Skill of synthetic superensemble hurricane forecasts for the Canadian maritime provinces

Summary From 1994 to 2003, fifty-five tropical cyclones entered the Canadian Hurricane Centre (CHC) Response Zone, or about 42% of all named Atlantic tropical cyclones in this ten-year period, and 2003 was the fourth consecutive year for a tropical cyclone to make landfall in Canada. The CHC forecasts all tropical cyclones that enter the CHC Response Zone and assumes the lead in forecasting once the cyclone enters its area of forecast responsibility. This study acknowledges the challenges of forecasting such tropical cyclones at extratropical latitudes. If a tropical cyclone has been declared extratropical, global models may no longer use vortex bogussing to carry the cyclone, and even if it is modeled, large model errors often result. The purpose of this study is to develop a new version of the Florida State University (FSU) hurricane superensemble with greater skill in tracking tropical cyclones, especially at extratropical latitudes. This has been achieved from the development of the synthetic superensemble, which is similar to the operational version of the multi-model superensemble that is used at FSU. The synthetic superensemble differs in that is has a larger set of member models consisting of regular member models, synthetic versions of these models, and the operational superensemble and its synthetic version. This synthetic superensemble is being used here to forecast hurricane tracks from the 2001, 2002, and 2003 hurricane seasons. The track forecasts from this method have generally less error than those of the member models, the operational superensemble, and the ensemble mean. This study shows that the synthetic superensemble performs consistently well and would be an asset to operational hurricane track forecasting.     

Rainfall anomaly prediction using statistical downscaling in a multimodel superensemble over tropical South America

This study addresses the predictability of rainfall variations over South America and the Amazon basin. A primary factor leading to model inaccuracy in precipitation forecasts is the coarse resolution data utilized by coupled models during the training phase. By using MERRA reanalysis and statistical downscaling along with the superensemble methodology, it is possible to obtain more precise forecast of rainfall anomalies over tropical South America during austral fall. Selective inclusion (and exclusion) of member models also allows for increased accuracy of superensemble forecasts. The use of coupled atmospheric–ocean numerical models to predict the rainfall anomalies has had mixed results. Improvement in individual member models is also possible on smaller spatial scales and in regions where substantial topographical changes were not handled well under original model initial conditions. The combination of downscaling and superensemble methodologies with other research methods presents the potential opportunity for increased accuracy not only in seasonal forecasts but on shorter temporal scales as well.     

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。     

Reduction of forecast error for global numerical weather prediction by The Florida State University (FSU) Superensemble

