Hasil untuk "Ocean engineering"

Liling Chu, Chao Xu, Yanwen Wang et al.

Compound precipitation and wind speed extremes (CPWE) pose significant threats to the sustainable development of urban areas. This study investigated the spatial evolution characteristics, potential population exposure risk, and multidimensional inequality of CPWE within nine urban agglomerations in China, each containing at least one city with a GDP exceeding one trillion yuan, utilizing spatiotemporal statistics and attribution analysis. The results indicated that the intensity of CPWE in these urban agglomerations decreased from southeast to northwest, and the population exposed to mild, moderate, severe, and extreme levels accounted for 58 %, 28.3 %, 11.4 %, and 2.3 % of the total, respectively. Changes in exposure risk were driven by climate effect (58.29 % ± 12.77 %), followed by population (32.15 % ± 6.20 %) and interaction effect (9.55 % ± 5.14 %). Pearl River Delta (PRD) and Yangtze River Delta (YRD), identified as particularly vulnerable, experienced an increase in CPWE intensity exceeding 0.015 /10a. An increase of approximately 0.62 × 104 people per decade was observed for exposure risk, with over 20 % of the population facing severe or extreme levels, mainly due to the climate effect. CPWE exposure risk was significantly unequal across various dimensions (spatial autocorrelation: Moran’s I = 0.3798, P = 0.001; Gini coefficient: 0.08–0.5). Areas characterized by high-risk and balanced development (e.g., PRD, YRD) exhibited lower inequality, whereas regions featuring low-risk and concentrated development (e.g., GPZ) demonstrated higher inequality. The climate effect was the predominant influence in the low-risk areas as well as most high-risk areas. These findings support the targeted implementation of appropriate climate adaptation policies to promote regional sustainable development.

Zheng Li, Weibin You, Sivakumar Manickam et al.

To effectively eliminate excess antibiotics from aqueous environments and to mitigate the dissemination of antibiotic resistance genes (ARGs), this study proposes a novel degradation system that activates peroxymonosulfate (PMS) through a synergistic combination of hydrodynamic cavitation (HC) and divalent copper ions (Cu2+). Levofloxacin (LEV) is employed as the representative target contaminant to evaluate the system’s performance. HC has emerged as a promising technique for pollutant removal. In this study, the localized high-temperature and high-pressure conditions generated by HC not only partially activated PMS but also facilitated its interaction with Cu2+ ions, leading to a pronounced synergistic enhancement in sulfate radical (SO4·−) generation and efficient pollutant degradation. Under optimized HC/Cu2+/PMS conditions (Cu2+ = 5 mM, PMS = 2.5 mM, inlet pressure = 0.15 MPa, pH = 10), complete removal of LEV (30 mg/L) was achieved within 50 min. This study elucidates the degradation mechanisms and pathways of LEV within the coupled HC/Cu2+/PMS system and evaluates the ecological safety of its degradation intermediates using the U.S. EPA’s T.E.S.T. (Toxicity Estimation Software Tool). Furthermore, the system’s applicability was validated through degradation experiments involving a range of representative pollutants, demonstrating its broad-spectrum effectiveness. Crucially, the HC/Cu2+/PMS system demonstrated a superior cavitation yield (2.78 × 10−5 mg/J) and a low electrical energy per order (EE/O) of 229.48 kWh/m3, highlighting its high energy efficiency and practical potential for sustainable wastewater treatment. The experimental results emphasize the system’s strong potential for the effective removal of organic pollutants from water, offering a novel and sustainable approach for advanced water treatment.

Jinghan Wang, Zhen Cao, Shiyang Fu et al.

