Prediction Uncertainty Research Articles

Optimizing multi-time series forecasting for enhanced cloud resource utilization based on machine learning

Due to its flexibility, cloud computing has become essential in modern operational schemes. However, the effective management of cloud resources to ensure cost-effectiveness and maintain high performance presents significant challenges. The pay-as-you-go pricing model, while convenient, can lead to escalated expenses and hinder long-term planning. Consequently, FinOps advocates proactive management strategies, with resource usage prediction emerging as a crucial optimization category. In this research, we introduce the multi-time series forecasting system (MSFS), a novel approach for data-driven resource optimization alongside the hybrid ensemble anomaly detection algorithm (HEADA). Our method prioritizes the concept-centric approach, focusing on factors such as prediction uncertainty, interpretability and domain-specific measures. Furthermore, we introduce the similarity-based time-series grouping (STG) method as a core component of MSFS for optimizing multi-time series forecasting, ensuring its scalability with the rapid growth of the cloud environment. The experiments performed demonstrate that our group-specific forecasting model (GSFM) approach enabled MSFS to achieve a significant cost reduction of up to 44%.

Deep learning-based geological parameterization for history matching CO2 plume migration in complex aquifers

History matching is crucial for reliable numerical simulation of geological carbon storage (GCS) in deep subsurface aquifers. This study focuses on inferring highly complex aquifer permeability fields with multi- and intra-facies heterogeneity to improve the characterization of CO2 plume migration. We propose a deep learning (DL)-based parameterization strategy combined with the ensemble smoother with multiple data assimilation (ESMDA) algorithm to formulate an integrated inverse framework. The DL model is employed to parameterize non-Gaussian permeability fields using low-dimensional latent variables in a Gaussian distribution, thereby mitigating the non-Gaussianity issue faced by the ensemble-based ESMDA inverse method and simultaneously alleviating the computational burden of high-dimensional inversion. The efficacy of the integrated DL-ESMDA inverse framework is demonstrated using a 3-D GCS model, where it estimates the non-Gaussian permeability field characterized by multi- and intra-facies heterogeneity. Results show that the DL model is able to represent the highly complex and high-dimensional permeability fields using low-dimensional latent vectors. The DL-ESMDA framework sequentially updates these low-dimensional latent vectors instead of the original high-dimensional permeability field to obtain posterior estimations of the permeability field. The resulting CO2 plume migration closely matches historical measurements, suggesting a significantly improved model reliability after history matching. Additionally, a substantial reduction in uncertainty for future plume migration predictions beyond the history matching period is observed. The proposed framework provides an effective approach for reliable characterization of CO2 plume migration in highly heterogeneous aquifers, enhancing GCS project operation and risk analysis.

Neural encoding of melodic expectations in music across EEG frequency bands.

The human brain tracks regularities in the environment and extrapolates these to predict future events. Prior work on music cognition suggests that low-frequency (1-8Hz) brain activity encodes melodic predictions beyond the stimulus acoustics. Building on this work, we aimed to disentangle the frequency-specific neural dynamics linked to melodic prediction uncertainty (modelled as entropy) and prediction error (modelled as surprisal) for temporal (note onset) and content (note pitch) information. By using multivariate temporal response function (TRF) models, we re-analysed the electroencephalogram (EEG) from 20 subjects (10 musicians) who listened to Western tonal music. Our results show that melodic expectation metrics improve the EEG reconstruction accuracy in all frequency bands below the gamma range (< 30 Hz). Crucially, we found that entropy contributed more strongly to the reconstruction accuracy enhancement compared to surprisal in all frequency bands. Additionally, we found that the encoding of temporal, but not content, information metrics was not limited to low frequencies, rather it extended to higher frequencies (> 8Hz). An analysis of the TRF weights revealed that the temporal predictability of a note (entropy of note onset) may be encoded in the delta- (1-4Hz) and beta-band (12-30 Hz) brain activity prior to the stimulus, suggesting that these frequency bands associate with temporal predictions. Strikingly, we also revealed that melodic expectations selectively enhanced EEG reconstruction accuracy in the beta band for musicians, and in the alpha band (8-12 Hz) for non-musicians, suggesting that musical expertise influences the neural dynamics underlying predictive processing in music cognition.

