Pith. sign in

REVIEW 5 major objections 5 minor 21 references

Eelgrass beds and oyster farming at a lagoon before and after the Great East Japan Earthquake 2011: potential to apply deep learning at a coastal area

T0 review · 5 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper aims to show that an image-to-image translation model (pix2pix) maps seagrass and sand from aerial photos more accurately and at higher resolution than a fully convolutional network, enough to track before-and-after change from…

desk verdict A useful, honest proof-of-concept for deep-learning coastal mapping, but the pix2pix-vs-FCN comparison rests on training accuracy and one test image, so it is suggestive rather than established. read the letter →

arxiv 1909.02747 v1 pith:J3OVODD2 submitted 2019-09-06 eess.IV cs.CVcs.LGstat.ML

Takehisa Yamakita This is my paper · ORCID
classification eess.IVcs.CVcs.LGstat.ML
keywords GreatEastJapanEarthquake2011tsunamiseagrassbedsoysterfarmingraftsdeeplearningpix2pixfullyconvolutionalnetworkaerialimageclassification
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that deep-learning image translation can map shallow coastal habitats from aerial photographs quickly enough to support disaster response. Its test case is Mangoku-ura Lagoon, where aerial images are available from 2008, before the Great East Japan Earthquake, and from June 2011, just after the tsunami. After training a pix2pix image-to-image translation model on the 2011 image and its manual classification, the study applies it to the independent 2008 image and reports 67% overall accuracy and roughly 69% producer's accuracy for vegetation (the fraction of manually mapped vegetation the model recovered), compared with 58% overall accuracy for a fully convolutional network. The same comparison is used to argue that seagrass and oyster-farming areas decreased while sand increased, matching the expected tsunami effects. The contribution is evidence that this class of model can extract ecologically meaningful spatial patterns from historical aerial photos in a coastal disaster context.

What carries the argument

The load-bearing mechanism is pix2pix, an image-to-image translation model built on a deep convolutional generative adversarial network: a generator learns to turn a $256 \times 256$ aerial image patch into a classified land-cover patch while a discriminator tries to tell the output apart from the manually traced supervision. The adversarial competition forces the generator to reproduce not just pixel colors but textures, shapes, and contrasts, which is what lets it recognize underwater seagrass and the outlines of oyster rafts. The comparison model is a fully convolutional network based on a standard image-classification convolutional architecture, which classifies with lower spatial resolution. Slicing images into $256 \times 256$ pixel grids, training on the 2011 image pair, and merging the predicted slices back into the full scene is the procedure that carries the independent 2008 test.

What would settle it

Trace independent ground-truth maps for a second pre-tsunami aerial image (or another coastal lagoon), run the same 2011-trained pix2pix model without retuning or color-matching, and compare; if overall accuracy falls to roughly the FCN level or vegetation producer's accuracy drops well below the reported 69%, the claim of general potential is not supported.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that pix2pix, a conditional adversarial image-to-image translation model, outperforms a fully convolutional network at extracting seagrass beds and sandy areas from aerial images of a coastal lagoon. Trained on the 2011 post-tsunami image and tested on the 2008 pre-tsunami image, pix2pix reaches 67% overall accuracy and roughly 69% producer's accuracy for total vegetation, while the FCN reaches 58% overall; pix2pix also preserves finer spatial detail and outlines of oyster farming rafts, although its per-pixel labeling of the rafts is less reliable than the FCN's. The paper uses the resulting maps to show increases of sand and decreases of dense seagrass and oyster rafts after the earthquake and tsunami, consistent with qualitative regional surveys.

Load-bearing premise

The whole demonstration rests on treating the manually traced 2008 classification as correct ground truth and on the single color-adjusted 2008 image being a fair test of how the model will perform elsewhere; if either assumption gives way, the reported accuracy and the conclusion about general potential do not transfer.

