Atmospheric and Oceanic Sciences

Global food security is increasingly threatened by population growth, climate change, and declining wild fish stocks, positioning aquaculture as a critical solution to meet rising fish demand. Effective breeding in captive fish is hindered by reproductive dysfunctions, such as incomplete oocyte maturation and insufficient sperm production, due to absent natural environmental cues. Induced breeding through hormonal manipulation, targeting the hypothalamus-pituitary-gonad (HPG) axis with gonadotropin-releasing hormone analogs (GnRHa) and recombinant gonadotropins, has significantly improved spawning outcomes in species. Emerging technologies, including CRISPR-Cas9, artificial intelligence, and multiomics profiling, offer precision and sustainability in reproductive management. However, challenges persist, including species-specific responses, scalability, environmental risks from hormone effluents, and ethical concerns. This paper proposes a precision endocrinology framework integrating omics, automation practices to optimize reproductive efficiency and ensure sustainable aquaculture growth addressing food security while minimizing environmental impact.

This study evaluated the potential of VIS–NIR–SWIR leaf spectroscopy to predict N, P, and K contents in sugarcane, considering spectral variability across two growing seasons (2023/2024 and 2024/2025), phenological stages, and five varieties. Spectral signatures (350–2500 nm) were acquired using a FieldSpec 3 spectroradiometer and processed with MSC, smoothing, and the first Savitzky–Golay derivative. Diagnostic bands were identified by Spearman correlation, and prediction was performed using partial least squares regression (PLSR). PCA showed that spectral variability was driven mainly by seasonal and phenological factors, whereas varietal effects were secondary. In 2023/2024, greater spectral homogeneity was associated with lower predictive performance. In 2024/2025, greater spectral heterogeneity was associated with improved prediction for N (R2 = 0.83; RMSE = 0.78 g kg−1) and P (R2 = 0.80; RMSE = 0.10 g kg−1). Potassium remained the most challenging nutrient to predict (maximum R2 = 0.25), mainly due to its ionic nature and the resulting lack of significant correlation with specific VIS, NIR, and SWIR spectral features. These results indicate strong potential for predicting N and P in fresh sugarcane leaves, although model robustness depends on the extent of spectral variability in the dataset.

Abstract Situated in the Cabo Verde Archipelago, Fogo is among the most active oceanic volcanoes in the Atlantic, hosting frequent eruptions some of which were highly explosive and at least one gravitational flank collapse in the last 100 kyr. This study presents new volcanic glass shard geochemical data with high spatial distribution from 54 sites comprising samples from both pre- and post-collapse times. The analyzed glasses comprise basanites, tephrites, and foidites with a subset extending into the phonolite field. The glass compositions complement bulk rock data particularly in the range between 6 and 1 wt.% MgO, thus enabling more detailed inter-dataset comparison. Major and trace element data reveal five geochemical Groups characterized by incompatible element contents partly delineating rock series. Compositional diversity is primarily controlled by fractional crystallization, and to lesser extents by mantle-source heterogeneity and degrees of partial melting. A geochemical framework is established for investigating provenance of ash and volcaniclastics in the Cabo Verde region and beyond by integrating new glass data with published bulk rock and tephra records. We validate the classification scheme through comparison with primary marine tephra deposits, demonstrating robust sample assignment to the observed compositional Groups. The new provenance data go beyond simple attribution to Fogo by resolving distinct compositional Groups, enabling improved discrimination within Fogo-derived material. Group assignment and characterization of differentiation trends provide a useful first-order indication of pre- versus post-collapse affinity. The glass dataset serves as a reference for future provenance and source correlation studies across the Cabo Verde Archipelago.

Barrier islands along Florida’s Atlantic coast are increasingly threatened by sea-level rise, intensified hurricanes, shoreline armoring, and rapid coastal development. This study examined how beach and dune configurations varied in relation to coastal elevation patterns, NDVI-based surface greenness, and stakeholder perceptions across the East Central Florida Atlantic coast. Light Detection and Ranging (LiDAR) elevation datasets (2016, 2022, 2024), National Agriculture Imagery Program (NAIP)-derived Normalized Difference Vegetation Index (NDVI) analyses (2015, 2019, 2023), and stakeholder survey data from two coastal resilience workshops conducted in Volusia County in November 2024 were assessed to evaluate geomorphic change, vegetation-greenness patterns, and public perceptions of shoreline management strategies. Results showed descriptive differences among shoreline-type groups. Seawall-backed sites experienced the greatest net elevation loss (−0.529 m averaged over two sites) and a small negative mean transect-level NDVI change (−0.034) between 2015 and 2023, while natural dune sites showed an overall elevation gain (0.255 m averaged over three sites), despite some site-level loss after the 2022 hurricanes, and no net mean transect-level NDVI change (0.000) over the same NDVI period. Because the LiDAR and NDVI datasets are not temporally matched, these patterns are interpreted as complementary rather than causal lines of evidence. Stakeholder survey responses demonstrated that most respondents recognized the importance of dunes and coastal vegetation for resilience, but also expressed concerns about effectiveness, long-term maintenance, and cost of natural or hybrid solutions. Overall, the findings suggest that natural and minimally armored shorelines may retain greater capacity for elevation and vegetation-greenness recovery than hardened coastal systems, while also emphasizing the need for adaptive, conservation-based coastal management strategies that account for both physical shoreline conditions and stakeholder concerns.

