Probabilistic Ensemble Research Articles

Constraining stochastic 3-D structural geological models with topology information using approximate Bayesian computation in GemPy 2.1

Abstract. Structural geomodeling is a key technology for the visualization and quantification of subsurface systems. Given the limited data and the resulting necessity for geological interpretation to construct these geomodels, uncertainty is pervasive and traditionally unquantified. Probabilistic geomodeling allows for the simulation of uncertainties by automatically constructing geomodel ensembles from perturbed input data sampled from probability distributions. But random sampling of input parameters can lead to construction of geomodels that are unrealistic, either due to modeling artifacts or by not matching known information about the regional geology of the modeled system. We present a method to incorporate geological information in the form of known geomodel topology into stochastic simulations to constrain resulting probabilistic geomodel ensembles using the open-source geomodeling software GemPy. Simulated geomodel realizations are checked against topology information using an approximate Bayesian computation approach to avoid the specification of a likelihood function. We demonstrate how we can infer the posterior distributions of the model parameters using topology information in two experiments: (1) a synthetic geomodel using a rejection sampling scheme (ABC-REJ) to demonstrate the approach and (2) a geomodel of a subset of the Gullfaks field in the North Sea comparing both rejection sampling and a sequential Monte Carlo sampler (ABC-SMC). Possible improvements to processing speed of up to 10.1 times are discussed, focusing on the use of more advanced sampling techniques to avoid the simulation of unfeasible geomodels in the first place. Results demonstrate the feasibility of using topology graphs as a summary statistic to restrict the generation of geomodel ensembles with known geological information and to obtain improved ensembles of probable geomodels which respect the known topology information and exhibit reduced uncertainty using stochastic simulation methods.

Reproducing complex simulations of economic impacts of climate change with lower-cost emulators

Abstract. Process-based models are powerful tools for simulating the economic impacts of climate change, but they are computationally expensive. In order to project climate-change impacts under various scenarios, produce probabilistic ensembles, conduct online coupled simulations, or explore pathways by numerical optimization, the computational and implementation cost of economic impact calculations should be reduced. To do so, in this study, we developed various emulators that mimic the behaviours of simulation models, namely economic models coupled with bio/physical-process-based impact models, by statistical regression techniques. Their performance was evaluated for multiple sectors and regions. Among the tested emulators, those composed of artificial neural networks, which can incorporate non-linearities and interactions between variables, performed better particularly when finer input variables were available. Although simple functional forms were effective for approximating general tendencies, complex emulators are necessary if the focus is regional or sectoral heterogeneity. Since the computational cost of the developed emulators is sufficiently small, they could be used to explore future scenarios related to climate-change policies. The findings of this study could also help researchers design their own emulators in different situations.

FaIRv2.0.0: a generalized impulse response model for climate uncertainty and future scenario exploration

Abstract. Here we present an update to the FaIR model for use in probabilistic future climate and scenario exploration, integrated assessment, policy analysis, and education. In this update we have focussed on identifying a minimum level of structural complexity in the model. The result is a set of six equations, five of which correspond to the standard impulse response model used for greenhouse gas (GHG) metric calculations in the IPCC's Fifth Assessment Report, plus one additional physically motivated equation to represent state-dependent feedbacks on the response timescales of each greenhouse gas cycle. This additional equation is necessary to reproduce non-linearities in the carbon cycle apparent in both Earth system models and observations. These six equations are transparent and sufficiently simple that the model is able to be ported into standard tabular data analysis packages, such as Excel, increasing the potential user base considerably. However, we demonstrate that the equations are flexible enough to be tuned to emulate the behaviour of several key processes within more complex models from CMIP6. The model is exceptionally quick to run, making it ideal for integrating large probabilistic ensembles. We apply a constraint based on the current estimates of the global warming trend to a million-member ensemble, using the constrained ensemble to make scenario-dependent projections and infer ranges for properties of the climate system. Through these analyses, we reaffirm that simple climate models (unlike more complex models) are not themselves intrinsically biased “hot” or “cold”: it is the choice of parameters and how those are selected that determines the model response, something that appears to have been misunderstood in the past. This updated FaIR model is able to reproduce the global climate system response to GHG and aerosol emissions with sufficient accuracy to be useful in a wide range of applications and therefore could be used as a lowest-common-denominator model to provide consistency in different contexts. The fact that FaIR can be written down in just six equations greatly aids transparency in such contexts.