Editorial extensions

If this is right

  • A model trained on one post-disaster aerial image can be applied to an older image of the same site, producing a before-and-after map without new manual tracing for the older date.
  • pix2pix's vegetation maps on independent data exceed 60% producer's accuracy, so the method could support rapid assessments of seagrass extent after future tsunamis.
  • Because pix2pix keeps finer spatial detail than the FCN, its output can be post-processed to isolate oyster-raft shapes, a route the paper explicitly suggests.
  • The measured area changes—sand increase, dense seagrass decrease, and oyster-raft decrease—match qualitative regional surveys, so the model output can be read ecologically rather than as pure image statistics.
  • For classes like oyster rafts, the FCN still gives better whole-area classification, so a practical pipeline may combine the two models rather than choose one.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's 2008 test is a single image whose colors were adjusted to match the 2011 image; if the model were tested on a second pre-disaster image without color normalization, the true generality of its texture-based extraction could be measured.
  • The same pix2pix setup may transfer to other lagoons on the same coastline, but only one site is demonstrated here, so cross-site transfer remains an extrapolation.
  • Because marine debris was included in the manual labels but its accuracy is low, a testable extension is to retrain with more debris samples or to treat debris as residual anomalies in the pix2pix output rather than as a direct class.
  • A stronger test of the disaster-response value would quantify the time to map a new image automatically compared with manual tracing, which the paper asserts qualitatively but does not measure.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The manuscript presents a case study comparing fully convolutional networks (FCN-AlexNet) and pix2pix (DCGAN) for classifying seagrass beds, sand, oyster farming rafts, and debris in aerial images of Mangoku-ura Lagoon before and after the 2011 Great East Japan Earthquake. Models are trained on a 2011 aerial image with manual labels and tested on an independent 2008 image. The paper reports overall accuracy of 67% for pix2pix and 58% for FCN on the 2008 test image, and uses the model outputs to estimate land-cover change between 2008 and 2011. The conclusion is that pix2pix achieves higher accuracy for vegetation extraction and better output resolution, while FCN is better for oyster raft categorization.

Significance. The case study addresses an important application—rapid coastal habitat and infrastructure mapping after a tsunami—and provides one of the few deep learning comparisons on marine remote sensing data. The inclusion of an independent 2008 test image and the explicit statement that the 2011 rows in Table I are training-data agreement are commendable and show awareness of validation pitfalls. If the results were supported by multiple independent test scenes and error bars, the claim that image-to-image translation (pix2pix) is preferable for vegetation extraction would be a useful contribution to coastal remote sensing. As it stands, the evidence is suggestive but not yet convincing at the level claimed.

major comments (5)
  1. [Section 2, Table I] The 2011 accuracy rows are explicitly computed against the supervised data used to produce the model, so they measure training-set agreement and cannot by themselves demonstrate that pix2pix generalizes better than FCN. The generalization comparison therefore rests on a single independent 2008 image, with no repeated experiments or statistical test of the pix2pix-FCN difference. Please provide additional independent test images or at least bootstrap confidence intervals for the accuracies, and restrict the superiority claim to the holdout setting.
  2. [Section 3, Table I] The abstract's 'more than 69% accuracy for vegetation classification' refers to the producer's accuracy for the aggregated total vegetation class on the 2008 image, but the same table reports only 20% producer's accuracy for sparse vegetation and 3% for oyster rafts under pix2pix (13% and 14% for FCN). This class-specific breakdown contradicts the unqualified phrase 'vegetation classification,' so the claims should be narrowed to dense/total vegetation and the poor performance on sparse vegetation and oyster rafts should be acknowledged in the conclusion.
  3. [Section 3, Table II] The before/after change assessment compares model outputs for 2008 and 2011, but the 2011 output is produced by a model trained on the 2011 manual labels. The 2011 area estimates in Table II therefore largely reproduce the training data rather than providing an independent estimate, making the reported sand increase and vegetation decrease partly circular. Please either train and test in both temporal directions with appropriate independent years, or explicitly state that the 2011 model output is a training-data-dependent estimate and interpret the change result accordingly.
  4. [Section 2, accuracy evaluation] The accuracy comparison relies on point estimates computed from more than 100,000 random points on a single test image, but no confidence intervals or standard errors are reported. Without an uncertainty measure, the 9-point overall accuracy advantage of pix2pix over FCN on the 2008 image cannot be distinguished from sampling noise. Please report binomial confidence intervals or standard errors for the reported proportions, and ideally repeat training with different random seeds.
  5. [Section 2, image preprocessing] All images, including the 2008 test image, were 'adjusted to the same color level of 2011 imagery' prior to analysis. This preprocessing may remove genuine spectral differences between the pre- and post-disaster scenes and makes the test condition dependent on the later image's color statistics; the sensitivity of the conclusions to this choice is not investigated. Please justify or test the color-adjustment step, for example by also evaluating on the original 2008 radiometry.
minor comments (5)
  1. [Abstract] The abstract contains formatting artifacts such as 'ABSTR ACT' and 'Inde x Terms' that should be corrected in the final version.
  2. [Table I] The header 'Total vegitation' contains a typo, and 'Rand accuracy' is a nonstandard abbreviation; please use 'random-point accuracy' or define the term explicitly.
  3. [Section 2] The description of the supervised data generation is brief; please specify how many class categories were used, how the manual tracing was performed, and whether field survey data informed the labeling.
  4. [Section 2] The train/test split is described only as 'the remaining aerial image was used as test data'; please clarify which portion of the image was used for training and which for testing, and how many 256x256 patches were generated.
  5. [Section 3] The statement 'the areas of oyster farming were correctly classified in the case of the test data 2008 imagery by the FCN model' appears inconsistent with Table I, which reports only 14% producer's accuracy for oyster rafts for FCN on the 2008 image; please clarify the intended meaning.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor circularity: training-set agreement on 2011 imagery is reported as comparative accuracy, but the central 69% claim rests on an independent 2008 holdout.

