Summarize Paper logo SummarizePaper.com

AI summary · arXiv: 2610.33609v1 · cs.LG

Satellite-derived bathymetry of Arabian Gulf ports under leakage-free validation: a multi-site benchmark of empirical, machine-learning and deep-learning models against 285,000 multibeam soundings, with an outlook for UAE waters

Muhammad Saeed, Fatema Al Hajri, Mohamed H. Alhosani, Dmitry Mikhaylov, Yasser F. Al Wahedi
Original paper Submitted to arXiv on 24 Oct. 2026 54 pages, 28 figures, 25 tables License: NONEXCLUSIVE-DISTRIB 1.0 Also in: fr

How to read this summary? Illustrated guide

The essentials

Spatially separated validation reveals larger satellite-bathymetry errors than random splits; local depth controls improve Arabian Gulf mapping, while complex models do not outperform empirical transfer.

Question

How accurately can satellite-derived bathymetry map turbid Arabian Gulf port waters when evaluated away from calibration soundings? The study examines whether model complexity, local controls and spaceborne lidar improve operational accuracy and coast-wide applicability.

Method

The benchmark paired 285,619 multibeam soundings from twelve surveys at six port areas with 72 Sentinel-2 images, comparing Stumpf and Lyzenga empirical methods, monotonic LightGBM and a U-Net ensemble. Validation used 450 m spatial blocks with 225 m buffers, leave-one-site-out transfer, local calibration using about 20% of site soundings, two independent sites and a temporal test. The processing chain included multi-image fusion and conformal uncertainty estimation, then extended to a 60 m UAE coastal mosaic checked and locally calibrated with ICESat-2 lidar.

Findings

  • Result For the same LightGBM model, RMSE increased from 0.57 m under random pixel splitting to 1.30 m under buffered spatial blocks and 1.66 m at unseen sites.
    See the passage in the paper
    Under a random 80/20 split of the pixels the RMSE was 0.57 m with $R^{2}=0.93$, and 73 % of the pixels were within the Order 1a/1b TVU. Under spatial blocks of 450 m with a 225 m buffer the RMSE rose to 1.30 m ($R^{2}=0.60$, 36 % within Order 1a/1b), and under leave-one-site-out to 1.66 m ($R^{2}=0.34$, 26 %).
  • Result The empirical Stumpf ratio achieved the lowest pooled transfer RMSE, although the U-Net underwent stricter leave-region-out evaluation.
    See the passage in the paper
    In transfer mode, i.e. when each model was applied to a training site it had never seen (Table 5; Fig. 3a), the Stumpf ratio gave the lowest pooled RMSE, 1.45 m (bias $+0.15$ m, $R^{2}=0.52$), followed by LightGBM at 1.66 m, the U-Net ensemble at 1.77 m (leave-region-out) and the Lyzenga transform at 2.00 m.
  • Result With local control, the hybrid achieved 1.11 m pooled RMSE but only 67% of training-site pixels fell within Order 2 tolerance, below the study's 95% compliance criterion.
    See the passage in the paper
    Pooled over the training sites, 32 % of the Stumpf transfer pixels and 39 % of the hybrid local-control pixels were within the Order 1a/1b TVU, and 57 and 67 % within Order 2; the Exclusive and Special Orders were met by 11–13 and 17–20 %.
  • Result Summer turbidity was five to six times higher than winter turbidity, with hybrid local-control coverage falling to 70% at the median site.
    See the passage in the paper
    The pooled turbidity of the images kept per season was 0.003 in winter, 0.004 in spring, 0.017 in summer and 0.005 in autumn, i.e. five to six times higher in summer than in winter.
  • Result Lidar calibration of 24 coastal cells reduced held-out mosaic RMSE from 2.89 m to 1.42 m and bias from +2.01 m to +0.24 m.
    See the passage in the paper
    On the held-out blocks of the 24 cells ($n=87,982$ lidar pixels within the limit) the error of the mosaic fell from 2.89 m to 1.42 m and its bias from $+2.01$ to $+0.24$ m (Table 19; Fig. 21).

