pith. sign in

arxiv: 2606.02764 · v1 · pith:NJJW3ZIJnew · submitted 2026-06-01 · 💻 cs.CV · physics.comp-ph

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 This is my paper

Pith reviewed 2026-06-28 14:47 UTC · model grok-4.3

classification 💻 cs.CV physics.comp-ph
keywords satellite-derived bathymetrydeep learningtransferabilitySentinel-2convolutional neural networksRandom Forestcoastal bathymetrymulti-temporal imagery
0
0 comments X

The pith

Deep learning models maintain robust performance for satellite bathymetry across different coastal regions while Random Forest degrades.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tests if machine learning or deep learning better supports satellite-derived bathymetry that works in new locations without retraining. Models are trained on Sentinel-2 data from Pratas Island and parts of the Great Barrier Reef, then checked on separate test zones within and across those areas. Deep learning networks keep errors between 2.46 and 2.98 meters in cross-regional use, whereas Random Forest errors increase to between 2.99 and 3.78 meters. Training with whole contiguous reef blocks instead of scattered patches proves key, along with a loss function that weights shallow depths more heavily. Using multiple images over time at the same site further improves results by averaging out variations in conditions.

Core claim

The central claim is that four CNN architectures trained with spatially continuous blocks and a Smooth Weight Function weighted RMSE loss produce transferable bathymetry estimates from Sentinel-2 multispectral imagery, achieving cross-regional RMSE of 2.46-2.98 m over 0-20 m depths compared to Random Forest degradation, and 0.19-0.22 m RMSE on the MagicBathyNet benchmark.

What carries the argument

The Smooth Weight Function (SWF)-weighted RMSE loss that prioritizes near-surface depths, paired with training on contiguous spatial blocks to preserve reef continuity.

If this is right

  • Intra-regional tests show RMSE from 1.15 to 1.92 m over 0-20 m depths, dropping to 0.26 m for depths under 3 m.
  • Median aggregation of predictions from multiple image passes reduces noise from sun angle, atmosphere, water, and tides.
  • The networks outperform a U-Net baseline and a task-specific transformer on the MagicBathyNet aerial benchmark with fewer parameters.
  • Optimized architectures and pretrained weights are released to support application at new sites.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Such models could support repeated mapping of shallow waters worldwide using freely available Sentinel-2 data without needing local training data each time.
  • The emphasis on spatial continuity in training data may apply to other remote sensing tasks where random sampling breaks important patterns.
  • Extending the approach to include additional sensors or wavelengths might address limitations in very turbid or complex waters.

Load-bearing premise

The training regions chosen represent a broad enough sample of coastal optical conditions to ensure the models generalize to arbitrary new areas.

What would settle it

A test on a coastal region with water properties or bottom reflectance outside the range seen in the Pratas Island and Great Barrier Reef training data where deep model RMSE rises above 3.5 m while Random Forest stays comparable.

Figures

Figures reproduced from arXiv: 2606.02764 by Hsiao-Jou Hsu, Joachim Moortgat.

Figure 1
Figure 1. Figure 1: Study areas considered in this work. Panel (a) shows all site locations. Panels (b) and (c) [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Relationship between RMSE and number of bands used for the RF model determined [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Depth-bin coverage comparison across the four study regions. [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Spatial comparison of depth predictions ( [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Cross-regional bathymetry predictions for Ashmore Reef. Top row: reference bathymetry [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Cross-regional bathymetry predictions for Cartier Reef. Top row: reference bathymetry [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Depth-dependent median RMSE (solid lines) and 95% confidence intervals (1.96 [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Depth-dependent RMSE for ResNet101 under SWF, RMSE, and RPE losses at Ashmore [PITH_FULL_IMAGE:figures/full_fig_p026_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Depth-dependent RMSE for ConvNeXt-Large under SWF, RMSE, and RPE losses. SWF [PITH_FULL_IMAGE:figures/full_fig_p027_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Depth-binned RMSE averaged across Ashmore and Cartier reefs under SWF, RMSE, [PITH_FULL_IMAGE:figures/full_fig_p027_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Median signed error as a function of depth bin centre for varying [PITH_FULL_IMAGE:figures/full_fig_p028_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Hexbin density plots of predicted versus ground truth bathymetric depth for four models [PITH_FULL_IMAGE:figures/full_fig_p030_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Spatial prediction maps under Strategy 1 (random patch split). Row (a): Ashmore Reef; [PITH_FULL_IMAGE:figures/full_fig_p031_13.png] view at source ↗
read the original abstract