Early warning system for floods at estuarine areas: combining artificial intelligence with process-based models

AbstractFloods are among the most common natural disasters, causing countless losses every year worldwide and demanding urgent measures to mitigate their impacts. This study proposes a novel combination of artificial intelligence and process-based models to construct a flood early warning system (FEWS) for estuarine regions. Using streamflow and rainfall data, a deep learning model with long short-term memory layers was used to forecast the river discharge at the fluvial boundary of an estuary. Afterwards, a hydrodynamic process-based model was used to simulate water levels in the estuary. The river discharge predictors were trained using different forecasting windows varying from 3 h to 36 h to assess the relationship between the time window and accuracy. The insertion of attention layers into the network architecture was evaluated to enhance forecasting capacity. The FEWS was implemented in the Douro River Estuary, a densely urbanised flood-prone area in northern Portugal. The results demonstrated that the Douro Estuary FEWS is reliable for discharges up to 5000 m3/s, with predictions made 36 h in advance. For values higher than this, the uncertainties in the model predictions increased; however, they were still capable of detecting flood occurrences.

Hydrological streamflow modeling at the inlet of Gilgel Gibe III dam of Ethiopia using soil and water assessment tool

AbstractComprehensive approach is a pre‐requisite to improve the estimation of variability in streamflow data for designing water supply systems, generating hydropower, determining environmental flows, allocating water, and conducting pollution studies that need a comprehensive approach. There is a need to model streamflow data to develop effective water policy plans when sufficient data are unavailable. This study aims at assessing the accuracy and applicability of the Soil and Water Assessment Tool (SWAT) model in predicting streamflow at the inlet of the Gilgel Gibe III dam watershed in Ethiopia. Historical meteorological data from nine available stations between 1987 and 2017 were used to predict the streamflow. The SWAT model was calibrated and validated for two different time periods (1997–1999 and 2003–2004) using SWATCUP program and SUFI‐2 technique. Calibration and validation were based on monthly observed discharges at the Abelti and Gojeb stations. The results showed acceptable values of R2, Nash–Sutcliffe modeling efficiency (NSE), and percent bias (PBIAS). At the Abelti station and Gojeb station, the calibration and validation results were 0.82, 0.79, and −10.1%; 0.88, 0.84, and −14.4%; and 0.86, 0.79, and 14.1% for calibration and 0.88, 0.8, and 11.1% for validation, respectively. The prediction uncertainty of 95% was consistent with the observed discharge. The mean annual runoff was 440.08 m3/s, demonstrating the model's effectiveness in simulating streamflow at the dam's inlet. The calibrated and validated model can be used to study the effects of climate and land use changes; analyze water quality and sediment yield; and inform future water policy plans, dam construction planning, and flood disaster risk management.

Saltwater intrusion simulations in coastal karstic aquifers related to climate change scenarios

Coastal zones are crucial ecosystems supporting significant biodiversity and pertinent socio-economic activities. However, anthropogenic development contributes to socio-environmental complexities, particularly public water supply threats caused by climate change. This research presents a case study on the north-western coast of Yucatan, Mexico, which models potential saltwater intrusion in groundwater for multiple projections of sea level rise and recharge change due to climate change and its implications for the public water supply of the regional population and ecosystem. For this purpose, a previously calibrated and validated numerical model is employed, adapting its boundary conditions, keeping its calibrated hydrogeologic parameters, and considering the 2040 and 2100 climate change projections. Simulation results show that under these projections, significant saltwater intrusion may occur, reducing freshwater thickness due to increased salinity in groundwater and a loss of freshwater sources resulting from brackish-saline wedge intrusion. These scenarios are of particular concern as freshwater in this coastal region is the main source for public water supply and for freshwater input in coastal ecosystems. Moreover, this study underscores the susceptibility of karstic aquifers to salinization, especially in the face of rising sea levels, given their unique hydrogeological characteristics and substantial responsiveness to marine forcings. In spite of the uncertainties in global climate change predictions, this study enhances our understanding of the dynamics of these unique aquifers, and presents future saltwater intrusion projections that offer valuable technical insights to design and implement pertinent and resilient coastal aquifer management strategies.

