Ensemble Of Parameters Research Articles

Constraining Clouds and Convective Parameterizations in a Climate Model Using Paleoclimate Data

AbstractCloud and convective parameterizations strongly influence uncertainties in equilibrium climate sensitivity. We provide a proof‐of‐concept study to constrain these parameterizations in a perturbed parameter ensemble of the atmosphere‐only version of the Goddard Institute for Space Studies Model E2.1 simulations by evaluating model biases in the present‐day runs using multiple satellite climatologies and by comparing simulated δ18O of precipitation (δ18Op), known to be sensitive to parameterization schemes, with a global database of speleothem δ18O records covering the Last Glacial Maximum (LGM), mid‐Holocene (MH) and pre‐industrial (PI) periods. Relative to modern interannual variability, paleoclimate simulations show greater sensitivity to parameter changes, allowing for an evaluation of model uncertainties over a broader range of climate forcing and the identification of parts of the world that are parameter sensitive. Certain simulations reproduced absolute δ18Op values across all time periods, along with LGM and MH δ18Op anomalies relative to the PI, better than the default parameterization. No single set of parameterizations worked well in all climate states, likely due to the non‐stationarity of cloud feedbacks under varying boundary conditions. Future work that involves varying multiple parameter sets simultaneously with coupled ocean feedbacks will likely provide improved constraints on cloud and convective parameterizations.

Comparative Study on the Synthesis of Path-Generating Four-Bar Linkages Using Metaheuristic Optimization Algorithms

Four-bar linkages are one of the most widely used mechanisms in industries. This paper presents a comparative study on the accuracy and efficiency of the optimum synthesis of path-generating four-bar linkages using five metaheuristic optimization algorithms. The utilized metaheuristic methods included two swarm intelligence-based algorithms, i.e., particle swarm optimization and hybrid particle swarm optimization, and three evolutionary-based algorithms, i.e., differential evolution, ensemble of parameters and mutation strategies in differential evolution, and linearly ensemble of parameters and mutation strategies in differential evolution. The objective function to be minimized is the sum of squares of the distance between the generated points and the precision points of a coupler point. The optimal design of four-bar linkages must meet the Grashof’s criteria and exhibit sequential constraints that can prevent the occurrence of order defect. This study investigated five representative cases of the dimensional synthesis of four-bar path generators with and without prescribed timing and compared the optimal solutions of the utilized five metaheuristic methods to those of previously reported algorithms in literature. The improved metaheuristic methods exhibited superior optimal solution and enhanced reliability compared to the original methods. Moreover, three improved metaheuristic methods were not only easy implemented, but also more efficient for solving the optimal synthesis problems, particularly for high dimensional problems.

Toward Efficient Calibration of Higher‐Resolution Earth System Models

AbstractProjections of future climate change to support decision‐making require Earth system models (ESMs) running at high spatial resolution, but at present this is computationally prohibitive. A major challenge is the calibration (parameter tuning) during the development of ESMs, which requires running large numbers of simulations to identify optimal values for parameters that are poorly constrained by observations. Here, we train a convolutional neural network (CNN) to emulate perturbed parameter ensembles from two lower‐resolution (and thus much less expensive) versions of the same ESM, and a smaller number of higher‐resolution simulations. Cross‐validated results show that the CNN's skill exceeds that of a climatological baseline for most variables with as few as 5–10 examples of the higher‐resolution ESM, and for all variables (including precipitation) with at least 20 examples. This proof‐of‐concept study demonstrates a machine learning based approach that makes the process of constructing a higher‐resolution emulator 20%–40% more computationally efficient, and thus offers the prospect of significantly more efficient calibration of ESMs.

Optimizing process-based models to predict current and future soil organic carbon stocks at high-resolution

From hillslope to small catchment scales (< 50 km2), soil carbon management and mitigation policies rely on estimates and projections of soil organic carbon (SOC) stocks. Here we apply a process-based modeling approach that parameterizes the MIcrobial-MIneral Carbon Stabilization (MIMICS) model with SOC measurements and remotely sensed environmental data from the Reynolds Creek Experimental Watershed in SW Idaho, USA. Calibrating model parameters reduced error between simulated and observed SOC stocks by 25%, relative to the initial parameter estimates and better captured local gradients in climate and productivity. The calibrated parameter ensemble was used to produce spatially continuous, high-resolution (10 m2) estimates of stocks and associated uncertainties of litter, microbial biomass, particulate, and protected SOC pools across the complex landscape. Subsequent projections of SOC response to idealized environmental disturbances illustrate the spatial complexity of potential SOC vulnerabilities across the watershed. Parametric uncertainty generated physicochemically protected soil C stocks that varied by a mean factor of 4.4 × across individual locations in the watershed and a − 14.9 to + 20.4% range in potential SOC stock response to idealized disturbances, illustrating the need for additional measurements of soil carbon fractions and their turnover time to improve confidence in the MIMICS simulations of SOC dynamics.