Satellite-derived bathymetry (SDB) from multispectral imagery is cost-effective but scales poorly across regions, especially in optically complex coastal environments. We evaluate machine learning and deep learning for transferable SDB over the 0-20 m depth range using Sentinel-2 imagery. A Random Forest baseline and four CNNs (ResNet-50, ResNet-101, EfficientNet-B4, ConvNeXt-Large) are trained on Pratas Island and selected Great Barrier Reef regions, then evaluated on spatially independent intra- and cross-regional test areas. Preserving spatial continuity during training, by keeping contiguous reef blocks rather than random patches, is the single most impactful design choice; we further introduce a Smooth Weight Function (SWF)-weighted RMSE loss that emphasizes near-surface depths. With these choices, intra-regional RMSE ranges from 1.15 to 1.92 m over 0-20 m and is as low as 0.26 m for depths <= 3 m. Random Forest degrades sharply under cross-regional transfer (RMSE 1.53 m -> 2.99-3.78 m), while the deep models stay more robust (2.46-2.98 m). On the public MagicBathyNet aerial-RGB benchmark (0-16 m) the proposed networks reach 0.19-0.22 m RMSE, outperforming a U-Net baseline and a task-specific transformer architecture with substantially fewer parameters. We further exploit multi-temporal repeat imagery: training on it broadens diversity, and median-aggregating predictions across passes at inference reduces noise from changing sun angles, atmospheric conditions, water properties, and tides. We release optimized architectures and pretrained weights to enable scalable transfer to new sites.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript evaluates Random Forest and four CNN architectures (ResNet-50, ResNet-101, EfficientNet-B4, ConvNeXt-Large) for satellite-derived bathymetry from Sentinel-2 imagery over 0-20 m depths. Models are trained on Pratas Island and selected Great Barrier Reef blocks with contiguous spatial blocks preserved and a Smooth Weight Function (SWF)-weighted RMSE loss; they are tested on spatially independent intra- and cross-regional areas. The central empirical claims are that DL models retain RMSE of 2.46-2.98 m under cross-regional transfer while RF degrades to 2.99-3.78 m, intra-regional RMSE reaches 1.15-1.92 m (0.26 m for ≤3 m), and the networks achieve 0.19-0.22 m RMSE on the MagicBathyNet aerial-RGB benchmark, outperforming a U-Net and a transformer baseline. Multi-temporal aggregation is shown to reduce noise, and pretrained weights are released.

Significance. If the cross-regional robustness claims hold after addressing site-diversity characterization, the work would be significant for practical SDB scaling, as it directly targets the known limitation of poor generalization across optically complex coastal sites. The release of optimized architectures and pretrained weights is a clear strength for reproducibility. The emphasis on preserving spatial continuity during training and the SWF loss are useful methodological contributions that could be adopted more broadly.

