Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 126 of the 126 references it cites were found in a registry or an index; none of the 105 sources its claims cite is checked for support yet.

A Review on Bathymetric Inversion Research Based on Deep Learning Models and Remote Sensing Images

Open access
Liu, Delong, Shi, Yufeng, Fang, Hong · 2026 · Remote Sensing

Abstract

High-precision inversion of shallow-water depth is crucial to marine resource development, ecological protection, and national defense security. Traditional acoustic detection, LiDAR, and empirical models are limited by high cost, low efficiency, or water quality dependence, struggling to meet people’s growing demand for shallow-water depth. With the rapid development of theories and technologies such as remote sensing information, computer science, and artificial intelligence, bathymetric inversion based on remote sensing images and deep learning models has become a research hotspot. In this study, journal articles and conference papers were searched in the Web of Science (WOS) and Google Scholar databases using keywords such as “remote sensing image”, “bathymetry”, and “deep learning model”. The publication time of the papers ranges from January 2021 to September 2025. A total of 309 relevant studies were retrieved and, after screening and quality control, 132 core studies were finally selected as the research objects for this review. These studies were classified according to deep learning models, including CNN, U-Net, MLP, and RNN. The study analyzed and summarized the characteristics of different deep learning models in bathymetric inversion, as well as their data source selection, inversion accuracy, and limitations. Additionally, the future development trends were discussed in combination with the latest research results.

Overview

CoChat Summary

TL;DR

This review synthesizes 132 core studies (2021–2025) on deep learning-based satellite-derived bathymetry (SDB), organizing them by architecture family (CNN, U-Net, MLP, RNN, Transformer) and comparing their reported performance. It concludes that deep learning substantially outperforms traditional empirical SDB models but remains constrained by scarce in situ depth labels, water clarity, bottom type, sensor characteristics, and cross-region/cross-sensor domain shift.

Key findings

  • Deep learning improves shallow-water bathymetry over traditional empirical approaches, but performance is "strongly constrained by water clarity, bottom types, sensor characteristics, and training-label quality"; the dominant bottlenecks are limited in situ depth labels and domain shift across regions, sensors, and turbidity.
  • Reported accuracies vary widely across the reviewed models: e.g., Bathymetry Transformer RMSE = 0.375 m (0–12 m, clear coral reef), BOA-CNN-BiLSTM RMSE 0.10–0.38 m (0–28 m), PI-CNN RMSE 1.39–1.56 m with R² > 0.95 (0–40 m), GRU RMSE = 3.69 m with R² = 0.88 (0–32 m), and RT-CNN RMSE = 1.46 m with R² = 0.91 (0–15 m, turbid inland river).
  • Architecture-specific strengths: CNNs and the U-Net family excel at multi-scale feature extraction and pixel-level depth mapping; MLPs and SVMs remain strong baselines under small-sample and shallow-water conditions; RNNs (LSTM, BiGRU) model multi-temporal dynamics; Transformers enhance large-area spatial consistency and cross-domain generalization.
  • The review identifies future directions: physics–data synergy, uncertainty quantification (epistemic vs. aleatoric), inland water bathymetry, low-data regimes with transfer learning and self-supervision, and computational efficiency for edge deployment.

Methods

  • Systematic literature review using Web of Science and Google Scholar, combining controlled subject terms with free-text keywords (deep learning/CNN/U-Net/Transformer; bathymetry/water depth/depth inversion; remote sensing/Sentinel-2/Landsat), plus snowballing from reference lists.
  • Screening: an initial retrieval of 540 publications was refined to 309 records explicitly employing deep learning; after duplicate removal and multi-stage screening against predefined inclusion/exclusion criteria, 132 core studies were retained for in-depth analysis. The review focuses on 2021–2025.
  • Synthesis is qualitative and comparative: studies are categorized by model architecture, and two summary tables are compiled — Table 1 (research highlights and limitations per model) and Table 2 (quantitative comparison by water type, dataset/labels, generalization test, RMSE/R², and depth range).

