Ensemble Prediction Research Articles

Surveying Nearshore Bathymetry Using Multispectral and Hyperspectral Satellite Imagery and Machine Learning

Nearshore bathymetric data are essential for assessing coastal hazards, studying benthic habitats and for coastal engineering. Traditional bathymetry mapping techniques of ship-sounding and airborne LiDAR are laborious, expensive and not always efficient. Multispectral and hyperspectral remote sensing, in combination with machine learning techniques, are gaining interest. Here, the nearshore bathymetry of southwest Puerto Rico is estimated with multispectral Sentinel-2 and hyperspectral PRISMA imagery using conventional spectral band ratio models and more advanced XGBoost models and convolutional neural networks. The U-Net, trained on 49 Sentinel-2 images, and the 2D-3D CNN, trained on PRISMA imagery, had a Mean Absolute Error (MAE) of approximately 1 m for depths up to 20 m and were superior to band ratio models by ~40%. Problems with underprediction remain for turbid waters. Sentinel-2 showed higher performance than PRISMA up to 20 m (~18% lower MAE), attributed to training with a larger number of images and employing an ensemble prediction, while PRISMA outperformed Sentinel-2 for depths between 25 m and 30 m (~19% lower MAE). Sentinel-2 imagery is recommended over PRISMA imagery for estimating shallow bathymetry given its similar performance, much higher image availability and easier handling. Future studies are recommended to train neural networks with images from various regions to increase generalization and method portability. Models are preferably trained by area-segregated splits to ensure independence between the training and testing set. Using a random train test split for bathymetry is not recommended due to spatial autocorrelation of sea depth, resulting in data leakage. This study demonstrates the high potential of machine learning models for assessing the bathymetry of optically shallow waters using optical satellite imagery.

Skillful subseasonal ensemble predictions of heat wave onsets through better representation of land surface uncertainties

Uncertainties in land surface processes notably limit subseasonal heat wave (HW) onset predictions. A better representation of the uncertainties in land surface processes using ensemble prediction methods may be an important way to improve HW onset predictions. However, generating ensemble members that adequately represent land surface process uncertainties, particularly those related to land surface parameters, remains challenging. In this study, a conditional nonlinear optimal perturbation related to parameters (CNOP-P) approach was employed to generate ensemble members for representing the uncertainties in land surface processes resulting from parameters. Via six strong and long-lasting HW events over the middle and lower reaches of the Yangtze River (MLYR), HW onset ensemble forecast experiments were conducted with the Weather Research and Forecasting (WRF) model. The performance of the CNOP-P approach and the traditional random parameter perturbation ensemble prediction method was evaluated. The results demonstrate that the deterministic and probabilistic skills of HW onset predictions show greater excellence using the CNOP-P approach, leading to much better predictions of extreme air temperatures than those using the traditional method. This occurred because the ensemble members generated by the CNOP-P method better represented the uncertainties in important land physical processes determining HW onsets over the MLYR, notably vegetation process uncertainties, whereas the ensemble members generated by the random parameter perturbation method could not. This finding suggests that the CNOP-P method is suitable for producing ensemble members that more appropriately represent model uncertainties through more reasonable parameter error characterization.

Robust RNA secondary structure prediction with a mixture of deep learning and physics-based experts.

A mixture-of-experts (MoE) approach has been developed to mitigate the poor out-of-distribution (OOD) generalization of deep learning (DL) models for single-sequence-based prediction of RNA secondary structure. The main idea behind this approach is to use DL models for in-distribution (ID) test sequences to leverage their superior ID performances, while relying on physics-based models for OOD sequences to ensure robust predictions. One key ingredient of the pipeline, named MoEFold2D, is automated ID/OOD detection via consensus analysis of an ensemble of DL model predictions without requiring access to training data during inference. Specifically, motivated by the clustered distribution of known RNA structures, a collection of distinct DL models is trained by iteratively leaving one cluster out. Each DL model hence serves as an expert on all but one cluster in the training data. Consequently, for an ID sequence, all but one DL model makes accurate predictions consistent with one another, while an OOD sequence yields highly inconsistent predictions among all DL models. Through consensus analysis of DL predictions, test sequences are categorized as ID or OOD. ID sequences are subsequently predicted by averaging the DL models in consensus, and OOD sequences are predicted using physics-based models. Instead of remediating generalization gaps with alternative approaches such as transfer learning and sequence alignment, MoEFold2D circumvents unpredictable ID-OOD gaps and combines the strengths of DL and physics-based models to achieve accurate ID and robust OOD predictions.