Extracting rampart crater ejecta blankets is crucial for understanding impact crater formation and material transport processes, offering key insights into the distribution of subsurface water and ice on Mars. However, traditional methods often fail to extract rampart crater ejecta blankets due to complex terrain, noise interference, and blurred boundaries. To overcome these challenges, we propose an edge-aware segment anything model (SAM) sputter analysis (EASSA) framework for the rapid and accurate extraction of rampart crater ejecta contours. EASSA comprises three key components. First, Wiener filtering and multiscale Retinex preprocessing are applied to suppress terrain noise and enhance the visual distinction between rampart features and the background. Second, an SAM-based unsupervised segmentation module is employed to automatically identify rampart crater boundaries. Finally, we refined the extracted edges by designing a contour optimization pipeline that applies classical image operators, such as Sobel to enhance gradients, Suzuki–Abe to correct contours, and Douglas–Peucker to simplify and smooth shapes. To validate EASSA, we construct a multiscale rampart crater dataset using context camera (targeting craters <1 km) and Thermal Emission Imaging System imagery (for craters >1 km) in the Chryse Planitia and Arabia Terra regions. Experimental results demonstrate that our method achieves a detection accuracy of 97.36%, a recall of 93.36%, and an intersection over union of 0.93, significantly outperforming the baseline SAM segmentation. Morphological analysis further reveals that rampart ejecta in both regions exhibit mobilities greater than 2, an average lobateness coefficient of 1.06, and relatively shallow excavation depths. Additionally, we observe that elevated terrains exhibit lower ejecta flow mobility under similar latitudinal conditions, while geomorphic evidence of past fluvial activity remains evident. These findings provide new insights into Martian subsurface water dynamics and the mobility characteristics of ejecta under varied geologic settings.

Giovanni Anconitano, Elena Arabini, Alessandro Patacchini et al.

The new Copernicus Radar Observing System for Europe in L-band (ROSE-L), expected to work in synergy with the C-band Sentinel-1 mission, will create a multiplatform Synthetic Aperture Radar (SAR) facility acquiring data in a systematic and coordinated way. This paper investigates the performance of a novel soil moisture retrieval scheme, extending the capability of a previously proposed multitemporal and multipolarization algorithm to the case of multifrequency SAR data. It relies on a Bayesian statistical criterion to invert a forward electromagnetic model based on the hypothesis that soil moisture can change abruptly, whereas soil roughness remains stable over time. The algorithm is applied to simulated data to compare two possible operational scenarios of ROSE-L and Sentinel-1 observations: L-band and C-band coincident (LC) or alternate (L-C) acquisitions. The case of single frequency (L or C) data is also considered in the analysis. In addition, quad-polarization (VV, VH, HH) and dual-polarization (VV, VH) data for ROSE-L are compared when combined with dual-polarization (VV, VH) data for Sentinel-1. The simulated multipolarization C-band and L-band SAR data are generated considering time variant scenarios of bare soil and crop covered fields. The algorithm is also tested on a time-series of non-coincident L-band SAOCOM-1A and C-band Sentinel-1A data to evaluate the improvements of the soil moisture retrieval against in-situ data when the two frequencies are merged in the multitemporal scheme. For the simulated case, results for bare soils show that the alternate configuration reaches a retrieval accuracy higher than that of single frequency, with an average percentage improvement in RMSE of approximately 18% compared to single C-band and 5% compared to single L-band. In many cases, it approaches the performance of the coincident acquisitions, maintaining a key advantage in terms of revisit time. The experiment on real data further confirms the advantage of alternating the acquisitions from the two frequency bands when exploited within a multitemporal framework.

Hoi Ming Lam, Torsten Geldsetzer, Stephen E. L. Howell et al.

The snow cover on first-year sea ice is a critically underobserved parameter in the Arctic sea ice system, recognized by the World Meteorological Organization as an Essential Climate Variable due to its influence on energy exchange between the atmosphere and ocean through turbulent, radiative, and conductive processes. Measurements of snow properties are limited, but present and future satellite missions provide opportunities for regular and continuous monitoring. This study explores the sub-km- to km-scale spatial association between in-situ snow depth measurements, Cryo2ice satellite altimetric measurements, and RADARSAT Constellation Misson compact polarimetric (CP) synthetic aperture radar (SAR) parameters over snow-covered landfast first-year sea ice in the Canadian Arctic Archipelago. In-situ snow depth measurements were collected at four unique sites along a 75-km track of near-coincidental Cryo2ice acquisitions in early spring 2022. We find that the Cryo2ice-retrived elevation difference (altimetric snow depth) can provide an estimate of snow depth on sea ice that is within one standard deviation (∼6 to 14 cm) of the in-situ measured values, particularly where the snow salinity is concentrated in the basal snow layer (bottom 2 cm). A statistically significant correlation (−0.77 to −0.85; p<0.01) is found between CP SAR backscatter coefficients at C-band and altimetric backscatter coefficients at Ku-band, particularly where there is low snow depth and smooth ice surface topography. Inconsistent statistical relationships are found between altimetric snow depth and CP SAR backscatter parameters that vary with snow and ice topography observed at the in-situ measurement sites.

Guilin Liu, Tao Ji, Guoxiang Wu et al.