Microbiological contamination of surface water represents a critical public health concern, while conventional disinfectants face limitations such as the generation of toxic by-products and the emergence of microbial resistance. In this study, the application of an ethanolic peel extract of Citrus paradisi (grapefruit), obtained by sonication at 40 kHz for 90 min at 55 °C using a 1:4 (w/v) solvent-to-solid ratio, was evaluated as a natural alternative for bacterial reduction in contaminated waters. The Minimum Inhibitory Concentration (MIC) and Minimum Bactericidal Concentration (MBC) of the extract were first determined against Staphylococcus aureus and Escherichia coli. The extract was then applied to samples of slightly contaminated surface water with bacterial loads between 103–104 CFU/mL and turbidity of 150 NTU, as well as to highly contaminated surface water with bacterial loads ≥105 CFU/mL and turbidity of 250 NTU. Bacterial removal was assessed at 6, 12, and 24 h. FTIR and UV-Vis characterization of the extract confirmed the presence of flavonoids (naringin), terpenes (limonene), and phenolic compounds. Results showed MIC/MBC values of 2.5/5.0 mg/mL for S. aureus and 5.0/10.0 mg/mL for E. coli. In slightly contaminated water, the extract at 5.0 mg/mL achieved complete (100%) removal of both microorganisms after 12 h, whereas in highly contaminated water, removals ranged from 80–90% for Staphylococcus spp. and E. coli. Statistical analysis (ANOVA, Bonferroni) demonstrated significant differences (p < 0.05) between the extract and ethanol. These findings indicate that Citrus paradisi extract constitutes an effective, sustainable, and low-cost natural alternative for bacterial reduction in surface waters, contributing to the valorization of agro-industrial residues and to the achievement of Sustainable Development Goal (SDG) 6.

Satellite Synthetic Aperture Radar (SAR) is an established technology for studying and monitoring archaeological landscapes, providing insights into surface morphology and the presence of near subsurface features. However, its application in large-scale archaeological prospection is limited by the lack of robust, automated methods for SAR data analysis. This study introduces a novel Deep Learning pipeline to automatically detect and segment archaeological settlement mounds, known as tells, in central Iraq on satellite SAR data. The pipeline leverages a state-of-the-art supervised method for instance segmentation, YOLOv8-Seg, and medium-resolution satellite SAR products, specifically the Copernicus Sentinel-1 Interferometric Wide Swath Mode Ground Range Detected and Copernicus Global 30-m Digital Elevation Model products. The model identifies tell sites with an Average Precision of 0.495±0.010 and a pixel-wise Intersection over Union of 0.361±0.048 over the test areas. Archaeological interpretation of the model’s inferences confirms its reliability in locating and segmenting archaeological sites, leading also to the identification of previously unmapped potential sites. After a main test in central Iraq, the proposed workflow demonstrates promising transferability to a nearby test area in Iran, although with a need for regional fine-tuning to account for inherent variations in feature morphology and environmental context. This research establishes a baseline for future Deep Learning applications in Synthetic Aperture Radar-based archaeological prospection.

Synthetic aperture radar (SAR) target images are highly sensitive to aspect angle, while practical data acquisition usually provides only sparse observations over limited viewpoints. This leads to severe data scarcity at unseen aspect angles and makes cross-angle generation prone to scattering-structure distortion and background statistical mismatch. Existing SAR image generation methods either focus on distribution matching without sufficiently exploiting scattering-related structural cues, or emphasize angle conditioning while failing to jointly preserve physically plausible dominant scattering-response variations and realistic background speckle statistics at unseen aspect angles. To address this issue, we propose a physically consistent framework for SAR image generation at unseen aspect angles. The proposed method introduces an ASC-inspired sparse scattering-structure prior to approximate the dominant scattering responses in the SAR image plane. Rather than performing full parametric ASC inversion, this prior serves as a differentiable and angle-aware structural proxy that guides the generator toward synthesizing SAR images with structurally plausible scattering layouts. In addition, a dual-consistency scheme is introduced to jointly enforce target-region scattering consistency and background-region statistical consistency, thereby improving the physical realism of the generated results in both the target and background regions. Extensive experiments under strict unseen-angle interpolation and hold-out protocols demonstrate that the proposed method consistently outperforms representative baselines in image fidelity, target-region scattering consistency, background statistical consistency, and angle-conditioned consistency. Further visualization and ablation studies verify the critical role of the ASC-inspired sparse scattering-structure prior in physically consistent SAR view completion.

The southern Kenyan shelf is amongst the narrowest in the world, averaging ca. 1 km. It has an irregular topography with alternating areas of topographically high hardground and mobile sediment in topographic lows. The shelf edge is characterised by multiple slumps. Seismic surveys reveal the stratigraphy to comprise a basal erosional surface, overlain by a coral reef unit (average interpreted thickness ca. 20 m) that spans the entire shelf width and contains two discrete ridges where interpreted reef thickness exceeds 50 m. The sequence is topped by a highly irregular surface that reflects the contemporary reef morphology. The surface bathymetry, backscatter and reef stratigraphy indicate periodic growth interrupted by backstepping that is attributed to variable rates of sea-level rise. Surface sediment comprises transgressive siliciclastic sand derived from lowstand river deltas. Carbonate sediment input from the reef is limited by low energy conditions, although coarse debris is present at the base of the contemporary reef slope. Reef evolution was accompanied by deposition of transgressive sands and the two are intimately associated, with sand being trapped on the seaward edge of growing reef crests and deposited in inter-reef depressions.

BACKGROUND: Elevated pancreatic enzymes and pancreatitis-like imaging findings have been reported in patients with accidental hypothermia; however, their clinical significance remains unclear. This study aimed to explore a serum amylase threshold that may support consideration of computed tomography (CT) for evaluating pancreatitis-like findings in accidental hypothermia and describe the clinical course of affected patients. METHODS: We conducted a retrospective single-center observational study of adult patients with accidental hypothermia admitted to a tertiary emergency and critical care center in Japan between November 2011 and April 2023. Accidental hypothermia was defined as a core body temperature <35°C. Receiver operating characteristic (ROC) analysis was performed to evaluate the ability of serum amylase levels to identify pancreatitis-like CT findings. Patients with hyperamylasemia and pancreatitis-like CT findings were descriptively analyzed. RESULTS: Among 169 patients included in the study, 36 (21.3%) had hyperamylasemia. Pancreatitis-like CT findings were observed in 14 patients, of whom 13 had hyperamylasemia. ROC analysis among patients who underwent CT evaluation identified 428 IU/L as a serum amylase threshold associated with pancreatitis-like CT findings (area under the curve, 0.91; sensitivity, 93%; specificity, 86%). The positive and negative predictive values were 44.8% and 99.0%, respectively. Most CT abnormalities consisted of localized peripancreatic fat stranding, fluid collection, or pancreatic enlargement. No patients developed pancreatic necrosis or required invasive pancreatic intervention. Most patients were managed conservatively with fluids and nutritional support, and short-term outcomes were generally favorable. CONCLUSIONS: In patients with accidental hypothermia, serum amylase levels ≥428 IU/L may support consideration of CT evaluation of pancreatitis-like findings. Although hyperamylasemia alone showed limited positive predictive value, most patients with pancreatitis-like CT findings had favorable short-term outcomes with conservative management.

