Explainable AI for Biodiversity Monitoring and Ecological Image Analysis

Brinnae Bent; David W. Johnston; G\"unel Aghakishiyeva; Holly R. Houliston; Jiayi Zhou

arxiv: 2606.27667 · v1 · pith:JZ2UC4LXnew · submitted 2026-06-26 · 💻 cs.CV · cs.AI· q-bio.QM

Explainable AI for Biodiversity Monitoring and Ecological Image Analysis

Brinnae Bent , Holly R. Houliston , Jiayi Zhou , G\"unel Aghakishiyeva , David W. Johnston This is my paper

Pith reviewed 2026-06-29 05:04 UTC · model grok-4.3

classification 💻 cs.CV cs.AIq-bio.QM

keywords explainable AIbiodiversity monitoringecological image analysiscomputer visionmodel validationconservationcamera trapsaerial imagery

0 comments

The pith

Explainable AI should become a standard part of validating models that analyze ecological images for biodiversity monitoring.

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

The paper contends that AI models for tasks like classifying species in camera-trap photos or detecting animals in drone imagery often succeed or fail for reasons that are hard to inspect. It shows that applying explanation techniques to three standard computer-vision tasks can expose whether a model is responding to biologically relevant patterns or to background clutter, lighting artifacts, and sampling biases. Two concrete case studies on harbor-seal detection and cetacean body-part segmentation illustrate how these explanations can flag false positives, reveal occlusion problems, and suggest targeted improvements in data collection or retraining. If the argument holds, conservation decisions that rest on automated image analysis become more transparent and therefore more defensible.

Core claim

We argue that explainable artificial intelligence (XAI) should become a standard component of ecological model validation because conservation practitioners increasingly depend on understanding not only whether a model is accurate, but why it is accurate. We provide practical guidance for applying XAI to image classification, object detection, and image segmentation, and we illustrate the approach with two case studies using aerial imagery that demonstrate how explanation methods can identify biologically meaningful cues, reveal false positives driven by background and shape confounds, uncover edge and occlusion effects, and guide data collection, augmentation, and retraining strategies.

What carries the argument

Explanation methods applied to the outputs of image-classification, object-detection, and segmentation models trained on ecological imagery, used to audit whether the model attends to biologically relevant visual features.

If this is right

Model refinement can target removal of confounds identified by explanations rather than blind accuracy tuning.
Data-augmentation and collection strategies can be chosen to reduce the edge and occlusion effects revealed by XAI.
Conservation reports can include explicit checks that model reasoning aligns with established ecological knowledge.
Deployment decisions for new monitoring platforms can incorporate XAI audits before scaling.

Where Pith is reading between the lines

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

The same auditing process could be applied to other sensor-derived ecological data such as acoustic recordings or satellite time series.
Training programs for field biologists may need to include basic interpretation of XAI outputs to maintain oversight of automated pipelines.
Regulatory or funding requirements for AI-assisted conservation projects could eventually reference XAI documentation as a standard deliverable.

Load-bearing premise

That currently available XAI techniques can reliably separate ecologically meaningful signals from spurious correlations and image artifacts in the setting of ecological imagery.

What would settle it

A controlled test in which XAI heatmaps or saliency maps for a high-accuracy ecological model consistently highlight non-biological regions (such as uniform background water or sensor glare) rather than animal morphology or habitat features, while the model still passes independent biological validation.