Detection of deep myometrial invasion in endometrial cancer MR imaging based on multi-feature fusion and probabilistic support vector machine ensemble

The depth of myometrial invasion affects the treatment and prognosis of patients with endometrial cancer (EC), conventionally evaluated using MR imaging (MRI). However, only a few computer-aided diagnosis methods have been reported for identifying deep myometrial invasion (DMI) using MRI. Moreover, these existing methods exhibit relatively unsatisfactory sensitivity and specificity. This study proposes a novel computerized method to facilitate the accurate detection of DMI on MRI. This method requires only the corpus uteri region provided by humans or computers instead of the tumor region. We also propose a geometric feature called LS to describe the irregularity of the tissue structure inside the corpus uteri triggered by EC, which has not been leveraged for the DMI prediction model in other studies. Texture features are extracted and then automatically selected by recursive feature elimination. Utilizing a feature fusion strategy of strong and weak features devised in this study, multiple probabilistic support vector machines incorporate LS and texture features, which are then merged to form the ensemble model EPSVM. The model performance is evaluated via leave-one-out cross-validation. We make the following comparisons, EPSVM versus the commonly used classifiers such as random forest, logistic regression, and naive Bayes; EPSVM versus the models using LS or texture features alone. The results show that EPSVM attains an accuracy, sensitivity, specificity, and F1 score of 93.7%, 94.7%, 93.3%, and 87.8%, all of which are higher than those of the commonly used classifiers and the models using LS or texture features alone. Compared with the methods in existing studies, EPSVM exhibits high performance in terms of both sensitivity and specificity. Moreover, LS can achieve an accuracy, sensitivity, and specificity of 89.9%, 89.5%, and 90.0%. Thus, the devised geometric feature LS is significant for DMI detection. The fusion of LS and texture features in the proposed EPSVM can provide more reliable prediction. The computer-aided classification based on the proposed method can assist radiologists in accurately identifying DMI on MRI.

Emulating Ocean Dynamic Sea Level by Two-Layer Pattern Scaling.

Ocean dynamic sea level (DSL) change is a key driver of relative sea level (RSL) change. Projections of DSL change are generally obtained from simulations using atmosphere‐ocean general circulation models (GCMs). Here, we develop a two‐layer climate emulator to interpolate between emission scenarios simulated with GCMs and extend projections beyond the time horizon of available simulations. This emulator captures the evolution of DSL changes in corresponding GCMs, especially over middle and low latitudes. Compared with an emulator using univariate pattern scaling, the two‐layer emulator more accurately reflects GCM behavior and captures non‐linearities and non‐stationarity in the relationship between DSL and global‐mean warming, with a reduction in global‐averaged error during 2271–2290 of 36%, 24%, and 34% in RCP2.6, RCP4.5, and RCP8.5, respectively. Using the emulator, we develop a probabilistic ensemble of DSL projections through 2300 for four scenarios: Representative Concentration Pathway (RCP) 2.6, RCP 4.5, RCP 8.5, and Shared Socioeconomic Pathway (SSP) 3–7.0. The magnitude and uncertainty of projected DSL changes decrease from the high‐to the low‐emission scenarios, indicating a reduced DSL rise hazard in low‐ and moderate‐emission scenarios (RCP2.6 and RCP4.5) compared to the high‐emission scenarios (SSP3‐7.0 and RCP8.5).