Microcystin-LR (MC-LR), a natural toxin produced by cyanobacteria, poses a significant threat to human health and ecological systems. Rapid and accurate quantification of MC-LR plays a critical role in the early warning systems and environmental risk assessment of eutrophic water bodies. Herein, we identified a unique UV-induced degradation mechanism of MC-LR mediated by a newly developed ratiometric fluorescence probe ASS12 . Under UV irradiation, the probe undergoes a specific reaction with MC-LR, during which its fluorescence color changes sequentially from yellow to red and ultimately to blue within 20 min. This dynamic response enables visual detection of MC-LR by the naked eye and significantly enhances the ratiometric fluorescence signal for MC-LR at the I 465 / I 555 ratio by 77-fold. For the first time, we found that the long alkyl chain and aldehyde group of the probe serve as key molecular recognition sites during MC-LR detection. Furthermore, the ratiometric fluorescence signal of the probe exhibits a strong correlation with the concentration of MC-LR across different regions of plateau lakes, effectively reflecting the spatial distribution patterns of MC-LR within these aquatic environments. Using a multivariate linear parametrization model linking the fluorescence ratio and water quality parameters, the distribution of MC-LR in lakes can be determined. This study provides a valuable theoretical foundation and technical approach for the development of novel MC-LR monitoring tools and the visual assessment of water eutrophication.

Soil contamination poses a significant challenge to agricultural production, ecosystem stability, and food safety (1). Heavy metals, organic pollutants, salt stress, and nutrient-related contaminants can persist for long periods and move through soil-water-plant systems (2)(3)(4)(5), with recent estimates suggesting that up to 14-17% of cropland is affected by toxic metal pollution globally (6). Their behavior is often controlled by complex interactions among soil minerals, organic matter, microorganisms, and plant roots (5,7), which makes remediation outcomes strongly site-specific. These challenges require remediation strategies that are effective, affordable, and compatible with sustainable land management.Biochar has become one of the most promising materials for this purpose because it can be produced from plant biomass and other organic residues through pyrolysis under oxygenlimited conditions (8,9). Its remediation value stems from its porous structure, surface functional groups, mineral components, alkalinity, and capacity to interact with different contaminants (10,11). Depending on the feedstock, pyrolysis temperature, activation method, and post-treatment process, biochar can immobilize pollutants, improve soil properties, increase water and nutrient retention, and support carbon sequestration (2,5,10). However, this flexibility also creates uncertainty. A biochar that performs well in a given soil or for a specific contaminant may not show the same effect under other conditions, which highlights the need for mechanism-based and data-driven assessment (2,5). This special issue gathered contributions from the research topic "Recent Advances in Biochar Synthesis for Remediation of Soil Contamination" and was launched to capture recent progress in biochar production, modification, characterization, and application for contaminated or degraded soils. Figure 1 summarizes the contributions included in this research topic, illustrating how each study addresses a distinct aspect of engineered biochar researchfrom data-driven design and feedstock-specific modification to environmental behavior assessment and targeted contaminant removal. The collection includes four scholarly papers, consisting of one systematic review and three original research articles. Together, these studies cover data-driven biochar design, modified carbon-based amendments for saline-alkali soils, the environmental behavior of biochar-derived dissolved organic matter, and biochar-mineral composites for organic contaminant adsorption. The contributions show that biochar research is shifting from general amendment use toward engineered, mechanism-guided, and sitespecific remediation strategies. FIGURE 1 Schematic overview of the four contributions in this research topic, illustrating how engineered biochar connects diverse biomass feedstocks, modification strategies, mechanisms, and remediation outcomes. Each quadrant represents a specific study: machine learning-aided biochar design (12), cotton straw biochar combined with gypsum for saline-alkali soil improvement (13), dissolved organic matter behavior from invasive plant-derived biochar (14), and biochar-montmorillonite composite for berberine adsorption (15).Ge et al. (2025) (12) provided a systematic review of machine learning-aided design of engineered biochar for contaminant removal from soil and water (Figure 1a). The review highlights how machine learning can help connect feedstock characteristics, pyrolysis conditions, surface area, pore volume, functional groups, and electrochemical properties with biochar performance. This is important because conventional biochar development still depends strongly on trial-and-error experiments. Models, such as random forest and gradient boosting regression, and interpretation tools, such as SHAP analysis, can help identify the main factors that control adsorption and transformation processes. The review also emphasizes the need to combine data-driven prediction with mechanistic knowledge, including molecular simulation and experimental validation. In this way, machine learning can reduce experimental time and guide the design of more targeted biochar materials.Zhang et al. (2025) (13) investigated the co-application of a superabsorbent carbon-based material and flue gas desulfurization gypsum for improving saline-alkali soil (Figure 1b). The superabsorbent carbon-based material was prepared from cotton straw biochar, thereby linking agricultural waste valorization with saline soil remediation. Saline soils are difficult to manage due to their poor structure, high sodium content, and weak water balance, which can limit plant growth and reduce amendment efficiency. The study showed that the combined treatment promoted sodium leaching, reduced salt accumulation, and improved important indicators, such as electrical conductivity, total soluble salts, and sodium adsorption ratio. The results suggest that modified carbon-based materials can work together with mineral amendments to enhance both salt leaching and water retention. This provides a useful example of converting agricultural waste into functional materials for soil restoration. 1c). This study adds an important environmental safety perspective. Biochar is often evaluated mainly based on its capacity for contaminant removal; however, its mobile fractions can also influence carbon cycling, nutrient release, and pollutant mobility. The authors showed that the extraction method strongly affected the quantity and composition of biochar-derived dissolved organic matter. Strong acidic and alkaline conditions resulted in the release of more dissolved organic carbon, while milder water and calcium chloride extractions preserved more aromatic components with larger molecular characteristics. These results indicate that biochar application should be assessed under realistic environmental conditions, not only under aggressive laboratory extraction conditions. (15) reported the synthesis of a ball-milled biochar-montmorillonite composite derived from traditional Chinese medicine residues for berberine adsorption (Figure 1d). Berberine accumulation in soil is a concern related to continuous cropping. The composite showed high adsorption capacity and retained a large portion of its performance after repeated regeneration cycles. The adsorption process involved several mechanisms, including pore filling, hydrogen bonding, π-π stacking, cation exchange, and electrostatic attraction. This study demonstrates how biochar-mineral composites can be designed for specific agricultural contaminants. This study also links waste valorization with targeted remediation, which is an important direction for circular and sustainable agriculture.The papers in this research topic point to several shared messages. First, feedstock selection and modification methods are central to biochar performance. Cotton straw, invasive plant biomass, and traditional Chinese medicine residues can all be converted into valuable remediation materials; however, their properties depend on the production and treatment processes. Second, biochar performance should be interpreted through mechanisms rather than only through removal efficiency. The pore structure, surface chemistry, ion exchange, redox activity, dissolved organic matter release, and soil solution chemistry all affect remediation outcomes. Third, the field is becoming increasingly interdisciplinary. It now connects soil science, environmental chemistry, materials engineering, spectroscopy, agronomy, and artificial intelligence. Despite these advances, several challenges remain. More long-term field studies are needed to test whether promising laboratory results can be maintained under real soil conditions. Biochar aging, wet-dry cycles, plant root activity, microbial processes, and coexisting ions may alter the behavior of biochar after application. Standardized characterization and testing methods are also needed to compare biochars prepared from different feedstocks and at different pyrolysis temperatures. Future studies should also consider both the benefits and possible risks, including contaminant immobilization, nutrient cycling, dissolved organic matter release, greenhouse gas emissions, crop response, and life-cycle impacts.Overall, this research topic shows that biochar-based remediation is becoming more precise and application-oriented. The collected studies move the field toward engineered, mechanismguided, and context-specific solutions for contaminated and degraded soils. By linking material design with soil processes and environmental safety, they provide useful guidance for future research and practical remediation. We thank all contributing authors, reviewers, and the Frontiers editorial team for their valuable support in developing this research topic.

