pith. sign in

arxiv: 2606.23615 · v1 · pith:T2C36J6Tnew · submitted 2026-06-22 · 💻 cs.CV · cs.LG

Hedgementation = Hedgerow Segmentation: A Remote Sensing Benchmark

Nathan Senyard , Salem Hamdani , Astrid Zhang , Derek Wang , Evan Shelhamer , Mathias L\'ecuyer , Jos\'ephine Gantois This is my paper

Pith reviewed 2026-06-26 09:18 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords hedgerow segmentationremote sensingbenchmark datasetmachine learninggeneralizationagricultural mappingself-supervised learningFrance
0
0 comments X

The pith

Hedgementation is a benchmark that tests machine learning models for mapping hedgerows from remote sensing data at country scale and 10 square meter resolution.

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

The paper introduces Hedgementation as a new benchmark dataset for evaluating how well machine learning models perform hedgerow segmentation on remote sensing imagery. It harmonizes multiple remote sensing products with ground truth labels drawn from a French national hedgerow inventory to enable tests at national scale. The evaluation focuses on model generalization across spatial distances and climatic zones, which is presented as a harder test than standard random splits. Both supervised and self-supervised approaches are assessed on this task of tracking fine-scale agricultural landscape features. A reader would care because accurate large-scale hedgerow maps can inform biodiversity and agricultural policy.

Core claim

We propose Hedgementation as a benchmark that combines harmonized remote sensing data products and ground truth labels from the French hedgerow inventory to evaluate machine learning models for hedgerow mapping at country scale and 10m² spatial resolution, measuring generalization across spatial distance and climatic zones for both supervised and self-supervised learning.

What carries the argument

The Hedgementation benchmark, a harmonized dataset and evaluation protocol built from French remote sensing products and hedgerow inventory labels for testing segmentation models.

If this is right

  • Models can be ranked by their ability to generalize across space and climate rather than just within similar regions.
  • Supervised and self-supervised methods become directly comparable on a fine-scale remote sensing task of agricultural importance.
  • The benchmark supplies a reproducible protocol and code for testing new models at national mapping scales.
  • Performance on the benchmark indicates suitability for tracking linear landscape features relevant to policy and monitoring.

Where Pith is reading between the lines

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

  • Models that succeed on this benchmark may transfer better to mapping similar narrow features in other countries once labels are available.
  • Self-supervised pretraining on the remote sensing data could reduce reliance on expensive ground labels for related segmentation tasks.
  • The dataset structure could be adapted to test generalization for other agricultural or ecological features beyond hedgerows.
  • If climatic zone splits prove harder than spatial splits, future work might prioritize climate-aware data augmentation techniques.

Load-bearing premise

The French hedgerow inventory labels and harmonized remote sensing data are accurate and representative enough to support reliable tests of model generalization.

What would settle it

An independent test set of hedgerow maps from outside France or from a different sensor showing that top-performing models on the benchmark drop sharply in accuracy would falsify reliable generalization testing.

Figures

Figures reproduced from arXiv: 2606.23615 by Astrid Zhang, Derek Wang, Evan Shelhamer, Jos\'ephine Gantois, Mathias L\'ecuyer, Nathan Senyard, Salem Hamdani.

Figure 1
Figure 1. Figure 1: Hedgerow label rasterization process, applied to one [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Dataset creation process: (2a) France is split into large tiles, randomly assigned to folds [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Input data for the FTW model: the model requires two input images taken from two [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative examples of predictions for our best model in each class. We show input RGB [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The construction of the RPG agriculture mask. The mask is shown in yellow, and the 10-m [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
read the original abstract

We propose Hedgementation: a new benchmark to evaluate machine learning models for hedgerow mapping from remote sensing data at country scale and 10m$^2$ spatial resolution. We combine and harmonize multiple remote sensing data products and ground truth labels sourced from a hedgerow inventory in France. We measure the ability of three baseline models to generalize across spatial distance, and across climatic zones, a more explicitly challenging task. Our benchmark tests both supervised and self-supervised learning approaches for remote sensing, applied to tracking fine-scale features of high agricultural importance. The code to reproduce the benchmark and baselines results is available at https://github.com/hedgementation/hedgementation.

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 paper proposes Hedgementation, a new benchmark for evaluating machine learning models on hedgerow segmentation from remote sensing data at country scale and 10 m² spatial resolution. It combines and harmonizes multiple remote sensing products with ground-truth labels from the French hedgerow inventory, then measures generalization of three baseline models (supervised and self-supervised) across spatial distance and climatic zones. Code to reproduce the benchmark and results is released.