Evaluation, Calibration, and Application of Probabilistic 100-m Wind Speed Forecasts Produced by the WRF Ensemble Prediction System in Taiwan

Abstract The use of renewable energy is on the rise; however, it brings more challenges for grid operations and unit scheduling due to the intermittent nature and limitations in predicting renewable power generation. High-quality meteorological forecasts play a critical role in the effective application and management of renewable energy resources. This study evaluates the quality and economic benefits of probabilistic forecasts for 100-m wind speeds from the Weather Research and Forecasting Model ensemble prediction system (WEPS). A multiple linear regression (MLR) method for calibration is also used to improve forecast quality. Additionally, this study explores the application of probabilistic hub-height wind speed forecasts in conjunction with an economic value analysis to determine whether wind turbines should be shut down during typhoon periods. The findings reveal that the forecast in the central offshore region of Taiwan exhibits the highest reliability and discrimination ability compared to those in the northern and southern regions. Additionally, employing the MLR calibration method significantly improved the reliability and discrimination ability of the probabilistic forecasts compared to the raw forecasts. Furthermore, during the typhoon seasons, almost all users, regardless of their cost–loss ratio, can benefit from basing their decisions on WEPS forecasts compared to using a single deterministic forecast. Importantly, decision-makers can benefit more from probabilistic forecasts than the ensemble mean forecasts when both are derived from the same ensemble prediction system, since the former incorporates forecast uncertainty. Significance Statement High-quality meteorological forecasts significantly influence the efficient utilization and management of renewable energy resources. Therefore, this study uses a simple yet effective approach to correct systematic bias in probabilistic wind speed forecasts over Taiwan, aiming to provide decision-makers with more reliable forecasts for hub-height wind speeds. Given that strong winds during typhoon periods pose safety concerns for wind turbine operation, we use an economic value analysis to assist turbine operators in deciding whether to shut down wind turbines. Furthermore, our research demonstrates that ensemble probabilistic forecasts can yield greater economic benefits for decision-makers when compared to the commonly used ensemble mean forecasts.

Advancing river flood forecasting with a collaborative integrated modeling method.

River flood forecasting and assessment are crucial for reducing flood risks, as they offer early alerts and allow for proactive actions to safeguard individuals from possible flood-related damage. Effective modeling in this field often multiple interconnected aspects of the hydrologic cycle, such as precipitation, infiltration, runoff, and evaporation, requiring collaboration among hydrology experts. Such collaboration enables experts to handle and manage their specialized processes more effectively, thereby enhancing the efficiency of the development of integrated flood forecasting models. Tight integration and loose integration are two common strategies for integrating different hydrologic cycle process models in river flood forecasting. However, most integration strategies rely on centralized models, necessitating experts to configure models and data on local computers. Currently, there is a deficiency in the capacity for effective collaboration in the integrated modeling of river flood forecasts. This issue arises from multiple obstacles: the complexity of understanding heterogeneous data and hydrologic cycle process models; the difficulty of integrating models with diverse runtime environments; and the challenge of synchronizing forecasting model changes among experts in real time. Therefore, we propose a web-based collaborative integrated modeling method, designed to support both tightly and loosely integrated modes, to enhance collaborative river flood forecasting and assessment. This method includes three core modules: (1) data and model description for providing a structured description of the execution logic of forecasting models and the internal structure of forecast data for expert understanding; (2) model access and integration for access and integration of data and multi-source heterogeneous models of hydrologic cycle processes; and (3) modeling scenario configuration for collaborative development of forecasting models and the execution of simulation tasks. Finally, we illustrate the application of the proposed method by utilizing the GEFS v12 (Global Ensemble Forecast System) rainfall ensemble forecasting dataset with the CREST (Coupled Routing and Excess STorage) hydrologic model. The results show enhanced efficiency in the collaborative development of river flood forecasts by hydrology experts, particularly in model accessibility, data processing, simulation, and evaluation, thereby potentially aiding decision-making.