Underground mining operations depend heavily on vertical shafts for access, ventilation, and ore transport, making their structural integrity and safety critical to overall mine performance. Traditional shaft inspections, though rigorous, are limited by human accessibility, environmental hazards, and subjective evaluation. This study presents the development and initial testing of a novel unmanned aerial vehicle (UAV) system designed specifically for shaft inspections in deep mining environments. The research focuses on the GG-1 shaft in Kwielice, Poland—the country’s deepest operational shaft—where challenging conditions such as high ventilation airflow, confined geometry, and absence of GNSS signals necessitated innovative solutions. A custom-built hexacopter equipped with high-resolution cameras and photogrammetric capabilities was deployed to capture detailed visual and spatial data. This article presents complementary path of UAV evolution, from concept, early development stage and results without positioning system through to the description of final results including positioning system and all six cameras until results of high-altitude flights. Results demonstrate that UAV-based inspection can deliver sufficient precision for identifying structural irregularities, documenting shaft infrastructure, and enhancing safety monitoring. The findings highlight the potential of UAV technology as a complementary tool to conventional inspections, offering improved data quality, reduced risk to personnel, and a new approach to shaft maintenance.

Hydraulic engineering plays an indispensable role in flood control, water supply, irrigation, hydropower generation, ecological restoration, and water environment protection [...]

Carbonated water flooding can enhance oil recovery from low-permeability sandstone reservoirs while supporting CO2 geological sequestration; however, the coupled effects of carbonated water–rock interactions on pore-scale fluid redistribution remain unclear. This study used online nuclear magnetic resonance (NMR), magnetic resonance imaging (MRI), and mineralogical analysis to evaluate wettability-related water redistribution, mineralogical alteration, and oil mobilization in low-permeability sandstone cores exposed to carbonated water for 0, 5, 10, and 15 days, followed by immiscible CO2 flooding. With increasing exposure duration, NMR-derived water saturation increased from 0.490 to 0.571, indicating an apparent increase in pore-scale water affinity under the same saturation protocol. XRD results showed carbonate and clay/zeolite-related mineral alteration, including calcite falling below the detection or quantification limit and marked decreases in chlorite and laumontite, which were associated with modified pore-wall properties and improved water-phase access. During subsequent immiscible CO2 flooding, oil was preferentially mobilized from well-connected migration pores, while carbonated water treatment enhanced oil recovery from capillary-controlled percolation pores. The overall recovery factor increased by 2.8 percentage points, reaching 53.8% after 15 days of treatment. These results indicate that carbonated water improves CO2 flooding performance through coupled mineral alteration, pore-connectivity modification, wettability-related water redistribution, and multi-scale oil mobilization. The study provides NMR-based pore-scale evidence for interpreting carbonated water-assisted CO2 utilization and enhanced oil recovery.

Abstract. Effective science communication plays a crucial role in enhancing public understanding of Quaternary science. One promising strategy involves highlighting the interconnectedness of Quaternary sites, archaeology, and human culture. Despite the recent increased focus on science communication within the geosciences, the significance and effectiveness of highlighting such geocultural connections in communicating about Quaternary geoheritage sites have rarely been examined experimentally. This study evaluates the effectiveness of including geocultural context in educational videos for communicating the significance of Quaternary geoheritage sites in United Arab Emirates (UAE) and Oman. An online experiment was conducted to evaluate the effects of videos produced with input from academics, museum professionals, and heritage administrators from the region. The study compares the impact of two different 9 min videos developed in collaboration with academics, museum professionals, and heritage administrators from the region – one emphasising the geocultural context, and the other focusing solely on Quaternary science –. The impacts on participants' knowledge, interest, and perception of Quaternary geoheritage sites were assessed. The videos enhanced participants' self-reported knowledge of Quaternary geoheritage sites and increased their interest. Although the statistical results remain tempered by uncertainty, the overall pattern suggests that geocultural framing can foster a stronger and more durable sense of the importance of conserving Quaternary geoheritage, especially among people with less prior knowledge of such sites. The number of participants of this study is small and demographically limited to highly educated, relatively young adults with pro-nature attitudes, but this study demonstrates the value of integrating geocultural context in communicating the importance of Quaternary science and raising awareness of Quaternary geoheritage.