Limitations

  • Brightness-based optical screening poorly distinguished deep water from shallow water, allowing saturated depths into the uncontrolled coastal mosaic.
    See the passage in the paper
    Deep water in the Gulf, and the water of the turbid basins, is brighter than the darkest 2 % of a context window, so that no test on the brightness of a single image can separate it from water of 8–10 m, and the Stumpf ratio, which saturates, gives it a depth of 8–10 m; within the surveyed sites this is caught by the seabed-not-readable test, which needs control depths that do not exist offshore.
  • The coast-wide lidar check primarily represented the western coast rather than providing geographically uniform validation.
    See the passage in the paper
    They cover the Al Dhafra and Abu Dhabi coasts densely (33 of the 36 cells; up to 18,300 pixels per cell) but only three cells east of 54.7∘E with a few passes each, so that the check describes mainly the western two thirds of the coast.
  • Pooled uncertainty intervals did not transfer reliably to locally calibrated predictions at Mirfa, where hybrid interval coverage was only 75%.
    See the passage in the paper
    At the independent sites the pooled intervals of the transfer methods were conservative (98–100 % at Mirfa and Old Musaffah) but those of the local methods were too narrow at Mirfa, 74–85 % for the three ratio-based methods and 75 % for the hybrid.

Why it matters

The authors argue that spatially separated validation, rather than random pixel splitting, is essential for estimating operational bathymetric accuracy. Their results support locally controlled Sentinel-2 bathymetry as a reconnaissance and change-monitoring complement to vessel surveys to about 10 m, not as a general replacement for hydrographic surveying.

Written by AI from the paper. Passages are checked word for word against the text we analysed; the rest may contain errors.

AI-generated keywords: Satellite-derived bathymetry; Spatial validation; Sentinel-2; ICESat-2; Arabian Gulf

Read further

  • Comprehensive Summary
  • Layman's Summary
  • Blog article
  • Original abstract
This preprint evaluates how accurately satellite-derived bathymetry estimates depth in the shallow, turbid, tidal port waters of the Arabian Gulf when tested in genuinely unseen areas. It also examines whether machine-learning and deep-learning models outperform traditional empirical methods, how much local depth control improves accuracy, and whether public satellite imagery and spaceborne lidar can support a UAE-wide mapping service. Its central argument is that conventional random-pixel validation substantially overstates operational accuracy because nearby training and test pixels share spatially correlated information. The benchmark combines 285,619 multibeam soundings from twelve surveys across six Abu Dhabi port areas with 72 selected Sentinel-2 images. Depths were referenced to Lowest Astronomical Tide and corrected for water levels at satellite overpass. The authors compared the Stumpf band-ratio method, the Lyzenga transform, physically constrained LightGBM trees, and a U-Net ensemble. Transfer tests withheld entire sites, while local-control tests calibrated each image using approximately 20% of a site’s soundings, supplemented, where available, by verified ICESat-2 lidar depths. Spatial validation used 450-metre blocks with a 225-metre separation buffer. Two sites were reserved for independent testing, and an earlier survey provided a temporal test. Predictions from multiple images were combined and accompanied by conformal uncertainty intervals. Validation design affected accuracy more than model choice. The same LightGBM model achieved a root-mean-square error (RMSE) of 0.57 metres under random splitting, 1.30 metres under buffered spatial validation, and 1.66 metres when transferred to an unseen site within the training region. In pooled transfer testing, Stumpf performed best, with an RMSE of 1.45 metres, compared with 1.66 metres for LightGBM, 1.77 metres for U-Net, and 2.00 metres for Lyzenga. With local control, a hybrid combining tree predictions, the band ratio, and a shallow-water adjustment achieved an RMSE of 1.11 metres and covered 75.7% of eligible shallow-water pixels. Its pooled 95% uncertainty intervals contained 95% of reference depths. However, only 67% of predictions fell within the IHO Order 2 vertical tolerance, below the 95% criterion used to claim compliance. Performance varied markedly across environments. Local calibration achieved RMSEs of 0.29–0.53 metres in clear water at the temporal test site, whereas the hybrid’s RMSE reached 1.84 metres at the independent Mirfa Bay site. No image could be successfully calibrated in the turbid Old Musaffah channel. Practical optical reach was approximately 8–9 metres in dredged basins and 11–12 metres in clearer water; the benchmark adopted an 11-metre transfer limit. Summer turbidity indices were five to six times their winter values, increasing local-control RMSE from approximately 1.02–1.04 metres to 1.38 metres and reducing coverage. Approximately 60–125 well-distributed control pixels per image and the fusion of three images provided most of the attainable benefit. A UAE reconnaissance mosaic at 60-metre resolution covered approximately 17,233 square kilometres. Comparison with ICESat-2 yielded an RMSE of 2.91 metres in shallow water and indicated that roughly 15% of the mapped area lay beyond the optical depth limit. Cell-by-cell lidar calibration reduced held-out RMSE to 1.42 metres and bias to +0.24 metres across 24 cells, covering roughly half the product. The authors conclude that Sentinel-2 bathymetry is useful for reconnaissance, survey planning, and change monitoring at depths of approximately 10 metres, but cannot replace navigational surveys. Limitations include the small number of independent sites, timing differences between imagery and surveys, sparse lidar verification, limited lidar coverage along the eastern coast, and uncertain performance under unfamiliar water conditions. The two maps of unsurveyed areas lack independent sounding validation, while the very-high-resolution comparison relied on an uncalibrated display image acquired years after the survey.
Created on 02 Nov. 2026