Towards a Better Uncertainty Quantification in Automated Valuation Models

AbstractThis study introduces a novel framework for quantifying prediction uncertainty in automated valuation models (AVMs), crucial tools in modern real estate finance. While non-linear AVMs excel in predictive performance, their limited methods for assessing prediction uncertainty reduces reliability and practical utility. We address this gap by proposing an approach for quantifying the uncertainty associated with predicted house prices and by introducing a model-specific AVM uncertainty estimate (AVMU) for AVM comparisons. Using a dataset of 51,747 historical apartment transactions in Oslo, Norway, we train three AVMs (XGBoost, random forest, support vector machine) to predict sales prices. Thereafter, we develop three base uncertainty estimators (direct loss estimation, bootstrap ensemble, quantile regression) and three meta estimators (average regressor, voting regressor, stacked generalization) for uncertainty quantification. Conformal calibration aligns the outputted uncertainty estimates from the six estimators with standard deviations of corresponding prediction errors. Having strong positive correlations with observed absolute prediction errors, the calibrated uncertainty estimators are shown to effectively capture prediction uncertainty. While the direct loss estimation excels among base estimators, the voting regressor and stacked generalization meta estimators consistently outperform it. Furthermore, by using the AVMU estimate from the stacked generalization meta estimator we can successfully identify the best-performing AVM for three separate apartment portfolios without knowing true sales prices. This alignment of the mean estimated prediction uncertainty with observed deviations underlines the utility of pre-factual AVMU estimates for model comparisons. In conclusion, our framework helps bridge prediction accuracy and uncertainty for AVMs, enhancing their reliability and supporting informed decision making for stakeholders.

Uncertainty Qualification for Deep Learning-Based Elementary Reaction Property Prediction.

The prediction of the thermodynamic and kinetic properties of elementary reactions has shown rapid improvement due to the implementation of deep learning (DL) methods. While various studies have reported the success in predicting reaction properties, the quantification of prediction uncertainty has seldom been investigated, thus compromising the confidence in using these predicted properties in practical applications. Here, we integrated graph convolutional neural networks (GCNN) with three uncertainty prediction techniques, including deep ensemble, Monte Carlo (MC)-dropout, and evidential learning, to provide insights into the uncertainty quantification and utility. The deep ensemble model outperforms others in accuracy and shows the highest reliability in estimating prediction uncertainty across all elementary reaction property data sets. We also verified that the deep ensemble model showed a satisfactory capability in recognizing epistemic and aleatoric uncertainties. Additionally, we adopted a Monte Carlo Tree Search method for extracting the explainable reaction substructures, providing a chemical explanation for DL predicted properties and corresponding uncertainties. Finally, to demonstrate the utility of uncertainty qualification in practical applications, we performed an uncertainty-guided calibration of the DL-constructed kinetic model, which achieved a 25% higher hit ratio in identifying dominant reaction pathways compared to that of the calibration without uncertainty guidance.

Optimization of underground mining production layouts considering geological uncertainty using deep reinforcement learning

Mineral extraction plays a key role in the global raw materials supply chain, however the exhaustion of shallow deposits and typical scarcity of sampled data during exploration activities creates challenges in mine planning and design, where decision-making is highly sensitive to uncertainty in geology and mineral grade prediction. Geostatistical techniques are commonly used to generate a set of equiprobable simulated numerical models to capture these uncertainties, however incorporating these simulated models within a mine planning and design framework remains a major challenge. The purpose of this paper is to propose a novel approach to decision-making in underground mine design that can use information from an ensemble of numerical realizations of a mineral resource to improve the financial performance of the asset. A deep reinforcement learning (DRL) framework, using the proximal policy optimization (PPO) algorithm, is developed for the design of underground mining production level layouts. A case study is presented using a gold mineral resource characterized by an ensemble of 100 numerical realizations to verify the advantages of the proposed method, considering a baseline consisting of an industry standard automated design method. The DRL approach achieved an 8.3% higher expected profit, a 3.4% more gold mined than the baseline, and has the added functionality of considering uncertainty in mineral grades.

Multifractal Analysis for Evaluating the Representation of Clouds in Global Kilometer‐Scale Models

AbstractClouds are one of the largest sources of uncertainty in climate predictions. Global km‐scale models need to simulate clouds and precipitation accurately to predict future climates. To isolate issues in their representation of clouds, models need to be thoroughly evaluated with observations. Here, we introduce multifractal analysis as a method for evaluating km‐scale simulations. We apply it to outgoing longwave radiation fields to investigate structural differences between observed and simulated anvil clouds. We compute fractal parameters which compactly characterize the scaling behavior of clouds and can be compared across simulations and observations. We use this method to evaluate the nextGEMS ICON simulations via comparison with observations from the geostationary satellite GOES‐16. We find that multifractal scaling exponents in the ICON model are significantly lower than in observations. We conclude that too much variability is contained in the small scales leading to less organized convection and smaller, isolated anvils.

