arxiv: 2604.13240 · v1 · submitted 2026-04-14 · 💻 cs.CV · cs.LG

Recognition: unknown

A High-Resolution Landscape Dataset for Concept-Based XAI With Application to Species Distribution Models

Augustin de la Brosse , Damien Garreau , Thomas Houet , Thomas Corpetti

Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:20 UTC · model grok-4.3

classification 💻 cs.CV cs.LG

keywords species distribution modelsexplainable AIconcept activation vectorslandscape conceptsdrone imageryaquatic insectsdeep learningecological modeling

0 comments

The pith

A new high-resolution landscape dataset enables the first concept-based explainable AI application to species distribution models.

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

This paper claims that species distribution models built with deep learning can now be explained using predefined landscape concepts from drone photos, rather than remaining opaque. By creating a dataset of 15 specific landscape types and applying Robust TCAV, the authors show how to measure each concept's effect on predictions for where species live. In tests with two groups of aquatic insects, this method confirmed what ecologists already knew about their habitats while also pointing to unexpected links that could lead to new research questions. Such an approach matters because it lets complex predictive models also deliver trustworthy ecological understanding needed for conservation and management decisions.

Core claim

The central claim is that the first implementation of concept-based XAI for SDMs, using Robust TCAV on a new dataset of high-resolution multispectral and LiDAR drone imagery patches covering 15 landscape concepts, successfully quantifies concept influences on CNN and Vision Transformer predictions for Plecoptera and Trichoptera distributions. This validates the models against expert knowledge and uncovers novel associations that generate new ecological hypotheses, while also providing landscape-level information useful for policy-making.

What carries the argument

Robust TCAV (Testing with Concept Activation Vectors) applied to deep neural networks for species distribution modeling, supported by a custom dataset of 653 landscape concept patches and 1,450 random reference patches extracted from drone imagery.

If this is right

Models can be checked for alignment with known ecological drivers of species presence.
Unexpected concept influences can suggest new hypotheses about species-habitat relationships.
XAI outputs at landscape scale can support decisions in conservation policy and land use planning.
The open dataset allows similar analyses for other species or regions.

Where Pith is reading between the lines

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

Extending the same concept set and TCAV workflow to other environmental prediction tasks could increase transparency beyond ecology.
If the 15 concepts generalize across more species and geographies, they could serve as a reusable benchmark for ecological explainability.
Pairing the generated hypotheses with targeted field surveys would test whether the novel associations reflect real biological patterns.

Load-bearing premise

The 15 predefined landscape concepts extracted from the drone imagery are ecologically meaningful and sufficient to explain the species distributions, and Robust TCAV provides faithful, unbiased quantification of their influence.

What would settle it

If Robust TCAV applied to the trained SDMs shows no significant influence from concepts that experts know strongly affect Plecoptera and Trichoptera distributions, such as water features or vegetation types, the claim that this XAI approach validates and extends the models would not hold.

Figures

Figures reproduced from arXiv: 2604.13240 by Augustin de la Brosse, Damien Garreau, Thomas Corpetti, Thomas Houet.

**Figure 2.** Figure 2: Left panel: global overview of one of the acquisition sites. Yellow dots correspond to the locations where Road concept patches were extracted. Right panel: The patch is extracted around the point vector from the multispectral image and the digital elevation models to obtain a 7-band patch. a good representation of a diversity of agricultural landscapes, ranging from extensive dairy farming systems to high… view at source ↗

**Figure 3.** Figure 3: Robust TCAV: Robust TCAV requires a set of examples for a concept (e.g., Road), and random examples ○a , labeled test examples for the studied class (.e.g., Presence of Plecoptera) ○b and a trained network ○c . We obtain the CAV (v ℓ C , orange arrow) by computing the difference between the mean activations for the concept examples and for the random examples ○d . We use the directional derivative to quan… view at source ↗