  1. fitted input called prediction [Section 3, Results, Table I and first paragraph (2011 rows)]
    "In the rows labeled “2011”, the result indicates the agreement with the model output against the supervised data used to produce the model. There was 80 % accuracy overall and more than 6 0 % agreement in the categories of sand , dense vegetation, sparse vegetation and the total of both vegetation in both models."

    The 2011 rows compare model outputs with the exact manual labels on which both models were trained. These are training-fit accuracies, not independent predictions, so using them in Table I and in the Results paragraph to compare pix2pix and FCN cannot by itself establish that pix2pix generalizes better. The paper does explicitly disclose this, and the subsequent 2008 rows are genuinely independent, so the circularity is limited to the way the 2011 training agreement is included as evidence in the model comparison.

full rationale

The central numerical claim in the abstract and conclusion — more than 69% vegetation-producer accuracy on independent test data — is based on the 2008 image, which was not used to train either model. That evaluation is not circular: the manual 2008 classification is a separate holdout label set, and the models were trained only on 2011 imagery and supervision. The main circularity concern is the 2011 rows of Table I, which are agreements against the same supervised data used to produce the models; the paper openly labels them that way, so the flaw is transparency-limited rather than hidden. There is also no load-bearing self-citation chain: the cited prior seagrass work [4] motivates the method and the manual tracing protocol, but the present conclusions are evaluated against the independent 2008 image and standard architectures (FCN, pix2pix). The single-test-image and color-adjustment limitations are statistical robustness concerns, not circularity. Overall, the finding is a minor, acknowledged training-accuracy issue rather than a derivation that reduces to its inputs, so a score of 2 is appropriate.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the reliability of manual labels, on the ability of a 2011-trained model to generalize to a color-adjusted 2008 image, and on the choice of default network architectures. No new physical entities are introduced. Model weights could be seen as free parameters, but their values are not reported, so the audit records the absence of that information.

assumptions (3)
  • domain assumption Manual tracing of aerial images provides an accurate ground truth for seagrass, sand, oyster rafts, and debris.
    The paper uses manual classification to train the models and to evaluate accuracy (Section 2, Table I). If the manual labels are noisy or biased, all accuracy numbers are affected. The paper does not quantify inter-observer variability.
  • domain assumption A model trained on 2011 imagery can generalize to 2008 imagery after color normalization.
    The models are trained on the 2011 image and applied to the 2008 image, which was resampled and color-adjusted to match 2011. The paper does not test robustness to different image acquisition conditions.
  • domain assumption The chosen network architectures (FCN-AlexNet and pix2pix) are appropriate for this segmentation task without additional tuning.
    The paper uses 'standard network structures' without reporting hyperparameter optimization, so the comparison depends on these default configurations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Eelgrass beds and oyster farming at a lagoon before and after the Great East Japan Earthquake 2011: potential to apply deep learning at a coastal area." pith.science (2026). https://pith.science/paper/J3OVODD2

@misc{pith2026190902747,
  author       = {Pith},
  title        = {Pith review of: Eelgrass beds and oyster farming at a lagoon before and after the Great East Japan Earthquake 2011: potential to apply deep learning at a coastal area},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J3OVODD2}},
  note         = {Machine review of arXiv:1909.02747}
}
read the original abstract