Genetic Prediction Modeling in Large Cohort Studies via Boosting Targeted Loss Functions.

Polygenic risk scores (PRS) aim to predict a trait from genetic information, relying on common genetic variants with low to medium effect sizes. As genotype data are high-dimensional in nature, it is crucial to develop methods that can be applied to large-scale data (large and large ). Many PRS tools aggregate univariate summary statistics from genome-wide association studies into a single score. Recent advancements allow simultaneous modeling of variant effects from individual-level genotype data. In this context, we introduced snpboost, an algorithm that applies statistical boosting on individual-level genotype data to estimate PRS via multivariable regression models. By processing variants iteratively in batches, snpboost can deal with large-scale cohort data. Having solved the technical obstacles due to data dimensionality, the methodological scope can now be broadened-focusing on key objectives for the clinical application of PRS. Similar to most methods in this context, snpboost has, so far, been restricted to quantitative and binary traits. Now, we incorporate more advanced alternatives-targeted to the particular aim and outcome. Adapting the loss function extends the snpboost framework to further data situations such as time-to-event and count data. Furthermore, alternative loss functions for continuous outcomes allow us to focus not only on the mean of the conditional distribution but also on other aspects that may be more helpful in the risk stratification of individual patients and can quantify prediction uncertainty, for example, median or quantile regression. This work enhances PRS fitting across multiple model classes previously unfeasible for this data type.

Contrail altitude estimation using GOES-16 ABI data and deep learning

Abstract. The climate impact of persistent aircraft contrails is currently estimated to be comparable to that due to aviation-emitted CO2. A potential near-term and low-cost mitigation option is contrail avoidance, which involves rerouting aircraft around ice-supersaturated regions, preventing the formation of persistent contrails. Current forecasting methods for these regions of ice supersaturation have been found to be inaccurate when compared to in situ measurements. Further assessment and improvements of the quality of these predictions can be realized by comparison with observations of persistent contrails, such as those found in satellite imagery. In order to further enable comparison between these observations and contrail predictions, we develop a deep learning algorithm to estimate contrail altitudes based on GOES-16 Advanced Baseline Imager (ABI) infrared imagery. This algorithm is trained using a dataset of 3267 contrails found within Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) data and achieves a root mean square error (RMSE) of 570 m. The altitude estimation algorithm outputs probability distributions for the contrail top altitude in order to represent predictive uncertainty. The 95 % confidence intervals constructed using these distributions, which are shown to contain approximately 95 % of the contrail data points, are found to be 2.2 km thick on average. These intervals are found to be 34.1 % smaller than the 95 % confidence intervals constructed using flight altitude information alone, which are 3.3 km thick on average. Furthermore, we show that the contrail altitude estimates are consistent in time and, in combination with contrail detections, can be used to observe the persistence and three-dimensional (3D) evolution of contrail-forming regions from satellite images alone.

Comment on Krüger, L. Decreasing Trends of Chinstrap Penguin Breeding Colonies in a Region of Major and Ongoing Rapid Environmental Changes Suggest Population Level Vulnerability. Diversity 2023, 15, 327

Historical data on chinstrap penguin (Pygoscelis antarctica) breeding population sizes are sparse and sometimes highly uncertain, making it hard to estimate true population trajectories. Yet, information on population trends is desirable as changes in population size can help inform conservation assessments. Recently, Krüger (2023) (Diversity 2023, 15, 327) used chinstrap penguin nest count data to predict breeding colony size trends between 1960 and 2020, to estimate whether the level of population change within three generations exceeded IUCN Red List Criteria for “Vulnerable” populations. Chinstrap penguin population trends are an important research topic, but we caution that Krüger (2023)’s statistical analyses (intended to form the foundation for drawing valid, evidence-based inferences from sparse data) contain fundamental errors that invalidate that paper’s findings. We discuss oversights in several key steps (data processing, exploratory data analysis, model fitting, model evaluation, and prediction) of that paper’s analysis to help others detect and avoid some of the pitfalls associated with estimating population trends via mixed models. We also show through reanalysis that improved statistical modelling can yield better predictions of chinstrap penguin population trends, at least within the range of observed data. This case study highlights (1) the profound influence that seemingly minor differences in modelling procedures (both unintentional errors and other decisions) can have on predictions of population trends, and (2) the substantial inherent uncertainty in population trend predictions derived from sparse, heterogenous data.