In the field of underwater acoustics, two methods are usually used to study the propagation of acoustic signals in seawater media. The first method is wave theory, which applies rigorous mathematical methods, combined with known fixed solution conditions, to solve the wave equation and study the change of amplitude and phase of acoustic signals in space. The second method is ray theory, in which the propagation of sound waves in a seawater medium is regarded as the propagation of sound lines in the medium in the high-frequency case; the change of sound intensity, the propagation time, and the propagation range of the sound lines in space are studied. Due to the approximation of ray theory, it is difficult to apply it in low-frequency shallow water conditions, and the importance of wave theory is particularly prominent in the context of the increasingly low frequency of sonar action. How to be able to solve the fluctuation equations accurately and quickly has become the focus of research by scholars in various countries.As shown in Figure 1, under the assumption of linear acoustics, the basic control equation of underwater sound propagation, the wave equation (WE), can be obtained according to the equation of motion, continuity equation, and state equation. Due to the spatial and temporal complexity of the wave equation, it is difficult to solve it directly. Generally, the strategy adopted in practice is to convert the Fourier transform to the frequency domain to obtain the Helmholtz equation (HE) (Jensen et al., 2011).Approaches to solving the Helmholtz equation are divided into two methods: direct and indirect. Solving the Helmholtz equation directly is difficult in practical applications, and the amount of computation is still staggering even today, despite the rapid development of computers. Liu et al. (2021) used the secondorder and fourth-order difference formats to solve the Helmholtz equation directly for Lloyd's mirror example with analytical solution, which took more than 1 hour and more than 1,000 iterations to reach convergence on a computer with 360 CPU cores, and the accuracy basically meets the requirements compared with the analytical solution, which is less usable in the actual sound field calculation.Most of the solution ideas are solved by various simplified theories of Helmholtz equations; in the process of long-term exploration, simplified models based on the wavenumber integration (WI) theory (Schmidt and Glattetre, 1985), the normal mode (NM) theory (Godin, 1992), and the parabolic equation (PE) theory (Collis et al., 2008) are used. At present, most of the numerical solution methods are based on the development of the above theories or a combination of each other; these methods have their own advantages and disadvantages, and restrictions on the use of conditions, and there is no model that can be applied in any case.Traditional numerical methods for solving partial differential equations (PDEs) are essentially solved discretely. For example, the finite difference method (FDM) (Stephen, 1988) solves partial differential equations by replacing differentiation with a difference approximation of the derivatives at grid points. The finite element method (FEM) (Thompson, 2006) solves the problem by dividing the solution domain into a finite number of small elements and then approximating the functional form of the solution on the elements. The finite volume method (FVM) (Fogarty and LeVeque, 1999) is based on the concept of control volume and divides the solution domain into multiple control volumes. The spectrum method (SM) (Tu et al., 2023) uses global basis functions (e.g., sine and cosine functions) to approximate the solution. Methods such as the boundary element method (BEM) (Lu et al., 2008) are widely used for solving PDEs, transforming the problem into integral equations on the boundary and discretizing only the boundary. These numerical discretization methods are currently widely used in underwater acoustic propagation model calculations and have an irreplaceable role for a short period of time.With the continuous development of artificial intelligence (AI) and deep learning (DL), the application of AI is not only limited to traditional tasks such as computer vision, natural language processing, and speech recognition; the application of AI in various disciplines is more and more extensive, and the combination of AI and other disciplines has become one of the main research directions in this field. The combination of artificial intelligence and other disciplines has become one of the main research directions in this field. One of the applications of combining AI with mathematics, physics, and other disciplines is solving partial differential equations (Han et al., 2018).In 1989, Cybenko (Cybenko, 1989) proved that a perceptron neural network with hidden layers has the ability to approximate any function when the activation function is a Sigmoid function, and in 1991, Hornik et al. (1989) further proved that the same applies when the activation function is any non-constant function. This is the fundamental theoretical basis of deep learning, universal approximation theorem (UAT), which describes the property that feedforward neural networks with a sufficient number of hidden units can approximate any continuous function to arbitrary accuracy, which is also known as the theoretical basis for neural networks to be able to solve PDEs.With the rapid development of computer hardware technology and the arrival of the era of big data and deep learning, deep learning frameworks such as TensorFlow (Dean and Monga, 2015), PyTorch (Paszke, 2019), and PaddlePaddle (Ma et al., 2019) have been gradually generated, perfected, and developed, and AI is rapidly becoming a powerful tool for solving complex scientific problems. In particular, the application of AI is opening up new possibilities in the field of solving PDEs. The core of the emerging field of artificial intelligence for partial differential equations (AI4PDE) lies in the use of neural networks and other machine learning techniques to approximate the solution of PDEs, a method called a deep learning solver. Deep learning solvers are able to learn from data and automatically capture complex patterns and features of the problem to provide faster and more accurate solutions than traditional numerical methods. Raissi et al. (2019) proposed physics-informed neural network (PINN), compared with the traditional purely data-driven neural network, by adding the difference between the equations and the to the function of the neural network that the equations are also in the that the neural network not only the network also the difference between each of the equations the and that the is to the at the same with the traditional purely data-driven neural network, by adding the difference of the equations and the to the function of the neural network, the equations are also in the process that the neural network not only own function also the difference of the equations in each of the and that the is to the of at the same time, use data to learn the model with a more This is known as the and has been applied to a number of scientific and such as solving the and of In can solve not only also problems. the problem to the of the sound field at the and the problem to the of the from the of the sound field and shown in Figure in there is also a of methods that use data for which are to as data-driven methods. This of methods not to use the equations at uses neural networks as to solve PDEs, of data to learn a direct data-driven Methods for solving underwater acoustic wave the to the equation This Deep proposed by et al. (2019) and Fourier proposed by et al. in the actual complex underwater acoustic the of data is and the data usually Due to the main on data-driven when the data are small or a of the model has low accuracy and and at the same time, purely on models also these methods have more in practical applications and are currently only used for solving with a amount of How to use of to solve with accuracy when there are or no data is a to be the of there have been scholars have solved the underwater acoustic propagation problem by of deep and combined with ray to a ray basis neural network and compared it with traditional data-driven machine learning process and artificial neural network the data numerical and one and for the data and the and of the method are numerical and one et al. proposed a physics-informed neural network the to a at the of the and and to directly the acoustic field of a in the and is at the of the and the to directly solve the Helmholtz equation of the sound and the of the method in the sound field is by et al. proposed a network which is a data-driven deep learning model for directly learning acoustic in the field. is also proposed that the ability of can be by the in the and the of et al. proposed a spatial domain physics-informed neural network a sound field for the by adding and by a for the of sound and on the ability of to learn the sound field by of and compared it with the spatial domain with spatial domain the of et al. proposed a method for underwater acoustic field based on in which the Helmholtz which the underwater acoustic propagation are as differentiation and combined with a small amount of data as to a and data-driven neural network to solve the underwater acoustic field. et al. proposed a method for underwater acoustic field based on in which the Helmholtz which the underwater acoustic propagation are as differentiation and combined with a small amount of data as to a and data-driven neural network to solve the underwater acoustic field. et al. proposed a method a physics-informed neural network by the equations of normal as a in the function of the neural network, which of the underwater acoustic in the of frequency of the research the research the combination of numerical solution methods or deep learning solution methods with underwater acoustic propagation the solution method and theory each to a and the combination of a solution method and a theory can a new underwater acoustic propagation shown in Figure the research of this on the combination of with the theory of normal and the theory of parabolic in the the field of underwater acoustic field core traditional numerical methods (e.g., finite difference method and based on from and to complex data-driven models (e.g., on which are to and in to and only a equation (e.g., Helmholtz to the of underwater acoustic propagation and in accuracy in complex these this study a on a which theory and are as theory is used to data functions and of each the the problem of by theory rapidly sound signals into the of neural network learning and solving the convergence of direct sound the fluctuation equation from theory is adopted as the control equation of which not only the advantages of parabolic approximation also the of theory, the approximation or complexity of control equations (e.g., Helmholtz the function is the for boundary for conditions, and for the of the fluctuation that the is usually two of than the other two is to it from data and which is from the of the theory to to and on the by and theories to the for and the of This not only the also a more and accurate for practical applications such as sound and sonar range the of a deep learning to use for the underwater sound field. the a is to a of the actual amplitude data at for data which is data-driven In to the data at the the the parabolic equations the and the boundary conditions, to the sound field to points. The network a sound and sound and as a and is used to the actual sound data at the points. to the by et al. in the network, it can be that it is difficult to the neural network directly sound data to to an solution to the rapidly of sound data in a with a range of convert the rapidly sound into a function and adding it to the neural network for the can the sound field at any The sound field is used in a range of applications, such as sound and sonar range the the of sound by normal in the between sound and function and the fluctuation the network of and the basic of to and the function by normal mode theory first proposed by in to solve the acoustic propagation problem in a and it is one of the theories for solving acoustic propagation in shallow the development of computer of the research of this the based on the finite difference which has the advantages of accuracy, and and this model is one of the widely used numerical computation for normal in underwater to the normal mode theory for solving the sound field is according to of (Jensen et al., and with the Helmholtz equation, the sound and only on the 1 the of 1 and the 1 is the most equation in the theory of normal the equation, is the is the is the mode function, and the above equation is the the is to and sound can be obtained by the sound field by the function in with the use to the functions and to each of from the and such as sound and as as of various that the number of in the and the field such as the frequency the phase low and the phase The function and wavenumber to each mode are and the sound is obtained which is used as the in to of (Jensen et al., the Helmholtz equation in the 1 is the sound is the and is the the solution of in the the function is with range and the differential 1 and the property of in 1 the approximation 1, at which can be by simplified fluctuation equation can be obtained by into the most control equation in the and is also from the direct use of 1 the control equation et al., et al., in boundary uses the first of boundary conditions, also known as the boundary as shown in difference between ability to accurately solve and numerical methods is the application of the neural differentiation traditional differentiation or numerical differentiation methods, differentiation is able to automatically derivatives with arbitrary accuracy, which has a for complex function differentiation is used in deep learning to the of the function with to the network for and shown in Figure the of the of an hidden and an with a of is the layers 1 to are the hidden and 1 is the The number of in the is which are range and The hidden layers are the number of in the is the function is the neural network be directly on the complex it to be into the and the and then the number of in the is which are and to the of et al. for solving Helmholtz equation, it can be that the use of activation function can be a solution to the use as the activation the of feedforward neural network as shown in and the and from the to the the can be by only the and boundary under the control of and boundary the network to learn the and for for the complex acoustic propagation it is difficult to use only the and in the for purely to be and in to the accuracy of it is to use a of the solution as the data to the as the data-driven function of as shown in each function is as shown in with the of the data-driven the of the control and the of the control The of each are and 1 1 1 is less than the in or when the number of the in the is and the model are is and the model are the model for sound field the of the model are and the range of the is the and and of the function of the in range are obtained by model and to obtain the function, which is then to obtain the sound by by the and to the accuracy of the method and the of the is compared with the numerical solutions by to the accuracy of the method and the of the and with of sound in the water the water sound a of shallow with as an example and the water as the as an acoustic space with a of the sound as the of the as and the frequency as is shown in Figure are shown in The range is based on the acoustic of from the and by at the and are is in the by is in the propagation is with is in the by the the is and the phase range is to of are for the of the sound and the and mode functions are shown in the approximation 1 and with the actual sound field application a of the are to the solution as the data to be used for data the field the range of of to a a number of range is of to a with a number of The number of of the solution data is The sound is into an function the number of is The sound data are into the function equation is obtained by from The function is then into and to be used as for the of the neural to the of in the the domain is in the range of for and for the complexity of the the the of the and the accuracy the number of the domain is to be The to range from 1,000 to at the boundary of the to and the number of to to the of et al. it can be that the of the function have a on the model is or and are by their to the of the function as 1 1 is on with The deep learning and used is TensorFlow and the used is with a learning of The number of hidden is used with a and the number of in each the the is to range from 1,000 to and is to range from to the field in the by in the same are obtained for to the accuracy of the method and the of the the number of is the model is and a is as shown in Figure the the can be that with a small number of accurate as are with accuracy, and most of the are basically to the accuracy of the numerical solutions with an of In only on purely in the and boundary can only the of the the accuracy of the is and it is only for use when the accuracy requirements are not also the of the of the network of this in on the process for is less the accuracy of the is adding the data-driven which the importance of on data for the data-driven deep learning of the complex the to and the the from as shown in Figure to the it can be that has learning ability for learning ability for the at is than that at and and the is and basically with the of in the range of for on the that the be to model the number of to with other and the and this the model is and a is as shown in Figure the the can be from Figure compared to Figure the sound field is at the the accurate data are not and by it can be that when the number of is not the model be and it is difficult to the accurate at the which a on the of the the to and the the by as shown in Figure can be from Figure compared with Figure the of the acoustic field is at the of and the is and the accuracy is for the model is still and it is to gradually the to a model to the a of the in Figure is to the as shown in Figure in Figure is able to the and the combined with Figure that the is at which is by the neural in learning for the these for a small of the range of with the of the practical the at the of is and the small is the up and the of the which can be used for sound field the actual that the of is two of than and when the function in the is by the other two it is small to be of to the of the functions to 1 1 and still The model is and a is as shown in Figure the the can be from Figure compared to Figure the sound field at the no accurate data is and the is by the which can be that when the between the the role of the is more and the of the is more the to and the the by as shown in Figure can be from Figure compared with Figure the of the acoustic field is in of and the is and the accuracy is for the model is still and it is to gradually each to a model to the of with propagation the acoustic propagation of mode and a with in the field mode and field is the of normal the number of gradually from partial to the and the fluctuation on mode and ability for signals is in the field is further the field and is the propagation of normal of are with fixed and sound waves in the form of only the wave with a and This process is with the learning of the at in this is the field and in field is the of normal In to wave the volume of seawater the to the for by the of the of the is to the problem of when signals low-frequency to a in a of the in Figure is to the as shown in Figure in Figure is able to the of the and the combination of the with Figure that the is at which is by the of the neural network to learn the these for a small in the and in combination with the practical application the at the of the is and small and are the of the which can be used for sound field the accuracy of under the of and 1 1 core are and These are used to the of with the numerical solutions of The first the and of each then in and the of the are based on The are as The the propagation from 1,000 to with a of and the The of for from to an of is and the for each is are as the between and the the the the The as shown in 1 is the by and is the by the between and and has to The as shown in 1 the of the of The as shown in is the of by and is the of by of with is essentially the of boundary to the model as at at the a of to the sound waves form of direct waves and waves to the boundary. The acoustic a in fluctuation frequency and to the layers and activation function of The network is to for high-frequency to at in the the water is by in phase than in the further at propagation boundary a of in the of the water and from the and the a it is a with boundary The normal have a at this with no or The acoustic and has a the learning of the at this have the with the numerical solution of and the is the at by a of to the it is by the acoustic difference between and the one normal at the the and the fluctuation to from a the other have a to rapid of sound waves and a in in the The combined of these two accuracy for signals and of the model in this study is based on the of the and the control The core and their are as The sine function is to the research of et al. this function for compared to used such as and and can the and learning The a learning of is which is widely used in deep learning for solving et al., that when the learning is than to and when the learning is than the convergence is be even For the number of hidden layers and the number of of and are the control the ability is when or the model The of hidden layers with an between accuracy and function The of the core to data and and be in in the of are to the of on model are shown in a a by the data and boundary it difficult for the model to learn the of acoustic propagation, and the the model the and are the the convergence is the and the can data the of the the model to the fluctuation equation, to from sound data and convergence the core of in the two data-driven models are for feedforward has the same as the in this study hidden layers sine activation the is function only neural network a of layers The is the of with no and the sound field is only data-driven number of is to and and activation the only difference lies in are are shown in on the data in the model with is to purely data-driven models such as feedforward neural networks and neural networks in of underwater acoustic field the of the and of the are than of the two of purely data-driven more accurate of acoustic field propagation and In purely data-driven to the of from acoustic are to in complex acoustic field in of various the of the of the is than that of purely data-driven models and to the of 1, a of between and the acoustic as as ability to more accurately the of acoustic field data the role of by from the fluctuation equation, boundary conditions, and acoustic the for the of purely data-driven which on features and This the to even in complex acoustic field the advantages of the proposed compared to purely on the in obtain the The is on solutions as data and be for complex that are difficult to learn for deep only a In the deep learning have a such as the learning the of the activation function the number of hidden the number of in each and in the of the such as the of the function have an on the of the model and and to be for In of the above For the problem of sound field it is to use network that can be by the features of the data a small amount of For example, et al. proposed an physics-informed neural network based on and which the and of in with complex partial differential et al. proposed a physics-informed neural network based on learning in which a network for the by the which has the of the For the problem that have a on the network, it is to use the method of control and it with the of in the for for a one at a the most are in can also be adopted to the For example, et al. (2021) proposed a learning to solve the problem of in For the and in the it is to and the in the function in the according to the of it is to in the the function is and the is and it is to use the to in the

