A Probabilistic Forecast for Multi-year ENSO Using Bayesian Convolutional Neural Network

Concrete cracks significantly impact structural durability by allowing harmful elements such as carbon dioxide and chloride ions to penetrate. Thus, the cracks should be prevented and treated to maintain the lifespan of concrete structures. Convolutional neural network (CNN) has been widely applied in crack detection. This study investigated the use of a Bayesian convolutional neural network (BCNN) for concrete crack classification and compared the performance of BCNN with UNet, a deep CNN-based model, and CNN. BCNN weight uncertainty through a Bayesian method, improved prediction reliability and prevented overfitting. BCNN demonstrated superior expected calibration error (0.0030) and Brier score (0.0025) compared to CNN, highlighting its robustness in prediction reliability. Furthermore, Bayesian entropy analysis confirmed the ability of BCNN to quantify uncertainty, leading to more precise crack classification. These results imply that BCNN provides a balance between computational efficiency and performance, offering strong potential for smart construction applications. Validation results indicated that BCNN achieved a validation accuracy of 92.22%, higher than CNN (85.78%) and slightly lower than UNet (94.34%). However, BCNN demonstrated a superior intersection over union (IoU) of 0.8049, outperforming CNN (0.6685) and approaching UNet (0.9135). These results implied that BCNN provides a balance between computational efficiency and performance, offering strong potential for smart construction applications.

<div class="section abstract"><div class="htmlview paragraph">Self-piercing riveting (SPR) is a key joining method in multi/thin-material automotive structures, yet accurately predicting the mechanical strength of SPR joints remains challenging due to numerous influencing factors. Empirical engineering equations [<span class="xref">1</span>] provide a foundation for estimating lap-shear and cross-tension strength but require several geometric parameters that are often unavailable in the design phase.</div><div class="htmlview paragraph">To address this limitation, we extract and leverage the core physical relationships embedded in these formulas. By reformulating the dependence of joint strength on the yield strength and total thickness of the sheet stack as practical regression models, we enable strength prediction using only commonly available material properties.</div><div class="htmlview paragraph">Furthermore, a Bayesian convolutional neural network (BCNN) model is developed to incorporate additional material features, offering improved prediction accuracy and uncertainty quantification.</div></div>

Probabilistic SAR-based water segmentation with adapted Bayesian convolutional neural network

This study provides the first monthly resolved, 41‐year record of geochemical variations (δ18O and Sr/Ca) in a fast‐growing Diploria strigosa brain coral from Guadeloupe, Caribbean Sea. Linear regression yields a significant correlation of coral Sr/Ca (δ18O) with instrumental sea surface temperature (SST) on both monthly and mean annual scales (e.g., r = −0.59 for correlation between Simple Ocean Data Assimilation (SODA) SST and Sr/Ca, and r = −0.66 for δ18O; mean annual scale, p < 0.0001). The generated coral Sr/Ca (δ18O)‐SST calibration equations are consistent with each other and with published equations using other coral species from different regions. Moreover, a high correlation of coral Sr/Ca and δ18O with local air temperature on a mean annual scale (r = −0.78 for Sr/Ca; r = −0.73 for δ18O; p < 0.0001) demonstrates the applicability of geochemical proxies measured from Diploria strigosa corals as reliable recorders for interannual temperature variability. Both coral proxies are highly correlated with annual and seasonal mean time series of major SST indices in the northern tropical Atlantic (e.g., r = −0.71 for correlation between the index of North Tropical Atlantic SST anomaly and Sr/Ca, and r = −0.70 for δ18O; mean annual scale, p < 0.001). Furthermore, the coral proxies capture the impact of the El Niño–Southern Oscillation on the northern tropical Atlantic during boreal spring. Thus fast‐growing Diploria strigosa corals are a promising new archive for the Atlantic Ocean.