Marine and atmospheric transport modeling supporting nuclear preparedness in Norway: Recent achievements and remaining challenges.

Numerical transport models are important tools for nuclear emergency decision makers in that they rapidly provide early predictions of dispersion of released radionuclides, which is key information to determine adequate emergency protective measures. They can also help us understand and describe environmental processes and can give a comprehensive assessment of transport and transfer of radionuclides in the environment. Transport of radionuclides in air and ocean is affected by a number of different physico-chemical processes. Along with uncertainty arising from the input data, the model estimates will therefore involve a combination of numerous uncertain factors, caused by knowledge gaps and assumptions in the model system. As discussed in this paper, the major sources to uncertainty affecting the model results are release descriptions, driving data, process descriptions and parameters. Here, we give a synthesis of the most important improvements in atmospheric and marine models achieved through the CERAD programme. In the atmospheric transport model, an important improvement has been inclusion of uncertainties in the dispersion estimates. Recent developments also include adaption to high resolution forcing data and ensemble forecasts, inversion methods and long term analyses. Case studies clearly show improved predictions from ensemble mean values compared to single deterministic runs, and promises for future upgrades of preparedness decision support systems. A major improvement in the marine model system was implementation of dynamic speciation including transformation of species, identifying particle size and parameterizations to be key factors affecting radionuclide distribution. The model system was further developed in a case study involving the impact of changing environmental factors on the transport of aluminium river run-off to an estuary in southeastern Norway. Suggestions for future improvements include implementation of an operational preparedness model for marine transport, better quantification of uncertainties using ensemble methods and improved source identification with further development of inverse transport.

Regularized Multi-Decoder Ensemble for an Error-Aware Scene Representation Network.

Feature grid Scene Representation Networks (SRNs) have been applied to scientific data as compact functional surrogates for analysis and visualization. As SRNs are black-box lossy data representations, assessing the prediction quality is critical for scientific visualization applications to ensure that scientists can trust the information being visualized. Currently, existing architectures do not support inference time reconstruction quality assessment, as coordinate-level errors cannot be evaluated in the absence of ground truth data. By employing the uncertain neural network architecture in feature grid SRNs, we obtain prediction variances during inference time to facilitate confidence-aware data reconstruction. Specifically, we propose a parameter-efficient multi-decoder SRN (MDSRN) architecture consisting of a shared feature grid with multiple lightweight multi-layer perceptron decoders. MDSRN can generate a set of plausible predictions for a given input coordinate to compute the mean as the prediction of the multi-decoder ensemble and the variance as a confidence score. The coordinate-level variance can be rendered along with the data to inform the reconstruction quality, or be integrated into uncertainty-aware volume visualization algorithms. To prevent the misalignment between the quantified variance and the prediction quality, we propose a novel variance regularization loss for ensemble learning that promotes the Regularized multi-decoder SRN (RMDSRN) to obtain a more reliable variance that correlates closely to the true model error. We comprehensively evaluate the quality of variance quantification and data reconstruction of Monte Carlo Dropout (MCD), Mean Field Variational Inference (MFVI), Deep Ensemble (DE), and Predicting Variance (PV) in comparison with our proposed MDSRN and RMDSRN applied to state-of-the-art feature grid SRNs across diverse scalar field datasets. We demonstrate that RMDSRN attains the most accurate data reconstruction and competitive variance-error correlation among uncertain SRNs under the same neural network parameter budgets. Furthermore, we present an adaptation of uncertainty-aware volume rendering and shed light on the potential of incorporating uncertain predictions in improving the quality of volume rendering for uncertain SRNs. Through ablation studies on the regularization strength and decoder count, we show that MDSRN and RMDSRN are expected to perform sufficiently well with a default configuration without requiring customized hyperparameter settings for different datasets.

