Representation Of Uncertainty Research Articles

Power market models for the clean energy transition: State of the art and future research needs

As power systems around the world are rapidly evolving to achieve decarbonization objectives, it is crucial that power system planners and operators use appropriate models and tools to analyze and address the associated challenges. This paper provides a detailed overview of the properties of power market models in the context of the clean energy transition. We review common power market model methodologies, their readiness for low- and zero‑carbon grids, and new power market trends. Based on the review, we suggest model improvements and new designs to increase modeling capabilities for future grids. The paper highlights key modeling concepts related to power system flexibility, with a particular focus on hydropower and energy storage, as well as the representation of grid services, price formation, temporal structure, and the importance of uncertainty. We find that a changing resource mix, market restructuring, and growing price uncertainty require more precise modeling techniques to adequately capture the new technology constraints and the dynamics of future power markets. In particular, models must adequately represent resource opportunity costs, multi-horizon flexibility, and energy storage capabilities across the full range of grid services. Moreover, at the system level, it is increasingly important to consider sub-hourly time resolution, enhanced uncertainty representation, and introduce co-optimization for dual market clearing of energy and grid services. Likewise, models should capture interdependencies between multiple energy carriers and demand sectors.

The USGS 2023 Conterminous U.S. Time-Independent Earthquake Rupture Forecast

ABSTRACT We present the 2023 U.S. Geological Survey time-independent earthquake rupture forecast for the conterminous United States, which gives authoritative estimates of the magnitude, location, and time-averaged frequency of potentially damaging earthquakes throughout the region. In addition to updating virtually all model components, a major focus has been to provide a better representation of epistemic uncertainties. For example, we have improved the representation of multifault ruptures, both in terms of allowing more and less fault connectivity than in the previous models, and in sweeping over a broader range of viable models. An unprecedented level of diagnostic information has been provided for assessing the model, and the development was overseen by a 19-member participatory review panel. Although we believe the new model embodies significant improvements and represents the best available science, we also discuss potential model limitations, including the applicability of logic tree branch weights with respect different types of hazard and risk metrics. Future improvements are also discussed, with deformation model enhancements being particularly worthy of pursuit, as well as better representation of sampling errors in the gridded seismicity components. We also plan to add time-dependent components, and assess implications with a wider range of hazard and risk metrics.

Hierarchical Representation of Measurement Data, Metrological Uncertainty and Metadata for Calibrated Battery Tests

AbstractWe present an interoperable hierarchical data representation for battery tests, leading to improved scalability of data transmission and enhanced data accessibility and comprehensibility for both human interpretation and machine processing. The hierarchical data format includes the raw trace electrical measurement data, the metrological calibration and uncertainty data, the metadata such as experimental settings, instruments and software versions, as well as post‐processed data such as electrochemical model fit parameters. This data representation allows repetition of the battery test under the exact same conditions such that identical results are achieved within defined error bounds. This is in line with the general F. A. I. R. data approach and provides repeatability and traceability in the battery value chain. As an application of the hierarchical data representation, we show the classification of cells as pass/fail being performed with quantitative confidence levels. We demonstrate the complete workflow of establishing the hierarchical data structure for electrochemical impedance spectroscopy (EIS), starting from metrological traceability of the calibration and uncertainty analysis towards the storage of the structured data as a single integrated file that preserves the hierarchical data format. The structured data file is provided in JSON format in the Zenodo repository, as well as the program scripts to generate and read the JSON EIS files.

Enhanced prediction of parking occupancy through fusion of adaptive neuro-fuzzy inference system and deep learning models