Summary The skill of the FSU Superensemble technique as applied to global numerical weather prediction is evaluated extensively in this paper. The global mass and motion fields for year 2000 and precipitation over the domain 55S to 55N for year 2001, as predicted by the Superensemble, the ensemble member models, and the mean of the ensemble members, are evaluated by standard statistical measures of skill to determine the performance of the Superensemble in relation to the other models. The member models are global forecast models from 5 of the worlds operational forecast centers in addition to the FSU global spectral model. For precipitation 5 additional versions of the FSU global model are utilized in the ensemble, as defined by different initial conditions provided by various physical initialization algorithms. Statistical parameters calculated for the mass and motion fields include root mean square (RMS) error, systematic error (or bias), and anomaly correlation. These are applied to the mean sea level pressure, 500hPa heights, and the wind fields at 850hPa and 200hPa. Statistical parameters that were calculated for precipitation include RMS error, correlation, equitable threat score (ETS), and a special definition of bias appropriate for the precipitation field. For the mass and motion fields the performance of the Superensemble was considered for the annual global case, as well as for each hemisphere (north and south) and for each of the four seasons. For precipitation only the annual case was considered over the domain cited above.For the mass and motion fields the RMS calculations showed the Superensemble to be superior (to have the smallest total forecast error) in all comparisons to the ensemble member models, and to be superior to the ensemble mean in the vast majority of comparisons. Performance in comparison to the other models was generally better in the Southern Hemisphere than in the Northern Hemisphere, and better in the transition seasons of fall and spring than in the extreme seasons of winter and summer. The Superensemble had the best success with mean sea level pressure, followed in order by 500hPa geopotential heights, 850hPa winds, and 200hPa winds.In the calculations of 500hPa geopotential height anomaly correlation the Superensemble had higher scores in all comparisons to the ensemble member models, as well as higher scores in the majority of comparisons to the ensemble mean. As with the RMS error results, the Superensemble performed better in the Southern Hemisphere than in the Northern Hemisphere, and better in fall than in summer, in comparison to the other models. The superior anomaly correlation scores of the Superensemble attest to the ability of the model to forecast daily perturbations from the climatological means, perturbations that are associated with transient synoptic scale features, given the horizontal resolution in the forecast models.In terms of systematic error reduction the Superensemble produces its most impressive results. Annual global mean sea-level pressure systematic errors for day 5 forecasts are generally in the range of ±1hPa (compared to errors as high as 8hPa in other models), and day 2 forecasts of 500hPa geopotential height produced systematic errors generally in the range of ±10 meters (compared to errors as high as 60 meters in other models). The Superensemble was able to reduce systematic errors in forecasts of a variety of important features in the global mass and motion fields: surface equatorial trough, wave amplitude in geopotential heights at 500hPa, trade winds and Somali Jet at 850hPa, mid-latitude westerlies, subtropical jet, and Tropical Easterly Jet (TEJ) at 200hPa.In terms of forecasting precipitation the Superensemble outperforms all ensemble member models and the ensemble mean in terms of RMS error, correlation coefficient, equitable threat score, and bias. The superior correlation scores indicate that the Superensemble is more reliable than the other models in predicting perturbations in the area distribution of precipitation, perturbations that are essentially associated with migrant synoptic scale disturbances, considering the horizontal resolution of the forecast models.The Superensemble is a valuable tool for significantly improving upon the global model forecasts of the worlds operational forecast centers. These forecasts are used daily as important guidance in making weather forecasts in all regions of the world. This paper will demonstrate that the Superensemble improves upon the ensemble member model forecasts: (1) in a statistical sense considering broad areas of the globe, (2) in a synoptic climatology sense through focus on the improved forecasts of climatological features seen in the global mass and motion fields, (3) in a synoptic sense through use of anomaly correlation and correlation coefficient where improvement is demonstrated in the forecasts of perturbations from mean fields which are essentially associated with transient synoptic scale disturbances.     

北半球中纬度地区地面气温的超级集合预报

基于TIGGE资料中的ECMWF、JMA、NCEP和UKMO四个中心2007年6月1日-8月31日北半球中纬度地区地面气温24～168 h集合预报资料,分别利用固定训练期超级集合(SUP, Superensemble)和滑动训练期超级集合(R-SUP, Running Training Period Superensemble )对2007年8月8-31日预报期24 d进行超级集合预报试验.采用均方根误差对预报结果进行检验评估,比较了两种超级集合方法与最好的单个中心模式预报、多模式集合平均的预报效果.结果表明,SUP预报有效降低了预报误差,24～144 h的预报效果优于多模式集合平均(EMN, Ensemble Mean)和最好的单个中心预报,168 h的预报效果略差于EMN.R-SUP预报进一步改善了预报效果.对于24～168 h的预报,R-SUP预报效果都要优于EMN.尤其对于168 h的预报,R-SUP改进了预报效果,优于EMN.     

江苏—南黄海地区M≥6强震有序网络结构及其预测研究

基于TIGGE（THORPEX Interactive Grand Global Ensemble,全球交互式大集合）资料中欧洲中期天气预报中心（European Centre for Medium-Range Weather,ECMWF）、日本气象厅（Japan Meteorological Agency,JMA）、美国国家环境预报中心（National Centers for Environmental Prediction,NCEP）和英国气象局（United Kingdom Met Office,UKMO）4个中心的北半球地面2m气温集合平均预报资料,利用插值技术与回归分析,并引入了消除偏差集合平均（bias-removed ensemble mean,BREM）和多模式超级集合（superensemble,SUP）方法进行统计降尺度预报研究.结果表明,在2007年夏季3个月中,4个单中心的降尺度预报明显地改善了预报效果.引入SUP和BREM两种集成预报方法后,预报误差得到进一步减小.对比综合表现最好的单中心ECMWF的预报,1～7d的降尺度预报误差改进率均达20％以上.研究还发现,引入SUP方法的降尺度预报效果优于引入BREM方法的降尺度预报,利用双线性插值方法在上述两方案中的预报效果优于其他3种插值方法.     