**Figure 4.** Figure 4: Distribution maps obtained with the CerberusCNN for the five study sites. The blue vectors represent the water bodies. The color scheme is as follows: dark red is used to denote high probability, and white is used to indicate low probability. Study site names are on the left [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗

read the original abstract

Mapping the spatial distribution of species is essential for conservation policy and invasive species management. Species distribution models (SDMs) are the primary tools for this task, serving two purposes: achieving robust predictive performance while providing ecological insights into the driving factors of distribution. However, the increasing complexity of deep learning SDMs has made extracting these insights more challenging. To reconcile these objectives, we propose the first implementation of concept-based Explainable AI (XAI) for SDMs. We leverage the Robust TCAV (Testing with Concept Activation Vectors) methodology to quantify the influence of landscape concepts on model predictions. To enable this, we provide a new open-access landscape concept dataset derived from high-resolution multispectral and LiDAR drone imagery. It includes 653 patches across 15 distinct landscape concepts and 1,450 random reference patches, designed to suit a wide range of species. We demonstrate this approach through a case study of two aquatic insects, Plecoptera and Trichoptera, using two Convolutional Neural Networks and one Vision Transformer. Results show that concept-based XAI helps validate SDMs against expert knowledge while uncovering novel associations that generate new ecological hypotheses. Robust TCAV also provides landscape-level information, useful for policy-making and land management. Code and datasets are publicly available.

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.

Desk Editor's Note private letter to a colleague

The paper gives ecologists a new open drone-derived landscape concept dataset and shows Robust TCAV can interpret deep SDMs for two insect taxa.

read the letter

The main deliverable is an open dataset of 653 high-resolution patches labeled with 15 landscape concepts, drawn from multispectral and LiDAR drone imagery, plus 1,450 reference patches. They pair it with Robust TCAV to measure concept influence on CNN and ViT predictions for Plecoptera and Trichoptera distributions. Code and data are released publicly, which is straightforward to use for follow-up work. The case study reports that the approach aligns with expert knowledge on some drivers while flagging new associations that could spark hypotheses, and it produces landscape-scale outputs relevant to management. That combination of reusable data and a first application of concept-based XAI to SDMs is the concrete advance. The setup avoids circular claims and applies established tools to fresh data without hidden fitting issues. The soft spot is the lack of detailed quantitative results or error analysis in the summary: no specific TCAV scores, confidence intervals, or validation steps against held-out expert judgments are shown, so the strength of the novel associations remains hard to judge from the abstract alone. The 15 concepts are presented as sufficient, but their ecological completeness for these species is an external question rather than an internal contradiction. Readers working on interpretable models for biodiversity or conservation modeling will get the most from the dataset and the worked example. It deserves a serious referee because the data release is real and the method extension fills a documented gap, even if the results section will need close scrutiny on quantification and robustness.

Referee Report

2 major / 3 minor

Summary. The paper introduces a new open-access high-resolution landscape concept dataset (653 patches across 15 classes plus 1450 reference patches) derived from drone multispectral and LiDAR imagery. It applies Robust TCAV to two CNNs and one Vision Transformer for species distribution models of Plecoptera and Trichoptera, claiming this constitutes the first concept-based XAI implementation for SDMs. The approach is reported to validate model predictions against expert knowledge while identifying novel landscape associations that generate new ecological hypotheses, with code and data released publicly.

Significance. If the quantitative results and validation hold, the work supplies a reusable resource for interpreting complex deep-learning SDMs in ecology, addressing the tension between predictive performance and ecological insight. The public dataset and code are explicit strengths that support reproducibility and extension. Landscape-level concept influence scores could inform conservation policy, though external validity depends on the ecological relevance of the 15 concepts.

major comments (2)

[Abstract] Abstract: the claim that results 'show that concept-based XAI helps validate SDMs against expert knowledge while uncovering novel associations' is presented without any quantitative metrics, confidence intervals, or error analysis; this weakens assessment of the central claim that the method both validates and generates hypotheses.
[Results] Results / case study section: the paper must detail how novel associations were distinguished from known ecology (e.g., literature cross-check or independent confirmation) and report statistical support for directional TCAV scores; without this the hypothesis-generation component rests on unverified interpretation.

minor comments (3)

[Methods] Methods: expand justification for the specific choice of 15 landscape concepts and their sufficiency for the target taxa, including any sensitivity checks.
Figures: ensure all concept patch visualizations include scale bars, resolution metadata, and clear legends for TCAV activation maps.
Notation: confirm consistent expansion of 'Robust TCAV' on first use and clarify reference patch sampling procedure.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the presentation of our results.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that results 'show that concept-based XAI helps validate SDMs against expert knowledge while uncovering novel associations' is presented without any quantitative metrics, confidence intervals, or error analysis; this weakens assessment of the central claim that the method both validates and generates hypotheses.