Guest Editorial: Situational awareness of integrated energy systems

Guest Editorial: Situational awareness of integrated energy systems

Future projections of daily haze-conducive and clear weather conditions over the North China Plain using a perturbed parameter ensemble

Abstract. We examine past and future changes in both winter haze and clear weather conditions over the North China Plain (NCP) using a perturbed parameter ensemble (PPE) and elucidate the influence of model physical parameterizations on these future projections for the first time. We use a large-scale meteorology-based haze weather index (HWI) with values &gt;1 as a proxy for haze-conducive weather and HWI &lt;-1 for clear weather conditions over the NCP. The PPE generated using the UK Met Office's HadGEM-GC3 model shows that under a high-emission (RCP8.5) scenario, the frequency of haze-conducive weather (HWI &gt;1) is likely to increase whereas the frequency of clear weather (HWI &lt;-1) is likely to decrease in the future with a growing influence of climate change over the 21st century. Nevertheless, a reduction in the frequency of haze-conducive weather and increment in the frequency of clear weather, though less likely, is also possible. In the future, the frequency of haze-conducive weather for a given winter could be as much as ∼3.5 times higher than the frequency of clear weather over the NCP. More frequent haze-conducive weather (HWI &gt;1) during winter over the NCP is found to be associated with an enhanced warming of the troposphere and weaker northwesterlies in the mid-troposphere over the NCP. We also examined the changes in the interannual variability of the haze-conducive and clear weather and found no marked changes in the variability during future periods. We find a clear influence of model physical parametrizations on climatological mean frequencies for both haze-conducive and clear weather. For the mid- to late 21st century (2033–2086), the parametric effect can explain up to ∼80 % of the variance in the climatological mean frequencies of PPE members. This shows that different model physical parameterizations lead to a different evolution of the model's mean climate, particularly towards the end of the 21st century. Therefore, it is desirable to consider the PPE in addition to the initialized and multimodel ensembles to obtain a more comprehensive range of plausible future projections.

Neutron-capture elements record the ordered chemical evolution of the disc over time

ABSTRACT An ensemble of chemical abundances probing different nucleosynthetic channels can be leveraged to build a comprehensive understanding of the chemical and structural evolution of the Galaxy. Using GALAH DR3 data, we seek to trace the enrichment by the supernovae Ia, supernovae II, asymptotic giant branch stars, and neutron-star mergers and/or collapsars nucleosynthetic sources by studying the [Fe/H], [α/Fe], [Ba/Fe], and [Eu/Fe] chemical compositions of ∼50 000 red giant stars, respectively. Employing small [Fe/H]–[α/Fe] cells, which serve as an effective reference-frame of supernovae contributions, we characterize the abundance-age profiles for [Ba/Fe] and [Eu/Fe]. Our results disclose that these age–abundance relations vary across the [Fe/H]–[α/Fe] plane. Within cells, we find negative age–[Ba/Fe] relations and flat age–[Eu/Fe] relations. Across cells, we see the slope of the age–[Ba/Fe] relations evolve smoothly and the [Eu/Fe] relations vary in amplitude. We subsequently model our empirical findings in a theoretical setting using the flexible Chempy Galactic chemical evolution (GCE) code, using the mean [Fe/H], [Mg/Fe], [Ba/Fe], and age values for stellar populations binned in [Fe/H], [Mg/Fe], and age space. We find that within a one-zone framework, an ensemble of GCE model parameters vary to explain the data. Using present day orbits from Gaia EDR3 measurements we infer that the GCE model parameters, which set the observed chemical abundance distributions, vary systematically across mean orbital radii. Under our modelling assumptions, the observed chemical abundances are consistent with a small gradient in the high-mass end of the initial mass function (IMF) across the disc, where the IMF is more top heavy towards the inner disc and more bottom heavy in the outer disc.

Winding number statistics of a parametric chiral unitary random matrix ensemble* *Dedicated to the Memory of Fritz Haake.