Ensemble learning prediction model for lithium-ion battery remaining useful life based on embedded feature selection

Ensemble learning prediction model for lithium-ion battery remaining useful life based on embedded feature selection

ПОТОЧЕЧНЫЕ И КОМПЛЕКСНЫЕ МЕРЫ КАЧЕСТВА В ИССЛЕДОВАНИЯХ АТМОСФЕРЫ И ОКЕАНА: ОБЗОР МЕТОДОВ И ПОДХОДОВ

In the oceanic and atmospheric sciences, various general quantitative indicators, or quality metrics, describe the quality of the various modeling products, including numerical weather prediction, statistical correction, and downscaling. Metrics provide the level of accuracy of model processes reproduction and allow for comparison of models by estimating the uncertainty of their results. This paper presents a classification of the most frequently encountered quality metrics in the scientific literature. Examples are given for each group of quality metrics. In addition to assessing traditional point-by-point metrics, complex metrics that consider various aspects of modeling results are studied. Such specific metrics have an emphasis on the spatial structure, internal correlations, and heterogeneity of the predicted variable fields, ensemble forecasts etc. Special attention in this paper is also devoted to the object-oriented metrics or metrics based for rare and extreme events.

Simple GCM Simulations of Rainfall Over Northeast Brazil, Part 1: Systematic Effect of Canonical Sea Surface Temperature Patterns

ABSTRACTThe response of February to April (FMA) Northeast Brazil precipitation to tropical sea surface temperature anomaly (SSTA) patterns is analysed by making ensemble seasonal forecasts with a simple GCM. The physical and dynamical link between the SSTAs and regional precipitation is examined for a set of tropical Pacific and Atlantic SSTAs, both separately and in combination. Five canonical SSTA patterns are considered: El Niño (strong eastern Pacific and moderate central Pacific); La Niña and the tropical North and South Atlantic. The model's anomaly responses to persisted SSTAs correspond quite well to observed precipitation regressions against SSTA timeseries: dry for El Niño and warm North Atlantic, wet for La Niña and warm South Atlantic. Experiments with reversed SSTAs show that model nonlinearity has a drying effect over northeastern South America. Combined tropical Atlantic SSTA patterns result in wet conditions over northern Northeast Brazil for a warm Atlantic or a negative dipole, and dry conditions for the contrary, so the South Atlantic has a greater impact than the North Atlantic. Combining SSTA patterns in the Pacific and Atlantic gives results that are similar to the sum of individual effects, with the stronger Pacific SSTAs exerting more influence despite the model's sensitivity to Atlantic conditions. Dry dynamical experiments with fixed basic states and heating perturbations confirm that interactive moist thermodynamics is crucial for accurately representing continental rainfall responses, even over short timescales. These results are generally promising for applications in operational seasonal forecasting, and this is reported in the companion paper, part 2.

Improving Model Chain Approaches for Probabilistic Solar Energy Forecasting through Post-processing and Machine Learning

Weather forecasts from numerical weather prediction models play a central role in solar energy forecasting, where a cascade of physics-based models is used in a model chain approach to convert forecasts of solar irradiance to solar power production. Ensemble simulations from such weather models aim to quantify uncertainty in the future development of the weather, and can be used to propagate this uncertainty through the model chain to generate probabilistic solar energy predictions. However, ensemble prediction systems are known to exhibit systematic errors, and thus require post-processing to obtain accurate and reliable probabilistic forecasts. The overarching aim of our study is to systematically evaluate different strategies to apply post-processing in model chain approaches with a specific focus on solar energy: not applying any post-processing at all; post-processing only the irradiance predictions before the conversion; post-processing only the solar power predictions obtained from the model chain; or applying post-processing in both steps. In a case study based on a benchmark dataset for the Jacumba solar plant in the U.S., we develop statistical and machine learning methods for post-processing ensemble predictions of global horizontal irradiance (GHI) and solar power generation. Further, we propose a neural-network-based model for direct solar power forecasting that bypasses the model chain. Our results indicate that postprocessing substantially improves the solar power generation forecasts, in particular when post-processing is applied to the power predictions. The machine learning methods for post-processing slightly outperform the statistical methods, and the direct forecasting approach performs comparably to the post-processing strategies.