major comments (2)
  1. [cross-regional transfer evaluation] Cross-regional transfer evaluation (abstract and corresponding results section): The claim that DL models maintain robustness (RMSE 2.46-2.98 m) while RF degrades (2.99-3.78 m) is load-bearing for the transferability conclusion, yet no quantitative metrics are provided characterizing differences between training sites (Pratas Island, selected GBR blocks) and test areas in optical properties, water clarity (e.g., Secchi depth), chlorophyll-a, sediment load, or bottom-type distributions. Without such evidence, retained performance could reflect site similarity rather than model generalization.
  2. [MagicBathyNet benchmark] MagicBathyNet benchmark results (abstract): The reported 0.19-0.22 m RMSE on aerial RGB imagery (0-16 m) is presented as supporting evidence, but this dataset and sensor type do not directly test the Sentinel-2 multispectral cross-regional transfer scenario that forms the paper's primary claim; the two evaluations therefore address distinct transfer problems.
minor comments (1)
  1. [abstract] Abstract: The intra-regional RMSE range (1.15-1.92 m) is given without per-model or per-depth-band breakdown, making it difficult to assess which architecture drives the best results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below, agreeing where the manuscript can be strengthened and clarifying the role of each evaluation.

read point-by-point responses

  1. Referee: [cross-regional transfer evaluation] Cross-regional transfer evaluation (abstract and corresponding results section): The claim that DL models maintain robustness (RMSE 2.46-2.98 m) while RF degrades (2.99-3.78 m) is load-bearing for the transferability conclusion, yet no quantitative metrics are provided characterizing differences between training sites (Pratas Island, selected GBR blocks) and test areas in optical properties, water clarity (e.g., Secchi depth), chlorophyll-a, sediment load, or bottom-type distributions. Without such evidence, retained performance could reflect site similarity rather than model generalization.

    Authors: We agree that explicit quantitative characterization of optical and environmental differences would strengthen the generalization argument. The training (Pratas Island, selected GBR blocks) and cross-regional test areas are drawn from geographically distinct locations known to differ in water properties, but the original manuscript did not include a dedicated comparison table. In revision we will add a supplementary table compiling available Secchi depth, chlorophyll-a, and bottom-type information from public datasets and literature for all sites to quantify the degree of difference. revision: yes

  2. Referee: [MagicBathyNet benchmark] MagicBathyNet benchmark results (abstract): The reported 0.19-0.22 m RMSE on aerial RGB imagery (0-16 m) is presented as supporting evidence, but this dataset and sensor type do not directly test the Sentinel-2 multispectral cross-regional transfer scenario that forms the paper's primary claim; the two evaluations therefore address distinct transfer problems.

    Authors: We agree that the MagicBathyNet results address transfer to aerial RGB rather than Sentinel-2 multispectral imagery and therefore constitute supplementary rather than primary evidence for the Sentinel-2 cross-regional claims. The core transferability results remain the Sentinel-2 intra- and cross-regional tests. We will revise the abstract and discussion to explicitly label the MagicBathyNet evaluation as additional validation on a different sensor and data modality. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical evaluation on independent test sets

full rationale

The paper reports RMSE values from training CNNs and Random Forest on Pratas Island/Great Barrier Reef imagery and evaluating on spatially independent intra- and cross-regional test areas plus the external MagicBathyNet benchmark. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text. All performance claims are direct measurements on held-out data; the SWF-weighted loss is an explicit design choice, not a self-referential fit. This is standard supervised learning evaluation and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that the chosen training and test splits allow generalization claims; no free parameters explicitly fitted in the reported results beyond standard ML training.

free parameters (1)
  • SWF weighting parameters
    The Smooth Weight Function is introduced to weight the RMSE loss but its exact parameterization or optimization is not specified in the abstract.
axioms (2)
  • domain assumption Spatially contiguous reef blocks during training improve generalization without introducing data leakage
    This is presented as the single most impactful design choice for transferability.
  • domain assumption The intra- and cross-regional test areas are truly independent of the training data
    Required for the transferability claims.

pith-pipeline@v0.9.1-grok · 5862 in / 1465 out tokens · 44171 ms · 2026-06-28T14:47:17.593848+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 42 canonical work pages

  1. [1]