Authors: We agree that the abstract would benefit from quantitative support to better substantiate the central claims. The results section of the manuscript already reports specific TCAV scores, directional influences, and alignments with expert knowledge for the Plecoptera and Trichoptera case studies. In the revised manuscript, we will add concise quantitative metrics (e.g., average concept influence scores and key directional findings) to the abstract while maintaining its brevity. revision: yes
Referee: [Results] Results / case study section: the paper must detail how novel associations were distinguished from known ecology (e.g., literature cross-check or independent confirmation) and report statistical support for directional TCAV scores; without this the hypothesis-generation component rests on unverified interpretation.

Authors: We acknowledge the need for greater rigor in distinguishing novel associations and providing statistical backing. The current manuscript draws on expert input and initial literature checks to identify novel landscape associations, but we will expand this in revision by adding a systematic literature cross-reference (e.g., a table or subsection) that classifies each association as previously documented or novel. We will also report statistical support for directional TCAV scores, including confidence intervals or significance measures as permitted by the Robust TCAV framework, to strengthen the hypothesis-generation claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper introduces a new high-resolution landscape concept dataset (653 patches across 15 classes plus references) and applies the established Robust TCAV method to CNN and ViT models trained on species distribution data for two insect taxa. No load-bearing steps reduce by construction to self-defined quantities, fitted inputs renamed as predictions, or self-citation chains; the central claims rest on empirical application and open data release rather than any internal derivation that equates outputs to inputs. The methodology is presented as an extension of prior XAI work without uniqueness theorems or ansatzes smuggled via author-overlapping citations.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No specific free parameters, axioms, or invented entities identifiable from the abstract. The work relies on standard deep learning architectures and the established Robust TCAV method applied to new imagery data.

pith-pipeline@v0.9.0 · 5534 in / 1133 out tokens · 111820 ms · 2026-05-10T16:20:51.777768+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

54 extracted references · 7 canonical work pages · 2 internal anchors

[1]

Allan, J.D.: Landscapes and riverscapes: the influence of land use on stream ecosys- tems. Annu. Rev. Ecol. Evol. Syst.35(1), 257–284 (2004)

2004
[2]

arXiv preprint arXiv:2412.07408 (2024)

Amara, J., König-Ries, B., Samuel, S.: Explainability of deep learning-based plant disease classifiers through automated concept identification. arXiv preprint arXiv:2412.07408 (2024)

work page arXiv 2024
[3]

In: Proceedings of the 18th International Joint Conference on Com- puter Vision, Imaging and Computer Graphics Theory and Applications (VISI- GRAPP 2023) - Volume 4: VISAPP

Amara, J., König-Ries, B., Samuel, S.: Concept explainability for plant diseases classification. In: Proceedings of the 18th International Joint Conference on Com- puter Vision, Imaging and Computer Graphics Theory and Applications (VISI- GRAPP 2023) - Volume 4: VISAPP. pp. 246–253. INSTICC, SciTePress (2023). https://doi.org/10.5220/0011667900003417

work page doi:10.5220/0011667900003417 2023
[4]

Frontiers in Ecology and the Environment11(3), 138–146 (2013)

Anderson, K., Gaston, K.J.: Lightweight unmanned aerial vehicles will revolu- tionize spatial ecology. Frontiers in Ecology and the Environment11(3), 138–146 (2013)

2013
[5]

Journal of Unmanned Vehicle Systems7(1), 54–75 (2018)

Assmann, J.J., Kerby, J.T., Cunliffe, A.M., Myers-Smith, I.H.: Vegetation moni- toring using multispectral sensors—best practices and lessons learned from high latitudes. Journal of Unmanned Vehicle Systems7(1), 54–75 (2018)

2018
[6]

Science of the Total Environment966, 178728 (2025)

Bergerot, B., Piscart, C., Roussel, J.: Tightly intertwined: Waterscapes prompt urgent reconsideration of aquatic insects and their role in agricultural landscapes. Science of the Total Environment966, 178728 (2025)

2025
[7]

arXiv preprint arXiv:2509.25816 (2025)

Botella, C., Deneu, B., Marcos, D., Servajean, M., Larcher, T., Leblanc, C., Estopinan, J., Bonnet, P., Joly, A.: Overview of geolifeclef 2023: Species com- position prediction with high spatial resolution at continental scale using remote sensing. arXiv preprint arXiv:2509.25816 (2025)

work page arXiv 2023
[8]