While predicting parking occupancy is crucial for managing urban congestion, existing models often exhibit gaps in accuracy, uncertainty handling, and integration potential. This study introduces an innovative combination of adaptive neuro-fuzzy inference system (ANFIS) and deep learning (DL) techniques to address these shortcomings. Specifically, ANFIS is utilized for its proficiency in uncertainty representation via fuzzy set theory, whereas DL models excel in automatic feature learning, non-linear modeling, and identifying long-term dependencies in time-series parking data. By integrating ANFIS with recurrent neural networks (RNN), long short-term memory (LSTM), and gated recurrent units (GRU), we formulated the ANFIS-RNN, ANFIS-LSTM, and ANFIS-GRU fusion models, testing them on real-world parking datasets. Subsequent experiments highlighted the dominance of these fusion models over individual and benchmark counterparts. ANFIS-RNN achieved a 30.61% improvement in MSE, 16.70% in RMSE, 21.21% in MAE, 21.58% in MAPE, and a 1.03% elevation in R2 over the standalone RNN. The ANFIS-LSTM surpassed LSTM by 34.04% in MSE, 18.76% in RMSE, 26.16% in MAE, 27.71% in MAPE, with a 1.04% R2 increment. ANFIS-GRU exceeded GRU metrics by 27.54% in MSE, 14.85% in RMSE, 19.27% in MAE, 20.01% in MAPE, and boosted R2 by 1.03%. These outcomes underline the potential of integrated models in refining prediction precision. By leveraging the combined strengths of ANFIS and DL, this research offers a significant leap in parking occupancy forecasting. Its implications extend to data-centric urban planning and traffic regulation, marking a pivotal step for future endeavors in hybrid predictive modeling incorporating soft computing and deep learning paradigms.

Incremental updating fuzzy tolerance rough set approach in intuitionistic fuzzy information systems with fuzzy decision

As information technology develops rapidly and data is constantly updated, efficient mining knowledge from dynamic intuitionistic fuzzy information systems with fuzzy decision (IFFD) is a significant topic. In the dynamic data environment, the corresponding IFFD is always changing with time when object sets of datasets may evolve in time. Fuzzy tolerance rough set (FTRS), as one of extended rough set models, has a strong capacity for the expression of information and better representation of uncertainty. Therefore, it is very effective and necessary to use FTRS to acquire valuable knowledge from dynamic IFFD. In this paper, we investigate the dynamic fuzzy tolerance rough set approach for IFFD. Firstly, a new fuzzy tolerance similarity is defined to describe fuzzy tolerance relation between objects in the IFFD. Second, we construct a fuzzy tolerance rough set model in the IFFD, discuss some properties, and propose the corresponding static algorithm. Subsequently, incremental approaches which update fuzzy tolerance rough approximations with the insertion and deletion of objects in the IFFD are investigated and the corresponding dynamic algorithms are also designed. Finally, the feasibility of the FTRS model and the effectiveness and efficiency of the dynamic algorithm in dynamic environments are validated through a series of comparative experiments on nine datasets.

Demand forecasting and predictability identification of ride-sourcing via bidirectional spatial-temporal transformer neural processes

Understanding the spatial–temporal stochasticity in shared mobility is crucial for ride-sourcing demand forecasting, supply–demand management, and vehicle dispatch optimization. In contrast to conventional deep learning methods that typically provide point predictions or deterministic predictions, this paper introduces the bidirectional spatial–temporal Transformer neural processes (Bi-STTNP) prediction model, which stands out from conventional deep learning methods by providing probabilistic predictions and uncertainty estimations for ride-sourcing demand. Bi-STTNP captures the multivariate spatial–temporal Gaussian distribution of demand, offering not only demand expectations but also comprehensive uncertainty representations. We propose a predictability identification process based on predictive distributions to assess varying predictability across time slots and regions, improving interpretability. Our model, consisting of the bidirectional supply–demand attention module and spatial–temporal Transformer module, maintains interpretability while ensuring accurate demand expectation predictions. Extensive experiments on a real-world dataset of 15 million ride-sourcing orders in Hangzhou, China, demonstrate that Bi-STTNP outperforms baseline models in predicting demand expectation and quantifying demand uncertainty. Furthermore, we compute loose spatial–temporal predictability lower bounds and categorize regions by predictability, providing insights for optimizing passenger pricing strategies, driver incentives, and vehicle dispatching in ride-sourcing platforms.

Mediterranean tropical-like cyclone forecasts and analysis using the ECMWF ensemble forecasting system with physical parameterization perturbations