IMPACT OF PREPROCESSING AND COMPARISON OF NEURAL NETWORK ENSEMBLE METHODS FOR SEGMENTATION OF THE THORACIC SPINE IN X-RAY IMAGES

Context. Automatic segmentation of medical images plays an important role in the process of automating the detection of various diseases in the spine and the use of radiography is the most accessible means of predicting diseases. Over the years many studies have been conducted on the topic of image segmentation. One of the many methods for improving image segmentation is the use of neural network ensembles. Objective. The aims of this study were to investigate the impact of preprocessing and compare the main methods of neural network ensembles and their effect on the segmentation of the thoracic region, in this study the area was considered which consists of the vertebrae: Th8, Th9, Th10, Th11. Method. To begin with, the influence of preprocessing of X-ray images was considered, which included the following methods: histogram equalization for contrast enhancement, contrast-limited adaptive histogram equalization, logarithmic transform method, median filter, Gaussian filter, and bilateral filter. To study the influence of neural network ensemble on segmentation quality, several methods were used. Averaging method – a simple half-averaging method. Weighted averaging method – an improved version of the averaging method which uses weights for each network, the higher the network weight, the greater its influence on averaging. Method of cumulative averaging – a modified averaging method in which each ensemble receives an averaged image, after which all the results of the ensembles are averaged. Bagging – method of averaging networks trained on different data, n networks are used, the training sample is divided into n parts, and each neural network is trained on its own subset of data, as a result, the averaging method is used for predictions. Averaging method for a large number of networks – in this method, 100 neural networks were trained, after which the averaging method was used. Method of averaging mask shapes – this method uses a distance transform to average multiple masks into one shape average. Results. It was investigated that the use of different methods of image preprocessing does not guarantee an improvement in the quality of segmentation of the spine region on X-ray images, but even on the contrary worsens the quality of segmentation. Different methods of combining predictions of neural network ensembles were considered, which made it possible to find out the pros and cons of specific methods for the task of segmentation of X-ray images. Conclusions. The experiments conducted allowed us to conclude that the use of any preprocessing methods should not be used for segmentation of X-ray images. Also, due to a large number of architectures and methods for combining predictions, the behavior of ensemble methods was studied, which will allow us to further determine the necessary approach for segmentation of X-ray images. Further study of the weighted averaging method and the mask shape averaging method will make it possible to improve the obtained result and achieve even greater success in segmentation.

ENSEMBLE METHOD BASED ON AVERAGING SHAPES OF OBJECTS USING THE PYRAMID METHOD

Context. Image segmentation plays a key role in computer vision. The quality of segmentation is affected by many factors: noise, artifacts, complex shapes of objects. Classical methods cannot always guarantee good success, depending on the quality of the image and the existing noise, they cannot always achieve the desired result. The proposed method uses an ensemble of neural networks, which makes it possible to increase the accuracy and stability of segmentation. Objective. The goal of the work is to develop a new method of combining predictions of neural network ensembles, which can improve segmentation accuracy by combining images of different image sizes. Method. A method is proposed that averages the shapes of objects depicted on prediction masks. A pyramid of images is used to improve segmentation quality, each level of the pyramid corresponds to an increased size of the original image. This approach allows obtaining image characteristics at different levels. For a test image, a prediction is obtained from each neural network in the ensemble, after which a pyramid is built for the image. All pyramid levels are combined into the final image using SAAMC. All obtained final images for each neural network are also combined at the end using SAAMC. The use of an ensemble of neural networks combined with the pyramid method allows for reducing the impact of noise and artifacts on the segmentation results. Results. The use of this method was compared with the usual use of individual neural networks and the ensemble averaging method. The obtained results show that the proposed method outperforms its competitors. Application of the proposed method improved the accuracy and quality of segmentation. Conclusions. The conducted research confirmed the sense of using an ensemble of neural networks and creating a new method of combining predictions. The use of an ensemble of neural networks makes it possible to compensate for the errors and shortcomings of individual neural networks. Using the proposed method can significantly reduce the impact of noise and artifacts on segmentation. Further study and modification of this method will make it possible to further improve the quality of segmentation.

Ensemble interval forecasts of mortality

In this paper, we investigate the use of ensemble interval forecasts in mortality modelling. The construction of ensemble interval forecasts involves combining the prediction intervals generated from different mortality models in order to improve the interval coverage. We consider a wide range of combination methods, including the simple and weighted averaging, median, mode, envelope, interior and exterior trimming, and probability averaging of endpoints and simple averaging of midpoints. For the mortality models, we adopt the Lee-Carter and Cairns-Blake-Dowd families of models, as well as autoregressive models. Using the mortality data of six populations, we find that the prediction intervals produced by the proposed ensemble approach generally outperform those generated individually from a single mortality model.