Was this synthesis useful?

Score: 0

Why do we need votes?

Votes are used to determine whether we need to re-run our summarizing tools. If the count reaches -10, our tools can be restarted.

Ask questions about this paper to our AI assistant

How to chat with this paper? Illustrated guide

Compare with the PDF

You can also chat with multiple papers at once here.

Figures and plots

Ask about a figure by number or use its question button to request visual analysis. Up to two figures and four image panels per question. Values read from plots are approximate; check the original data.
AI assistant instructions?

Tools for this paper

Explain it differently 3 free rewrites left today
Save to a collection Group papers in collections to compare them, chat with them and export them. Create a free account Log in
Export BibTeX · RIS (Zotero, Mendeley, EndNote) · Obsidian (Markdown) · Notion (Markdown)

Advanced tools Premium

For researchers who follow a field closely: citation alerts, journal-club slides, version comparison and Zotero export. Discover Premium

Zotero
Citation alerts Citation alerts guide
Who works on it? The people and institutions who publish on its topic
Weekly letter Presented in "Artificial intelligence this week" of 2 November 2026. Artificial intelligence this week: when better scores hide the wrong answer
Journal club Premium+ Slides to present it Propose for the journal club
Versions What changed between versions
Talk about it in public Premium+ Press kit: release, questions, hype check Press kit guide

Papers around this one

See them on a map

Ranked by how close their title and abstract are to this paper’s. Select some to read them together.

Closest papers

  1. Very close Same question
    Comparative Analysis of SBES Bathymetry and Machine Learning-Based Satellite-Derived Bathymetry from Sentinel-2 in the Coastal Waters of Dubibir, Situbondo
    M. Fuad, Rizky Wisudawan Zannuar, Seftiawan Samsu Rijal et al. · 2026 · JFMR-Journal of Fisheries and Marine Research · Not summarized yet

    The Dubibir study compares Sentinel-2-derived depths with SBES measurements, as this study compares satellite-derived bathymetry with multibeam soundings.

    Abstract

    Accurate Bathymetryc information in shallow coastal waters is critical for marine navigation, coastal zone management, habitat assessment, and environmental monitoring. Conventional Bathymetryc surveys are limited by high operational costs, restricted spatial coverage, and time-consuming fieldwork. To overcome these challenges, the present study assesses the potential of Satellite-Derived Bathymetry (SDB) using Sentinel-2 imagery as an alternative, comparing it with in situ Bathymetryc measurements obtained via Single Beam Echosounder (SBES) in the coastal waters of Dubibir, Situbondo, Indonesia. Bathymetryc data were collected with an SBES GPSMAP 585c, and Sentinel-2 multispectral imagery w…

  2. Very close Another approach to the problem
    Sentinel-2 Satellite-Derived Bathymetry with Data-Efficient Domain Adaptation
    Christos G. E. Anagnostopoulos, Vassilios Papaioannou, Konstantinos Vlachos et al. · 2025 · Journal of Marine Science and Engineering · cited 3 times · Not summarized yet

    The Agia Napa study tests U-Net transfer learning from Poland to reduce local training data, while this study benchmarks several models and local-control strategies across ports.