Abstract. Mediterranean tropical-like cyclones, called medicanes, present a multi-scale nature, and their track and intensity have been recognized as highly sensitive to large-scale atmospheric forcing and diabatic heating as represented by the physical parameterizations in numerical weather prediction. Here, we analyse the structure and investigate the predictability of medicanes with the aid of the European Centre for Medium-Range Weather Forecast (ECMWF) Integrated Forecast System (IFS) ensemble forecasting system with 25 perturbed members at 9 km horizontal resolution (compared with the 16 km operational resolution). The IFS ensemble system includes the representation of initial uncertainties from the ensemble data assimilation (EDA) and a recently developed uncertainty representation of the model physics with perturbed parameters (stochastically perturbed parameterizations, SPP). The focus is on three medicanes, Ianos, Zorbas, and Trixie, among the strongest in recent years. In particular, we have carried out separate ensemble simulations with initial perturbations, full physics SPP, with a reduced set of SPP, where only convection is perturbed to highlight the convective nature of medicanes and an operational ensemble combining the SPP and the initial perturbations. It is found that compared with the operational analysis and satellite rainfall data, the forecasts reproduce the tropical-like features of these cyclones. Furthermore, the SPP simulations compare to the initial-condition perturbation ensemble in terms of tracking, intensity, precipitation, and, more generally, in terms of ensemble skill and spread. Moreover, the study confirms that similar processes are at play in the development of the investigated three medicanes, in that the predictability of these cyclones is linked not only to the prediction of the precursor events (namely the deep cutoff low) but also to the interaction of the upper-level advected potential vorticity (PV) streamer with the tropospheric PV anomaly that is diabatically produced by latent heat.

Q-Multi Cubic Pythagorean Fuzzy Sets and Their Correlation Coefficients for Multi-Criteria Group Decision Making

Q-multi cubic Pythagorean fuzzy sets (Q-mCPFSs) are influential, effective and symmetrical for representing uncertain and imprecise information in decision making processes. Q-mCPFSs extend the concept of Q-multi fuzzy sets by introducing the notion of cubic Pythagorean membership functions, which provide a more flexible and accurate representation of uncertainty. First, we will introduce the concepts of Q-mPFSs and Q-mIVPFSs. With the combination of Q-mPFSs and Q-mIVPFSs, we will present the concept of Q-mCPFSs. Then, we propose two correlation coefficients for Q-mCPFSs. Furthermore, multi-criteria GDM methods using Q-mCPFSs will be discussed, highlighting their advantages in handling uncertain and imprecise information. Finally, we will provide an illustrative example, to demonstrate the effectiveness of Q-mCPFSs in decision making processes.The main contributions of the Q-mCPFS information expression, correlation coefficients and GDM methods in the Q-mCPFS setting of both uncertainty and certainty are thus highlighted in this study. These contributions provide valuable insights into the application of Q-mCPFSs in decision making processes, allowing decision makers to make more informed and effective choices. Additionally, the illustrative example serves as a practical demonstration of how these methods can be applied in real-world scenarios, further emphasizing their effectiveness and relevance.

Improved computational methods for probabilistic liquefaction hazard analysis

Current procedures for analysis of and design against liquefaction hazards focus primarily on the use of probabilistic ground motions at a single ground-shaking hazard level, with the cyclic loading represented by a peak ground acceleration (PGA) corresponding to a target return period and a single representative moment magnitude Mw. These parameters are typically used in conjunction with deterministic simplified procedures for estimating liquefaction triggering and consequences that largely neglect the considerable uncertainties inherent to liquefaction problems. Motivated by these limitations and the resulting inconsistencies in liquefaction design levels, early methods for probabilistic liquefaction hazard analysis (PLHA) were proposed that incorporate the full ground motion hazard space, integrated with probabilistic liquefaction triggering models. Such methods provide liquefaction factor of safety (FSL) hazard curves for standard penetration test (SPT) data. Recognizing the increased use of higher-resolution cone penetrometer test (CPT) data in engineering analysis and design and the potential computational challenges it presents, an expanded suite of probabilistic triggering models, and wider availability of more detailed seismic hazard data, an improved PLHA computational methodology is presented in this study. The methods described utilize the U.S. Geological Survey National Seismic Hazard Model directly through web services to obtain PGA hazard and disaggregation calculations for any site and average shear wave velocity in the upper 30 m of the site (Vs30) in the conterminous United States, to reconstruct the PGA hazard space for use in PLHA calculations, and to employ array calculations for efficient liquefaction hazard curve estimates for the thousands of CPT measurements in a given profile. The framework presented here is modular in nature, and discussion on the use of alternative models and extension to probabilistic liquefaction consequence evaluation is presented. The importance of appropriate representation of probabilistic liquefaction model uncertainties is also highlighted, along with the impacts of different levels of uncertainty on the PLHA calculation. Finally, a potential roadmap for incorporation of the PLHA framework in seismic provisions is presented, with an illustration of how it can address current limitations and impacts in liquefaction hazard analysis and design.