There is a small number of case studies of automatic land cover classification on the coastal area. Here, I test extraction of seagrass beds, sandy area, oyster farming rafts at Mangoku-ura Lagoon, Miyagi, Japan by comparing manual tracing, simple image segmentation, and image transformation using deep learning. The result was used to extract the changes before and after the earthquake and tsunami. The output resolution was best in the image transformation method, which showed more than 69% accuracy for vegetation classification by an assessment using random points on independent test data. The distribution of oyster farming rafts was detected by the segmentation model. Assessment of the change before and after the earthquake by the manual tracing and image transformation result revealed increase of sand area and decrease of the vegetation. By the segmentation model only the decrease of the oyster farming was detected. These results demonstrate the potential to extract the spatial pattern of these elements after an earthquake and tsunami. Index Terms: Great East Japan Earthquake of 2011, Land use land cover (LULC), Zosteracea seagrass, cultured oyster, deep learning, Mangoku Bay

Figures

Figures reproduced from arXiv: 1909.02747 by the authors.

Figure 1
Figure 1. Original images and classified images of the Mangoku-ura Lagoon, Miyagi, Japan. Images taken in 2008 and 2011 were used to represent before and after the earthquake. The pair of images in the orange rectangle was used to build the models using fully convolutional network (FCN) and pix2pix which implement Deep Convolutional Generative Adversarial Network (DCGAN). Earthquake and subsequent tsunami in 2011. I especiall… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 19 canonical work pages

  1. [1]

    Automated analysis using deep learning is expected to allow faster, effortles s, an d more accurate analysis of images [1]

    INTRO DUCTION In recent years, aerial images have become available at finer resolution and lower prices, leading to new applications of remote sensing to perform object detection and recognition that are beyond pi xel color -based classification and extraction. Automated analysis using deep learning is expected to allow faster, effortles s, an d more accu...

    work page 2011

  2. [2]

    fully convolutional networks

    METHODS As the study site, I selected Mangoku -ura Lagoon, near the Oshika Peninsula, within the re gion damaged by the Great East Japan Earthquake. The lag oon, which is one of the largest enclosed bodies o f water in this region, includes a wide distribution of seagrass bed and oyster farming areas [9]–[11]. Because of the ear thquake, this area was aff...

    work page 2008

  3. [3]

    Automatically identifying, counting, and describing wild animals in cam era-trap images with deep learning,

    M. S. Norouzzadeh et al., “Automatically identifying, counting, and describing wild animals in cam era-trap images with deep learning,” Proc. Natl. Acad. Sci., p. 201719367, 2018

    work page 2018

  4. [4]

    The comparison of the two techniques revealed that the pix2pix technique produces higher accuracy for the vegetation extraction

    CONCLUSION In this study, we proposed methods for extracting seagrass beds and oyster farming rafts from aerial photographs using automatic image analysis techniques applying deep learning. The comparison of the two techniques revealed that the pix2pix technique produces higher accuracy for the vegetation extraction . The resolution of the result image is...

    work page

  5. [5]

    I also would like to thank Tomonori Mita ( SKYMAP Co., Ltd.) and Fumiaki Sodeyama ( JAMSTEC) for tec hnical assistance

    ACKNOWLEDGMENT I thank Fisherman’s Association of Miyagi Prefecture, Miyagi Prefecture Fishery Technology Institute , Masakazu Hori, Daisuke Muraoka, Goro Yoshida, Hitoshi Tamaki, and other members of the F isheries Research and Education Agency ( FRA) for the opportunity of field observati on. I also would like to thank Tomonori Mita ( SKYMAP Co., Ltd.) ...

    work page

  6. [6]

    Deep learning: adaptive computation and machine learning,

    I. J. Goodfellow, Y. Beng io, and A. Courville, “ Deep learning: adaptive computation and machine learning,” MIT Press, 2016

    work page 2016

  7. [7]

    Immediate ecological impacts o f the 2011 Tohoku earthquake tsu nami on intertidal flat communities.,

    J. Urabe, T. Suzuki, T. Nishita, and W. Maki no, “Immediate ecological impacts o f the 2011 Tohoku earthquake tsu nami on intertidal flat communities.,” PLoS One, vol. 8, no. 5, p. e62779, 2013

    work page 2011

  8. [8]

    Application of deep learning techniques for determining the aerial extent and classification of seag rass beds, Trang, Thailand,

    T. Yamakita, F. Sodeyama, N. Whanpetch, K. Watanabe, and M. Nakaoka, “Application of deep learning techniques for determining the aerial extent and classification of seag rass beds, Trang, Thailand,” Bot. Mar ., vol. 62, no. 4 , in press, 2019. DOI: 10.1515/bot-2018-0017

    doi:10.1515/bot-2018-0017 2019

Show all 21 references
  1. [9]