Biologia futura76(4), 585–595 (2025)

Buebos-Esteve,D.E.,Dagamac,N.H.A.:Evaluatingmodel-agnosticpost-hocmeth- ods in explainable artificial intelligence: augmenting species distribution models. Biologia futura76(4), 585–595 (2025)

2025
[9]

Advances in Neural Information Processing Systems35, 2590–2607 (2022)

Crabbé, J., van der Schaar, M.: Concept activation regions: A generalized frame- work for concept-based explanations. Advances in Neural Information Processing Systems35, 2590–2607 (2022)

2022
[10]

In: International Conference on Learning Representations (2021)

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An image is worth 16x16 words: Transformers for image recognition at scale. In: International Conference on Learning Representations (2021)

2021
[11]

Annual review of ecology, evolution, and system- atics40(1), 677–697 (2009)

Elith, J., Leathwick, J.R.: Species distribution models: ecological explanation and prediction across space and time. Annual review of ecology, evolution, and system- atics40(1), 677–697 (2009)

2009
[12]

Global Change Biology30(12), e70005 (2024) A High-Resolution Landscape Dataset for Concept-Based XAI 15

Fan, S., Newbold, T., Tscharntke, T., Tang, W., Yu, Z., Liu, Y.: Impact of crop type on biodiversity globally. Global Change Biology30(12), e70005 (2024) A High-Resolution Landscape Dataset for Concept-Based XAI 15

2024
[13]

Landscape Ecology38(11), 2917–2929 (2023)

Gerber, R., Piscart, C., Roussel, J.M., Georges, R., Houet, T., Royer, J., Berg- erot, B.: Landscape models can predict the distribution of aquatic insects across agricultural areas. Landscape Ecology38(11), 2917–2929 (2023)

2023
[14]

Environmental Monitoring and As- sessment194(10), 697 (2022)

Gomes, P.G.S., Lima, E.L., Silva, S.R., Juen, L., Brasil, L.S.: Does land use and land cover affect adult communities of ephemeroptera, plecoptera and trichoptera (ept)? a systematic review with meta-analysis. Environmental Monitoring and As- sessment194(10), 697 (2022)

2022
[15]

Methods in Ecology and Evolution9(6), 1614–1625 (2018)

Guélat, J., Kéry, M.: Effects of spatial autocorrelation and imperfect detection on species distribution models. Methods in Ecology and Evolution9(6), 1614–1625 (2018)

2018
[16]

Cambridge University Press (2017)

Guisan, A., Thuiller, W., Zimmermann, N.E.: Habitat suitability and distribution models: with applications in R. Cambridge University Press (2017)

2017
[17]

Ecological Indicators144, 109523 (2022)

He, B., Zhao, Y., Mao, W.: Explainable artificial intelligence reveals environmental constraints in seagrass distribution. Ecological Indicators144, 109523 (2022)

2022
[18]

He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

2016
[19]

Jurnal JTIK (Jurnal Teknologi Informasi dan Komunikasi)7(4), 758–768 (2023)

Hindarto, D.: Use resnet50v2 deep learning model to classify five animal species. Jurnal JTIK (Jurnal Teknologi Informasi dan Komunikasi)7(4), 758–768 (2023)

2023
[20]

Hole, D.G., Perkins, A., Wilson, J., Alexander, I., Grice, P., Evans, A.D.: Does or- ganic farming benefit biodiversity? Biological conservation122(1), 113–130 (2005)

2005
[21]

Huang, Y., Hou, S., Horve, Z.N., Fei, S.: Barkxai: A lightweight post-hoc explain- ablemethodfortreespeciesclassificationwithquantifiableconcepts.In:ICLR2025 Workshop: XAI4Science: From Understanding Model Behavior to Discovering New Scientific Knowledge (2025)

2025
[22]

Explainable Artificial Intelligence for Cyber Security: Next Generation Ar- tificial Intelligence pp

Islam, M.U., Mozaharul Mottalib, M., Hassan, M., Alam, Z.I., Zobaed, S., Fa- zle Rabby, M.: The past, present, and prospective future of xai: A comprehensive review. Explainable Artificial Intelligence for Cyber Security: Next Generation Ar- tificial Intelligence pp. 1–29 (2022)

2022
[23]

In: 2024 3rd International Conference on Automation, Computing and Renewable Systems (ICACRS)