基于TIGGE资料的北半球地面气温预报的统计降尺度研究

基于TIGGE(THORPEX Interactive Grand Global Ensemble,全球交互式大集合)资料中欧洲中期天气预报中心(European Centre for Medium-Range Weather,ECMWF)、日本气象厅(Japan Meteorological Agency,JMA)、美国国家环境预报中心(National Centers for Environmental Prediction,NCEP)和英国气象局(United Kingdom Met Office,UKMO)4个中心的北半球地面2 m气温集合平均预报资料,利用插值技术与回归分析,并引入了消除偏差集合平均(bias-removed ensemble mean,BREM)和多模式超级集合(superensemble,SUP)方法进行统计降尺度预报研究。结果表明,在2007年夏季3个月中,4个单中心的降尺度预报明显地改善了预报效果。引入SUP和BREM两种集成预报方法后,预报误差得到进一步减小。对比综合表现最好的单中心ECMWF的预报,1~7 d的降尺度预报误差改进率均达20%以上。研究还发现,引入SUP方法的降尺度预报效果优于引入BREM方法的降尺度预报,利用双线性插值方法在上述两方案中的预报效果优于其他3种插值方法。     

Assessment of the APCC coupled MME suite in predicting the distinctive climate impacts of two flavors of ENSO during boreal winter

Forecast skill of the APEC Climate Center (APCC) Multi-Model Ensemble (MME) seasonal forecast system in predicting two main types of El Niño-Southern Oscillation (ENSO), namely canonical (or cold tongue) and Modoki ENSO, and their regional climate impacts is assessed for boreal winter. The APCC MME is constructed by simple composite of ensemble forecasts from five independent coupled ocean-atmosphere climate models. Based on a hindcast set targeting boreal winter prediction for the period 1982–2004, we show that the MME can predict and discern the important differences in the patterns of tropical Pacific sea surface temperature anomaly between the canonical and Modoki ENSO one and four month ahead. Importantly, the four month lead MME beats the persistent forecast. The MME reasonably predicts the distinct impacts of the canonical ENSO, including the strong winter monsoon rainfall over East Asia, the below normal rainfall and above normal temperature over Australia, the anomalously wet conditions across the south and cold conditions over the whole area of USA, and the anomalously dry conditions over South America. However, there are some limitations in capturing its regional impacts, especially, over Australasia and tropical South America at a lead time of one and four months. Nonetheless, forecast skills for rainfall and temperature over East Asia and North America during ENSO Modoki are comparable to or slightly higher than those during canonical ENSO events.     

Multimodel ensemble forecasts for precipitations in China in 1998

Different multimodel ensemble methods are used to forecast precipitations in China, 1998, and their forecast skills are compared with those of individual models. Datasets were obtained from monthly simulations of eight models during the period of January 1979 to December 1998 from the “Climate of the 20th Century Experiment” （20C3M） for the Fourth IPCC Assessment Report. Climate Research Unit （CRU） data were chosen for the observation analysis field. Root mean square （RMS） error and correlation coeffi-cients （R） are used to measure the forecast skills. In addition, superensemble forecasts based on different input data and weights are analyzed. Results show that for original data, superensemble forecasting based on multiple linear regression （MLR） performs best. However, for bias-corrected data, the superensemble based on singular value decomposition （SVD） produces a lower RMS error and a higher R than in the MLR superensemble. It is an interesting result that the SVD superensemble based on bias-corrected data performs better than the MLR superensemble, but that the SVD superensemble based on original data is inferior to the corresponding MLR superensemble. In addition, weights calculated by different data formats are shown to affect the forecast skills of the superensembles. In comparison with the MLR superensemble, a slightly significant effect is present in the SVD superensemble. However, both the SVD and MLR superensembles based on different weight formats outperform the ensemble mean of bias-corrected data.     

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.     

Calibration and combination of dynamical seasonal forecasts to enhance the value of predicted probabilities for managing risk

Seasonal probability forecasts produced with numerical dynamics on supercomputers offer great potential value in managing risk and opportunity created by seasonal variability. The skill and reliability of contemporary forecast systems can be increased by calibration methods that use the historical performance of the forecast system to improve the ongoing real-time forecasts. Two calibration methods are applied to seasonal surface temperature forecasts of the US National Weather Service, the European Centre for Medium Range Weather Forecasts, and to a World Climate Service multi-model ensemble created by combining those two forecasts with Bayesian methods. As expected, the multi-model is somewhat more skillful and more reliable than the original models taken alone. The potential value of the multimodel in decision making is illustrated with the profits achieved in simulated trading of a weather derivative. In addition to examining the seasonal models, the article demonstrates that calibrated probability forecasts of weekly average temperatures for leads of 2–4 weeks are also skillful and reliable. The conversion of ensemble forecasts into probability distributions of impact variables is illustrated with degree days derived from the temperature forecasts. Some issues related to loss of stationarity owing to long-term warming are considered. The main conclusion of the article is that properly calibrated probabilistic forecasts possess sufficient skill and reliability to contribute to effective decisions in government and business activities that are sensitive to intraseasonal and seasonal climate variability.     