read the original abstract

Artificial intelligence is transforming biodiversity monitoring by enabling automated analysis of ecological imagery collected from camera traps, drones, satellites, underwater platforms, and other sensing systems. These tools can expand the scale and speed of conservation assessments, yet many computer vision models remain difficult to inspect, making it challenging to determine whether predictions are based on ecologically meaningful signals or on spurious correlations, sampling biases, and other artifacts that may undermine conservation decisions. We argue that explainable artificial intelligence (XAI) should become a standard component of ecological model validation because conservation practitioners increasingly depend on understanding not only whether a model is accurate, but why it is accurate. We provide practical guidance for applying XAI to three common ecological computer vision tasks: image classification, object detection, and image segmentation. To illustrate how XAI can support ecological model auditing, refinement, and deployment, we present two case studies using aerial imagery: harbor seal detection and cetacean anatomical segmentation. These examples demonstrate how explanation methods can identify biologically meaningful cues, reveal false positives driven by background and shape confounds, uncover edge and occlusion effects, and guide data collection, augmentation, and retraining strategies. More broadly, they show how explainability can help assess whether model reasoning aligns with ecological understanding. We conclude by identifying key challenges and opportunities. By making model behavior more transparent and scientifically interrogable, XAI can help ensure that AI-supported ecological evidence is more reliable, understandable, and actionable for biodiversity conservation.

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

This is a clear advocacy piece for routine XAI in ecological CV work, with useful illustrations but no new empirical test of the recommendation.

read the letter

The core takeaway is that the authors want XAI treated as standard validation practice for computer vision models in biodiversity monitoring. They make the case with guidance on three standard tasks and two aerial-imagery examples rather than any new method or controlled experiment.

The paper does a decent job laying out how saliency maps and similar tools can surface when a seal detector is latching onto water texture or when a cetacean segmentation model is confused by occlusion. Those concrete walkthroughs are the most useful part for someone who actually runs these models in the field. The advice on using explanations to guide data augmentation and retraining is straightforward and matches what practitioners already do informally.

The main limitation is that the argument stays at the level of demonstration. The case studies show XAI can reveal confounds, but there is no measurement of whether acting on those explanations improves model accuracy, reduces false positives in deployment, or changes conservation decisions. The assumption that current XAI techniques will reliably separate ecological signal from artifact is stated but not stress-tested against known failure modes of the methods themselves. Because the paper is normative rather than quantitative, the evidential bar is lower than for a methods paper, yet the central claim still needs more than illustration to land with force.

This is written for ecologists and conservation technologists who already use off-the-shelf CV models and want a practical checklist for auditing them. It is not aimed at XAI researchers looking for new techniques. The writing is coherent and the examples are relevant, so it clears the threshold for a serious referee even though the contribution is incremental and qualitative. I would send it out for review with the expectation that the authors strengthen the link between the XAI outputs and downstream outcomes.

Referee Report

0 major / 2 minor

Summary. The manuscript argues that explainable artificial intelligence (XAI) should become a standard component of ecological model validation for biodiversity monitoring using computer vision. It supplies practical guidance for applying XAI to image classification, object detection, and image segmentation, and illustrates the value of the approach with two case studies on harbor seal detection and cetacean anatomical segmentation in aerial imagery.

Significance. If the recommendation is followed, XAI could improve the scientific credibility and actionability of AI-supported conservation decisions by surfacing whether predictions rest on ecologically meaningful signals rather than artifacts. The position paper usefully connects existing XAI techniques to the specific validation needs of ecological imagery analysis.

minor comments (2)

[Abstract] Abstract: the claim that the case studies show how XAI can 'guide data collection, augmentation, and retraining strategies' is stated at a high level; the case-studies section would be strengthened by one or two concrete examples of such guidance derived from the explanations.
[Case studies] Case studies: the descriptions remain qualitative; adding a short table or paragraph summarizing any before/after performance metrics (even if modest) after XAI-informed refinements would make the practical utility of the illustrations easier to evaluate.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of the manuscript, its significance for ecological AI validation, and the recommendation of minor revision. The report does not list any specific major comments.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The manuscript is a position/advocacy paper with no equations, derivations, fitted parameters, or quantitative predictions. Its central claim is normative (XAI should become standard for ecological validation) and rests on general principles plus illustrative case studies. No load-bearing step reduces to its own inputs by construction, self-citation, or renaming. The case studies are explicitly framed as demonstrations of existing XAI methods rather than as falsifiable empirical claims whose validity depends on internal fitting.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a perspective and guidance paper with no mathematical content; it introduces no free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5812 in / 1063 out tokens · 37019 ms · 2026-06-29T05:04:05.043911+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

14 extracted references · 13 canonical work pages

[1]

Abnar, S., & Zuidema, W

cited 38 in this work appear in conference proceedings that serve as primary archival venues within these fields. Abnar, S., & Zuidema, W. (2020). Quantifying attention flow in transformers . In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL) , 4190–4197. Adebayo, J., Gilmer, J., Muelly, M., Goodfellow, I., Ha...