Impact of ensemble‐based hybrid background‐error covariances in ECMWF's next‐generation ocean reanalysis system

AbstractAn ensemble of data assimilations (EDA) can provide valuable information on analysis and short‐range forecast uncertainties. The present European Centre for Medium‐Range Weather Forecasts (ECMWF) operational ocean analysis and reanalysis system, called the Ocean Reanalysis System 5 (ORAS5), produces an ensemble but does not exploit it for the specification of the background‐error covariance matrix , a key component of the data assimilation system. In this article, we describe EDA developments for the ocean, which take advantage of the short‐range forecast ensemble for specifying, in two distinct ways, parameters of a covariance model representation of . First, we generate a climatological ensemble over an extended period to produce seasonally varying climatological estimates of background‐error variances and horizontal correlation length‐scales. Second, in each assimilation cycle, we diagnose flow‐dependent variances from the ensemble and blend them with the climatological estimates to form hybrid variances. We also use the ensemble to diagnose flow‐dependent vertical correlation length‐scales. We demonstrate for the Argo‐rich period that this new, ensemble‐based hybrid formulation of results in a significant reduction of background errors compared with the parameterized formulation of used in ORAS5, with the globally averaged reduction ranging between 5% and 15% for subsurface temperature and salinity and up to and for sea‐surface temperature and sea‐surface height, respectively. The new ocean EDA system will be employed in ORAS6, ECMWF's next‐generation ocean reanalysis system.

BIPE: A Bi-Layer Predictive Ensemble Framework for Forest Fire Susceptibility Mapping in Germany

Forest fires diminish forests’ ecological services, including carbon sequestration, water retention, air cooling, and recreation, while polluting the environment and endangering habitats. Despite considerable economic advancements, firefighting strategies remain less than optimal. This paper introduces the Bi-layer Predictive Ensemble (BIPE), an innovative machine learning model designed to enhance the accuracy and generalization of forest fire susceptibility mapping. BIPE integrates model-centric and data-driven strategies, employing automated methods such as 10-fold cross-validation and meta-learning to improve stability and generalization. During its 10-fold cross-validation, BIPE demonstrated excellent performance, with the Area Under the Curve (AUC) values ranging from 0.990 to 0.996 and accuracy levels consistently high, around 97%, underscoring its robust class separation ability and strong generalization across different datasets. Our results confirm that BIPE outperforms traditional high-performance models like Support Vector Machine (SVM), Multilayer Perceptron (MLP), Extreme Gradient Boosting (XGBoost), Deep Neural Network (DNN), and Convolutional Neural Network (CNN), showcasing its practical effectiveness and reliability on the data of nonlinear, high-dimensional, and complex interactions. Additionally, our forest fire susceptibility maps offer valuable complementary information for German forest fire management authorities, enhancing their ability to assess and manage fire risks more effectively.

Assessing predictive attribution in NMME forecasts of summer precipitation in eastern china using deep learning

Due to systematic errors in models and the special geographic location of eastern China, most global climate models exhibit significant biases in predicting summer precipitation in this region. This study evaluates the North American Multi-Model Ensemble (NMME) forecasts for eastern China, with a lead time of six months.While NMME simulates precipitation climatology well, it poorly predicts anomalies. Using the Res34-Unet deep learning post-processing method, which has been proven to enhance NMME’s forecasts, we explore that Western Pacific Subtropical High (WPSH) and sea surface temperature (SST) are critical in enhancing forecast accuracy. Among the four models evaluated, only GEM-NEMO (correlation of 0.538 with the WPSH) and CanSIPS-IC3 (which partly captured the impact of SST anomalies on precipitation) partially reflected the key factors identified by deep learning. Simulating these factors more accurately could greatly enhance NMME’s predictive skill for summer precipitation.