The winding number is a concept in complex analysis which has, in the presence of chiral symmetry, a physics interpretation as the topological index belonging to gapped phases of fermions. We study statistical properties of this topological quantity. To this end, we set up a random matrix model for a chiral unitary system with a parametric dependence. We analytically calculate the discrete probability distribution of the winding numbers, as well as the parametric correlations functions of the winding number density. Moreover, we address aspects of universality for the two-point function of the winding number density by identifying a proper unfolding procedure. We conjecture the unfolded two-point function to be universal.

Investigating Parametric Dependence of Climate Feedbacks in the Atmospheric Component of CNRM‐CM6‐1

AbstractWe sample 30 calibration parameters of the CNRM‐CM6‐1 atmospheric component to produce a perturbed parameter ensemble. An error score aggregating four annual‐averaged metrics is used here as an example to represent the model performance. We propose a method using statistical emulators of climatic fields and feedback parameters to produce model candidates spanning the maximal range of net feedback while minimizing the error score. Optimal candidates are found for a range of climate sensitivity chosen to span 3.8–10.1 K. The candidates show large errors in the extremes of the climate sensitivity range and none of them have better scores than the released version of CNRM‐CM6. However, 65% of the optimal candidates have errors within the inter‐model standard deviation of the seven models participating in the Cloud Feedback Model Intercomparison Project, spanning and ECS range of (4.14–7.22 K). These optimal calibrations will be used to assess feedback diversity in coupled experiments.

Hammerstein-Wiener nonlinear system identification by using honey badger algorithm hybridized Sage-Husa adaptive Kalman filter with real-time applications

This paper introduces a hybrid Sage-Husa adaptive Kalman filter (SHAKF) with a metaheuristic algorithm to estimate the parameters of the Hammerstein-Wiener (H-W) model. The initial state parameters of the SHAKF must be appropriately tuned based on the estimation problem; otherwise, it would lead to divergence. To get the global optimal tuned SHAKF parameters for modelling the H-W system, this proposed work has formulated an objective function based on the identification problem optimised with a recently introduced metaheuristic technique called the Honey Badger algorithm (HBA). Finally, the unknown parameters of the H-W system are identified by using the basic SHAKF technique by considering the achieved optimised SHAKF parameters as the initial state inputs. The efficiency of the proposed technique is verified on numerical problems, and its practical feasibility is tested on mechanical systems, namely arm robots and ball-and-beam system identification based on real data sets. The experimental results warrant that the proposed technique possesses high robustness to additive noise, quick convergence speed and low steady-state error compared to the other recently reported methods. For comparison study, different metaheuristic algorithms, namely, comprehensive learning particle swarm optimiser (CLPSO), an ensemble of mutation strategies and control parameters with the differential evolution (EPSDE) and colliding bodies optimisation (CBO), are incorporated into SHAKF to estimate the H-W systems and the obtained results for various standard metrics show the superiority of the proposed HBA-aided SHAKF method over others.

Deep Learning for Latent Space Data Assimilation in Subsurface Flow Systems

Summary We present a new deep learning architecture for efficient reduced-order implementation of ensemble data assimilation in learned low-dimensional latent spaces. Specifically, deep learning is used to improve two important aspects of data assimilation workflows: (i) low-rank representation of complex reservoir property distributions for geologically consistent feature-based model updating, and (ii) efficient prediction of the statistical information that are required for model updating. The proposed method uses deep convolutional autoencoders (AEs) to nonlinearly map the original complex and high-dimensional parameters onto a low-dimensional parameter latent space that compactly represents the original parameters. In addition, a low-dimensional data latent space is constructed to predict the observable response of each model parameter realization, which can serve as a proxy model in the latent space to compute the statistical information needed for data assimilation. The two mappings are developed as a joint deep learning architecture with two variational AEs (VAEs) that are connected and trained together. The training procedure uses an ensemble of model parameters and their corresponding production response predictions. Simultaneous training of the two mappings leads to a joint data-parameter manifold that captures the most salient information in the two spaces for effective data assimilation, where only relevant data and parameter features are included. Moreover, the parameter-to-data mapping provides a fast forecast model that can be used to significantly increase the ensemble size in data assimilation, without the corresponding computational overhead. We apply the developed approach to a series of numerical experiments, including a 3D example based on the Volve field in the North Sea. For data assimilation methods that involve iterative schemes, such as the ensemble smoother with multiple data assimilation (ESMDA) or iterative forms of the ensemble Kalman filter (EnKF), the proposed approach offers a computationally competitive alternative. Our results suggest that a fully low-dimensional implementation of ensemble data assimilation in effectively constructed latent spaces using deep learning architectures could offer several advantages over the standard algorithms, including joint data-parameter reduction that respects the salient features in each space, geologically consistent feature-based updates, as well as increased ensemble size to improve the accuracy and computational efficiency of calculating the required statistics for the update step.