    URL https://www.hydro-international.com/content/news/vast-new-areas-of-the-s eafloor-mapped-in-latest-seabed-2030-update

    Seabed 2030 Project, Vast new areas of the seafloor mapped in latest seabed 2030 update, hydro International, accessed 2026-01-09 (2025). URL https://www.hydro-international.com/content/news/vast-new-areas-of-the-s eafloor-mapped-in-latest-seabed-2030-update

    2030

  2. [2]

    Mayer, M

    L. Mayer, M. Jakobsson, G. Allen, B. Dorschel, R. Falconer, V. Ferrini, G. Lamarche, H. Snaith, P. Weatherall, The nippon foundation – gebco seabed 2030 project: The quest to see the world’s oceans completely mapped by 2030, Geosciences 8 (2) (2018) 63.doi:10.3390/geosciences8 020063. URLhttps://doi.org/10.3390/geosciences8020063

    work page doi:10.3390/geosciences8 2030

  3. [3]

    Lyzenga, Passive remote sensing techniques for mapping water depth and bottom features, Appl

    D. Lyzenga, Passive remote sensing techniques for mapping water depth and bottom features, Appl. Opt. 17 (1978) 379–383.doi:10.1364/AO.17.000379

    work page doi:10.1364/ao.17.000379 1978

  4. [4]

    Lyzenga, Shallow-water bathymetry using combined lidar and passive multispectral scanner data, Int

    D. Lyzenga, Shallow-water bathymetry using combined lidar and passive multispectral scanner data, Int. J. Remote Sens. 6 (1985) 115–125.doi:10.1080/01431168508948428

    work page doi:10.1080/01431168508948428 1985

  5. [5]

    Lyzenga, N

    D. Lyzenga, N. Malinas, F. Tanis, Multispectral bathymetry using a simple physically based algorithm, IEEE Trans. Geosci. Remote Sens. 44 (8) (2006) 2251–2259.doi:10.1109/TGRS.2 006.872909

    work page doi:10.1109/tgrs.2 2006

  6. [6]

    Stumpf, K

    R. Stumpf, K. Holderied, M. Sinclair, Determination of water depth with high-resolution satellite imagery over variable bottom types, Limnol. Oceanogr. 48 (2003) 547–556.doi: 10.4319/lo.2003.48.1_part_2.0547

    work page doi:10.4319/lo.2003.48.1_part_2.0547 2003

  7. [7]

    B. Chen, Y. Yang, D. Xu, E. Huang, A dual band algorithm for shallow water depth retrieval from high spatial resolution imagery with no ground truth, ISPRS J. Photogramm. Remote Sens. 151 (2019) 1–13.doi:10.1016/j.isprsjprs.2019.02.012

    work page doi:10.1016/j.isprsjprs.2019.02.012 2019

  8. [8]

    Mabula, D

    M. Mabula, D. Kisanga, S. Pamba, Application of machine learning algorithms and sentinel-2 satellite for improved bathymetry retrieval in lake victoria, tanzania, Egypt. J. Remote Sens. Space Sci. 26 (2023) 619–627.doi:10.1016/j.ejrs.2023.07.003

    work page doi:10.1016/j.ejrs.2023.07.003 2023

  9. [9]

    M. Ye, C. Yang, X. Zhang, S. Li, X. Peng, Y. Li, T. Chen, Shallow water bathymetry inversion based on machine learning using icesat-2 and sentinel-2 data, Remote Sens. 16 (23) (2024) 4603. doi:10.3390/rs16234603

    work page doi:10.3390/rs16234603 2024

  10. [10]

    Figliomeni, C

    F. Figliomeni, C. Parente, Methods for satellite derived bathymetry from sentinel-2 images: A comparison, Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. XLVIII-4/W9-2024 (2024) 165–173.doi:10.5194/isprs-archives-XLVIII-4-W9-2024-165-2024

    work page doi:10.5194/isprs-archives-xlviii-4-w9-2024-165-2024 2024

  11. [11]