A shallow water equations-based model with an improved local time-stepping (LTS) scheme is developed for modeling coastal hydrodynamics across multiple scales, from large areas to detailed local regions. To enhance the stability of the shallow water model for long-duration simulations and at larger LTS gradings, a prediction-correction method using a single-layer interface that couples coarse and fine time discretizations is adopted. The proposed scheme improves computational efficiency with an acceptable additional computational burden and ensures accurate conservation of time truncation errors in a discrete sense. The model performance is verified with respect to conservation and computational efficiency through two idealized tests: the spreading of a drop of shallow water and a tidal flat/channel system. The results of both tests demonstrate that the improved LTS scheme maintains precision as the LTS grading increases, preserves conservation properties, and significantly improves computational efficiency with a speedup ratio of up to 2.615. Furthermore, we applied the LTS scheme to simulate tides at grid scales of 40,000 m to 200 m for a portion of the Northwest Pacific. The proposed model shows promise for modeling cross-scale hydrodynamics in complex coastal and ocean engineering problems.

Jian Cao, Jianhua Zhang, Chang Wang et al.

Abstract The wearable exoskeleton system is a typical strongly coupled human–robotic system. Human–robotic is the environment for each other. The two support each other and compete with each other. Achieving high human–robotic compatibility is the most critical technology for wearable systems. Full structural compatibility can improve the intrinsic safety of the exoskeleton, and precise intention understanding and motion control can improve the comfort of the exoskeleton. This paper first designs a physiologically functional bionic lower limb exoskeleton based on the study of bone and joint functional anatomy and analyzes the drive mapping model of the dual closed-loop four-link knee joint. Secondly, an exoskeleton dual closed-loop controller composed of a position inner loop and a force outer loop is designed. The inner loop of the controller adopts the PID control algorithm, and the outer loop adopts the adaptive admittance control algorithm based on human–robot interaction force (HRI). The controller can adaptively adjust the admittance parameters according to the HRI to respond to dynamic changes in the mechanical and physical parameters of the human–robot system, thereby improving control compliance and the wearing comfort of the exoskeleton system. Finally, we built a joint simulation experiment platform based on SolidWorks/Simulink to conduct virtual prototype simulation experiments and recruited volunteers to wear rehabilitation exoskeletons to conduct related control experiments. Experimental results show that the designed physiologically functional bionic exoskeleton and adaptive admittance controller can significantly improve the accuracy of human–robotic joint motion tracking, effectively reducing human–machine interaction forces and improving the comfort and safety of the wearer. This paper proposes a dual-closed loop four-link knee joint exoskeleton and a variable admittance control method based on HRI, which provides a new method for the design and control of exoskeletons with high compatibility.

Xumei Sun, Yu-Lei Chen, Fan Xin et al.

Abstract Background It has been reported that the most common post-transcriptional modification of eukaryotic RNA is N6-methyladenosine (m6A). Previous studies show m6A is a key regulator for viral infection and immune response. However, whether there is a pathogen stimulus-dependent m6A regulation in invertebrate shrimp has not been studied. Results In this study, we performed a transcriptome-wide profiling of mRNA m6A methylation in shrimp (Marsupenaeus japonicus) after white spot syndrome virus (WSSV) infection by methylated RNA immunoprecipitation sequencing (MeRIP-seq). A total of 15,436 m6A peaks were identified in the shrimp, distributed in 8,108 genes, mainly enriched in the CDS, 3′ UTR region and near the stop codon. After WSSV infection, we identified 2,260 m6A peaks with significantly changes, of which 1,973 peaks were significantly up-regulated and 287 peaks were significantly down-regulated. 1,795 genes were identified as differentially methylated genes. GO and KEGG analysis showed that hyper-methylated genes or hypo-methylated genes were highly associated with innate immune process and related to metabolic pathways including HIF-1 signaling pathway, lysine degradation and Wnt signaling pathway. Combined analysis showed a positive correlation between m6A methylation levels and mRNA expression levels. In addition, computational predictions of protein-protein interaction indicated that genes with altered levels of m6A methylation and mRNA expression clustered in metabolism, DNA replication, and protein ubiquitination. ZC3H12A and HIF-1 were two hub genes in protein-protein interaction (PPI) network that involved in immune and metabolism processes, respectively. Conclusion Our study explored the m6A methylation pattern of mRNA in shrimp after WSSV infection, exhibited the first m6A map of shrimp at the stage of WSSV induced metabolic reprogramming. These findings may reveal the possible mechanisms of m6A-mediated innate immune response in invertebrates.