Fault diagnosis is a vital technique to pinpoint the machine malfunctions in manufacturing systems. In recent years, the deep learning techniques greatly improve the fault detection accuracy, but there still remain some problems. If one fault is absent in the training data or the fault signal is disturbed by severe noise interference, the fault classifier may misjudge the health state. This problem limits the reliability of the fault diagnosis in real applications. In this paper, we enhance the fault diagnosis method by using Bayesian Convolutional Neural Network (BCNN). A Shannon entropy-based method is presented to quantify the prediction uncertainty. The BCNN turns the deterministic predictions to probabilistic distributions and enhances the robustness of the fault diagnosis. The uncertainty quantification method helps to indicate the wrong predictions, detect unknown faults, and discover the strong disturbances. Then, a fine-tuning strategy is applied to enhance the model performance further. The potential usability of the proposed method in monitoring the motors of 3D printers is studied. And the experiment is conducted on a motor bearing dataset provided by Case Western Reserve University. The proposed BCNN achieves 99.82% fault classification accuracy over nine health conditions. Its robustness is verified by comparing the testing accuracy with three other methods on the noisy datasets. And the uncertainty quantification method successfully detects the outlier inputs.

In this study, we compare the simulation of El Nino and the Southern Oscillation (ENSO) in the historical integrations of 17 Coupled Model Intercomparison Project 5 (CMIP5) models with corresponding observations. The mean state and ENSO variations are analyzed in both the atmosphere and ocean and it is found that most of the CMIP5 models exhibit cold (warm) biases in the equatorial (subtropical eastern) Pacific Ocean sea surface temperature that are reminiscent of the split intertropical convergence zone phenomenon found in previous studies. There is, however, a major improvement in the representation of the power spectrum of the Nino3.4 sea surface temperature variations, which shows that, as in the observations, a majority of the models display a spectral peak in the 2–7 year range, have a near-linear relationship with the displacement of the equatorial thermocline and exhibit a robust atmospheric response to ENSO variations. Several issues remain such as erroneous amplitudes in the Nino3.4 sea surface temperature spectrum’s peak and a width of the spectral peak that is either too broad or too narrow. It is also seen that most CMIP5 models unlike the observations extend the ENSO variations in the equatorial Pacific too far westward beyond the dateline and there is very little asymmetry in event duration between the warm and cold phases. ENSO variability forces a dominant mode of rainfall variability in the southeastern United States, especially in the boreal winter season. The CMIP5 exhibited a wide range of response in this metric with several displaying weak to nonexistent, some showing relatively strong, and one indicating excessively zonally symmetric teleconnection over the southeastern United States.

Recent studies have been shown the vulnerability of Deep neural networks (DNNs) by a small adversarial perturbation in images that humans cannot distinguish and a well-trained neural network misclassifies. There are many defense methods, e.g. randomness, ensemble, and adversarial training that can improve the robustness of neural networks with respect to adversaries. Among them, adversarial training has been being the most outstanding defense. However, the training cost becomes a big challenge to adversarial learning. In this paper, we propose a new defense algorithm called Bayes without Bayesian Learning, which does not add the training phase in resisting adversarial attacks. Our method has based on the stochastic components of Bayesian Convolutional Neural Network (BCNN) to prevent the forceful gradient based attacks and generate the ensemble model to enhance the performance of models. In order to avoid adversarial training, we utilize pretrained CNN models on both natural and perturbed data. Therefore, this approach can be applied to any model with learned parameters to accelerate the computation of the defensive method. Our defensive algorithm has significantly improved the accuracy of pretrained CNN models under the high level of attacks. For instance, on ImageNet with Resnet50 network, we achieve 57% prediction accuracy in the top5, under strong PGD attack. We also confirm 3% improvement when combining our method with adversarial training. Experimental results prove the efficiency of our BCNN model that does not need Bayesian learning but still resists adversarial perturbation on both small and large scale datasets as CIFAR10, CIFAR100, and ImageNet.

Systematic and compensating errors can lead to degraded predictive skill in climate models. Such errors may be identified by comparing different models in an analysis of individual physical processes. We examine model simulations of El Nino–Southern Oscillation (ENSO) in five Coupled Model Intercomparison Project (CMIP) models, using transfer functions to analyze nine processes critical to ENSO's dynamics. The input and output of these processes are identified and analyzed, some of which are motivated by the recharge oscillator theory. Several errors and compensating errors are identified. The east-west slope of the equatorial thermocline is found to respond to the central equatorial Pacific zonal wind stress as a damped driven harmonic oscillator in all models. This result is shown to be inconsistent with two different formulations of the recharge oscillator. East Pacific sea surface temperature (SST) responds consistently to changes in the thermocline depth in the eastern Pacific in the five CMIP models examined here. However, at time scales greater than 2 years, this consistent model response disagrees with observations, showing that the SST leads thermocline depth at long time scales. Compensating errors are present in the response of meridional transport of water away from the equator to SST: two different models show different response of the transport to off-equatorial wind curl and wind curl response to East Pacific SST. However, these two models show the same response of meridional transport to East Pacific SST. Identification of errors in specific physical processes can hopefully lead to model improvement by focusing model development efforts on these processes.