    Z. Wu, Z. Mao, W. Shen, D. Yuan, X. Zhang, H. Huang, Satellite-derived bathymetry based on machine learning models and an updated quasi-analytical algorithm approach, Opt. Express 30 (10) (2022) 16773–16793.doi:10.1364/OE.456094

    work page doi:10.1364/oe.456094 2022

  12. [12]

    Kwon, H.-k

    J.-y. Kwon, H.-k. Shin, D.-h. Kim, H.-g. Lee, J.-k. Bouk, J.-h. Kim, T.-h. Kim, Estimation of shallow bathymetry using sentinel-2 satellite data and random forest machine learning: A case study for cheonsuman, hallim, and samcheok coastal seas, J. Appl. Remote Sens. 18 (1) (2024) 014522.doi:10.1117/1.JRS.18.014522

    work page doi:10.1117/1.jrs.18.014522 2024

  13. [13]

    Tonion, F

    F. Tonion, F. Pirotti, G. Faina, D. Paltrinieri, A machine learning approach to multispectral satellite-derived bathymetry, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci. V-3-2020 (2020) 565–570.doi:10.5194/isprs-annals-V-3-2020-565-2020

    work page doi:10.5194/isprs-annals-v-3-2020-565-2020 2020

  14. [14]

    Sagawa, Y

    T. Sagawa, Y. Yamashita, H. Yamaguchi, T. Okumura, Satellite derived bathymetry using machine learning and multi-temporal satellite images, Remote Sens. 11 (10) (2019) 1158. doi:10.3390/rs11101158. 39

    work page doi:10.3390/rs11101158 2019

  15. [15]

    K. Peng, H. Xie, Q. Xu, P. Huang, Z. Liu, A physics-assisted convolutional neural network for bathymetric mapping using icesat-2 and sentinel-2 data, IEEE Trans. Geosci. Remote Sens. 60 (2022) 4210513.doi:10.1109/TGRS.2022.3213248

    work page doi:10.1109/tgrs.2022.3213248 2022

  16. [16]

    M. A. Najar, R. Benshila, Y. E. Bennioui, G. Thoumyre, R. Almar, E. W. J. Bergsma, J.- M. Delvit, D. G. Wilson, Coastal bathymetry estimation from sentinel-2 satellite imagery: Comparing deep learning and physics-based approaches, Remote Sens. 14 (5) (2022) 1196. doi:10.3390/rs14051196

    work page doi:10.3390/rs14051196 2022

  17. [17]

    C. Xie, P. Chen, Z. Zhang, D. Pan, Satellite-derived bathymetry combined with sentinel-2 and icesat-2 datasets using machine learning, Front. Earth Sci. 11 (2023) 1111817.doi: 10.3389/feart.2023.1111817

    work page doi:10.3389/feart.2023.1111817 2023

  18. [18]

    Abdul Gafoor, M

    F. Abdul Gafoor, M. R. Al-Shehhi, C.-S. Cho, H. Ghedira, Gradient boosting and linear regression for estimating coastal bathymetry based on sentinel-2 images, Remote Sens. 14 (19) (2022) 5037.doi:10.3390/rs14195037

    work page doi:10.3390/rs14195037 2022

  19. [19]

    Setiawan, N

    K. Setiawan, N. Suwargana, D. Ginting, M. Manessa, N. Anggraini, S. Adawiah, A. Julzarika, S. Surahman, M. Rosid, A. Supardjo, Bathymetry extraction from spot 7 satellite imagery using random forest methods, Int. J. Remote Sens. Earth Sci. (IJReSES) 16 (2019) 23.doi: 10.30536/j.ijreses.2019.v16.a3085

    work page doi:10.30536/j.ijreses.2019.v16.a3085 2019

  20. [20]