    Abstract

    Satellite-derived bathymetry (SDB) enables the efficient mapping of shallow waters such as coastal zones but typically requires extensive local ground truth data to achieve high accuracy. This study evaluates the effectiveness of transfer learning in reducing this requirement while keeping estimation accuracy at acceptable levels by adapting a deep learning model pretrained on data from Puck Lagoon (Poland) to a new coastal site in Agia Napa (Cyprus). Leveraging the open MagicBathyNet benchmark dataset and a lightweight U-Net architecture, three scenarios were studied and compared: direct inference to Cyprus, site-specific training in Cyprus, and fine-tuning from Poland to Cyprus with increm…

  3. Very close Same method, another problem
    Application of Backpropagation (BP) Neural Networks for Satellite-Derived Bathymetry Using Sentinel-2 Data: Accurate Depth Estimation in Coastal Environments
    Che Heng, Guo-Hao Liu, Zhong-Qiang Wu · 2025 · Marine Geodesy · cited 3 times · Not summarized yet

    The Nanshan Port study applies a backpropagation neural network to Sentinel-2 bathymetry, while this study includes a U-Net among its SDB models.

    Abstract

    Abstract Accurate bathymetry estimation is essential for coastal management, resource exploration, and ecological conservation. However, traditional depth measurement methods are often time-consuming, expensive, and logistically challenging, particularly for large-scale marine surveys. These limitations have driven the development of Satellite-Derived Bathymetry (SDB), which leverages high-resolution satellite imagery and advanced computational techniques to provide a more efficient, cost-effective, and scalable solution. This study utilizes Sentinel-2 satellite data combined with various algorithmic approaches to estimate bathymetry in the Nanshan Port area. Among these methods, the Backpro…

  4. Very close Another approach to the problem
    Satellite Derived Bathymetry with Sentinel-2 Imagery: Comparing Traditional Techniques with Advanced Methods and Machine Learning Ensemble Models
    T. Susa · 2022 · Marine Geodesy · cited 41 times · Not summarized yet

    This study benchmarks empirical, physics-based and machine-learning SDB methods, while the neighbour compares traditional techniques with advanced and ensemble methods.

    Abstract

    Abstract Accurate charting of nearshore bathymetry is critical to the safe and dependable use of coastal waterways frequented by the trading, fishing, tourism, and ocean energy industries. The accessibility of satellite imagery and the availability of various satellite-derived bathymetry (SDB) techniques have provided a cost-effective alternative to traditional in-situ bathymetric surveys. Furthermore, improved algorithms and the advancement of machine learning models have provided opportunity for higher quality bathymetric derivations. However, to date the relative accuracy and performance between traditional physics-based techniques, improved physics-based methods, and machine learning ens…

  5. Very close Another approach to the problem
    From Local Training to Large-Scale Mapping: A Comparative Assessment of Machine Learning and Deep Learning for Transferable Satellite-Derived Bathymetry
    Hsiao-Jou Hsu, Joachim Moortgat · 2026 · Remote Sensing · Not summarized yet

    The Pratas Island and Great Barrier Reef study tests model transfer across regions, while this study evaluates unseen-site transfer and local calibration in Arabian Gulf ports.