Parameter Estimation Using Adaptive Observations Toward Maximum Total Variance Reduction With Ensemble Adjustment Kalman Filter

In real applications, one common issue of parameter estimation using ensemble-based data assimilation methods is the accumulation of sampling errors when a large number of observations are used to update single-value parameters. In this article, a new parameter estimation method which assimilates a large number of observations to estimate the states while assimilates adaptive observations to update the parameters is introduced. The observations resulting in maximum total variance reduction to the parameter ensembles are identified to perform parameter estimation. To validate this new method, the two-scale Lorenz-96 model is used to generate true states, while a parameterized one-scale Lorenz-96 model is used to perform state and parameter estimation experiments. The comparison between state estimation and parameter estimation with fixed or adaptive observations shows the new method can be more effective in estimating the model parameters and providing more accurate analyses. This method also shows its potential to be used in the data assimilation with large general circulation models to better produce reanalyzes.

An ensemble of parameters from a robust Markov-based model reproduces L-type calcium currents from different human cardiac myocytes.

The development of modeling structures at the channel level that can integrate subcellular and cell models and properly reproduce different experimental data is of utmost importance in cardiac electrophysiology. In contrast to gate-based models, Markov Chain models are well suited to promote the integration of the subcellular level of the cardiomyocyte to the whole cell. In this paper, we develop Markov Chain models for the L-type Calcium current that can reproduce the electrophysiology of two established human models for the ventricular and Purkinje cells. In addition, instead of presenting a single set of parameters, we present a collection of set of parameters employing Differential Evolution algorithms that can properly reproduce very different protocol data. We show the importance of using an ensemble of a set of parameter values to obtain proper results when considering a second protocol that suppresses calcium inactivation and mimics a pathological condition. We discuss how model discrepancy, data availability, and parameter identifiability can influence the choice of the size of the collection. In summary, we have modified two cardiac models by proposing new Markov Chain models for the L-type Calcium. We keep the original whole-cell dynamics by reproducing the same characteristic action potential and calcium dynamics, whereas the Markov chain-based description of the L-type Calcium channels allows novel small spatial scale simulations of subcellular processes. Finally, the use of collections of parameters was crucial for addressing model discrepancy, identifiability issues, and avoiding fitting parameters overly precisely, i.e., overfitting.

Statistical Analysis and Kinematic Assessment of Upper Limb Reaching Task in Parkinson's Disease.

The impact of neurodegenerative disorders is twofold; they affect both quality of life and healthcare expenditure. In the case of Parkinson’s disease, several strategies have been attempted to support the pharmacological treatment with rehabilitation protocols aimed at restoring motor function. In this scenario, the study of upper limb control mechanisms is particularly relevant due to the complexity of the joints involved in the movement of the arm. For these reasons, it is difficult to define proper indicators of the rehabilitation outcome. In this work, we propose a methodology to analyze and extract an ensemble of kinematic parameters from signals acquired during a complex upper limb reaching task. The methodology is tested in both healthy subjects and Parkinson’s disease patients (N = 12), and a statistical analysis is carried out to establish the value of the extracted kinematic features in distinguishing between the two groups under study. The parameters with the greatest number of significances across the submovements are duration, mean velocity, maximum velocity, maximum acceleration, and smoothness. Results allowed the identification of a subset of significant kinematic parameters that could serve as a proof-of-concept for a future definition of potential indicators of the rehabilitation outcome in Parkinson’s disease.

SCARF: a biomedical association rule finding webserver.