    Z. Lv, J. Herman, E. Brewer, K. Nunez, D. Runfola, Bathyformer: A transformer-based deep learning method to map nearshore bathymetry with high-resolution multispectral satellite imagery, Remote Sens. 17 (7) (2025) 1195.doi:10.3390/rs17071195

    work page doi:10.3390/rs17071195 2025

  21. [21]

    Al Najar, Y

    M. Al Najar, Y. El Bennioui, G. Thoumyre, R. Almar, E. Bergsma, R. Benshila, J.-M. Delvit, D. Wilson, A combined color and wave-based approach to satellite-derived bathymetry using deep learning, in: Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., Vol. XLIII-B3-2022, 2022, pp. 9–16.doi:10.5194/isprs-archives-XLIII-B3-2022-9-2022

    work page doi:10.5194/isprs-archives-xliii-b3-2022-9-2022 2022

  22. [22]

    L. Yang, M. Liu, N. Liu, J. Guo, L. Lin, Y. Zhang, X. Du, Y. Xu, C. Zhu, Y. Wang, Recovering bathymetry from satellite altimetry-derived gravity by fully connected deep neural network, IEEE Geosci. Remote Sens. Lett. 20 (2023) 1–5.doi:10.1109/LGRS.2023.3302992

    work page doi:10.1109/lgrs.2023.3302992 2023

  23. [23]

    Zhong, J

    J. Zhong, J. Sun, Z. Lai, Y. Song, Nearshore bathymetry from icesat-2 lidar and sentinel-2 imagery datasets using deep learning approach, Remote Sens. 14 (17) (2022) 4229. doi: 10.3390/rs14174229

    work page doi:10.3390/rs14174229 2022

  24. [24]

    S. Qian, Y. Chen, W. Wang, G. Zhang, L. Li, Z. Hao, Y. Wang, Physics-guided deep neural networks for bathymetric mapping using sentinel-2 multispectral imagery, Front. Mar. Sci. 12 (2025) 1636124.doi:10.3389/fmars.2025.1636124

    work page doi:10.3389/fmars.2025.1636124 2025

  25. [25]

    Y. A. Lumban-Gaol, K. Ohori, R. Peters, Satellite-derived bathymetry using convolutional neural networks and multispectral sentinel-2 images, Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. XLIII-B3-2021 (2021) 201–207.doi:10.5194/isprs-archives-XLIII-B3-2 021-201-2021

    work page doi:10.5194/isprs-archives-xliii-b3-2 2021

  26. [26]

    International Hydrographic Organization, IHO Standards for Hydrographic Surveys (S-44), 6th Edition, International Hydrographic Organization, Monaco, 2020

    2020

  27. [27]

    O. B. Andersen, DTU23 Global Ocean Tide model (DTU23 OceanTide), Technical University of Denmark. Dataset (2023).doi:10.11583/DTU.23828874.v1. URLhttps://doi.org/10.11583/DTU.23828874.v1

    work page doi:10.11583/dtu.23828874.v1 2023

  28. [28]

    P. Shih, D. Arumugam, S. Shyue, Bathymetric lidar survey of penghu islands and dongsha atoll using an ellipsoidal height system for bathymetric mapping in shallow waters and difficult- to-navigate environments, Sea Technology 52 (11) (2011) 42–45. 40

    2011

  29. [29]

    R. J. Beaman, Ausbathytopo (great barrier reef) 30 m 2017 – a regional-scale depth model (20170025c), accessed: 2026-01-09 (2017).doi:10.4225/25/5a207b36022d2. URLhttp://dx.doi.org/10.4225/25/5a207b36022d2

    work page doi:10.4225/25/5a207b36022d2 2017

  30. [30]

    Gorelick, M

    N. Gorelick, M. Hancher, M. Dixon, S. Ilyushchenko, D. Thau, R. Moore, Google earth engine: Planetary-scale geospatial analysis for everyone, Remote Sens. Environ. 202 (2017) 18–27. doi:10.1016/j.rse.2017.06.031

    work page doi:10.1016/j.rse.2017.06.031 2017

  31. [31]