    Abstract

    Satellite-derived bathymetry (SDB) provides a cost-effective means for mapping shallow-water depths, yet its scalability and cross-regional generalizability remain challenging in optically complex coastal environments. This study systematically evaluates machine learning (ML) and deep learning (DL) approaches for transferable SDB over the 0–20 m depth range using multispectral Sentinel-2 imagery. A Random Forest model and four deep learning architectures–ResNet-50, ResNet-101, EfficientNet-B4, and ConvNeXt-Large–are developed and trained using data from Pratas Island (South China Sea) and selected reef regions of the Great Barrier Reef (GBR), and subsequently evaluated on spatially independe…

  6. Very close Same question
    Integrating Multiple Datasets and Machine Learning Algorithms for Satellite-Based Bathymetry in Seaports
    Zhong-Qiang Wu, Z. Mao, W. Shen · 2021 · Remote Sensing · cited 19 times · Not summarized yet

    The seaports study estimates depth from satellite imagery under turbid conditions, as this study assesses operational SDB accuracy in turbid ports.

    Abstract

    Water depth estimation in seaports is essential for effective port management. This paper presents an empirical approach for water depth determination from satellite imagery through the integration of multiple datasets and machine learning algorithms. The implementation details of the proposed approach are provided and compared against different existing machine learning algorithms with a single training set. For a single training set and a single machine learning method, our analysis shows that the proposed depth estimation method provides a better root-mean-square error (RMSE) and a higher coefficient of determination (R2) under turbid water conditions, with overall RMSE and R2 improvement…

  7. Close Same method, another problem
    Enhancing Water Depth Estimation from Satellite Images Using Online Machine Learning: A Case Study Using Baidu Easy-DL with Acoustic Bathymetry and Sentinel-2 Data
    Zhong-Qiang Wu, Shu-Lei Wu, Hai-Xia Yang et al. · 2023 · Remote Sensing · cited 6 times · Not summarized yet

    The Rushikonda study uses Sentinel-2 and compares machine-learning depth estimation with the Stumpf model, methods also evaluated in this study.

    Abstract

    Water depth estimation is paramount in various domains, including navigation, environmental monitoring, and resource management. Traditional depth measurement methods, such as bathymetry, can often be expensive and time-consuming, especially in remote or inaccessible areas. This study delves into the application of machine learning techniques, specifically focusing on the Baidu Easy DL model for water depth estimation leveraging satellite imagery. Utilizing Sentinel-2 satellite data over Rushikonda Beach in India and processing it into remote sensing reflectance using ACOLITE software, this research compares the performance of several machine learning algorithms, including the Stumpf model, …

  8. Close Same method, another problem
    A Comprehensive Evaluation of Machine Learning and Classical Approaches for Spaceborne Active-Passive Fusion Bathymetry of Coral Reefs
    Jian Cheng, Liang Cheng, Sensen Chu et al. · 2023 · ISPRS Int. J. Geo Inf. · cited 25 times · Not summarized yet

    The Anda Reef study fuses Sentinel-2 with ICESat-2 and compares classical and machine-learning bathymetry methods, as this study uses both data sources and model types.

    Abstract

    Satellite-derived bathymetry (SDB) techniques are increasingly valuable for deriving high-quality bathymetric maps of coral reefs. Investigating the performance of the related SDB algorithms in purely spaceborne active–passive fusion bathymetry contributes to formulating reliable bathymetric strategies, particularly for areas such as the Spratly Islands, where in situ observations are exceptionally scarce. In this study, we took Anda Reef as a case study and evaluated the performance of eight common SDB approaches by integrating Sentinel-2 images with Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2). The bathymetric maps were generated using two classical and six machine-learning algori…

Go further

How are these papers found?

We compare the title and abstract of this paper with those of every paper summarized on this site, and with the papers Semantic Scholar recommends, those it cites and those citing it (for a much-cited paper, its most recent citations and, from OpenAlex, the most cited ones). Among the prior work and the follow-ups, the papers that use each other in depth, a method it uses and the much-cited follow-ups come a little higher. The levels (very close, close, related) come from the similarity of the texts, not from a judgment of quality; * marks a paper without an abstract, compared by its title. The sentence under a paper is written by an AI from the two abstracts: check it in the papers. A paper marked not relevant by three readers leaves the list.

Updated on 2 November 2026.

Disclaimer: The AI-based summarization tool and virtual assistant provided on this website may not always provide accurate and complete summaries or responses. We encourage you to carefully review and evaluate the generated content to ensure its quality and relevance to your needs.