• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2023년도 학술발표회
• /
• pp.25-25
• /
• 2023

The ensemble optimal interpolation (EnOI) scheme is a sub-optimal alternative to the ensemble Kalman filter (EnKF) with a reduced computational demand making it potentially more suitable for operational applications. Since only one model is integrated forward instead of an ensemble of model realizations, online estimation of the background error covariance matrix is not possible in the EnOI scheme. In this study, we investigate two Gaussian noise based ensemble generation strategies to produce dynamic covariance matrices for assimilation of water level observations into a distributed hydrological model. In the first approach, spatially correlated noise, sampled from a normal distribution with a fixed fractional error parameter (which controls its standard deviation), is added to the model forecast state vector to prepare the ensembles. In the second method, we use an adaptive error estimation technique based on the innovation diagnostics to estimate this error parameter within the assimilation framework. The results from a real and a set of synthetic experiments indicate that the EnOI scheme can provide better results when an optimal EnKF is not identified, but performs worse than the ensemble filter when the true error characteristics are known. Furthermore, while the adaptive approach is able to reduce the sensitivity to the fractional error parameter affecting the first (non-adaptive) approach, results are usually worse at ungauged locations with the former.

• Asia-Pacific Journal of Atmospheric Sciences
• /
• 제54권4호
• /
• pp.623-637
• /
• 2018

This study investigates the effects of horizontal resolution, cumulus parameterization scheme (CPS), and probability forecasting on precipitation forecasts over the Korean Peninsula from 00 UTC 15 August to 12 UTC 14 September 2013, using the limited-area ensemble prediction system (LEPS) of the Korea Meteorological Administration. To investigate the effect of resolution, the control members of the LEPS with 1.5- and 3-km resolution were compared. Two 3-km experiments with and without the CPS were conducted for the control member, because a 3-km resolution lies within the gray zone. For probability forecasting, 12 ensemble members with 3-km resolution were run using the LEPS. The forecast performance was evaluated for both the whole study period and precipitation cases categorized by synoptic forcing. The performance of precipitation forecasts using the 1.5-km resolution was better than that using the 3-km resolution for both the total period and individual cases. The result of the 3-km resolution experiment with the CPS did not differ significantly from that without it. The 3-km ensemble mean and probability matching (PM) performed better than the 3-km control member, regardless of the use of the CPS. The PM complemented the defect of the ensemble mean, which better predicts precipitation regions but underestimates precipitation amount by averaging ensembles, compared to the control member. Further, both the 3-km ensemble mean and PM outperformed the 1.5-km control member, which implies that the lower performance of the 3-km control member compared to the 1.5-km control member was complemented by probability forecasting.

• 대기
• /
• 제25권1호
• /
• pp.1-18
• /
• 2015

The most objective way to overcome the limitation of numerical weather prediction model is to represent the uncertainty of prediction by introducing probabilistic forecast. The uncertainty of the numerical weather prediction system developed due to the parameterization of unresolved scale motions and the energy losses from the sub-scale physical processes. In this study, we focused on the growth of model errors. We performed ensemble forecast to represent model uncertainty. By employing the multi-physics scheme (PHYS) and the stochastic kinetic energy backscatter scheme (SKEBS) in simulating typhoon Rusa (2002), we assessed the performance level of the two schemes. The both schemes produced better results than the control run did in the ensemble mean forecast of the track. The results using PHYS improved by 28% and those based on SKEBS did by 7%. Both of the ensemble mean errors of the both schemes increased rapidly at the forecast time 84 hrs. The both ensemble spreads increased gradually during integration. The results based on SKEBS represented model errors very well during the forecast time of 96 hrs. After the period, it produced an under-dispersive pattern. The simulation based on PHYS overestimated the ensemble mean error during integration and represented the real situation well at the forecast time of 120 hrs. The displacement speed of the typhoon based on PHYS was closest to the best track, especially after landfall. In the sensitivity tests of the model uncertainty of SKEBS, ensemble mean forecast was sensitive to the physics parameterization. By adjusting the forcing parameter of SKEBS, the default experiment improved in the ensemble spread, ensemble mean errors, and moving speed.