2020
[2]

D., Houliston, H

Aghakishiyeva, G., Zhou, J., Arya, S., Dale, J., Poling, J. D., Houliston, H. R., Womble, J. N., Larsen, G. D., Johnston, D. W., & Bent, B. (2025). Photorealistic inpainting for perturbation-based explanations in ecological monitoring . NeurIPS 2025 Imageomics Workshop. Axford, D., Sohel, F., Vanderklift, M. A., & Hodgson, A. J. (2024). Collectively advan...

work page doi:10.1016/j.ecoinf.2024.102842 2025
[3]

https://doi.org/10.3390/f14091881 Fantozzi, P., & Naldi, M. (2024). The explainability of transformers: Current status and directions . Computers, 13(4),

work page doi:10.3390/f14091881 2024
[4]

https://doi.org/10.3390/computers13040092 Fernandes, A. C. M., Quintero González, R., Xie, H., & Azevedo, A. (2020). Machine learning for conservation planning in a changing climate. Sustainability, 12(18),

work page doi:10.3390/computers13040092 2020
[5]

C., & Vedaldi, A

https://doi.org/10.3390/su12187657 41 Fong, R. C., & Vedaldi, A. (2017). Interpretable explanations of black boxes by meaningful perturbation. In Proceedings of the IEEE International Conference on Computer Vision (ICCV) (pp. 3449–3457). https://doi.org/10.1109/ICCV.2017.371 Gálvez-Salido, A., Robles, F., Gonçalves, R. J., de la Herrán, R., Ruiz Rejón, C....

work page doi:10.3390/su12187657 2017
[6]

https://doi.org/10.3390/jimaging12020088 Geirhos, R., Jacobsen, J.-H., Michaelis, C., Zemel, R., Brendel, W., Bethge, M., & Wichmann, F. A. (2020). Shortcut learning in deep neural networks . Nature Machine Intelligence, 2, 665–673. https://doi.org/10.1038/s42256-020-00257-z Gevaert, C. M. (2022). Explainable AI for Earth observation: A review including s...

work page doi:10.3390/jimaging12020088 2020
[7]

C., Larsen, G

https://doi.org/10.3390/rs13193953 Gray, P. C., Larsen, G. D., & Johnston, D. W. (2022). Drones address an observational blind spot for biological oceanography. Frontiers in Ecology and the Environment, 20 (2), 93–94. https://doi.org/10.1002/fee.2472 González-Rivero, M., Beijbom, O., Rodriguez-Ramirez, A., Bryant, D. E. P., Ganase, A., Gonzalez-Marrero, Y...

work page doi:10.3390/rs13193953 2022
[8]

B., Valles, D., Hibbitts, T

https://doi.org/10.3390/rs12030489 Islam, S. B., Valles, D., Hibbitts, T. J., Ryberg, W. A., Walkup, D. K., & Forstner, M. R. J. (2023). Animal Species Recognition with Deep Convolutional Neural Networks from Ecological Camera Trap Images. Animals, 13(9),

work page doi:10.3390/rs12030489 2023
[9]

https://doi.org/10.3390/ani13091526 Jain, S., & Wallace, B.C. (2019). Attention is not Explanation. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages 3543–3556, Association for Computational Linguistics. https://doi.org...

work page doi:10.3390/ani13091526 2019
[10]

Why should I trust you?

https://doi.org/10.3390/bdcc9080193 Reichstein, M., Camps-Valls, G., Stevens, B., Jung, M., Denzler, J., Carvalhais, N., & Prabhat. (2019). Deep learning and process understanding for data-driven Earth system science. Nature, 566 (7743), 195–204. https://doi.org/10.1038/s41586-019-0912-1 Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster R-CNN: Towar...