    E. Twiggs, Ashmore reef and cartier island marine parks satellite-derived bathymetry acquisition (20220016s), Commonwealth of Australia (Geoscience Australia) (2023).doi:10.26186/149098

    work page doi:10.26186/149098 2023

  32. [32]

    Zhang, X

    X. Zhang, X. Chen, W. Han, X. Huang, Y. Chen, J. Li, L. Wang, Satellite-driven deep learning algorithm for bathymetry extraction, in: Web Information Systems Engineering – WISE 2024, Springer Nature Singapore, Singapore, 2024, pp. 313–325.doi:10.1007/978-981-96-0573-6 _23

    work page doi:10.1007/978-981-96-0573-6 2024

  33. [33]

    L.-C. Chen, Y. Zhu, G. Papandreou, F. Schroff, H. Adam, Encoder-decoder with atrous separable convolution for semantic image segmentation, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 801–818.doi:10.1007/978-3-030-01234-2_49

    work page doi:10.1007/978-3-030-01234-2_49 2018

  34. [34]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. VanderPlas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, É. Duchesnay, Scikit-learn: Machine learning in python, J. Mach. Learn. Res. 12 (2011) 2825–2830

    2011

  35. [35]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Köpf, E. Z. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, Pytorch: An imperative style, high-performance deep learning library, in: Adv. Neural Inf. Process. Syst., Vol. 32, 2019

    2019

  36. [36]

    Iakubovskii, Segmentation models pytorch, accessed: 2026-02-24 (2019)

    P. Iakubovskii, Segmentation models pytorch, accessed: 2026-02-24 (2019). URLhttps://github.com/qubvel/segmentation_models.pytorch

    2026

  37. [37]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, Imagenet: A large-scale hierarchical image database, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2009, pp. 248–255.doi:10.1109/CVPR.2009.5206848

    work page doi:10.1109/cvpr.2009.5206848 2009

  38. [38]

    K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV, USA, 2016, pp. 770–778. doi:10.1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  39. [39]

    M. Tan, Q. V. Le, Efficientnet: Rethinking model scaling for convolutional neural networks, in: Proc. 36th Int. Conf. Mach. Learn. (ICML), Long Beach, CA, USA, 2019, pp. 6105–6114. URLhttp://proceedings.mlr.press/v97/tan19a.html

    2019

  40. [40]

    Z. Liu, H. Mao, C.-Y. Wu, C. Feichtenhofer, T. Darrell, S. Xie, A convnet for the 2020s, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2022, pp. 11976–11986

    2022

  41. [41]

    W. Luo, Y. Li, R. Urtasun, R. Zemel, Understanding the effective receptive field in deep convolutional neural networks, in: Adv. Neural Inf. Process. Syst., Vol. 29, 2016, pp. 4898–4906

    2016

  42. [42]

    B. J. Kim, H. Choi, H. Jang, D. G. Lee, W. Jeong, S. W. Kim, Dead pixel test using effective receptive field, Pattern Recognit. Lett. 167 (2023) 149–156.doi:10.1016/j.patrec.2023.02 .018

    work page doi:10.1016/j.patrec.2023.02 2023

  43. [43]

    M. L. Richter, C. Pal, Receptive field refinement for convolutional neural networks reliably improves predictive performance, arXiv preprintarXiv:2211.14487. 41

    arXiv

  44. [44]

    Agrafiotis, B

    P. Agrafiotis, B. Demir, Deep learning-based bathymetry retrieval without in-situ depths using remote sensing imagery and sfm-mvs dsms with data gaps, ISPRS J. Photogramm. Remote Sens. 225 (2025) 341–361.doi:10.1016/j.isprsjprs.2025.04.020

    work page doi:10.1016/j.isprsjprs.2025.04.020 2025

  45. [45]