Tridacnine oxygen isotope thermometers (TOITs) commonly utilize gridded sea surface temperature (SST) and sea surface salinity (SSS) data without carefully checking whether they were appropriate substitutes for in situ monitoring records at the tridacnine growth site. Many SST and SSS datasets are available, but those best suited for establishing TOITs have yet to be examined sufficiently. In this study, we have established TOITs for shells collected at two islands (Okinoerabu Jima in the north and Ishigaki Jima in the south) in the northwestern Pacific using six combinations of SST (Extended Reconstructed Sea Surface Temperature [ERSST], Sea Surface Temperature data of Integrated Global Ocean Services System [IGOSS SST], and Hadley Centre Sea Ice and Sea Surface Temperature dataset [HadISST]) and SSS (Simple Ocean Data Assimilation ocean/sea ice reanalysis [SODA3.15.2] and Met Office Hadley Centre “EN” series [EN4.2.2]) datasets and evaluated the differences among the TOITs. Consequently, the slopes and intercepts of all TOITs fell within or near the range of TOITs reported in the previous studies, regardless of the combination of datasets. However, a significant difference in the coldest month SST at Okinoerabu Jima (~ 1.0 °C) exists between IGOSS SST and ERSST, caused by that an SST drop by the southward migration of the Kuroshio Front during winter is detected not in IGOSS SST (1° gridded data) but in ERSST (2° gridded data). Thus, the choice of SST dataset can lead to up to 1.0 °C difference in reconstructed winter SST. SSS differences exist between SODA SSS and EN SSS around the two islands up to 0.8 (equivalent to ~ 0.25 ‰ in oxygen isotope composition of seawater), practically leading to insignificant variations of the slopes and intercepts of the TOITs. To establish TOITs using the gridded SST and SSS datasets, we should select datasets with SSTs and SSSs similar to those at the growth sites of the studied tridacnine shells. Even if a gridded SST (SSS) dataset and an in situ monitoring SST (SSS) record show good correspondence, it could be a coincidence. We propose an averaged Ryukyu TOIT: SST = − 3.95 ± 1.02 × (δ 18 O shell [VPDB] − δ 18 O sw [VSMOW]) + 20.50 ± 2.75 (2σ).

High-resolution marine data and transient simulations support orbital forcing of ENSO amplitude since the mid-Holocene

Remaining useful life (RUL) is one of the essential ingredients in the battery management system. However, due to the characteristic of the dynamic and time-varying electrochemical system with nonlinear and complicated internal mechanisms, the uncertainty of RUL estimation has been expanded, and it is difficult to give an accurate time to reach the end of life. This article proposes the Bayesian mixture neural network (BMNN), a probabilistic deep learning method, to obtain more accurate RUL prediction and provide uncertainty estimation, while the quasi-Gramian angular field (Q-GAF) beneficial to identify prior distribution is utilized to transform time-series sequence into temporal images. BMNN consists of the Bayesian convolutional neural network (BCNN) extracting features in temporal images and Bayesian long short-term memory (B-LSTM) learning correlation between retention capacity and other degradation inducements. After concatenating two terms, the variational Bayesian neural network outputs the distribution of prediction results. In the experimental stage, the performance of the proposed method is validated on four different lithium-ion battery datasets and demonstrates higher stability, lower uncertainty, and more accuracy than other methods.