Nonconventional thermal states of interacting bosonic oligomers

There has recently been a growing effort to understand the physics and intricate dynamics of many-body and many-state (multimode) interacting bosonic systems in a comprehensive manner. For instance, in photonics, nonlinear multimode fibers are being intensely investigated nowadays due to their promise for ultrahigh-bandwidth and high-power capabilities. Similar prospects are being pursued in connection with magnon Bose-Einstein (BE) condensates, and ultracold atoms in periodic lattices for room-temperature quantum devices and quantum computation, respectively. While it is practically impossible to monitor the phase space of such complex systems (classically or quantum mechanically), thermodynamics has succeeded in predicting their thermal state: the Rayleigh-Jeans (RJ) distribution for classical fields and the BE distribution for quantum systems. These distributions are monotonic and promote either the ground state or the most excited mode. Here, we demonstrate the possibility to advance the participation of other modes in the thermal state of bosonic oligomers. The resulting nonmonotonic modal occupancies are described by a microcanonical treatment, while they deviate drastically from the RJ/BE predictions of canonical and grand-canonical ensembles. Our results provide a paradigm of ensemble equivalence violation and can be used for designing the shape of thermal states. Published by the American Physical Society 2024

Optimal distributions of growing‐type initial perturbations for ensemble forecasts: Theory and application in the Lorenz‐96 model

AbstractEnsemble forecasts are frequently utilized to assess the uncertainties of prediction systems. There is a consensus that generating initial perturbations with specific structures is more conducive to characterizing the growth dynamics of analysis errors, demonstrating higher forecast skills. However, the widely used methods, such as linear singular vectors (SVs) and orthogonal conditional nonlinear optimal perturbations (O‐CNOPs) exhibit strong linear‐assumption dependence and the overestimation of growing properties for analysis errors, respectively, severely limiting the ability to capture analysis errors and forecast skills. To tackle these challenges, a theoretical framework is established to solve the optimal distribution of the nonlinear growing‐type initial perturbations by variational inference (VI) incorporated with the concept of CNOPs, marked as VI‐CNOPs. As the distribution is obtained, diverse initial perturbations for ensemble forecasts can be easily sampled in it. To evaluate the reliability of VI‐CNOPs, a series of ensemble forecast experiments are then conducted using the Lorenz‐96 model. We compare the deterministic and probabilistic forecast skills of VI‐CNOPs, O‐CNOPs, and SVs under various optimization durations. The results reveal that, as the optimization durations extend, the forecast skills of VI‐CNOPs progressively improve, consistently outperforming O‐CNOPs and SVs. This trend remains consistent across various forecast lead times. Further analysis reveals that VI‐CNOPs more effectively capture the covariance matrix of analysis errors, aligning with the fundamental concept of perturbation generation methods for ensemble forecasts. Moreover, unlike O‐CNOPs and SVs, VI‐CNOPs do not require the utilization of adjoint and tangent linear models, largely expanding its application. These results indicate the novelty and efficacy of VI‐CNOPs for ensemble forecasts.

Dynamically downscaled seasonal ocean forecasts for North American east coast ecosystems

Abstract. Using a 1/12° regional model of the Northwest Atlantic Ocean (MOM6-NWA12), we downscale an ensemble of retrospective seasonal forecasts from a 1° global forecast model. To evaluate whether downscaling improved the forecast skill for surface temperature and salinity and bottom temperature, the global and downscaled forecasts are compared with each other and with a reference forecast of persistence using anomaly correlation. Both sets of forecasts are also evaluated on the basis of mean bias and ensemble spread. We find that downscaling significantly improved the forecast skill for monthly sea surface temperature anomalies in the Northeast US Large Marine Ecosystem, a region that global models have historically struggled to skillfully predict. The downscaled sea surface temperature (SST) predictions for this region were also more skillful than the persistence baseline across most initialization months and lead times. Although some of the SST prediction skill in this region stems from the recent rapid warming trend, prediction skill above persistence is generally maintained after removing the contribution of the trend, and patterns of skill suggestive of predictable processes are also preserved. While downscaling mainly improved the SST anomaly prediction skill in the Northeast US region, it improved bottom temperature and sea surface salinity anomaly skill across many of the marine ecosystems along the North American east coast. Although improvements in anomaly prediction via downscaling were ubiquitous, the effects of downscaling on prediction bias were mixed. Downscaling generally reduced the mean surface salinity biases found in the global model, particularly in regions with sharp salinity gradients (the Northern Gulf of Mexico and the Northeast US). In some cases, however, downscaling amplified the surface and bottom temperature biases found in the global predictions. We discuss several processes that are better resolved in the regional model and contribute to the improved skill, including the autumn reemergence of temperature anomalies and advection of water masses by coastal currents. Overall, the results show that a downscaled high-resolution model can produce improved seasonal forecast skill by representing fine-scale processes that drive predictability.