Probabilistic Characterization of Extreme Storm Surges Induced by Tropical Cyclones

AbstractThe overtopping of flood defenses by extreme storm surges during tropical cyclones poses a significant threat to life and property in coastal and estuarine regions. Since little effort has been devoted to investigating the complex interaction of multiple mechanisms causing extreme storm surges, for the first time, we propose a robust data‐driven framework to explicitly uncover the complex interaction and thus to improve the characterization of extreme storm surges induced by tropical cyclones. The framework constructs a probabilistic ensemble of dependence structures of multiple climatological forcing factors that potentially cause extreme storm surges based on a multi‐structure regular vine copula approach. The uncertainty in the vine‐based model structure is explicitly addressed in a Bayesian framework. The climatological forcing factors sensitive to extreme storm surge levels are selected using partial correlations, including 10‐m wind, 850‐mb temperature, 700‐mb geopotential height, precipitation, sea level pressure, and its spatial gradient extracted from the ERA5 reanalysis data. We demonstrate the framework by an in‐depth analysis of the extreme storm surge levels observed over two tidal gauging stations in Hong Kong during 1979–2018. Our findings show that the proposed framework substantially improves the characterization of extreme storm surges by taking into account the joint evolution of multiple mechanisms causing extreme storm surges and underlying uncertainties. Furthermore, the framework not only demonstrates higher skill than previous single‐structure vine‐based models but also can outperform the principal components regression and the random forest regression in terms of characterizing extreme storm surges.

A Probabilistic Ensemble Prediction Method for PV Power in the Nonstationary Period

Due to the large number of grid connection of distributed power supply, the existing scheduling methods can not meet the demand gradually. The proposed virtual power plant provides a new idea to solve this problem. The photovoltaic power prediction provides the data basis for the scheduling of the virtual power plant. Prediction intervals of photovoltaic power is a powerful statistical tool used for quantifying the uncertainty of photovoltaic power generation in power systems. To improve the interval prediction accuracy during the non-stationary periods of photovoltaic power, this paper proposes a probabilistic ensemble prediction model, which combines the modules of data preprocessing, non-stationary period discrimination, feature extraction, deterministic prediction, uncertainty prediction, and optimization integration into a general framework. More specifically, in the non-stationary period discrimination module, the method of discriminating the difference of the power ratio difference is introduced and applied for identifying the non-stationary period of the data of photovoltaic output; in the deterministic point prediction module, a stacking- long-short-term memory neural network model is used for point forecasts; in the uncertainty interval prediction module, a BAYES neural network is introduced for probabilistic forecasts; in the optimization integration module, an optimization algorithm named Non-dominated Sorting Genetic Algorithm-II is applied for integrating and optimizing the results of the point forecast and probabilistic forecast. The proposed model is tested using two photovoltaic outputs and weather data measured from a grid-connected photovoltaic system. The results show that the proposed model outperforms conventional forecast methods to predict short-term photovoltaic power outputs and associated uncertainties. The interval width is reduced by 10–20%, and the prediction accuracy is improved by at least 10%; this can be a useful tool for photovoltaic power forecasting.

Automatic identification of abnormal blood smear images using color and morphology variation of RBCS and central pallor

The anatomy of red blood cells (RBCs) in blood smear images plays an important role in the detection of several diseases. The automated image-based technique is fast and accurate for the analysis of blood cells morphology that can save time of both pathologists as well as that of patients. In this paper, we propose a novel method which segment and identify varied RBCs in a given blood smear images. In the proposed method, the central pallor and whole cell information are used, after using color processing followed by double thresholding of blood smear images. The shape and size variances of cells are calculated for the identification of abnormalities in peripheral blood smear images. We used cross-validation accuracy weighted probabilistic ensemble (CAWPE). It is a heterogeneous ensembling technique of nearly equivalent classifiers produced on averagely significant better classifiers (regarding errors and probability estimates) as compared to a wide range of potential parent classifiers. The proposed method is tested on 3 sets of images. The sets of images were prepared in a local government hospital by expert pathologists. Each image set has varied photographic conditions. The method was found accurate in term of results, closer to the ground truth. The average accuracy of the proposed method is 97% for the segmentation of single cells and 96% for overlapped cells. The variance (σ2) of accuracy is 3.5 and the deviation (σ) is 1.87.