Light quality, particularly the spectral composition emitted by LEDs, modulate growth dynamics, physiological processes, and the nutritional attributes of microgreens cultivated under controlled environments. In this study, the effects of three distinct LED spectra on growth performance, pigmentation, and nutritional composition of basil (Ocimum basilicum L.) microgreens were assessed within an indoor vertical farming system. The lighting treatments included: (LED1) 70% red (660 nm) + 30% blue (450 nm), (LED 2) a full-spectrum PAR range (400-700 nm), and (LED 3) 65% red (660 nm), 25% blue (450 nm), 5% white (broad spectrum around 400-700 nm), and 5% far-red (730 nm). Key parameters assessed were plant height, hypocotyl length, stem diameter, individual plant fresh weight, yield, color values (L* = luminosity, a* = red-green axis, b* = blue-yellow axis), dry matter content, macro and micronutrient levels, pH (hydrogen ion concentration) and titratable acidity (total acidity), soluble solid content (SSC), electrical conductivity (EC), and concentrations of oil, total phenols, flavonoids, and vitamin C. Among the treatments, LED 3 significantly enhanced plant height, hypocotyl length, yield, total phenol content, acidity, EC, and color brightness (L*), highlighting its superior overall performance. LED 2 was most effective in increasing vitamin C, flavonoid, and oil content, while LED 1 promoted higher dry matter and mineral contents. These findings emphasize the importance of optimizing LED spectra to improve both productivity and nutritional quality. Based on its consistent advantages across multiple parameters, LED 3 is the most advantageous in terms of yield, biomass accumulation, and visual quality, highlighting its potential suitability for commercial microgreen cultivation in controlled environment systems. Further research is needed to validate these findings across different cultivar-specific responses and environmental conditions.

