed better skill in classifying deficient rainfall (< − 8%) over NW India and excess rainfall (> 3%) over Peninsular India. Received October 2, 1999 Revised December 28, 1999     

2010年8月环流形势的多模式超级集合预报

利用多模式超级集合预报法,以欧洲中期天气预报中心、日本气象厅、德国气象局、中国气象局和中国空军气象中心共5个决定性7 d预报产品为集合成员,对2010年8月500 hPa高度场和850 hPa温度场分别进行固定训练期和滑动训练期超级集合预报。采用均方根误差和相关系数对超级集合预报、单一模式预报和简单集合平均预报进行对比检验,同时对各预报结果的均方根误差空间分布进行对比分析。结果表明:超级集合预报在所有预报结果中最佳,且滑动集合预报对8月后期时段预报要略好于固定集合预报,两者预报效果均好于参与集合预报的各模式,也好于集合平均预报。但随着预报时效的延长,集合平均预报的优势也随之提升。从预报结果均方根误差的空间分布可知,多模式超级集合预报相比于单一模式预报效果提高的区域,500 hPa位势高度场主要位于印度半岛、印度洋、青藏高原及以西地区,而850 hPa温度场则主要位于蒙古、青藏高原、中国新疆及以西地区。     

Projection of Extreme Temperatures in Hong Kong in the 21st Century

The possible changes in the frequency of extreme temperature events in Hong Kong in the 21st century were investigated by statistically downscaling 26 sets of the daily global climate model projections (a combination of 11 models and 3 greenhouse gas emission scenarios, namely A2, A1B, and B1) of the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. The models’ performance in simulating the past climate during 1971–2000 has also been verified and discussed. The verification revealed that the models in general have an acceptable skill in reproducing past statistics of extreme temperature events. Moreover, the models are more skillful in simulating the past climate of the hot nights and cold days than that of the very hot days. The projection results suggested that, in the 21st century, the frequency of occurrence of extremely high temperature events in Hong Kong would increase significantly while that of the extremely low temperature events is expected to drop significantly. Based on the multi-model scenario ensemble mean, the average annual numbers of very hot days and hot nights in Hong Kong are expected to increase significantly from 9 days and 16 nights in 1980–1999 to 89 days and 137 nights respectively in 2090–2099. On the other hand, the average annual number of cold days will drop from 17 days in 1980–1999 to about 1 day in 2090–2099. About 65 percent of the model-scenario combinations indicate that there will be on average less than one cold day in 2090–2099. While all the model-emission scenarios in general have projected consistent trends in the change of temperature extremes in the 21st century, there is a large divergence in the projections between difierent model/emission scenarios. This reflects that there are still large uncertainties in the model simulation of the future climate of extreme temperature events.     

基于TIGGE资料的地面气温多模式超级集合预报

基于TIGGE资料, 采用均方根误差分别对欧洲中期天气预报中心、日本气象厅、美国国家环境预报中心和英国气象局4个中心集合预报的地面气温场集合平均结果进行检验评估, 比较各中心地面气温的预报效果。并利用超级集合、多模式集合平均和消除偏差集合平均3种方法对4个中心的地面气温预报进行集成, 同时对预报结果进行分析。结果表明: 2007年夏季日本气象厅与欧洲中期天气预报中心在北半球大部分地区预报效果最好, 各中心在不同地区预报效果不同。超级集合与消除偏差集合平均降低了预报误差, 预报效果优于最好的单个中心预报和多模式集合平均。对于较长的预报时效, 消除偏差集合平均表现出了更好的预报性能。     

基于深度学习的中国地面气温的多模式集成预报研究

基于TIGGE资料中的欧洲中期天气预报中心、英国气象局、美国国家环境预报中心、韩国气象厅和日本气象厅2015年1月1日—9月30日中国及周边地区地面2 m气温24～168 h集合预报资料,利用长短期记忆神经网络(Long Short-Term Memory,LSTM)、浅层神经网络(Neural Networks,NN)、滑动训练期消除偏差集合平均(BREM)和滑动训练期多模式超级集合(SUP)方法对2015年9月5—30日26 d预报期进行集成预报试验。结果表明,BREM对5个单模式进行等权集成,预报结果易受预报效果较差模式的影响,整体预报技巧略低于单个最优模式ECMWF的预报技巧。其中在新疆南部,等权集成后的预报技巧更低。SUP的预报结果比所有单个模式预报更为准确。在144 h之前,SUP的误差明显小于ECMWF的预报误差,但随预报时效增加,误差增长幅度增大。NN对地面气温的预报效果与SUP的预报效果相当。LSTM整体预报效果最好,特别是在预报时效较长(超过72 h)时,比其他方法预报准确率明显提高。LSTM神经网络方法明显改进了我国西北、华北、东北、西南和华南大部分地区的气温预报,但在南疆部分地区误差较大。