Limitations

  • Stated by the paper: direct quantitative comparison across studies is problematic because datasets, study regions, water optical conditions, depth ranges, sensor characteristics, training sample sizes, and validation strategies differ substantially; the authors caution that "absolute RMSE values do not necessarily reflect intrinsic model superiority" and that Table 2 is intended as a structured overview rather than a ranking.
  • Stated by the paper: systematic benchmarks evaluating transfer performance across cross-regional and cross-sensor scenarios remain limited, making robustness gains hard to quantify; standardized inland bathymetric datasets are scarce, and most benchmarks concentrate on coastal or reef environments.
  • Evident from the text: the review is a narrative synthesis without a meta-analysis or pooled effect size, and its conclusions rest on self-reported metrics from heterogeneous primary studies; several entries in Table 2 are marked "N (no cross-scene testing)," so generalization claims are unevenly supported across the surveyed models.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/rs18050720

Cites

126 of 126 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Statistics of the number of published papers from 2001 to 2025.
Figure 1 Statistics of the number of published papers from 2001 to 2025.CC-BY · From the PDF
Figure 2 PRISMA-style flow diagram illustrating the literature search and screening process adopted in this review.
Figure 2 PRISMA-style flow diagram illustrating the literature search and screening process adopted in this review.CC-BY · From the PDF
Figure 3 Keyword co-occurrence graph. The different colored lines represent automatically generated cluster relationships in the co-occurrence network.
Figure 3 Keyword co-occurrence graph. The different colored lines represent automatically generated cluster relationships in the co-occurrence network.CC-BY · From the PDF
Figure 4 Workflow of deriving water depth from remote sensing imagery using deep learning models, from remote sensing data to final bathymetric output.
Figure 4 Workflow of deriving water depth from remote sensing imagery using deep learning models, from remote sensing data to final bathymetric output.CC-BY · From the PDF
Figure 5 Architecture of the CNN model.
Figure 5 Architecture of the CNN model.CC-BY · From the PDF
Figure 6 Architecture of the U-net model.
Figure 6 Architecture of the U-net model.CC-BY · From the PDF
Figure 7 Architecture of the MLP mode.
Figure 7 Architecture of the MLP mode.CC-BY · From the PDF
Figure 8 Architecture of the LSTM/GRU-based model for water depth inversion: ( A ) internal structure of an LSTM unit; ( B ) internal structure of a GRU.
Figure 8 Architecture of the LSTM/GRU-based model for water depth inversion: ( A ) internal structure of an LSTM unit; ( B ) internal structure of a GRU.CC-BY · From the PDF

Verification

Claims the paper makes

0 Valid · 7 Not valid · 82 Not checked · 25 Own results and interpretations

43 more claims

Create a free account to see all 114 claims and to verify the 105 references still open.

See all claims — free account
  • It may state two separate claims as one. 50
  • Probably not a substantial claim. 21

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Research works printed but not linked: 5

  1. Annan, 2024, Refined bathymetric prediction based on feature extraction of gravity field signals: BATHY-FE, J. Geophys. Res. Mach. Learn. Comput., 1, e2024JH000205
  2. Hu, 2024, Water depth extraction of ICESat-2 and application to bathymetric inversion, Chin. J. Geophys., 67, 997
  3. Vaswani, 2017, Attention is all you need, Adv. Neural Inf. Process. Syst., 30, 5998
  4. Huang, 2025, Deep Learning Method Suitable for Airborne Laser Bathymetry of Different Water Qualities, Chin. J. Lasers, 52, 0110003
  5. Shiri, F.M., Perumal, T., Mustapha, N., and Mohamed, R. (2023). A comprehensive overview and comparative analysis on deep learning models: CNN, RNN, LSTM, GRU. arXiv.

Cited by

See all 4

Metadata

Authors
Liu, D., Shi, Y., & Fang, H.
Affiliations
Nanjing Forestry University (CN)
Corresponding author
Shi, Yufeng
Venue
Remote Sensing
Publisher
MDPI AG
Published
2026-02-27
Volume
18
Issue
5
Pages
720
ISSN
2072-4292 (ISSN-L)
OpenAlex
W7131819416
Cited by
4
References
126
Citations by year
2026: 4
FWCI
8.09
Type
journal article
Language
en
Open access
Locations
  • Remote Sensing (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
High-level Scientific Research Foundation of Nanjing Forestry University (163050254)
Topics
  • Physical Sciences › Earth and Planetary Sciences › Oceanography › Underwater Acoustics Research
  • Physical Sciences › Earth and Planetary Sciences › Earth-Surface Processes › Coastal and Marine Dynamics
  • Life Sciences › Agricultural and Biological Sciences › Forestry › Remote Sensing and LiDAR Applications
Keywords
bathymetric inversion, shallow water depth, remote sensing images, deep learning models, convolutional neural networks, U-Net, recurrent neural networks, marine resource development

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-19, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-19 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.