Confidence Contours: Uncertainty-Aware Annotation for Medical Semantic Segmentation

Medical image segmentation modeling is a high-stakes task where understanding of uncertainty is crucial for addressing visual ambiguity. Prior work has developed segmentation models utilizing probabilistic or generative mechanisms to infer uncertainty from labels where annotators draw a singular boundary. However, as these annotations cannot represent an individual annotator's uncertainty, models trained on them produce uncertainty maps that are difficult to interpret. We propose a novel segmentation representation, Confidence Contours, which uses high- and low-confidence ``contours’’ to capture uncertainty directly, and develop a novel annotation system for collecting contours. We conduct an evaluation on the Lung Image Dataset Consortium (LIDC) and a synthetic dataset. From an annotation study with 30 participants, results show that Confidence Contours provide high representative capacity without considerably higher annotator effort. We also find that general-purpose segmentation models can learn Confidence Contours at the same performance level as standard singular annotations. Finally, from interviews with 5 medical experts, we find that Confidence Contour maps are more interpretable than Bayesian maps due to representation of structural uncertainty.

Supplemental Material for Effects of Aging on Externally Cued and Internally Driven Uncertainty Representations

Supplemental Material for Effects of Aging on Externally Cued and Internally Driven Uncertainty Representations

83 Uncertainty Representation in Mild Cognitive Impairment: Comparing Internally Cued versus Externally Driven Uncertainty

Objective:Choice response time (RT) increases linearly with increasing information uncertainty, which can be represented externally or internally. Using a card-sorting task, we previously showed that Alzheimer’s disease (AD) dementia patients were more impaired relative to cognitively normal older adults (CN) under conditions that manipulated internally cued rather than externally driven uncertainty, but this study was limited by a between-subjects design that prevented us from directly comparing the two uncertainty conditions. The objective of this study was to assess internally cued and externally driven cued uncertainty representations in CN and mild cognitive impairment (MCI) patients.Participants and Methods:Older participants (age > 60 years; N=49 CN, N=33 MCI patients) completed a card-sorting task that separately manipulated externally cued uncertainty (i.e., the number of sorting piles with equal probability of each stimulus type) or internally cued uncertainty (i.e., the probability of each stimulus type with fixed number of sorting piles) at three different uncertainty loads (low, medium, high). Exploratory analyses separated MCI patients by etiology into possible/probable cortical neurodegenerative process (i.e., AD, frontotemporal dementia; N=13) or nonneurodegenerative process (i.e., vascular, psychiatric, sleep, medication effect; N=20).Results:CN and MCI patients maintained a high level of accuracy on both tasks (M accuracy > .94 across conditions). MCI patients performed more slowly than CN on the externally and internally cued tasks, and both groups showed a significant positive association between uncertainty load and RT (p’s < .05). There was a group x load x uncertainty condition interaction (p = .05). For CNs, the slope of the linear association between load and RT was significantly steeper in the externally cued compared to internally cued condition. For MCI patients in contrast, RTs increased with load to a similar degree in both conditions. Exploratory analyses showed the MCI-neurodegenerative patients were significantly slower than MCI-nondegenerative and CN (p < .001). While the group x load x condition interaction was significant when comparing all three groups (p < .05), this was driven by the differences between CN and MCI patients described above; the MCI-neurodegenerative and non-neurodegenerative groups did not significantly differ in the strength of the RT-load association between the externally or internally cued conditions.Conclusions:Overall, CN participants showed greater RT slowing with increasing load of externally driven than internally cued uncertainty. Though they were slower than CNs, MCI patients (even those with a possible/probable cortical neurodegenerative condition) were able to accurately perform an internally cued uncertainty task and did not show differential slowing compared to an externally driven task. This provides preliminary evidence that internal representations of probabilistic information are intact in patients with MCI due to a neurodegenerative condition, meaning they may not depend on cortical processes. Future work will increase the sample sizes of the MCI-neurodegenerative and non-degenerative groups.