Gleice Kelly Barbosa Souza, Samara Oliveira Silva Santos, André Luiz Carvalho Ottoni et al.

Reinforcement learning is an important technique in various fields, particularly in automated machine learning for reinforcement learning (AutoRL). The integration of transfer learning (TL) with AutoRL in combinatorial optimization is an area that requires further research. This paper employs both AutoRL and TL to effectively tackle combinatorial optimization challenges, specifically the asymmetric traveling salesman problem (ATSP) and the sequential ordering problem (SOP). A statistical analysis was conducted to assess the impact of TL on the aforementioned problems. Furthermore, the Auto_TL_RL algorithm was introduced as a novel contribution, combining the AutoRL and TL methodologies. Empirical findings strongly support the effectiveness of this integration, resulting in solutions that were significantly more efficient than conventional techniques, with an 85.7% improvement in the preliminary analysis results. Additionally, the computational time was reduced in 13 instances (i.e., in 92.8% of the simulated problems). The TL-integrated model outperformed the optimal benchmarks, demonstrating its superior convergence. The Auto_TL_RL algorithm design allows for smooth transitions between the ATSP and SOP domains. In a comprehensive evaluation, Auto_TL_RL significantly outperformed traditional methodologies in 78% of the instances analyzed.

Wenjin Sun, Wenjin Sun, Wenjin Sun et al.

The North Pacific Subtropical Countercurrent area (STCC) is high in mesoscale eddy activities. According to the rotation direction of the eddy flow field and the sign of temperature anomaly within the eddy, they can be divided into four categories: cyclonic cold-core eddy (CCE), anticyclonic warm-core eddy (AWE), cyclonic warm-core eddy (CWE) and anticyclonic cold-core eddy (ACE). CCE and AWE are called normal eddies, and CWE and ACE are named abnormal eddies. Based on the OFES data and vector geometry automatic detection method, we find that at the sea surface, the maximum monthly number of the CCE, AWE, CWE, and ACE occurs in December (765.70 ± 52.05), January (688.20 ± 82.53), August (373.40 ± 43.09) and August (533.00 ± 56.92), respectively. The number of normal eddies is more in winter and spring, and less in summer and autumn, while abnormal eddies have the opposite distribution. The maximum rotation velocity of the four types of eddies appears in June (11.71 ± 0.75 cm/s), June (12.24 ± 0.86 cm/s), May (10.63 ± 0.99 cm/s) and June (9.97 ± 0.91 cm/s), which is fast in winter and spring. The moving speed of the four types of eddies is almost similar (about 10 ~ 11 cm/s). The amplitude of normal and abnormal eddies is both high in summer and autumn, and low in winter and spring, with larger amplitudes in normal than abnormal eddies. The eccentricity (defined as the eccentricity of the ellipse obtained by fitting the eddy boundary) of the four types of eddies is also close to each other, and their variation ranges from 0.7 to 0.8, with no apparent seasonal variation. The vertical penetration depth, which has no significant seasonal difference, is 675.13 ± 67.50 m in cyclonic eddies (CCE and CWE), which is deeper than that 622.32 ± 81.85 m in anticyclonic eddies (ACE and AWE). In addition, increasing the defined temperature threshold for abnormal eddies can significantly reduce their numbers but does not change their seasonal variation trend.

Emily Ralston, Geoffrey Swain

The best way to stop the introduction of non-indigenous species (NISs) is by preventing their transport. In the case of ship hulls, this may be accomplished by managing entrainment onto the hull. This study was designed to examine the role of hull husbandry, i.e., cleaning and grooming, in fouling community structure and to determine the effect of husbandry on the recolonization of surfaces after a transplant was performed. A series of panels were placed at two locations along the east coast of Florida (Port Canaveral and Sebastian Inlet) that are typified by distinct fouling communities. Panels were subjected to one of three treatments: groomed weekly, cleaned every two months, or freely fouling. After four months, all panels were cleaned and transplanted between sites; no further husbandry was performed. Fouling community composition and coverage was characterized at monthly intervals both before and after transplantation. Hull husbandry was found to affect coverage and composition, with groomed panels carrying a lower cover of macrofouling in general. The effect of the original location on subsequent fouling composition and recolonization by specific organisms was confirmed for encrusting bryozoans, barnacles, sponges, and tunicates. Hull husbandry also affected subsequent fouling with specific preferences shown for surfaces that had been groomed, cleaned and undisturbed.