The analysis of enormous datasets with missing data entries is a standard task in biological and medical data processing. Large-scale, multi-institution clinical studies are the typical examples of such datasets. These sets make possible the search for multi-parametric relations since from the plenty of the data one is likely to find a satisfying number of subjects with the required parameter ensembles. Specifically, finding combinatorial biomarkers for some given condition also needs a very large dataset to analyze. For fast and automatic multi-parametric relation discovery association-rule finding tools are used for more than two decades in the data-mining community. Here we present the SCARF webserver for generalized association rule mining. Association rules are of the form: a AND b AND … AND x → y, meaning that the presence of properties a AND b AND … AND x implies property y; our algorithm finds generalized association rules, since it also finds logical disjunctions (i.e., ORs) at the left-hand side, allowing the discovery of more complex rules in a more compressed form in the database. This feature also helps reducing the typically very large result-tables of such studies, since allowing ORs in the left-hand side of a single rule could include dozens of classical rules. The capabilities of the SCARF algorithm were demonstrated in mining the Alzheimer’s database of the Coalition Against Major Diseases (CAMD) in our recent publication (Archives of Gerontology and Geriatrics Vol. 73, pp. 300–307, 2017). Here we describe the webserver implementation of the algorithm.

An Approach to Detect Phishing Websites with Features Selection Method and Ensemble Learning

Nowadays, phishing is a major problem on a global scale. Everyone must use the internet in today’s society in order to cope up in the real world. As a result, internet crime like phishing has become a serious issue throughout the world. This type of crime can be committed by anyone; all they need is a computer. Additionally, hacking may now be learned quickly by anyone with programming and mathematical skills. The adoption of various techniques by anti-phishing toolbars, such as machine learning, may enable users to quickly identify a fake website. As a result, researchers are now particularly interested in the problem of detecting fraudulent websites. Machine learning techniques have been offered throughout the entire process to more precisely identify fraudulent websites. To find the best accurate outcome, classification with random parameter tuning and ensemble based approaches are utilized. A user-friendly interface has also been suggested to make the system more accessible to the public.

Parameter uncertainty dominates C-cycle forecast errors over most of Brazil for the 21st century

Abstract. Identification of terrestrial carbon (C) sources and sinks is critical for understanding the Earth system as well as mitigating and adapting to climate change resulting from greenhouse gas emissions. Predicting whether a given location will act as a C source or sink using terrestrial ecosystem models (TEMs) is challenging due to net flux being the difference between far larger, spatially and temporally variable fluxes with large uncertainties. Uncertainty in projections of future dynamics, critical for policy evaluation, has been determined using multi-TEM intercomparisons, for various emissions scenarios. This approach quantifies structural and forcing errors. However, the role of parameter error within models has not been determined. TEMs typically have defined parameters for specific plant functional types generated from the literature. To ascertain the importance of parameter error in forecasts, we present a Bayesian analysis that uses data on historical and current C cycling for Brazil to parameterise five TEMs of varied complexity with a retrieval of model error covariance at 1∘ spatial resolution. After evaluation against data from 2001–2017, the parameterised models are simulated to 2100 under four climate change scenarios spanning the likely range of climate projections. Using multiple models, each with per pixel parameter ensembles, we partition forecast uncertainties. Parameter uncertainty dominates across most of Brazil when simulating future stock changes in biomass C and dead organic matter (DOM). Uncertainty of simulated biomass change is most strongly correlated with net primary productivity allocation to wood (NPPwood) and mean residence time of wood (MRTwood). Uncertainty of simulated DOM change is most strongly correlated with MRTsoil and NPPwood. Due to the coupling between these variables and C stock dynamics being bi-directional, we argue that using repeat estimates of woody biomass will provide a valuable constraint needed to refine predictions of the future carbon cycle. Finally, evaluation of our multi-model analysis shows that wood litter contributes substantially to fire emissions, necessitating a greater understanding of wood litter C cycling than is typically considered in large-scale TEMs.

Correspondence between Site Amplification and Topographical, Geological Parameters: Collation of Data from Swiss and Japanese Stations, and Neural Networks-Based Prediction of Local Response

ABSTRACTWe address the relation between seismic local amplification and topographical and geological indicators describing the site morphology. We focus on parameters that can be derived from layers of diffuse information (e.g., digital elevation models, geological maps) and do not require in situ surveys; we term these parameters as “indirect” proxies, as opposed to “direct” indicators (e.g., f0, VS30) derived from field measurements. We first compiled an extensive database of indirect parameters covering 142 and 637 instrumented sites in Switzerland and Japan, respectively; we collected topographical indicators at various spatial extents and focused on shared features in the geological descriptions of the two countries. We paired this proxy database with a companion dataset of site amplification factors at 10 frequencies within 0.5–20 Hz, empirically measured at the same Swiss and Japanese stations. We then assessed the robustness of the correlation between individual site-condition indicators and local response by means of statistical analyses; we also compared the proxy-site amplification relations at Swiss versus Japanese sites. Finally, we tested the prediction of site amplification by feeding ensembles of indirect parameters to a neural network (NN) structure. The main results are: (1) indirect indicators show higher correlation with site amplification in the low-frequency range (0.5–3.33 Hz); (2) topographical parameters primarily relate to local response not because of topographical amplification effects but because topographical features correspond to the properties of the subsurface, hence to stratigraphic amplification; (3) large-scale topographical indicators relate to low-frequency response, smaller-scale to higher-frequency response; (4) site amplification versus indirect proxy relations show a more marked regional variability when compared with direct indicators; and (5) the NN-based prediction of site response is the best achieved in the 1.67–5 Hz band, with both geological and topographical proxies provided as input; topographical indicators alone perform better than geological parameters.

