Editorial: Remote sensing applications in oceanography with deep learning
Bibliographic record
Abstract
Deep Learning and Remote Sensing for the Ocean: From Concept to Operational Impact Deep learning (DL) and remote sensing (RS) are transforming how we observe and manage the ocean. Modern algorithms, platforms, and multi-sensor data integration now deliver insights at scales and speeds that were impossible just a few years ago. This Research Topic gathers 17 contributions across seafloor geomorphology, Ship and hazard monitoring, water quality assessment, mesoscale dynamics, under-ice processes, sonar perception, and enabling methods—demonstrating a field that is both technically innovative and mission-driven. Seafloor to Shoreline Ocean science relies on accurate mapping of the seabed. Automation can quickly analyse broad regions while collecting characteristics that satellite altimetry misses, as demonstrated by a CNN + U-Net pipeline for recognising tiny seamounts in multibeam data. RipFinder, a mobile machine learning system for real-time rip current identification that also functions as a citizen-science tool in places with restricted connection, exemplifies "AI to edge" at the land–sea interface. Ships, Safety, and Hazards For marine awareness, synthetic aperture radar (SAR), is still essential. While previous research use AIS data, sea fog, and remote sensing to evaluate collision risk, a super-resolution Mask R-CNN architecture uses scale-aware fusion to improve ship detection in noisy SAR settings, providing evidence-based navigation management tools. Ecosystems and Water Quality Chlorophyll-a (Chl-a) variations and harmful blooms are important ecological markers. Green tide identification from MODIS images is enhanced by WaveNet (VGG16 + BiFPN + CBAM). The importance of physics-aware features is demonstrated by ResUNet models that relate ocean-atmosphere dynamics to Chl-a in the South China Sea. Long-term variability in the Persian Gulf and Arabian Sea is revealed by rebuilt MODIS datasets, and new techniques also yield transferable Chl-a products for estuaries. MarGEN, a GAN-based augmentation technique that enhances marine mammal call categorisation in situations where labelled audio is limited, is one example of an advancement in acoustic ecology. Mesoscale and Cryosphere Dynamics OIEDNet generates the first large-scale MIZ eddy catalogues by detecting under-ice eddies from Sentinel-1 dual-pol data, whereas Conv-LSTM GAN hybrids predict mesoscale eddy properties with high fidelity. Perception Underwater Sonar and visual sensing are crucial for autonomous systems. Forward-looking sonar object detection is improved by MLFANet, side-scan sonar small-object recognition is improved by SOCA-YOLO, and underwater optical imaging is improved by CUG-UIEF using edge-and attention-based fusion. Data, Platforms, and Decision Support New contributions also address scalable data management (LSH-based retrieval for ocean archives) and decision-making (multi-criteria approaches for underwater IoT and AUV deployments), underscoring the need to co-design sensing, connectivity, and computation. Cross-cutting Lessons Five themes emerge: (1) multi-scale architectures consistently boost detectability; (2) embedding physics-aware features enhances generalization; (3) translating models to edge-deployable tools enables real-world impact; (4) data efficiency strategies such as augmentation and self-supervision are critical in data-sparse regimes; and (5) benchmarking and openness will accelerate progress. Outlook This collection highlights a decisive shift from proof-of-concept to operational potential in ocean AI. Future priorities include embedding physical priors, advancing generative/self-supervised methods for sparse data, and ensuring scalability, efficiency, and usability for real-world applications. Together, these works show how DL and RS can protect mariners, monitor ecosystems, and reveal ocean dynamics— bringing us closer to truly actionable ocean intelligence.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.006 | 0.021 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.004 | 0.003 |
| Bibliometrics | 0.004 | 0.002 |
| Science and technology studies | 0.002 | 0.003 |
| Scholarly communication | 0.009 | 0.006 |
| Open science | 0.005 | 0.002 |
| Research integrity | 0.017 | 0.020 |
| Insufficient payload (model declined to judge) | 0.031 | 0.026 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".