Significance. If the harmonized labels prove reliable, the benchmark would be a useful contribution for testing model robustness on fine-scale agricultural features under distribution shift. The explicit open release of code and baselines is a clear strength that supports reproducibility.

major comments (2)
  1. [Dataset construction and harmonization] The central generalization claims rest on the assumption that the French hedgerow inventory labels are accurate and representative at the target 10 m² resolution. No quantitative validation, error rates, or cross-check against independent sources is reported in the data construction description, leaving open the possibility that measured performance differences reflect label noise rather than model capability.
  2. [Experimental setup and results] The evaluation of generalization across climatic zones is presented as more challenging, yet the manuscript provides no details on how climatic zones were delineated, how many zones are represented in the train/test splits, or any ablation showing that performance drops are attributable to climate rather than other covariates.
minor comments (1)
  1. [Abstract and introduction] The notation '10m$^2$ spatial resolution' is ambiguous; standard remote-sensing usage refers to pixel size (e.g., 10 m resolution), not area. Clarify the exact pixel footprint and whether the 10 m² figure is intended as area or a typographical shorthand.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on the Hedgementation benchmark. We address each major comment below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Dataset construction and harmonization] The central generalization claims rest on the assumption that the French hedgerow inventory labels are accurate and representative at the target 10 m² resolution. No quantitative validation, error rates, or cross-check against independent sources is reported in the data construction description, leaving open the possibility that measured performance differences reflect label noise rather than model capability.

    Authors: We acknowledge that the manuscript does not report quantitative validation or error rates for the labels. The hedgerow data originates from the official French national inventory, which is the authoritative reference for this task. In the revised version we will add a limitations subsection describing label provenance, collection methodology, and potential sources of noise at 10 m² resolution, together with a discussion of how such noise could affect measured generalization gaps. We cannot supply new quantitative cross-validation numbers because no independent high-resolution validation dataset was available during benchmark construction. revision: yes

  2. Referee: [Experimental setup and results] The evaluation of generalization across climatic zones is presented as more challenging, yet the manuscript provides no details on how climatic zones were delineated, how many zones are represented in the train/test splits, or any ablation showing that performance drops are attributable to climate rather than other covariates.

    Authors: We agree that the climatic-zone analysis requires more explicit documentation. In the revised manuscript we will expand the experimental-setup section to specify the climate classification source and criteria used for delineation, report the exact number of zones and their representation in the train/test splits, and include an ablation that compares performance drops when controlling for spatial distance versus climate covariates. revision: yes

Circularity Check

0 steps flagged

No circularity: benchmark dataset construction and baseline evaluation contain no derivations or self-referential reductions

full rationale

The paper constructs a remote-sensing benchmark by harmonizing existing French inventory data products and labels, then reports standard supervised and self-supervised baseline performance on generalization splits. No equations, fitted parameters, or predictions appear in the provided text. The central claims rest on the external validity of the source inventory rather than any internal derivation that reduces to the paper's own inputs by construction. No self-citation chains, ansatzes, or uniqueness theorems are invoked. This is the expected non-finding for a dataset/benchmark paper whose contribution is curation and empirical reporting rather than a closed mathematical argument.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the quality and representativeness of the sourced French inventory labels and harmonized remote sensing products; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption The harmonized remote sensing data products and ground truth labels from the French hedgerow inventory are accurate and representative enough to support reliable tests of model generalization across spatial distance and climatic zones.
    Invoked in the abstract description of the benchmark construction and evaluation protocol.

pith-pipeline@v0.9.1-grok · 5662 in / 1185 out tokens · 21641 ms · 2026-06-26T09:18:02.331676+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

53 extracted references · 1 canonical work pages

  1. [1]

    Science of the Total Environment , volume=

    Planting hedgerows: Biomass carbon sequestration and contribution towards net-zero targets , author=. Science of the Total Environment , volume=. 2023 , publisher=

    2023

  2. [2]

    Renewable Agriculture and Food Systems , volume=

    Targeting perennial vegetation in agricultural landscapes for enhancing ecosystem services , author=. Renewable Agriculture and Food Systems , volume=. 2014 , publisher=

    2014

  3. [3]

    Annual Review of Ecology, Evolution, and Systematics , volume=

    Hedgerows as ecosystems: service delivery, management, and restoration , author=. Annual Review of Ecology, Evolution, and Systematics , volume=. 2020 , publisher=

    2020

  4. [4]

    Ecology letters , volume=

    The effectiveness of flower strips and hedgerows on pest control, pollination services and crop yield: a quantitative synthesis , author=. Ecology letters , volume=. 2020 , publisher=

    2020

  5. [5]

    Remote Sensing of Environment , volume=

    Hedgerow mapping with high resolution satellite imagery to support policy initiatives at national level , author=. Remote Sensing of Environment , volume=. 2025 , publisher=

    2025

  6. [6]