    ImageNet Large Scale Visual Recognition Challenge,

    O. Russakovsky et al ., “ImageNet Large Scale Visual Recognition Challenge,” Int. J. Comput. Vis ., vol. 115, no. 3, pp. 211–252, 2015

    2015

  2. [10]

    T. Yamakita et al., “Image dataset of ophiuroid and other d eep sea benthic organisms in 2015 e xtracted from the survey off Sanriku, Japan , by the research following the Great East Japan Earthquake 2011,” Ecol. Res., vol. 33, no. 2, pp. 285–285, 2018

    2015

  3. [11]

    Asynchrono us local dynamics contributes to sta bility of a seagrass bed in Tokyo Bay,

    T. Yamakita, K. Watanabe, and M. Nakaoka, “Asynchrono us local dynamics contributes to sta bility of a seagrass bed in Tokyo Bay,” Ecography, vol. 34, no. 3, pp. 519–528, 2011

    2011

  4. [12]

    Scale dependency in seagrass dynamics: how does the neighboring effect vary with grain of observation?,

    T. Yamakita and M. Nakaoka, “Scale dependency in seagrass dynamics: how does the neighboring effect vary with grain of observation?,” Popul. Ecol., vol. 51, no. 1, pp. 33–40, 2009

    2009

  5. [13]

    Shoji and M

    J. Shoji and M. Morimoto, “Changes in fish community in seagrass beds in Mangoku -ura Bay from 2009 to 2014, the period before and after the tsunami following the 2011 off the Pacific coast of Tohoku earthquake,” J. Oceanogr., vol. 72, no. 1, pp. 91– 98, 2016

    2009

  6. [14]

    Survey Report on Vegetation and Marine Area in Pacific Coast Side of Tohoku Region in FY2015

    Biodiversity Center of Japan, Ministry of the Environment , “Survey Report on Vegetation and Marine Area in Pacific Coast Side of Tohoku Region in FY2015 ” Ministry of the Environment , p.281, 2016

    2016

  7. [15]

    Temporal changes in seagrass beds of Sanriku Coast before and after the Great East Japan Earthquake,

    M. Nakaoka, H. Tam aki, D. Muraoka, M. Tokuoka, T. Komatsu, and N. Tanaka, “Temporal changes in seagrass beds of Sanriku Coast before and after the Great East Japan Earthquake,” Nippon Suisan Gakkaishi, vol. 83, no. 4, pp. 659–663, 2017

    2017

  8. [16]

    Change of the ocean after the Great East Japan Earthquake –Utilization of geographic inform ation system and GIScience,

    T. Yamakita, “Change of the ocean after the Great East Japan Earthquake –Utilization of geographic inform ation system and GIScience,” E-Research Species Biol., vol. 2, no. 1, in press, 2018

    2018

  9. [17]

    The effects o f liquefaction, tsunami, and land subsidence on intertidal mollus ks following the Great East Jap an Earthquake,

    K. Okoshi, “The effects o f liquefaction, tsunami, and land subsidence on intertidal mollus ks following the Great East Jap an Earthquake,” in Ecological Impacts of Ts unamis on Coastal Ecosystems, pp. 165–178, Springer, 2016

    2016

  10. [18]

    Imag eNet classification with deep convolutional neura l networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imag eNet classification with deep convolutional neura l networks,” in Proceedings of the 25th International Conference on Neural Information Processing Systems, pp. 1097–1105 IEEE, 2012

    2012

  11. [19]

    New techniques for standardization of environmental impact assessment,

    Y. Furushima, T. Yamakita, T. M iwa, D. Lindsay, T. Fukushima, and Y. Shir ayama, “New techniques for standardization of environmental impact assessment,” in Environmental Issues of Deep-Sea Mining: Impacts, Consequences and Policy Perspectives , R. Sharma, Ed. p p.275-313, Sp...

    2019

  12. [20]

    Fully conv olutional networks for semantic segmentat ion,

    J. Long, E. Shelhamer, and T. Darrell, “Fully conv olutional networks for semantic segmentat ion,” in Proceedings of the I EEE Conference on Computer Vision and Patte rn Recognition, pp. 3431–3440, IEEE , 2015

    2015

  13. [21]

    Image -to- image translation with conditional adversa rial networks,

    P. Isola, J. -Y. Zhu, T. Zhou, and A. A. Efros, “Image -to- image translation with conditional adversa rial networks,” arXiv Prepr. arXiv1611.07004, 2016

    2016 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.