ERA5-based global meteorological wildfire danger maps

Forest fires are an integral part of the natural Earth system dynamics, however they are becoming more devastating and less predictable as anthropogenic climate change exacerbates their impacts. In order to advance fire science, fire danger reanalysis products can be used as proxy for fire weather observations with the advantage of being homogeneously distributed both in space and time. This manuscript describes a reanalysis dataset of fire danger indices based on the Canadian Fire Weather Index system and the ECMWF ERA5 reanalysis dataset, which supersedes the previous dataset based on ERA-Interim. The new fire danger reanalysis dataset provides a number of benefits compared to the one based on ERA-Interim: it relies on better estimates of precipitation, evaporation and soil moisture, it is available in a deterministic form as well as a probabilistic ensemble and it is characterised by a considerably higher spatial resolution. It is a valuable resource for forestry agencies and scientists in the field of wildfire danger modeling and beyond. The global dataset is produced by ECMWF, as the computational centre of the European Forest Fire information System (EFFIS) of the Copernicus Emergency Management Service, and it is made available free of charge through the Climate Data Store.

Probabilistic SVM classifier ensemble selection based on GMDH-type neural network

Support vector machine (SVM) provides a good classification and regression ability, especially, for small sample learning. However, in practice, the learning ability of implemented SVM is occasionally far from the expected level. Group method of data handling neural network (GMDH-NN) has been applied in various fields for pattern recognition and data mining. It makes it possible to automatically find interrelations in data, to select an optimal structure of network or model and to improve the accuracy of existing algorithms. In this work we propose to take the advantages of GMDH-NN for further increasing the classification performance of SVM. One weakness of the symmetric regularity criterion of GMDH-NN is that if one of the input attributes has a relatively big range, then it may overcome the other attributes. Thus, we first define a standardized symmetric regularity criterion (SSRC) to evaluate and select the candidate models, and optimize a classifier ensemble selection approach. Secondly, we define a novel structure of initial model of GMDH-NN which is from the posterior probability outputs of SVMs. These probabilistic outputs are generated from the improved Platt’s probabilistic outputs. Thirdly, in real classification tasks, different classifiers usually have different classification advantages. So we use probabilistic SVM as base learner and integrate the probabilistic SVMs with GMDH-NN, and then propose a special classifier ensemble selection approach for probabilistic SVM classifiers based on GMDH-NN called GMDH-PSVM. Moreover, we use the Borda sorting and Random weighted Borda sorting to discuss the results of our experiments. Experiments on standard UCI datasets demonstrate the effectiveness of our method.

Ensemble flood forecasting: Current status and future opportunities

AbstractEnsemble flood forecasting has gained significant momentum over the past decade due to the growth of ensemble numerical weather and climate prediction, expansion in high performance computing, growing interest in shifting from deterministic to risk‐based decision‐making that accounts for forecast uncertainty, and the efforts of communities such as the international Hydrologic Ensemble Prediction Experiment (HEPEX), which focuses on advancing relevant ensemble forecasting capabilities and fostering its adoption. With this shift, comes the need to understand the current state of ensemble flood forecasting, in order to provide insights into current capabilities and areas for improvement, thus identifying future research opportunities to allow for better allocation of research resources. In this article, we provide an overview of current research activities in ensemble flood forecasting and discuss knowledge gaps and future research opportunities, based on a review of 70 papers focusing on various aspects of ensemble flood forecasting around the globe. Future research directions include opportunities to improve technical aspects of ensemble flood forecasting, such as data assimilation techniques and methods to account for more sources of uncertainty, and developing ensemble forecasts for more variables, for example, flood inundation, by applying techniques such as machine learning. Further to this, we conclude that there is a need to not only improve technical aspects of flood forecasting, but also to bridge the gap between scientific research and hydrometeorological model development, and real‐world flood management using probabilistic ensemble forecasts, especially through effective communication.This article is categorized under: Engineering Water > Methods Science of Water > Water Extremes

Optimising the use of ensemble information in numerical weather forecasts of wind power generation