    Remote Sensing for Agriculture, Ecosystems, and Hydrology XXI , volume=

    A new Copernicus high resolution layer at pan-European scale: Small woody features , author=. Remote Sensing for Agriculture, Ecosystems, and Hydrology XXI , volume=. 2019 , organization=

    2019

  7. [7]

    Canadian Journal of Remote Sensing , volume=

    Shelterbelt agroforestry systems inventory and removal analyzed by object-based classification of satellite data in Saskatchewan, Canada , author=. Canadian Journal of Remote Sensing , volume=. 2019 , publisher=

    2019

  8. [8]

    Science Advances , volume=

    The overlooked contribution of trees outside forests to tree cover and woody biomass across Europe , author=. Science Advances , volume=. 2023 , publisher=

    2023

  9. [9]

    2024 , note=

    IGN, Institut national de l'information g\'eographique et foresti\`ere , title=. 2024 , note=

    2024

  10. [10]

    2025 , note=

    ESA, European Space Agency and Copernicus and Google Earth Engine , title=. 2025 , note=

    2025

  11. [11]

    Geoderma , volume=

    Carbon sequestration potential in hedgerow soils: Results from 23 sites in Germany , author=. Geoderma , volume=. 2024 , publisher=

    2024

  12. [12]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

  13. [13]

    and Osindero, Simon and Teh, Yee Whye , journal =

    Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

  14. [14]

    2016 , publisher=

    Deep learning , author=. 2016 , publisher=

    2016

  15. [15]

    Gorelick, M

    Google Earth Engine: Planetary-scale geospatial analysis for everyone , author=. Remote Sensing of Environment , year=. doi:10.1016/j.rse.2017.06.031 , url=

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

  16. [16]

    2021 , Eprint =

    Vivien Sainte Fare Garnot and Loic Landrieu , Title =. 2021 , Eprint =

    2021

  17. [17]

    Brown and Michal R

    Christopher F. Brown and Michal R. Kazmierski and Valerie J. Pasquarella and William J. Rucklidge and Masha Samsikova and Chenhui Zhang and Evan Shelhamer and Estefania Lahera and Olivia Wiles and Simon Ilyushchenko and Noel Gorelick and Lihui Lydia Zhang and Sophia Alj and Emily Schechter and Sean Askay and Oliver Guinan and Rebecca Moore and Alexis Bouk...

    2025

  18. [18]

    Lavista Ferres and Jennifer Marcus , Title =

    Hannah Kerner and Snehal Chaudhari and Aninda Ghosh and Caleb Robinson and Adeel Ahmad and Eddie Choi and Nathan Jacobs and Chris Holmes and Matthias Mohr and Rahul Dodhia and Juan M. Lavista Ferres and Jennifer Marcus , Title =. 2024 , Eprint =

    2024

  19. [19]

    Global Agro-Ecological Zones version 4 (

  20. [20]

    Photogrammetric Engineering and Remote Sensing , year=

    Landsat’s Enduring Legacy: Pioneering Global Land Observations from Space , author=. Photogrammetric Engineering and Remote Sensing , year=

  21. [21]

    Remote sensing of environment , volume=

    A first assessment of the Sentinel-2 Level 1-C cloud mask product to support informed surface analyses , author=. Remote sensing of environment , volume=. 2018 , publisher=

    2018

  22. [22]

    2023 , howpublished=

    Mission Ends for Copernicus Sentinel-1B Satellite , author=. 2023 , howpublished=

    2023

  23. [23]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Panoptic segmentation of satellite image time series with convolutional temporal attention networks , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  24. [24]

    Proceedings of the International Conference on Machine Learning Time Series Workshop , volume=

    Ru. Proceedings of the International Conference on Machine Learning Time Series Workshop , volume=

  25. [25]

    2023 , publisher=

    Van Tricht, Kristof and Degerickx, Jeroen and Gilliams, Sven and Zanaga, Daniele and Battude, Marjorie and Grosu, Alex and Brombacher, Joost and Lesiv, Myroslava and Bayas, Juan Carlos Laso and Karanam, Santosh and others , journal=. 2023 , publisher=

    2023

  26. [26]

    2021 , author=

    Mapping crops at global scale! What works and what doesn't? , howpublished =. 2021 , author=

    2021

  27. [27]

    2024 , month = jan, note =

    2024

  28. [28]

    IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , volume=

    Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification , author=. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , volume=. 2019 , publisher=

    2019

  29. [29]

    Advances in Neural Information Processing Systems , volume=

    Lacoste, Alexandre and Lehmann, Nils and Rodriguez, Pau and Sherwin, Evan and Kerner, Hannah and L. Advances in Neural Information Processing Systems , volume=

  30. [30]