Representing model uncertainties in brittle fracture simulations

This work focuses on the representation of model-form uncertainties in phase-field models of brittle fracture. Such uncertainties can arise from the choice of the degradation function for instance, and their consideration has been unaddressed to date. The stochastic modeling framework leverages recent developments related to the analysis of nonlinear dynamical systems and relies on the construction of a stochastic reduced-order model. In the latter, a POD-based reduced-order basis is randomized using Riemannian projection and retraction operators, as well as an information-theoretic formulation enabling proper concentration in the convex hull defined by a set of model proposals. The model thus obtained is mathematically admissible in the almost sure sense and involves a low-dimensional hyperparameter, the calibration of which is facilitated through the formulation of a quadratic programming problem. The relevance of the modeling approach is further assessed on one- and two-dimensional applications. It is shown that model uncertainties can be efficiently captured and propagated to macroscopic quantities of interest. An extension based on localized randomization is also proposed to handle the case where the forward simulation is highly sensitive to sample localization. This work constitutes a methodological development allowing phase-field predictions to be endowed with statistical measures of confidence, accounting for the variability induced by modeling choices.

Probabilistic finite element-based reliability of corroded pipelines with interacting corrosion cluster defects

The random distribution of pitting corrosion defects in pipelines normally leads to interacting cluster defects that behave noticeably different from single metal loss defects. Generally, the probabilistic approaches employ explicit burst pressure limit states for corroded pipelines subjected to only internal pressure. However, a significant level of conservatism is typically associated with the probabilistic assessments of corroded pipelines using the closed-form explicit limit state functions, which presents considerable challenges in maintenance planning and risk management. Therefore, this paper proposes a pathway for developing efficient performance functions for the assessment of interacting pipeline corrosion clustering defects using probabilistic finite element-based reliability method. This seven-stage framework combines multiple uncertainty representation schemes to evaluate the probability of failure. The impact of the critical design variables such as the elastic and plastic material properties, corrosion features, and interacting cluster defect characterisations are identified to guide the burst pressure design, operations and maintenance optimisation. The employed surrogate-based active learning reliability approach yielded an efficient probability estimate at a lesser computational cost than the simulation-based reliability methods. The proposed framework reduces the conservatism and computational cost related with explicit burst pressure limit state functions for corroded pipelines and offers informed decision-making on risk and maintenance management.

Prefrontal Cortex Codes Representations of Target Identity and Feature Uncertainty.

Many objects in the real world have features that vary over time, creating uncertainty in how they will look in the future. This uncertainty makes statistical knowledge about the likelihood of features critical to attention demanding processes such as visual search. However, little is known about how the uncertainty of visual features is integrated into predictions about search targets in the brain. In the current study, we test the idea that regions prefrontal cortex code statistical knowledge about search targets before the onset of search. Across 20 human participants (13 female; 7 male), we observe target identity in the multivariate pattern and uncertainty in the overall activation of dorsolateral prefrontal cortex (DLPFC) and inferior frontal junction (IFJ) in advance of the search display. This indicates that the target identity (mean) and uncertainty (variance) of the target distribution are coded independently within the same regions. Furthermore, once the search display appears the univariate IFJ signal scaled with the distance of the actual target from the expected mean, but more so when expected variability was low. These results inform neural theories of attention by showing how the prefrontal cortex represents both the identity and expected variability of features in service of top-down attentional control.SIGNIFICANCE STATEMENT Theories of attention and working memory posit that when we engage in complex cognitive tasks our performance is determined by how precisely we remember task-relevant information. However, in the real world the properties of objects change over time, creating uncertainty about many aspects of the task. There is currently a gap in our understanding of how neural systems represent this uncertainty and combine it with target identity information in anticipation of attention demanding cognitive tasks. In this study, we show that the prefrontal cortex represents identity and uncertainty as unique codes before task onset. These results advance theories of attention by showing that the prefrontal cortex codes both target identity and uncertainty to implement top-down attentional control.

Studying the neural representations of uncertainty.

The study of the brain's representations of uncertainty is a central topic in neuroscience. Unlike most quantities of which the neural representation is studied, uncertainty is a property of an observer's beliefs about the world, which poses specific methodological challenges. We analyze how the literature on the neural representations of uncertainty addresses those challenges and distinguish between 'code-driven' and 'correlational' approaches. Code-driven approaches make assumptions about the neural code for representing world states and the associated uncertainty. By contrast, correlational approaches search for relationships between uncertainty and neural activity without constraints on the neural representation of the world state that this uncertainty accompanies. To compare these two approaches, we apply several criteria for neural representations: sensitivity, specificity, invariance and functionality. Our analysis reveals that the two approaches lead to different but complementary findings, shaping new research questions and guiding future experiments.