work page doi:10.3390/bdcc9080193 2019
[11]

C., Torralba, A., Murphy, K

Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence, 1(5), pp.206-215. Russell, B. C., Torralba, A., Murphy, K. P., & Freeman, W. T. (2008). LabelMe: A database and web-based tool for image annotation. International Journal of Computer Vision, 77 (1–3), 157–173. http...

work page doi:10.1007/s11263-007-0090-8 2008
[12]

A., Norouzzadeh, M

https://doi.org/10.3390/rs16234440 Tabak, M. A., Norouzzadeh, M. S., Wolfson, D. W., Sweeney, S. J., Vercauteren, K. C., Snow, N. P., Halseth, J. M., Salvo, P. A. D., Lewis, J. S., White, M. D., Teton, B., Beasley, J. C., Schlichting, P. E., Boughton, R. K., Wight, B., Newkirk, E. S., Ivan, J. S., Odell, E. A., Brook, R. K., … Miller, R. S. (2019). Machin...

work page doi:10.3390/rs16234440 2019
[13]

M.-C., & Hedström, A

https://doi.org/10.3390/rs11111309 Wickstrøm, K., Höhne, M. M.-C., & Hedström, A. (2024). From flexibility to manipulation: The slippery slope of XAI evaluation. In Proceedings of the ECCV 2024 Workshop on Explainable Computer Vision: Where are We and Where are We Going? arXiv. https://arxiv.org/abs/2412.05592 Wiegreffe, S., & Pinter, Y. (2019). Attention...

work page doi:10.3390/rs11111309 2024
[14]

https://doi.org/10.3390/rs17111828 Zurell, D., et al. (2020). A standard protocol for reporting species distribution models. Ecography , 43(9), 1261–1277. https://doi.org/10.1111/ecog.04960 51

work page doi:10.3390/rs17111828 2020

[1] [1]

Abnar, S., & Zuidema, W

cited 38 in this work appear in conference proceedings that serve as primary archival venues within these fields. Abnar, S., & Zuidema, W. (2020). Quantifying attention flow in transformers . In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL) , 4190–4197. Adebayo, J., Gilmer, J., Muelly, M., Goodfellow, I., Ha...

2020

[2] [2]

D., Houliston, H

Aghakishiyeva, G., Zhou, J., Arya, S., Dale, J., Poling, J. D., Houliston, H. R., Womble, J. N., Larsen, G. D., Johnston, D. W., & Bent, B. (2025). Photorealistic inpainting for perturbation-based explanations in ecological monitoring . NeurIPS 2025 Imageomics Workshop. Axford, D., Sohel, F., Vanderklift, M. A., & Hodgson, A. J. (2024). Collectively advan...

work page doi:10.1016/j.ecoinf.2024.102842 2025

[3] [3]

https://doi.org/10.3390/f14091881 Fantozzi, P., & Naldi, M. (2024). The explainability of transformers: Current status and directions . Computers, 13(4),

work page doi:10.3390/f14091881 2024

[4] [4]

https://doi.org/10.3390/computers13040092 Fernandes, A. C. M., Quintero González, R., Xie, H., & Azevedo, A. (2020). Machine learning for conservation planning in a changing climate. Sustainability, 12(18),

work page doi:10.3390/computers13040092 2020

[5] [5]

C., & Vedaldi, A

https://doi.org/10.3390/su12187657 41 Fong, R. C., & Vedaldi, A. (2017). Interpretable explanations of black boxes by meaningful perturbation. In Proceedings of the IEEE International Conference on Computer Vision (ICCV) (pp. 3449–3457). https://doi.org/10.1109/ICCV.2017.371 Gálvez-Salido, A., Robles, F., Gonçalves, R. J., de la Herrán, R., Ruiz Rejón, C....