    Scientific Data , volume=

    Dynamic World, Near real-time global 10 m land use land cover mapping , author=. Scientific Data , volume=. 2022 , publisher=

    2022

  31. [31]

    Bonafilia, Derrick and Tellman, Beth and Anderson, Tyler and Issenberg, Erica , booktitle=

  32. [32]

    Detecting Marine pollutants and Sea Surface features with Deep learning in

    Kikaki, Katerina and Kakogeorgiou, Ioannis and Hoteit, Ibrahim and Karantzalos, Konstantinos , journal=. Detecting Marine pollutants and Sea Surface features with Deep learning in. 2024 , publisher=

    2024

  33. [33]

    Jaccard, Paul , journal=

  34. [34]

    Fully Convolutional Networks for Semantic Segmentation , year =

    Shelhamer*, Evan and Long*, Jonathan and Darrell, Trevor , journal =. Fully Convolutional Networks for Semantic Segmentation , year =

  35. [35]

    Environmental Research Letters , volume =

    Combining GEDI and Sentinel-2 for wall-to-wall mapping of tall and short crops , author =. Environmental Research Letters , volume =

  36. [36]

    TerraBytes-Towards global datasets and models for Earth Observation , pages=

    Using Multiple Input Modalities can Improve Data-Efficiency and OOD Generalization for ML with Satellite Imagery , author=. TerraBytes-Towards global datasets and models for Earth Observation , pages=. 2025 , organization=

    2025

  37. [37]

    Advances in Neural Information Processing Systems , volume=

    Satmae: Pre-training transformers for temporal and multi-spectral satellite imagery , author=. Advances in Neural Information Processing Systems , volume=

  38. [38]

    Forty-second International Conference on Machine Learning , year=

    Galileo: Learning Global & Local Features of Many Remote Sensing Modalities , author=. Forty-second International Conference on Machine Learning , year=

  39. [39]

    ICCV , eprint=

    Johannes Jakubik and Felix Yang and Benedikt Blumenstiel and Erik Scheurer and Rocco Sedona and Stefano Maurogiovanni and Jente Bosmans and Nikolaos Dionelis and Valerio Marsocci and Niklas Kopp and Rahul Ramachandran and Paolo Fraccaro and Thomas Brunschwiler and Gabriele Cavallaro and Juan Bernabe-Moreno and Nicolas Longépé , year=. ICCV , eprint=

  40. [40]

    arXiv preprint arXiv:2511.13655 , year=

    OlmoEarth: Stable Latent Image Modeling for Multimodal Earth Observation , author=. arXiv preprint arXiv:2511.13655 , year=

    arXiv

  41. [41]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Panopticon: Advancing any-sensor foundation models for earth observation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  42. [42]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    AnySat: One Earth Observation Model for Many Resolutions, Scales, and Modalities , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  43. [43]

    International Conference on Machine Learning , pages=

    Prioritized training on points that are learnable, worth learning, and not yet learnt , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  44. [44]

    Bastani, Favyen and Wolters, Piper and Gupta, Ritwik and Ferdinando, Joe and Kembhavi, Aniruddha , booktitle=

  45. [45]

    Fuller, Anthony and Millard, Koreen and Green, James , booktitle=

  46. [46]

    Planet Application Program Interface: In Space for Life on Earth , year =

    Planet Labs PBC , organization =. Planet Application Program Interface: In Space for Life on Earth , year =

  47. [47]

    arXiv preprint arXiv:2508.11739 , year=

    Scalable geospatial data generation using AlphaEarth foundations model , author=. arXiv preprint arXiv:2508.11739 , year=

    arXiv

  48. [48]

    arXiv preprint arXiv:2601.00857 , year=

    Harvesting AlphaEarth: Benchmarking the Geospatial Foundation Model for Agricultural Downstream Tasks , author=. arXiv preprint arXiv:2601.00857 , year=

    Pith/arXiv arXiv

  49. [49]

    La haie, levier de la planification

    de Menthi. La haie, levier de la planification. Rapport du CGAAER , volume=

  50. [50]

    Journal of environmental management , volume=

    Hedgerows: an international perspective on their origin, function and management , author=. Journal of environmental management , volume=. 2000 , publisher=

    2000

  51. [51]

    Le registre parcellaire graphique: des donn

    Cantelaube, Pierre and Carles, Marie , journal=. Le registre parcellaire graphique: des donn

  52. [52]

    2020 , month =

    IGN and OFB , title =. 2020 , month =

    2020

  53. [53]

    Dispositif National de Suivi des Bocages : mise à jour du référentiel des haies établi lors de la phase 1 , institution =

    Commagnac, Lo. Dispositif National de Suivi des Bocages : mise à jour du référentiel des haies établi lors de la phase 1 , institution =. 2024 , month =

    2024