    S. Sun, Y. Chen, L. Mu, Y. Le, H. Zhao, Improving shallow water bathymetry inversion through nonlinear transformation and deep convolutional neural networks, Remote Sens. 15 (17) (2023) 4247.doi:10.3390/rs15174247

    work page doi:10.3390/rs15174247 2023

  46. [46]

    Breiman, Random Forests, Machine Learning 45 (1) (2001) 5–32.doi:10.1023/A:101093 3404324

    L. Breiman, Random Forests, Machine Learning 45 (1) (2001) 5–32.doi:10.1023/A:101093 3404324

    work page doi:10.1023/a:101093 2001

  47. [47]

    Hastie, R

    T. Hastie, R. Tibshirani, J. Friedman, The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd ed., Springer, 2009.URL:https://hastie.su.domains/ElemSt atLearn/

    2009

  48. [48]

    Bonamutial, S

    M. Bonamutial, S. Y. Prasetyo, Exploring the Impact of Feature Data Normalization and Standardization on Regression Models for Smartphone Price Prediction, in: 2023 International Conference on Information Management and Technology (ICIMTech), IEEE, 2023, pp. 294–298. doi:10.1109/ICIMTech59029.2023.10277860

    work page doi:10.1109/icimtech59029.2023.10277860 2023

  49. [49]

    J. D. Hedley, A. R. Harborne, P. J. Mumby, Simple and robust removal of sun glint for mapping shallow-water benthos, International Journal of Remote Sensing 26 (10) (2005) 2107–2112. doi:10.1080/01431160500034086

    work page doi:10.1080/01431160500034086 2005

  50. [50]

    J. D. Hedley, C. Roelfsema, V. Brando, C. Giardino, T. Kutser, S. Phinn, P. J. Mumby, O. Bar- rilero, J. Laporte, B. Koetz, Coral reef applications of Sentinel-2: Coverage, characteristics, bathymetry and benthic mapping with comparison to Landsat 8, Remote Sens. Environ. 216 (2018) 598–614.doi:10.1016/j.rse.2018.07.014

    work page doi:10.1016/j.rse.2018.07.014 2018

  51. [51]

    Santos, R

    R. Santos, R. Quartau, L. Lamas, R. Cândido, J. Pinto, Application of Neural Networks for Bathymetric Estimation Using a short time-series of Sentinel-2. A Case Study: Setúbal–Troia, Portugal, Remote Sens. Appl.: Soc. Environ. 39 (2025) 101683.doi:10.1016/j.rsase.2025 .101683

    work page doi:10.1016/j.rsase.2025 2025

  52. [52]

    C. Liu, H. Xie, K. Luan, Q. Xu, Y. Sun, M. Ji, X. Tong, Generalized satellite-derived bathymetry across spatial and temporal domains: a domain-adaptive deep learning approach with multi-source remote sensing data, ISPRS J. Photogramm. Remote Sens. 230 (2025) 452–468.doi:10.1016/j.isprsjprs.2025.09.021

    work page doi:10.1016/j.isprsjprs.2025.09.021 2025

  53. [53]

    Roy Chowdhury, A

    S. Roy Chowdhury, A. Radhakrishnan, H. J. Hsu, H. Subramoni, J. Moortgat, From bands to depth: Understanding bathymetry decisions on Sentinel-2, arXiv preprint arXiv:2601.12636 (2026).arXiv:2601.12636

    arXiv 2026

  54. [54]

    C. E. Parrish, L. A. Magruder, A. L. Neuenschwander, N. Forfinski-Sarkozi, M. Alonzo, M. Jasinski, Validation of ICESat-2 ATLAS bathymetry and analysis of ATLAS bathymetric mapping performance, Remote Sens. 11 (14) (2019) 1634.doi:10.3390/rs11141634. 42

    work page doi:10.3390/rs11141634 2019