Introduction Coordinating multiple unmanned surface vehicles (USVs) in coastal waters requires simultaneous consideration of COLREGs compliance, real-time response, energy efficiency, fault tolerance, and semantic scene understanding. Existing approaches typically address only part of this problem and provide limited support for integrated fleet-level coordination. Methods This paper proposes NSC-Marine, a neuro-symbolic framework that conceptually addresses these coupled constraints. It combines multimodal causal perception, LLM-based rule reasoning, energy-aware motion planning, and distributed fault reconfiguration within a dual-rate control architecture. To improve deployment safety, semantic reasoning is used to generate structured constraints, while low-level motion execution remains under deterministic planning and control. Results The framework is evaluated strictly within a physics-based simulation environment with approximately 70% overall fidelity, involving 1,000 trials of 20–100 USVs under Beaufort Scale 3–5 conditions. Under these simulated conditions, NSC-Marine achieves 88.7% ± 2.3% COLREGs compliance, 82.3% mission success under compound faults, and 13.6% energy reduction relative to the RLCA baseline, while maintaining a 320 ms critical-path latency. Discussion These metrics reflect an idealized simulation baseline and must not be generalized to physical deployment readiness. Real-world performance remains unvalidated, and staged hardware-in-the-loop testing and field trials are required to characterize the system’s actual behavior under physical disturbances.