A charging-feature-based estimation model for state of health of lithium-ion batteries

Within battery management systems, the state of health of lithium-ion batteries is a key and vital enabler to ensure battery safety and efficiency. However, the accurate state of health estimation is still a critical but challenging task, and the complex electrochemical attributes underlying the degradation processes of lithium-ion batteries are not directly available. In response to this challenge, this study proposes a charging-feature-based model to realize state of health estimation by Gaussian process regression. In this approach, two features are extracted only from the monitoring parameters obtained from charging current and voltage curves. These extracted features have been demonstrated to be correlated with the state of health. Subsequently, a regression model with a 2-dimensional linear mean function and a new double-covariance function is developed to improve estimation performance. Consequently, the proposed model effectively tracks both global and local degradation trends synchronously. Finally, the reliability and accuracy of the proposed model are verified using two different batteries datasets. The results illustrate that the proposed model is capable of realizing accurate batteries’ state of health estimation, thereby outperforming other counterparts in uncertainty representation and estimation errors, whether under static profiles or dynamic profiles.

Analyzing system reliability using conflicting bifuzzy failure rates of the components

Since uncertainty and ambiguity are common in real-world problems, numerous methods based on different kinds of fuzzy sets have been widely used in the evaluation of system reliability over the years. However, these methods are unable to deal with conflict uncertainty, particularly when it comes to the uncertain failure rates of the system's components. This paper introduces the idea of the triangular conflicting bifuzzy set (TCBFS), which is a novel paradigm for dealing with conflicting uncertainty. A novel method is devised to determine the membership and non-membership functions for the system's reliability when subsystems/components exhibit conflicting bifuzzy failure rates. In order to generate the membership and non-membership functions for various system configurations, including series, parallel, series-parallel, and parallel-series systems, non-linear programming is used to derive a number of functions based on conflicting bifuzzy numbers. The validity of the suggested method in modeling and evaluating system reliability with different conflicting bifuzzy failure rates is demonstrated through illustrated scenarios. This novel approach improves the precision and resilience of reliability estimations for systems by providing a more precise representation of uncertainties.

MaskDUF: Data uncertainty learning in masked face recognition with mask uncertainty fluctuation

As an essential component of experts and intelligent systems, Masked Face Recognition (MFR) has been applied to various applications, but it is still a challenging task due to the aggravated uncertainty caused by mask occlusion. Most existing MFR methods are deterministic point embedding models that are limited in representing the uncertainty in masked face images. Data Uncertainty Learning (DUL) is an advanced uncertainty modeling method, but there are two problems when it is applied to MFR task: (1) the masked face tends to be regarded as noise due to the mask occlusion, which weakens its optimization; (2) the large representation difference between face and masked face results in a dispersed intra-class distribution. To solve the above problems, we propose a novel masked face data uncertainty learning method (MaskDUF) for MFR task, which can adaptively adjust the optimization weight by modeling the uncertainty and measuring the recognizability of samples, thus learning an ideal sample distribution with compact intra-class, discrepant inter-class and distant noise. Specifically, a Hard Kullback–Leibler Divergence (H-KLD) method is proposed to serve as an adaptive variance regularizer for masked faces, which contributes to learning more accurate uncertainty representations and avoiding overfitting noise. Moreover, by combining feature magnitude and variance uncertainty, Mask Uncertainty Fluctuation (MUF) is proposed to comprehensively measure sample recognizability, which contributes to enhancing the learning preference of masked faces and constructing a more compact cone-like intra-class distribution. Finally, compared with other advanced models, MaskDUF achieves an average accuracy improvement of 1.33% to 13.28%, and its effectiveness and strong robustness are also proved by ablation study, noise experiment and parametric analysis.

Leveraging Deep Learning for Climate Change Prediction Models : A Dive into Cutting-Edge Techniques

In recent years, deep learning has emerged as a promising approach for statistical downscaling, which involves generating high-resolution climate data from coarse low-resolution variables. However, concerns persist regarding the ability of these models to generalize to future climate change scenarios, primarily due to the assumption of stationarity. In this study, we propose the use of deep ensembles as a straightforward method to enhance the uncertainty quantification of statistical downscaling models. By improving the representation of uncertainty, these models offer superior planning capabilities against extreme weather events, which can have significant negative social and economic impacts. Given the absence of observational future data, we rely on pseudo-reality experiments to evaluate the effectiveness of deep ensembles in quantifying the uncertainty of climate change projections. The adoption of deep ensembles facilitates more robust risk assessment, addressing critical needs in various sectors for adapting to climate change challenges.