• Journal of Information Processing Systems
• /
• 제16권4호
• /
• pp.809-819
• /
• 2020

Gender prediction accuracy increases as convolutional neural network (CNN) architecture evolves. This paper compares voting and ensemble schemes to utilize the already trained five CNN models to further improve gender prediction accuracy. The majority voting usually requires odd-numbered models while the proposed softmax-based voting can utilize any number of models to improve accuracy. The ensemble of CNN models combined with one more fully-connected layer requires further tuning or training of the models combined. With experiments, it is observed that the voting or ensemble of CNN models leads to further improvement of gender prediction accuracy and that especially softmax-based voters always show better gender prediction accuracy than majority voters. Also, compared with softmax-based voters, ensemble models show a slightly better or similar accuracy with added training of the combined CNN models. Softmax-based voting can be a fast and efficient way to get better accuracy without further training since the selection of the top accuracy models among available CNN pre-trained models usually leads to similar accuracy to that of the corresponding ensemble models.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2009년도 학술발표회 초록집
• /
• pp.1147-1151
• /
• 2009

In this study, we address the problem of producing probability forecasts of summer seasonal rainfall, on the basis of Hindcast experiments from a ensemble of GCMs(cwb, gcps, gdaps, metri, msc_gem, msc_gm2, msc_gm3, msc_sef and ncep). An advanced Hierarchical Bayesian weighting scheme is developed and used to combine nine GCMs seasonal hindcast ensembles. Hindcast period is 23 years from 1981 to 2003. The simplest approach for combining GCM forecasts is to weight each model equally, and this approach is referred to as pooled ensemble. This study proposes a more complex approach which weights the models spatially and seasonally based on past model performance for rainfall. The Bayesian approach to multi-model combination of GCMs determines the relative weights of each GCM with climatology as the prior. The weights are chosen to maximize the likelihood score of the posterior probabilities. The individual GCM ensembles, simple poolings of three and six models, and the optimally combined multimodel ensemble are compared.

• 대기
• /
• 제20권3호
• /
• pp.287-306
• /
• 2010

An ensemble Kalman filter (EnKF) with Weather Research and Forecasting (WRF) Model is applied for Typhoon Wukong (200610) to investigate the performance of ensemble forecasts depending on experimental configurations of the EnKF. In addition, the ensemble sensitivity analysis is applied to the forecast and analysis ensembles generated in EnKF, to investigate the possibility of using the ensemble sensitivity analysis as the adaptive observation guidance. Various experimental configurations are tested by changing model error, ensemble size, assimilation time window, covariance relaxation, and covariance localization in EnKF. First of all, experiments using different physical parameterization scheme for each ensemble member show less root mean square error compared to those using single physics for all the forecast ensemble members, which implies that considering the model error is beneficial to get better forecasts. A larger number of ensembles are also beneficial than a smaller number of ensembles. For the assimilation time window, the experiment using less frequent window shows better results than that using more frequent window, which is associated with the availability of observational data in this study. Therefore, incorporating model error, larger ensemble size, and less frequent assimilation window into the EnKF is beneficial to get better prediction of Typhoon Wukong (200610). The covariance relaxation and localization are relatively less beneficial to the forecasts compared to those factors mentioned above. The ensemble sensitivity analysis shows that the sensitive regions for adaptive observations can be determined by the sensitivity of the forecast measure of interest to the initial ensembles. In addition, the sensitivities calculated by the ensemble sensitivity analysis can be explained by dynamical relationships established among wind, temperature, and pressure.