Electricity generation output forecasts for wind farms across Europe use numerical weather prediction (NWP) models. These forecasts influence decisions in the energy market, some of which help determine daily energy prices or the usage of thermal power generation plants. The predictive skill of power generation forecasts has an impact on the profitability of energy trading strategies and the ability to decrease carbon emissions. Probabilistic ensemble forecasts contain valuable information about the uncertainties in a forecast. The energy market typically takes basic approaches to using ensemble data to obtain more skilful forecasts. There is, however, evidence that more sophisticated approaches could yield significant further improvements in forecast skill and utility. In this letter, the application of ensemble forecasting methods to the aggregated electricity generation output for wind farms across Germany is investigated using historical ensemble forecasts from the European Centre for Medium-Range Weather Forecasting (ECMWF). Multiple methods for producing a single forecast from the ensemble are tried and tested against traditional deterministic methods. All the methods exhibit positive skill, relative to a climatological forecast, out to a lead time of at least seven days. A wind energy trading strategy involving ensemble data is implemented and produces significantly more profit than trading strategies based on single forecasts. It is thus found that ensemble spread is a good predictor for wind electricity generation output forecast uncertainty and is extremely valuable at informing wind energy trading strategy.

Consensus and uncertainty in the geographic range of Aedes aegypti and Aedes albopictus in the contiguous United States: Multi-model assessment and synthesis.

Aedes (Stegomyia) aegypti (L.) and Ae. (Stegomyia) albopictus (Skuse) mosquitoes can transmit dengue, chikungunya, yellow fever, and Zika viruses. Limited surveillance has led to uncertainty regarding the geographic ranges of these vectors globally, and particularly in regions at the present-day margins of habitat suitability such as the contiguous United States. Empirical habitat suitability models based on environmental conditions can augment surveillance gaps to describe the estimated potential species ranges, but model accuracy is unclear. We identified previously published regional and global habitat suitability models for Ae. aegypti (n = 6) and Ae. albopictus (n = 8) for which adequate information was available to reproduce the models for the contiguous U.S. Using a training subset of recently updated county-level surveillance records of Ae. aegypti and Ae. albopictus and records of counties conducting surveillance, we constructed accuracy-weighted, probabilistic ensemble models from these base models. To assess accuracy and uncertainty we compared individual and ensemble model predictions of species presence or absence to both training and testing data. The ensemble models were among the most accurate and also provided calibrated probabilities of presence for each species. The quantitative probabilistic framework enabled identification of areas with high uncertainty and model bias across the U.S. where improved models or additional data could be most beneficial. The results may be of immediate utility for counties considering surveillance and control programs for Ae. aegypti and Ae. albopictus. Moreover, the assessment framework can drive future efforts to provide validated quantitative estimates to support these programs at local, national, and international scales.

Online Fault Diagnosis for Industrial Processes With Bayesian Network-Based Probabilistic Ensemble Learning Strategy

The efficient mitigation of the detrimental effects of a fault in complex systems requires online fault diagnosis techniques that are able to identify the cause of an observable anomaly. However, an individual diagnosis model can only acquire a limited diagnostic effect and may be insufficient for a particular application. In this paper, a Bayesian network-based probabilistic ensemble learning (PEL-BN) strategy is proposed to address the aforementioned issue. First, an ensemble index is proposed to evaluate the candidate diagnosis models in a probabilistic manner so that the diagnosis models with better diagnosis performance can be selected. Then, based on the selected classifiers, the architecture of the Bayesian network can be constructed using the proposed three types of basic topologies. Finally, the advantages of different diagnosis models are integrated using the developed Bayesian network, and thus, the fault causes of the observable anomaly can be accurately inferred. In addition, the proposed method can effectively capture the mixed fault characteristics of multifaults (MFs) by integrating decisions derived from different diagnosis models. Hence, this method can also provide a feasible solution for diagnosing MFs in real industrial processes. A simulation process and a real industrial process are adopted to verify the performance of the proposed method, and the experimental results illustrate that the proposed PEL-BN strategy improves the diagnosis performance of single faults and is a feasible solution for MF diagnosis. Note to Practitioners —The focus of this paper is to develop a probabilistic ensemble learning strategy based on the Bayesian network (PEL-BN) to diagnose different kinds of faults in industrial processes. The PEL-BN strategy can automatically select the base classifiers to establish the architecture of the Bayesian network. In this way, the conclusions of these base classifiers can be effectively integrated to provide better diagnosis performance. In addition, the proposed method is also a feasible technique for diagnosing MFs resulted from the joint effects of multiple faults.