改进的超级集成预报方法在长江三角洲地区O3预报中的应用

针对当前单模式系统臭氧（O₃）预报的不确定性问题,提出了一种基于活动区间的多模式超级集成的、高效的预报方法。本研究基于长江三角洲（长三角）地区多模式空气质量预报系统,将改进后的超级集成预报方法（AR-SUP）运用到2015年长三角地区的O₃预报中,并与滑动训练期的超级集成预报（R-SUP）、多模式集成平均预报（EMN）、消除偏差的集成平均预报（BREM）对比,结果表明AR-SUP对预报效果的改善最明显,其在暖季和冷季的均方根误差（RMSE）较最优单模式平均下降了20%和23%。将AR-SUP运用到48 h和72 h预报中发现,当预报时效增加时该方法依旧保持较高的预报技巧。多项统计数据均证明AR-SUP在研究时段内所有站点均能显著减小O₃预报误差、提高整体相关性和一致性,有效提高当前短期（三天）预报准确率。     

An analysis of the difference between the multiple linear regression approach and the multimodel ensemble mean

An investigation of the difference in seasonal precipitation forecast skills between the multiple linear regression (MLR) ensemble and the simple multimodel ensemble mean (EM) was based on the forecast quality of individual models. The possible causes of difference in previous studies were analyzed. In order to make the simulation capability of studied regions relatively uniform, three regions with different temporal correlation coefficients were chosen for this study. Results show the causes resulting in the incapability of the MLR approach vary among different regions. In the Nino3.4 region, strong co-linearity within individual models is generally the main reason. However, in the high latitude region, no significant co-linearity can be found in individual models, but the abilities of single models are so poor that it makes the MLR approach inappropriate for superensemble forecasts in this region. In addition, it is important to note that the use of various score measurements could result in some discrepancies when we compare the results derived from different multimodel ensemble approaches.     

地面气象要素多模式集成预报研究进展

目前,集合预报已成为天气预报业务的主要支撑。然而,由于数值模式本身的限制与不完善以及集合系统存在初值扰动、集合大小等方面的局限,常存在预报偏差。不同预报模式通常具有不同的物理过程参数化方案、初始条件等,导致其预报能力各有不同。为此,如何纠正预报偏差以及如何充分有效地利用不同模式的预报信息以获得更加准确的天气预报广受关注。近年来,利用统计理论与预报诊断,基于多个集合预报系统的多模式集成预报技术得到快速发展,已成为有效消除预报偏差从而提高天气预报技巧的一种统计后处理方法。针对气温、降水和风3个最基本的地面气象要素,首先依据预报形式将应用范围较广的简单集合平均、消除偏差集合平均、超级集合、贝叶斯模式平均、集合模式输出统计等加权或等权平均多模式集成技术,分成确定性预报和概率预报两大类,并做系统介绍。最后,讨论使用和发展多模式集成技术需要关注的问题,包括考虑参与集成的模式个数、发展降水及风速分级预报模型和发展基于机器学习的多模式集成新技术。     

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.     

Improving multimodel weather forecast of monsoon rain over China using FSU superensemble

In this paper we present the current capabilities for numerical weather prediction of precipitation over China using a suite of ten multimodels and our superensemble based forecasts. Our suite of models includes the operational suite selected by NCARs TIGGE archives for the THORPEX Program. These are: ECMWF, UKMO, JMA, NCEP, CMA, CMC, BOM, MF, KMA and the CPTEC models. The superensemble strategy includes a training and a forecasts phase, for these the periods chosen for this study include the months February through September for the years 2007 and 2008. This paper addresses precipitation forecasts for the medium range i.e. Days 1 to 3 and extending out to Day 10 of forecasts using this suite of global models. For training and forecasts validations we have made use of an advanced TRMM satellite based rainfall product. We make use of standard metrics for forecast validations that include the RMS errors, spatial correlations and the equitable threat scores. The results of skill forecasts of precipitation clearly demonstrate that it is possible to obtain higher skills for precipitation forecasts for Days 1 through 3 of forecasts from the use of the multimodel superensemble as compared to the best model of this suite. Between Days 4 to 10 it is possible to have very high skills from the multimodel superensemble for the RMS error of precipitation. Those skills are shown for a global belt and especially over China. Phenomenologically this product was also found very useful for precipitation forecasts for the Onset of the South China Sea monsoon, the life cycle of the mei-yu rains and post typhoon landfall heavy rains and flood events. The higher skills of the multimodel superensemble make it a very useful product for such real time events.     