Abstract. Systematic, routine, and comprehensive evaluation of Earth system models (ESMs) facilitates benchmarking improvement across model generations and identifying the strengths and weaknesses of different model configurations. By gauging the consistency between models and observations, this endeavor is becoming increasingly necessary to objectively synthesize the thousands of simulations contributed to the Coupled Model Intercomparison Project (CMIP) to date. The Program for Climate Model Diagnosis and Intercomparison (PCMDI) Metrics Package (PMP) is an open-source Python software package that provides quick-look objective comparisons of ESMs with one another and with observations. The comparisons include metrics of large- to global-scale climatologies, tropical inter-annual and intra-seasonal variability modes such as the El Niño–Southern Oscillation (ENSO) and Madden–Julian Oscillation (MJO), extratropical modes of variability, regional monsoons, cloud radiative feedbacks, and high-frequency characteristics of simulated precipitation, including its extremes. The PMP comparison results are produced using all model simulations contributed to CMIP6 and earlier CMIP phases. An important objective of the PMP is to document the performance of ESMs participating in the recent phases of CMIP, together with providing version-controlled information for all datasets, software packages, and analysis codes being used in the evaluation process. Among other purposes, this also enables modeling groups to assess performance changes during the ESM development cycle in the context of the error distribution of the multi-model ensemble. Quantitative model evaluation provided by the PMP can assist modelers in their development priorities. In this paper, we provide an overview of the PMP, including its latest capabilities, and discuss its future direction.

The Western Pacific Warm Pool (WPWP) large warm water mass contributes to atmospheric heat and moisture, modulating El Niño-Southern Oscillation (ENSO) phase changes, potentially leading to global climate extremes. Researching WPWP climate variability will improve understanding of ENSO and the behavior of longer timescale climate drivers. However, the lack of long, continuous and high-resolution instrumental climate observations within the WPWP limits understanding of its past climate changes. Massive corals from the WPWP allows past climate reconstruction characterizing the warm pool spatiotemporal and hydrology changes beyond instrumental timescales. A better understanding of these processes and their underlying climate drivers allows for better evaluation of regional trends, climate connections to global regions, and climate predictions in response to global warming. Porites spp. coral cores sampled from Kosrae island (KOS) and Woleai atoll (WOL), located on separate sites of Federated States of Micronesia, capture the opposite dipole signal of ENSO. Each coral core was slabbed and x-rayed to reveal images of annual density banding to assist in determining sampling axis and age assignment. The slabbed coral cores were analyzed for δ18Oc and Sr/Ca proxies for sea surface temperature (SST) and salinity (SSS), respectively. Monthly resolved reconstructed SST and SSS spanning 165 and 277 years were generated from KOS and WOL, respectively. The Sr/Ca-based reconstructed SST from KOS and WOL display the expected ENSO dipole signal. The difference between these records generates the zonal SST gradient, KOSWOLSST, capturing the warm pool's east-west zonal migration during ENSO events. The extended KOSWOLSST century exhibit ENSO’s spatiotemporal SST influence on the WPWP dating back to the mid-19th century. At decadal timescales, separate KOS SST displays consistent variability to the IPO. However, WOL SST exhibit strong influence driven by the Northern Hemisphere SST. Extending the WOL SST record further ~110 years allow examination of centennial-scale SST variability. Significant power and centennial relationship between extended WOL SST to a South china sea paleo SST record indicates simulated Pacific Centennial Oscillation (PCO) observed in controlled coupled global climate simulation (cGCMs) exists in nature. These findings collectively establish a centennial-scale natural SST forcing in the WPWP that potentially impacts global teleconnections.The δ18Oc-based reconstructed KOS and WOL SSS reflect each site’s precipitation driven changes. The averaged SSS records generate KOSWOLSSS, an index representative of the regional WPWP precipitation. Separate KOS and WOL SSS records dating back to the mid-19th century indicates WPWP experienced increasing precipitation changes leading to freshening trends driven by global warming. Extending WOL SSS records further ~110 years allows examination of WPWP centennial driven SST variability (PCO) influence on WOL SSS. Significant WOL SSS and SST relationship indicate during SST increases saw a decline in precipitation. This and antiphase centennial SST relationships between Micronesia and Vietnam suggest a robust precipitation footprint of the PCO. This thesis derived a WPWP zonal SST gradient, regional hydrological records, established WPWP climate SST and SSS climate drivers and verified the presence of PCO and its precipitation footprint based on coral proxy records. These key findings underscore the need to investigate natural climate driver(s) variability within the WPWP. Furthermore, incorporating the PCO variability and its precipitation footprint in cGCMs is crucial to modelling future climate as climate changes within this region impact global teleconnections and tropical Pacific regional hydrology.

A reliable Bayesian neural network for the prediction of reservoir thickness with quantified uncertainty

Interannual to Decadal Rainfall Variability over Greater Horn of Africa: Non-Stationary Teleconnections, Physical Mechanisms, and Predictability