Kaur, J., Rani, S., Sharma, A., Dogra, A.: Enhanced bird species identification using resnet-50: A deep learning framework for high-performance classification. In: 2024 3rd International Conference on Automation, Computing and Renewable Systems (ICACRS). pp. 821–826. IEEE (2024)

2024
[24]

In: Proceedings of the 35th International Conference on Machine Learning

Kim, B., Wattenberg, M., Gilmer, J., Cai, C., Wexler, J., Viegas, F., sayres, R.: Interpretability beyond feature attribution: Quantitative testing with concept ac- tivation vectors (TCAV). In: Proceedings of the 35th International Conference on Machine Learning. Proceedings of Machine Learning Research, vol. 80, pp. 2668–
[25]

PMLR (10–15 Jul 2018)

2018
[26]

nature521(7553), 436–444 (2015)

LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. nature521(7553), 436–444 (2015)

2015
[27]

Science391(6788), 917–921 (2026)

Leroy, F., Jarzyna, M.A., Keil, P.: Acceleration hotspots of north american birds’ decline are associated with agriculture. Science391(6788), 917–921 (2026)

2026
[28]

John Wiley & Sons (2015)

Lillesand, T., Kiefer, R.W., Chipman, J.: Remote sensing and image interpretation. John Wiley & Sons (2015)

2015
[29]

CEUR-WS (2022)

Lorieul, T., Cole, E., Deneu, B., Servajean, M., Bonnet, P., Joly, A.: Overview of geolifeclef 2022: Predicting species presence from multi-modal remote sensing, bioclimatic and pedologic data. CEUR-WS (2022)

2022
[30]

Decoupled Weight Decay Regularization

Loshchilov, I., Hutter, F., et al.: Fixing weight decay regularization in adam. arXiv preprint arXiv:1711.051015(5), 5 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017
[31]

Advances in neural information processing systems30(2017) 16 A

Lundberg, S.M., Lee, S.I.: A unified approach to interpreting model predictions. Advances in neural information processing systems30(2017) 16 A. de la Brosse et al

2017
[32]

Applied Geography42, 63– 72 (2013)

Marcantonio, M., Rocchini, D., Geri, F., Bacaro, G., Amici, V.: Biodiversity, roads, & landscape fragmentation: Two mediterranean cases. Applied Geography42, 63– 72 (2013)

2013
[33]

Martin, T., Weller, A.: Interpretable machine learning. M. Phil. diss., Dept. of Engineering, University of Cambridge2(3), 5 (2019)

2019
[34]

Annual Review of Ecology, Evolution, and System- atics51(1), 81–102 (2020)

Montgomery, I., Caruso, T., Reid, N.: Hedgerows as ecosystems: service delivery, management, and restoration. Annual Review of Ecology, Evolution, and System- atics51(1), 81–102 (2020)

2020
[35]

CEUR-WS (2024)

Picek, L., Botella, C., Servajean, M., Leblanc, C., Palard, R., Larcher, T., Deneu, B., Marcos, D., Estopinan, J., Bonnet, P., et al.: Overview of geolifeclef 2024: Species composition prediction with high spatial resolution at continental scale using remote sensing. CEUR-WS (2024)

2024
[36]

ACM Computing Surveys (2023)

Poeta, E., Ciravegna, G., Pastor, E., Cerquitelli, T., Baralis, E.: Concept-based explainable artificial intelligence: A survey. ACM Computing Surveys (2023)

2023
[37]

Prevedello, J.A., Almeida-Gomes, M., Lindenmayer, D.B.: The importance of scat- teredtreesforbiodiversityconservation:Aglobalmeta-analysis.JournalofApplied Ecology55(1), 205–214 (2018)

2018
[38]

Agriculture, ecosystems & environment 270, 32–40 (2019)

Raitif,J.,Plantegenest,M.,Roussel,J.M.:Fromstreamtoland:Ecosystemservices provided by stream insects to agriculture. Agriculture, ecosystems & environment 270, 32–40 (2019)

2019
[39]

why should i trust you?