Data-constrained assessment of ocean circulation changes since the middle Miocene in an Earth system model

Abstract. Since the middle Miocene (15 Ma, million years ago), the Earth's climate has undergone a long-term cooling trend, characterised by a reduction in ocean temperatures of up to 7–8 ∘C. The causes of this cooling are primarily thought to be due to tectonic plate movements driving changes in large-scale ocean circulation patterns, and hence heat redistribution, in conjunction with a drop in atmospheric greenhouse gas forcing (and attendant ice-sheet growth and feedback). In this study, we assess the potential to constrain the evolving patterns of global ocean circulation and cooling over the last 15 Ma by assimilating a variety of marine sediment proxy data in an Earth system model. We do this by first compiling surface and benthic ocean temperature and benthic carbon-13 (δ13C) data in a series of seven time slices spaced at approximately 2.5 Myr intervals. We then pair this with a corresponding series of tectonic and climate boundary condition reconstructions in the cGENIE (“muffin” release) Earth system model, including alternative possibilities for an open vs. closed Central American Seaway (CAS) from 10 Ma onwards. In the cGENIE model, we explore uncertainty in greenhouse gas forcing and the magnitude of North Pacific to North Atlantic salinity flux adjustment required in the model to create an Atlantic Meridional Overturning Circulation (AMOC) of a specific strength, via a series of 12 (one for each tectonic reconstruction) 2D parameter ensembles. Each ensemble member is then tested against the observed global temperature and benthic δ13C patterns. We identify that a relatively high CO2 equivalent forcing of 1120 ppm is required at 15 Ma in cGENIE to reproduce proxy temperature estimates in the model, noting that this CO2 forcing is dependent on the cGENIE model's climate sensitivity and that it incorporates the effects of all greenhouse gases. We find that reproducing the observed long-term cooling trend requires a progressively declining greenhouse gas forcing in the model. In parallel to this, the strength of the AMOC increases with time despite a reduction in the salinity of the surface North Atlantic over the cooling period, attributable to falling intensity of the hydrological cycle and to lowering polar temperatures, both caused by CO2-driven global cooling. We also find that a closed CAS from 10 Ma to present shows better agreement between benthic δ13C patterns and our particular series of model configurations and data. A final outcome of our analysis is a pronounced ca. 1.5 ‰ decline occurring in atmospheric (and ca. 1 ‰ ocean surface) δ13C that could be used to inform future δ13C-based proxy reconstructions.

Estimation of Seismic Moment Tensors Using Variational Inference Machine Learning

AbstractWe present an approach for rapidly estimating full moment tensors of earthquakes and their parameter uncertainties based on short time windows of recorded seismic waveform data by considering deep learning of Bayesian Neural Networks (BNNs). The individual neural networks are trained on synthetic seismic waveform data and corresponding known earthquake moment‐tensor parameters. A monitoring volume has been predefined to form a three‐dimensional grid of locations and to train a BNN for each grid point. Variational inference on several of these networks allows us to consider several sources of error and how they affect the estimated full moment‐tensor parameters and their uncertainties. In particular, we demonstrate how estimated parameter distributions are affected by uncertainties in the earthquake centroid location in space and time as well as in the assumed Earth structure model. We apply our approach as a proof of concept on seismic waveform recordings of aftershocks of the Ridgecrest 2019 earthquake with moment magnitudes ranging from Mw 2.7 to Mw 5.5. Overall, good agreement has been achieved between inferred parameter ensembles and independently estimated parameters using classical methods. Our developed approach is fast and robust, and therefore, suitable for down‐stream analyses that need rapid estimates of the source mechanism for a large number of earthquakes.