work page doi:10.3390/su12187657 2017

[6] [6]

https://doi.org/10.3390/jimaging12020088 Geirhos, R., Jacobsen, J.-H., Michaelis, C., Zemel, R., Brendel, W., Bethge, M., & Wichmann, F. A. (2020). Shortcut learning in deep neural networks . Nature Machine Intelligence, 2, 665–673. https://doi.org/10.1038/s42256-020-00257-z Gevaert, C. M. (2022). Explainable AI for Earth observation: A review including s...

work page doi:10.3390/jimaging12020088 2020

[7] [7]

C., Larsen, G

https://doi.org/10.3390/rs13193953 Gray, P. C., Larsen, G. D., & Johnston, D. W. (2022). Drones address an observational blind spot for biological oceanography. Frontiers in Ecology and the Environment, 20 (2), 93–94. https://doi.org/10.1002/fee.2472 González-Rivero, M., Beijbom, O., Rodriguez-Ramirez, A., Bryant, D. E. P., Ganase, A., Gonzalez-Marrero, Y...

work page doi:10.3390/rs13193953 2022

[8] [8]

B., Valles, D., Hibbitts, T

https://doi.org/10.3390/rs12030489 Islam, S. B., Valles, D., Hibbitts, T. J., Ryberg, W. A., Walkup, D. K., & Forstner, M. R. J. (2023). Animal Species Recognition with Deep Convolutional Neural Networks from Ecological Camera Trap Images. Animals, 13(9),

work page doi:10.3390/rs12030489 2023

[9] [9]

https://doi.org/10.3390/ani13091526 Jain, S., & Wallace, B.C. (2019). Attention is not Explanation. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages 3543–3556, Association for Computational Linguistics. https://doi.org...

work page doi:10.3390/ani13091526 2019

[10] [10]

Why should I trust you?

https://doi.org/10.3390/bdcc9080193 Reichstein, M., Camps-Valls, G., Stevens, B., Jung, M., Denzler, J., Carvalhais, N., & Prabhat. (2019). Deep learning and process understanding for data-driven Earth system science. Nature, 566 (7743), 195–204. https://doi.org/10.1038/s41586-019-0912-1 Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster R-CNN: Towar...

work page doi:10.3390/bdcc9080193 2019

[11] [11]

C., Torralba, A., Murphy, K

Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence, 1(5), pp.206-215. Russell, B. C., Torralba, A., Murphy, K. P., & Freeman, W. T. (2008). LabelMe: A database and web-based tool for image annotation. International Journal of Computer Vision, 77 (1–3), 157–173. http...

work page doi:10.1007/s11263-007-0090-8 2008

[12] [12]

A., Norouzzadeh, M

https://doi.org/10.3390/rs16234440 Tabak, M. A., Norouzzadeh, M. S., Wolfson, D. W., Sweeney, S. J., Vercauteren, K. C., Snow, N. P., Halseth, J. M., Salvo, P. A. D., Lewis, J. S., White, M. D., Teton, B., Beasley, J. C., Schlichting, P. E., Boughton, R. K., Wight, B., Newkirk, E. S., Ivan, J. S., Odell, E. A., Brook, R. K., … Miller, R. S. (2019). Machin...

work page doi:10.3390/rs16234440 2019

[13] [13]

M.-C., & Hedström, A

https://doi.org/10.3390/rs11111309 Wickstrøm, K., Höhne, M. M.-C., & Hedström, A. (2024). From flexibility to manipulation: The slippery slope of XAI evaluation. In Proceedings of the ECCV 2024 Workshop on Explainable Computer Vision: Where are We and Where are We Going? arXiv. https://arxiv.org/abs/2412.05592 Wiegreffe, S., & Pinter, Y. (2019). Attention...

work page doi:10.3390/rs11111309 2024

[14] [14]

https://doi.org/10.3390/rs17111828 Zurell, D., et al. (2020). A standard protocol for reporting species distribution models. Ecography , 43(9), 1261–1277. https://doi.org/10.1111/ecog.04960 51

work page doi:10.3390/rs17111828 2020