Floating plastic debris can provide long-lived substrates for attached organisms, but reconstructing the drift history of small consumer items remains difficult. Here we report a colonized plastic bottle cap collected in the northwest Pacific. The cap hosted a tube-building polychaete and an associated assemblage including benthic foraminifera. We combined (1) a census of the fouling community, (2) chamber-level δ 18 O and δ 13 C measurements from two Rosalina globularis tests, and (3) Lagrangian drift simulations driven by surface currents to constrain the cap's likely trajectory and timescale. The assemblage comprised nine taxa and 307 individuals, and was strongly dominated by spirorbid tubes ( Spirorbis sp.; 76.5%). For specimen #021, δ 18 O-derived temperatures were 26.0 °C for the pooled early chambers (p–f-4), 26.8 °C for f-3, 29.9 °C for f-2, 23.3 °C for f-1, and 22.3 °C for the final chamber; the final-chamber estimate was close to the in situ sea-surface temperature at collection (21.7–21.8 °C). The final chamber of specimen #005 yielded an estimated temperature of 27.0 °C but should be treated as a reference value because of its very small carbonate mass. Drift simulations suggested that trajectories reaching the sampling site within ∼1–3 months most frequently originated from the Philippine region and were transported northward by the Kuroshio system. This multi-proxy approach illustrates how benthic biofoulers, including foraminifera, can help reconstruct the dispersal history of small plastic items and highlights the potential for occasional long-distance transport of coastal benthic taxa on tiny rafts.

Despite the success of COVID-19 mRNA vaccines, they still face challenges with high costs, complex manufacturing, off-target biodistribution, and systemic reactogenicity stemming from their inflammatory carriers: lipid nanoparticles (LNPs). While "naked" RNA delivery could in principle solve these issues, studies have suggested that it is infeasible due to rapid degradation by RNases and poor cellular entry, thereby necessitating formulations that enhance intracellular delivery and RNA stability. Now, we challenge this paradigm by showing that a simple and inexpensive (<$1), lighter-derived electroporator with microneedle electrodes (Piezopen) can augment gene expression and immunogenicity to naked mRNA leading to comparable responses to LNPs at low doses. We achieve robust responses in the absence of systemic inflammation and reactogenicity using skin-targeted delivery, administer diverse construct types (i.e., mRNA, self-amplifying RNA (saRNA), circular RNA (circRNA)), and demonstrate cross-species validation in live human skin to derisk subsequent clinical application. Our results introduce Piezopen as an inexpensive, well-tolerated, and efficacious alternative to LNPs for mRNA vaccine delivery, designed to facilitate routine vaccinations and pandemic response.

Understanding shear rate-dependent mechanical behavior and fracture evolution is essential for evaluating the stability of rock engineering structures under shear loading. In this study, direct shear tests and real-time acoustic emission (AE) monitoring were conducted on sandstone. Three shear rates were used: 0.015, 0.03, and 0.06 mm/s. All tests were performed under a constant normal stress of 5 MPa. The results show that shear rate strongly affects the shear response of sandstone. As shear rate increases, the shear strength and residual strength increase by approximately 15.28% and 24.52%, respectively. This rate-strengthening behavior is mainly related to the time-dependent propagation of microcracks and inertial effects under rapid loading. AE activity also shows clear rate-dependent characteristics. The cumulative AE event count decreases with increasing shear rate, whereas the average energy per event increases. This indicates that rapid loading suppresses the number of microcracking events but promotes more intense energy release from individual events. Frequency-domain analysis shows a transition from a multi-band frequency distribution at low shear rates to low-frequency dominance at high shear rates. This shift reflects the change from distributed microcracking to localized shear band development. The Gaussian Mixture Model (GMM) classification of RA-AF parameters further confirms this transition. With increasing shear rate, the proportion of shear cracks increases from 34.30% to 57.42%, while the proportion of tensile cracks decreases from 65.70% to 42.58%. AE source localization also shows a clear spatial transition. AE events change from a dispersed volumetric distribution at low shear rates to concentrated localization along the shear band at high shear rates.