Skewed logistic distribution for statistical temperature post-processing in mountainous areas

Abstract. Nonhomogeneous post-processing is often used to improve the predictive performance of probabilistic ensemble forecasts. A common quantity used to develop, test, and demonstrate new methods is the near-surface air temperature, which is frequently assumed to follow a Gaussian response distribution. However, Gaussian regression models with only a few covariates are often not able to account for site-specific local features leading to uncalibrated forecasts and skewed residuals. This residual skewness remains even if many covariates are incorporated. Therefore, a simple refinement of the classical nonhomogeneous Gaussian regression model is proposed to overcome this problem by assuming a skewed response distribution to account for possible skewness. This study shows a comprehensive analysis of the performance of nonhomogeneous post-processing for the 2 m temperature for three different site types, comparing Gaussian, logistic, and skewed logistic response distributions. The logistic and skewed logistic distributions show satisfying results, in particular for sharpness, but also in terms of the calibration of the probabilistic predictions.

A probabilistic classifier ensemble weighting scheme based on cross-validated accuracy estimates

Our hypothesis is that building ensembles of small sets of strong classifiers constructed with different learning algorithms is, on average, the best approach to classification for real-world problems. We propose a simple mechanism for building small heterogeneous ensembles based on exponentially weighting the probability estimates of the base classifiers with an estimate of the accuracy formed through cross-validation on the train data. We demonstrate through extensive experimentation that, given the same small set of base classifiers, this method has measurable benefits over commonly used alternative weighting, selection or meta-classifier approaches to heterogeneous ensembles. We also show how an ensemble of five well-known, fast classifiers can produce an ensemble that is not significantly worse than large homogeneous ensembles and tuned individual classifiers on datasets from the UCI archive. We provide evidence that the performance of the cross-validation accuracy weighted probabilistic ensemble (CAWPE) generalises to a completely separate set of datasets, the UCR time series classification archive, and we also demonstrate that our ensemble technique can significantly improve the state-of-the-art classifier for this problem domain. We investigate the performance in more detail, and find that the improvement is most marked in problems with smaller train sets. We perform a sensitivity analysis and an ablation study to demonstrate the robustness of the ensemble and the significant contribution of each design element of the classifier. We conclude that it is, on average, better to ensemble strong classifiers with a weighting scheme rather than perform extensive tuning and that CAWPE is a sensible starting point for combining classifiers.

Characterizing climate change risks by linking robust decision frameworks and uncertain probabilistic projections

There is increasing concern that avoiding climate change impacts will require proactive adaptation, particularly for infrastructure systems with long lifespans. However, one challenge in adaptation is the uncertainty surrounding climate change projections generated by general circulation models (GCMs). This uncertainty has been addressed in different ways. For example, some researchers use ensembles of GCMs to generate probabilistic climate change projections, but these projections can be highly sensitive to assumptions about model independence and weighting schemes. Because of these issues, others argue that robustness-based approaches to climate adaptation are more appropriate, since they do not rely on a precise probabilistic representation of uncertainty. In this research, we present a new approach for characterizing climate change risks that leverages robust decision frameworks and probabilistic GCM ensembles. The scenario discovery process is used to search across a multi-dimensional space and identify climate scenarios most associated with system failure, and a Bayesian statistical model informed by GCM projections is then developed to estimate the probability of those scenarios. This provides an important advancement in that it can incorporate decision-relevant climate variables beyond mean temperature and precipitation and account for uncertainty in probabilistic estimates in a straightforward way. We also suggest several advancements building on prior approaches to Bayesian modeling of climate change projections to make them more broadly applicable. We demonstrate the methodology using proposed water resources infrastructure in Lake Tana, Ethiopia, where GCM disagreement on changes in future rainfall presents a major challenge for infrastructure planning.