Yulin Shangguan, Xiaoxiao Min, Zhou Shi

As an essential climate variable, soil moisture (SM) exerts an indispensable influence on numerous disciplines. However, various degrees of data gaps exist in current microwave SM products. Therefore, this article proposed a spatiotemporal attention-based residual deep network (STARN) to reconstruct gaps of the daily SM data from the Climate Change Initiative program of the European Space Agency (ESA CCI) over the Qinghai&#x2013;Tibet Plateau (QTP) during unfrozen seasons (May to September) from 2001 to 2021. The developed model is an end-to-end residual network embedded with three attention modules to comprehensively consider the potential relationship between SM and surface variables. Evaluation results revealed that the proposed model could well reconstruct SM gaps with an overall median <italic>R</italic> and unbiased RMSE (ubRMSE) values of 0.52 and 0.054 m³&#x002F;m³, while the overall median <italic>R</italic> and ubRMSE values for the ESA CCI SM were 0.41 and 0.058 m³&#x002F;m³. Besides, comparison with five baseline methods (e.g., the artificial neural network, convolutional neural network, extreme gradient boosting, long-short term memory, and DCT-PLS model) indicated that the STARN model had certain advantages over the five baseline models with higher correlation and more reasonable distribution patterns. The <italic>R</italic>&#x002F;ubRMSE values for the five models were 0.38&#x002F;0.057, 0.34&#x002F;0.058, 0.40&#x002F;0.058, 0.41&#x002F;0.056, and 0.41&#x002F;0.058, respectively. The pretraining using the ERA5-Land SM data could further improve the accuracy of generated seamless SM data since the ERA5-Land and ESA CCI SM complemented each other to a certain extent on the QTP. In summary, by leveraging the spatiotemporal information and attention modules, the STARN model showed great potentials in SM gap filling.

Aryanto Adriansyah Bagus, Prabowo Aditya Rio, Muttaqie Teguh et al.

B. Pena, Luofeng Huang

Abstract Machine learning techniques have inspired reduced-order solutions in the fluid mechanics field that show benefits of unprecedented capability and efficiency. Targeting ocean-wave problems, this work has developed a novel data-driven computational approach, named Wave-GAN. This new tool is based upon the conditional Generative Adversarial Network (GAN) principle, and it provides the ability to predict three-dimensional nonlinear wave loads and run-up on a fixed structure. The paper presents the principle of Wave-GAN and an application example of regular waves interacting with a vertical fixed cylinder. Computational Fluid Dynamics (CFD) is used to provide training and testing datasets for the Wave-GAN deep learning network. Upon verification, Wave-GAN proved the ability to provide accurate results for predicting wave load and run-up for wave conditions that were not informed during training. Yet the CFD-comparative results were only obtained within seconds by the deep learning tool. The promising results demonstrate Wave-GAN's outstanding potential to act as a pioneering sample of applying machine learning techniques to wave-structural interaction problems. It is envisioned that the new approach could be extended to more complex shapes and wave conditions to facilitate the various design stages of marine and offshore engineering applications such as monopiles. As a result, enhanced reliability is expected to optimise structural performance and prevent environmental disasters.

P. Puerta, C. Johnson, M. Carreiro-Silva et al.

Circulation patterns in the North Atlantic Ocean have changed and re-organized multiple times over millions of years, influencing the biodiversity, distribution, and connectivity patterns of deep-sea species and ecosystems. In this study, we review the effects of the water mass properties (temperature, salinity, food supply, carbonate chemistry, and oxygen) on deep-sea benthic megafauna (from species to community level) and discussed in future scenarios of climate change. We focus on the key oceanic controls on deep-sea megafauna biodiversity and biogeography patterns. We place particular attention on cold-water corals and sponges, as these are ecosystem-engineering organisms that constitute vulnerable marine ecosystems (VME) with high associated biodiversity. Besides documenting the current state of the knowledge on this topic, a future scenario for water mass properties in the deep North Atlantic basin was predicted. The pace and severity of climate change in the deep-sea will vary across regions. However, predicted water mass properties showed that all regions in the North Atlantic will be exposed to multiple stressors by 2100, experiencing at least one critical change in water temperature (+2°C), organic carbon fluxes (reduced up to 50%), ocean acidification (pH reduced up to 0.3), aragonite saturation horizon (shoaling above 1000 m) and/or reduction in dissolved oxygen (>5%). The northernmost regions of the North Atlantic will suffer the greatest impacts. Warmer and more acidic oceans will drastically reduce the suitable habitat for ecosystem-engineers, with severe consequences such as declines in population densities, even compromising their long-term survival, loss of biodiversity and reduced biogeographic distribution that might compromise connectivity at large scales. These effects can be aggravated by reductions in carbon fluxes, particularly in areas where food availability is already limited. Declines in benthic biomass and biodiversity will diminish ecosystem services such as habitat provision, nutrient cycling, etc. This study shows that the deep-sea VME affected by contemporary anthropogenic impacts and with the ongoing climate change impacts are unlikely to withstand additional pressures from more intrusive human activities. This study serves also as a warning to protect these ecosystems through regulations and by tempering the ongoing socio-political drivers for increasing exploitation of marine resources.