Analysis and control of Hall effect thruster using optical emission spectroscopy and artificial neural network

This study presents a proof-of-principle for using optical emission spectroscopy and artificial neural networks for real-time monitoring and control of the operational parameters of a Hall effect thruster: the anode voltage, the anode xenon injection, the discharge current, and the coil current. In that regard, we build an optical database of 26 spectral lines across 6469 operating conditions to train and test the neural network. We then reduced the learning lines from 26 to 15 based on their statistical correlation with the target parameters. After tuning the hyperparameters of the network, the network predicted the thruster’s parameters with notable accuracies: 95% for the anode voltage, 84% for the coil current, and 99% for both the anode flow rate and the discharge current. The estimated uncertainty of predictions, at 3σ, is ±51V for voltage, ±1A for coil current, ±0.15A for discharge current, and ±0.15mgs−1 for anode flow rate. The prediction calculations were within milliseconds and enabled real-time monitoring of the thruster parameters. Therefore, a proportional-integrator-derivative controller (PID) controller was implemented to regulate the anode voltage and flow rate based on the optical emission of the plume. The PID showcased short settling times from 0.1 to 0.4 s and overshoot levels up to 3% of the target value for the voltage and 10% of the target value for the flow rate. These results were for a fixed coil current at 4A. The study showed that changing the coil current may necessitate more sophisticated prediction models and control strategies. Future work will expand the model’s generalizability to different thruster types, propellants, and magnetic field configurations.

GTP-24-1666 - Temperature Margin Calculation During Finite Element Creep Simulations (GT2024-121776)

Abstract Creep deformation in turbomachinery applications is a highly non-linear phenomenon where +/- 15°C may halve or double the average rupture life. Even in well-controlled laboratory conditions, this stochastic nature means that identical tests may differ by a factor of two. Analysts using implicit finite element user creep subroutines may not appreciate the uncertainty of their solution, or the design changes required to account for uncertainty in either the boundary conditions or the solution itself. Designing to a maximum strain limit does not consider the sudden acceleration of tertiary creep rates. Applying a time factor may extend the simulation time by multiple factors. This paper presents a methodology where two estimations of creep damage are made in parallel to the creep simulation at actual operating temperatures. These two estimations consider the predicted change in stress relaxation, at higher or lower temperatures, to estimate the temperature change required to reach the onset of creep, at any given node in the model, and at any given time in the analysis. This margin calculation is expanded to consider the uncertainty in creep prediction. A time-based scatter factor is incorporated into the temperature margin calculation to provide a minimum temperature margin that includes model uncertainty. The analyst can also consider the uncertainty of the temperature prediction and boundary conditions to produce a robust creep prediction in a real-world simulation. The methodology is validated through the FEA of example cases and applied to a creep-limited 2nd stage turbine blade.

Accelerating Optimal Synthesis of Atomically Thin MoS2: A Constrained Bayesian Optimization Guided Brachistochrone Approach

AbstractA machine learning (ML) guided approach is presented for the accelerated optimization of chemical vapor deposition (CVD) synthesis of 2D materials toward the highest quality, starting from low‐quality or unsuccessful synthesis conditions. Using 26 sets of these synthesis conditions as the initial training dataset, our method systematically guides experimental synthesis towards optoelectronic‐grade monolayer MoS2 flakes. A‐exciton linewidth (σA) as narrow as 38 meV could be achieved in 2D MoS2 flakes after only an additional 35 trials (reflecting 15% of the full factorial design dataset for training purposes). In practical terms, this reflects a decrease of the possible experimental time to optimize the parameters from up to one year to about two months. This remarkable efficiency was achieved by formulating a constrained sequencing optimization problem solved via a combination of constraint learning and Bayesian Optimization with the narrowness of σA as the single target metric. By employing graph‐based semi‐supervised learning with data acquired through a multi‐criteria sampling method, the constraint model effectively delineates and refines the feasible design space for monolayer flake production. Additionally, the Gaussian Process regression effectively captures the relationships between synthesis parameters and outcomes, offering high predictive capability along with a measure of prediction uncertainty. This method is scalable to a higher number of synthesis parameters and target metrics and is transferrable to other materials and types of reactors. This study envisions that this method will be fundamental for CVD and similar techniques in the future.