Probabilistic Snow Cover and Ensemble Streamflow Estimations in the Upper Euphrates Basin

Abstract Predicting snow cover dynamics and relevant streamflow due to snowmelt is a challenging issue in mountainous basins. Spatio-temporal variations of snow extent can be analyzed using probabilistic snow cover maps derived from satellite images within a relatively long period. In this study, Probabilistic Snow Depletion Curves (P-SDCs) and Probabilistic Snow Lines (P-SLs) are acquired from Moderate Resolution Imaging Spectroradiometer (MODIS) cloud-filtered daily snow cover images. Analyses of P-SDCs show a strong correlation with average daily runoff (R2 = 0.90) and temperature (R2 = 0.96). On the other hand, the challenge lies in developing noteworthy methods to use P-SDCs in streamflow estimations. Therefore, the main objective is to explore the feasibility of producing probabilistic runoff forecasts with P-SDC forcing in a snow dominated basin. Upper Euphrates Basin in Turkey has large snow extent and high snowmelt contribution during spring and summer periods. The melting characteristics are defined by P-SDCs using MODIS imagery for 2001-2012. The value of snow probability maps on ensemble runoff predictions is shown with Snowmelt Runoff Model (SRM) during 2013-2015 where the estimated runoff values indicate good consistency (NSE: 0.47-0.93) with forecasts based on the derived P-SDCs. Therefore, the probabilistic approach distinguishes the snow cover characteristics for a region and promotes a useful methodology on the application of probabilistic runoff predictions especially for snow dominated areas.

Ensemble sensitivity analysis of Greenland blocking in medium‐range forecasts

The North Atlantic Oscillation (NAO) is the leading mode of variability in the large‐scale circulation over the North Atlantic in winter, and strongly influences the weather and climate of Europe. On synoptic time‐scales, the negative phase of the NAO often corresponds to the occurrence of a blocking episode over Greenland. Hence, the dynamics and predictability of these blocking events is of interest for the prediction of the NAO and its related impacts over a wide region.Ensemble sensitivity analysis utilises the information contained in probabilistic forecast ensembles to calculate a statistical relationship between a forecast metric and some precursor condition. Here the method is applied to 15‐day forecasts of a set of 26 Greenland blocking events using the state‐of‐the‐art European Centre for Medium‐Range Weather Forecasts (ECMWF) forecasting system. The ensemble sensitivity analysis shows that Greenland blocking does not develop in isolation in these forecasts, but instead the blocking is sensitive to remote precursors, such as 500 and 50 hPa geopotential height, particularly in the low‐frequency flow. In general, there are more significant sensitivities to anomalies in the Tropics than in the polar regions. Stratospheric sensitivities tend to emerge at later lead times than tropospheric sensitivities. The strongest and most robust sensitivities correspond to a Rossby wave precursor reaching from the Pacific basin across North America.

Supervised latent Dirichlet allocation with a mixture of sparse softmax

Abstract Real data often show that from appearance within-class similarity is relatively low and between-class similarity is relatively high, which could increase the difficulty of classification. To classify this kind of data effectively, we learn multiple classification criteria simultaneously, and make different classification criterion be applied to classify different data for the purpose of relieving difficulty of fitting this kind of data and class label only by using a single classifier. Considering that topic model can learn high-level semantic features of the original data, and that mixture of softmax model is an efficient and effective probabilistic ensemble classification method, we embed a mixture of softmax model into latent Dirichlet allocation model, and propose a supervised topic model, supervised latent Dirichlet allocation with a mixture of softmax, and its improved version, supervised latent Dirichlet allocation with a mixture of sparse softmax. Next, we give their parameter estimation algorithms based on variational Expectation Maximization (EM) method. Moreover, we give an approximation method to classify unseen data, and analyze the convergence of the parameter estimation algorithm. Finally, we demonstrate the effectiveness of the proposed models by comparing them with some recently proposed approaches on two real image datasets and one text dataset. The experimental results demonstrate the good performance of the proposed models.