Ribeiro, M.T., Singh, S., Guestrin, C.: “why should i trust you?” explaining the pre- dictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. pp. 1135–1144 (2016)

2016
[40]

Ecography44(2), 199–205 (2021)

Ryo, M., Angelov, B., Mammola, S., Kass, J.M., Benito, B.M., Hartig, F.: Ex- plainable artificial intelligence enhances the ecological interpretability of black-box species distribution models. Ecography44(2), 199–205 (2021)

2021
[41]

In: Proc

Sankupellay, M., Konovalov, D.: Bird call recognition using deep convolutional neural network, resnet-50. In: Proc. Acoustics. pp. 1–8 (2018)

2018
[42]

Journal of Insect Conservation23(1), 175–186 (2019)

Silva, D.P., Andrade, A.F., Oliveira, J.P., Morais, D.M., Vieira, J.E., Engel, M.S.: Current and future ranges of an elusive north american insect using species distri- bution models. Journal of Insect Conservation23(1), 175–186 (2019)

2019
[43]

Environmental Pollution Series A, Ecological and Biological 32(3), 157–170 (1983)

Smith, M.E., Kaster, J.L.: Effect of rural highway runoff on stream benthic macroinvertebrates. Environmental Pollution Series A, Ecological and Biological 32(3), 157–170 (1983)

1983
[44]

arXiv preprint arXiv:2002.03549 (2020)

Soni, R., Shah, N., Seng, C.T., Moore, J.D.: Adversarial tcav–robust and ef- fective interpretation of intermediate layers in neural networks. arXiv preprint arXiv:2002.03549 (2020)

work page arXiv 2002
[45]

Revista Chilena de Historia Natural (2009)

Tognelli, M.F., Roig, S.A., Marvaldi, A., Flores, G.E., Lobo, J.M.: An evaluation of methods for modelling distribution of patagonian insects. Revista Chilena de Historia Natural (2009)

2009
[46]

Oikos pp

Ulfstrand, S.: Life cycles of benthic insects in lapland streams (ephemeroptera, plecoptera, trichoptera, diptera simuliidae). Oikos pp. 167–190 (1968)

1968
[47]

Journal of Applied Ecology57(1), 4–16 (2020)

Valdés, A., Lenoir, J., De Frenne, P., Andrieu, E., Brunet, J., Chabrerie, O., Cousins, S.A., Deconchat, M., De Smedt, P., Diekmann, M., et al.: High ecosystem service delivery potential of small woodlands in agricultural landscapes. Journal of Applied Ecology57(1), 4–16 (2020)

2020
[48]

arXiv preprint arXiv:2509.24058 (2025)

Wenkmann, J., Garreau, D.: On The Variability of Concept Activation Vectors. arXiv preprint arXiv:2509.24058 (2025)

work page arXiv 2025
[49]

Journal of vegetation Science23(4), 796–802 (2012) A High-Resolution Landscape Dataset for Concept-Based XAI 17

Wilson, J.B., Peet, R.K., Dengler, J., Pärtel, M.: Plant species richness: the world records. Journal of vegetation Science23(4), 796–802 (2012) A High-Resolution Landscape Dataset for Concept-Based XAI 17

2012
[50]

Methods in Ecology and Evolution17(1), 188– 206 (2026)

Zbinden, R., Van Tiel, N., Sumbul, G., Vanalli, C., Kellenberger, B., Tuia, D.: Masksdm with shapley values to improve flexibility, robustness and explainability in species distribution modelling. Methods in Ecology and Evolution17(1), 188– 206 (2026)

2026
[51]

Zedler, J.B., Kercher, S.: Wetland resources: status, trends, ecosystem services, and restorability. Annu. Rev. Environ. Resour.30, 39–74 (2005)

2005
[52]

In: European conference on computer vision

Zeiler, M.D., Fergus, R.: Visualizing and understanding convolutional networks. In: European conference on computer vision. pp. 818–833. Springer (2014)

2014
[53]

In: International Conference on Learning Representations (2018)

Zhang, H., Cisse, M., Dauphin, Y.N., Lopez-Paz, D.: mixup: Beyond empirical risk minimization. In: International Conference on Learning Representations (2018)

2018
[54]

Representation Engineering: A Top-Down Approach to AI Transparency

Zou, A., Phan, L., Chen, S., Campbell, J., Guo, P., Ren, R., Pan, A., Yin, X., Mazeika, M., Dombrowski, A.K., et al.: Representation engineering: A top-down approach to ai transparency. arXiv preprint arXiv:2310.01405 (2023) 18 A. de la Brosse et al. 7 Appendix 7.1 Implementation resources All resources necessary to the implementation of our paper are lis...

work page internal anchor Pith review arXiv 2023