Dynamics of standing dead wood and severe fire in north Australian savannas: implications for carbon management

Background Many fires in north Australian savannas are severe enough to cause canopy scorch, tree death and removal of stags. Better fire management may increase carbon sequestration in trees, perhaps including stags. Aims To describe and analyse dynamics of stags in tropical savannas (600–1000 mm annual rainfall) in relation to fire and better understand their role in biomass sequestration. Methods We monitored marked populations of live and dead trees over 12 years. Statistical models describing influences on stag creation and loss are applied in stag dynamics simulations. Key results Immediately following severe fire, stag biomass increases acutely because many more live trees are killed than stags removed. Between severe fires, stag losses exceed tree deaths, so peaks are quite short. Many ‘new’ stags are lost (fallen or consumed) quickly. Conclusions Between fires, stags comprise ~7.5–8.9% of standing above-ground biomass, more under dry conditions and during recovery from severe fire or other drivers of increased tree mortality. Fire management is unlikely to increase proportions of total woody biomass in stags unless it also reduces live biomass. Implications Reducing frequency of severe fires can increase total carbon sequestration in dry tropical savannas. Prediction uncertainties and management risks around sequestration present daunting challenges for policy-makers and fire management practitioners.

Monitoring of volcanic precursors using satellite data: the case of Taftan volcano in Iran

Abstract In recent weeks, there have been reports of gas emanations from the crater of the dormant Taftan volcano in Iran. In this study, due to the limitations of ground stations and the advantages of satellite remote sensing, it has been tried to detect possible anomalies using the plasma data measured by Swarm (A, B and C) and GPS (Global Positioning System) satellites around the location of the mentioned volcano. Also, lithospheric and atmospheric data including OLR (Outgoing Longwave Radiation), water vapor, ozone, relative humidity, surface and air temperature, AOD (Aerosol Optical Depth), sulfur dioxide (SO2) and nitrogen dioxide (NO2) using the Giovanni website in a period of about 5 months, were downloaded and analyzed. Using the median and interquartile method, possible anomalies were detected in the pre-processed time series of the desired parameters. To justify some of the non-volcanic anomalies, synoptic data including precipitation and temperature were prepared from the nearest ground station. By rejecting the possibility that some detected anomalies are related to volcanic activities, hypotheses were presented for other proposed anomalies. As a result of this research, the capabilities of Swarm satellites and GPS-TEC (Total Electron Content) are emphasized in studies related to the prediction, detection and tracking of volcanic activities and it is shown that by comparative comparison with other lithospheric and atmospheric precursors, uncertainty in eruption prediction can be reduced.

Remaining useful life prediction method for jointless track circuits based on multivariate feature fusion and nonlinear Wiener process

Abstract Predicting the Remaining Useful Life (RUL) of track circuits is essential to ensure the safe and reliable operation of high-speed railways. In response to the challenges faced by current machine-learning-based RUL prediction methods, which struggle to represent the uncertainty in the probability distribution of RUL predictions, this paper suggests a hybrid-driven method for estimating remaining life. Firstly, the track circuit Health Index (HI) is constructed by feature dimensionality reduction and fusion of the original multivariate monitoring data through Kernel Principal Component Analysis (KPCA) and Autoencoder (AE); Secondly, the degraded state of the rail circuit is modelled using a nonlinear Wiener degradation model. Finally, the principle of First Hitting Time (FHT) is used to derive the Probability Density Function (PDF) of the anticipated RUL. The efficacy and superiority of the approach presented in this paper are validated by experimental research on the track circuit monitoring dataset. The method enhances forecast accuracy and reduces prediction uncertainty, offering robust technical support for track circuit maintenance decision-making.

A hierarchical multistage holistic model for acoustic emission source monitoring in composites

Abstract This paper introduces a multistage smart structural health monitoring (SHM) model for carbon-fiber composites, with a focus on multiple types of acoustic emission (AE) source localization and classification. The SHM model uses time-frequency data from various AE events (such as tool drops, impact, and artificial debonding) across different zones of a composite structure. The SHM strategy demonstrates a robust smart monitoring of composites with high accuracy. Further, a hypothesis testing has been carried out that supports the superiority of a 2-stage identification process, revealing statistically significant higher accuracy and confidence intervals across all zones and AE source types. This research establishes a novel framework for solving a hierarchical multistage holistic damage source identification problem, offering robustness in identifying various damage scenarios and quantifying associated prediction uncertainties.&amp;#xD;