Numerical weather prediction over the Eastern Caribbean using Florida State University (FSU) global and regional spectral models and multi-model/multi-analysis super-ensemble

Summary A month-long short-range numerical weather prediction experiment using the Florida State University’s (FSU) global and regional models and the multi-model/multi-analysis super-ensemble over the Eastern Caribbean domain is presented in this paper. The paper also investigates weather prediction capabilities of FSU global and regional models by examining the root mean square errors (RMSE) for the wind and precipitation fields. Super-ensemble forecasting, a new statistical approach to weather forecasting, is used over this domain. Here, forecasts from a number of numerical models provide the input and statistical combinations of these forecasts produce the super-ensemble forecast. A similar approach is used for the precipitation field where one model using different rain rate algorithms is used to generate different model outputs. The results show that the super-ensemble method produces forecasts that are superior to those obtained from the ensemble members. Received May 29, 2000/Revised February 15, 2001     

Seasonal Climate Forecasts – Potential Agricultural-Risk Management Tools?

The importance of anchoring seasonal climate forecasts to user needs is examined in this paper. Although it is generally accepted that seasonal climate forecasts have potential value, many constraints preclude the optimal use of these forecasts, including the way forecasts are produced, interpreted and applied in a variety of decision-making processes. In South Africa, a variety of agricultural users exists, ranging from the small-scale farmer to larger commercial farming entities. Useful seasonal are those produced and disseminated with the end user in mind. A retroactive test period during the 1990s, evaluates the perceived impact of incorporating seasonal rainfall forecasts into decisions made by commercial crop farmers in the central parts of South Africa. Although a small sample of commercial farmers was interviewed, the results show some benefits to commercial agriculture if seasonal climate forecast information is continuously and effectively applied over the long-term.     

NCEP集合预报系统在亚欧和北美区域的预报效果对比

使用NCEP集合预报系统（EPS）输出的500hPa位势高度场预报资料和相应的NCEP／NCAR再分析资料,针对集合平均预报和概率预报,采用多种预报效果检验评价方法,对该系统在亚欧和北美区域的预报效果进行全面的分析比较。总体而言,NCEP—EPS对亚欧区域的环流集合预报效果不亚于其对北关区域的预报效果。1）ACC检验表明,亚欧区域的集合平均预报效果在除冬季外的三个季节都明显优于北美区域,可用预报的时效相差达0．6～1d,且夏季的差别最大。RMSE检验表明,亚欧区域的预报效果在四个季节里均优于北美区域。2）集合概率预报可靠性的季节差别不明显,均为预报时效较短（长）时,北关（亚欧）区域的可靠性更好。系统对亚欧区域的事件识别范围相对较小,但其预报可靠性较高,北美区域则正好相反。3）夏季亚欧区域的集合概率预报效果明显优于北美区域,秋季和冬季北关区域的预报效果较好,春季在预报时效小于5d时北美区域占优,而其后则是亚欧区域的预报分辨能力更好。     

Examination of multi-model ensemble seasonal prediction methods using a simple climate system

A simple climate model was designed as a proxy for the real climate system, and a number of prediction models were generated by slightly perturbing the physical parameters of the simple model. A set of long (240 years) historical hindcast predictions were performed with various prediction models, which are used to examine various issues of multi-model ensemble seasonal prediction, such as the best ways of blending multi-models and the selection of models. Based on these results, we suggest a feasible way of maximizing the benefit of using multi models in seasonal prediction. In particular, three types of multi-model ensemble prediction systems, i.e., the simple composite, superensemble, and the composite after statistically correcting individual predictions (corrected composite), are examined and compared to each other. The superensemble has more of an overfitting problem than the others, especially for the case of small training samples and/or weak external forcing, and the corrected composite produces the best prediction skill among the multi-model systems.