Zakria Zakria, Jianhua Deng, Rajesh Kumar et al.

Traditional target detection algorithms have difficulty to adapt complex environmental changes and have limited applicable scenarios. However, the deep-learning-based target detection model can automatically learn with strong generalization capability. In this article, we choose a single-stage deep-learning-based target detection model for research based on the model&#x2019;s real-time processing requirements and to improve the accuracy and the robustness of target detection in remote sensing images. In addition, we improve the YOLOv4 network and present a new approach. First, we propose a classification setting of the nonmaximum suppression threshold to increase the accuracy without affecting the speed. Second, we study the anchor frame allocation problem in YOLOv4 and propose two allocation schemes. The proposed anchor frame scheme also improves the detection performance, and experimental results on the DOTA dataset validate their effectiveness.

Da Huo, Da Huo, Da Huo et al.

Zhenwei Chen, Jialong Jiao, Qiang Wang et al.

In this paper, both numerical and experimental methods are adopted to study the fluid–structure interaction (FSI) problem of a wedge structure with stiffeners impacted with water during the free-falling water entry process. In the numerical model, a partitioned two-way couple of CFD and FEM solvers is applied to deal with the FSI problem, where the external fluid pressure exported from the CFD simulation is used to derive the structural responses in the FEM solver, and the structural deformations are fed back into the CFD solver to deform the mesh. Moreover, a tank experiment using a steel wedge model that has the same structural properties is also conducted to compare with the numerical results. Verification and validation of the numerical results indicate that the CFD-FEM coupled method is feasible and reliable. The slamming response results by numerical simulation and experiments, including displacement, velocity, acceleration, slamming pressure, deformation, structural stresses and total forces on the wedge, accounting for hydroelasticity effects in different free falling height conditions are comprehensively analyzed and discussed.

Pengfei Liu, Yun Li

In this article, an effectively variational pansharpening method with spectral gradient fidelity and spatial Hessian hyper-Laplacian sparsity constraints (PSGFSHHS) was proposed to fuse the low resolution multispectral (LRMS) and panchromatic (Pan) images to the high resolution multispectral (HRMS) image. First, the spectral feature correlation prior between LRMS and HRMS was modeled by the spectral gradient fidelity constraint. Second, the spatial correlation prior between Pan and HRMS was particularly modeled by the spatial Hessian hyper-Laplacian sparsity constraint from the statistical perspective, which clearly held strong novelty for pansharpening recently by the spatial Hessian hyper-Laplacian sparsity modeling. Third, by combining the spectral gradient fidelity constraint and the spatial Hessian hyper-Laplacian sparsity constraint, the PSGFSHHS model was formed and the alternating direction method of multipliers method was utilized for optimization. Finally, the experimental fusion examples clearly illustrated the effectiveness and capability of PSGFSHHS.

Stefan Wolff, Fearghal O'Donncha, Bei Chen

In situ and remotely sensed observations have huge potential to develop data-driven predictive models for oceanography. A suite of machine learning models, including regression, decision tree and deep learning approaches were developed to estimate sea surface temperatures (SST). Training data consisted of satellite-derived SST and atmospheric data from The Weather Company. Models were evaluated in terms of accuracy and computational complexity. Predictive skills were assessed against observations and a state-of-the-art, physics-based model from the European Centre for Medium Weather Forecasting. Results demonstrated that by combining automated feature engineering with machine-learning approaches, accuracy comparable to existing state-of-the-art can be achieved. Models captured seasonal trends in the data together with short-term variations driven by atmospheric forcing. Further, it demonstrated that machine-learning-based approaches can be used as transportable prediction tools for ocean variables -- a challenge for existing physics-based approaches that rely heavily on user parametrisation to specific geography and topography. The low computational cost of inference makes the approach particularly attractive for edge-based computing where predictive models could be deployed on low-power devices in the marine environment.