• 지능정보연구
• /
• 제19권2호
• /
• pp.55-71
• /
• 2013

부도예측을 위한 지식기반시스템에서 모델은 실적에 영향을 끼치는 주요한 요인이다. 예측 모형의 개발에 있어 초기 연구들은 통계기법 및 인공지능기법들을 이용하여 최고 실적을 가지는 단일 모델을 만드는데 주력하였다. 1980년대 중반 이후에는 다수 기술의 통합(하이브리드), 더 나아가, 다수 모델의 결과의 결합(앙상블) 기법이 수많은 실험에서 개별 모델들보다 더 나은 결과를 보여왔다. 다수 모델들의 출력값들을 결합하여 한 개의 최종 예측값을 산출하는 앙상블 모델링에서 결합기법은 앙상블의 예측 정확도에 영향을 끼치는 중요한 이슈이다. 본 논문은 부도예측을 위한 앙상블 결합기법으로서 앙상블 멤버들이 다른 유형의 연속형 수치 출력값들을 산출하더라도 통일된 확신을 측정할 수 있는 확신 기반의 선택 접근법을 제안하고 이에 대한 앙상블 멤버 사이즈의 영향을 연구하였다. 실험 결과는 앙상블 멤버들의 생성 타입에 따라 결합하는 모델 개수를 변화시켰을 때 가장 많은 기본 모델들을 가지는 앙상블에서의 제안 결합기법이 부도예측에 가장 자주 사용되는 다른 방법들에 비해서도 가장 높은 실적을 가진다는 것을 보였다.

• 한국해안·해양공학회논문집
• /
• 제30권2호
• /
• pp.84-94
• /
• 2018

해양파랑 예측에 있어 단일 수치모델의 불확실성을 보완하기 위하여 앙상블 기법을 적용한 지역 파랑예측시스템을 구축하였다. 기상청 전지구 대기 수치모델의 확률예측시스템에서 생산되는 24개 앙상블 해상풍을 입력자료로 이용, 87시간까지 파랑 예측자료를 생산하였으며, 기상청 계류부이 관측자료와 다양한 통계방법을 적용하여 검증을 수행하였다. 2일예측 이후의 앙상블 예측평균의 평균제곱근오차(RMSE)는 단일모델예측에 비하여 향상된 결과를 보였으며, 특히 3일예측의 경우 단일모델예측 대비 RMSE가 약 15% 정도 향상되었다. 이것은 앙상블 기법이 수치모델의 불확실성을 감소시켜 예측정확도 향상에 크게 기여한 것으로 보인다. ROC(Relative Operating Characteristic) 분석결과, 전체 예측시간에 대하여 ROC 영역이 모두 0.9 이상을 보여 확률예측 성능이 뛰어남을 보였으며, 앙상블 파랑예측 결과가 해상 확률예보에 유용하게 활용될 수 있을 것으로 판단된다.

• Journal of Information Processing Systems
• /
• 제10권3호
• /
• pp.443-458
• /
• 2014

An extreme learning machine (ELM) is a recently proposed learning algorithm for a single-layer feed forward neural network. In this paper we studied the ensemble of ELM by using a bagging algorithm for facial expression recognition (FER). Facial expression analysis is widely used in the behavior interpretation of emotions, for cognitive science, and social interactions. This paper presents a method for FER based on the histogram of orientation gradient (HOG) features using an ELM ensemble. First, the HOG features were extracted from the face image by dividing it into a number of small cells. A bagging algorithm was then used to construct many different bags of training data and each of them was trained by using separate ELMs. To recognize the expression of the input face image, HOG features were fed to each trained ELM and the results were combined by using a majority voting scheme. The ELM ensemble using bagging improves the generalized capability of the network significantly. The two available datasets (JAFFE and CK+) of facial expressions were used to evaluate the performance of the proposed classification system. Even the performance of individual ELM was smaller and the ELM ensemble using a bagging algorithm improved the recognition performance significantly.

• Journal of the Korean Data and Information Science Society
• /
• 제27권2호
• /
• pp.549-557
• /
• 2016

In this paper we propose multiclassification method for large data sets by ensembling least squares support vector machines (LS-SVM) with principal components instead of raw input vector. We use the revised one-vs-all method for multiclassification, which is one of voting scheme based on combining several binary classifications. The revised one-vs-all method is performed by using the hat matrix of LS-SVM ensemble, which is obtained by ensembling LS-SVMs trained using each random sample from the whole large training data. The leave-one-out cross validation (CV) function is used for the optimal values of hyper-parameters which affect the performance of multiclass LS-SVM ensemble. We present the generalized cross validation function to reduce computational burden of leave-one-out CV functions. Experimental results from real data sets are then obtained to illustrate the performance of the proposed multiclass LS-SVM ensemble.