An emerging tool for ecological surveillance: machine learning to identify small mammals from their footprints

Small mammals are important ecological indicators but remain under-represented in biodiversity monitoring because traditional survey methods can be labor-intensive, invasive, or difficult to scale. We introduce a non-invasive ecological surveillance approach that combines trackplates and computer vision to identify species from individual footprint images. By converting footprints into a standardized digital data stream, the method addresses a critical gap in biodiversity monitoring and ecological surveillance. Between October 2023 and March 2025, footprints from identified small mammal species were collected at five sites in South Africa. These data were curated into a reference library and processed into a benchmark dataset comprising 3,355 cropped footprint images from 122 usable trackplates representing 11 species. Each footprint received a primary species label and an auxiliary front-versus-hind label within a multi-task learning framework designed to promote anatomically meaningful feature learning. The final DINOv3-based Vision Transformer was evaluated using five-fold cross-validation grouped at the trackplate level to ensure assessment on previously unseen trackplates and prevent information leakage. The model achieved a mean cross-validation accuracy of 89.2% (SD = 2.4%) across five folds and a best-fold accuracy of 92.5%. Balanced species-level performance was also strong, with a mean macro F1 score of 88.7% (SD = 3.0%). Several species were classified with near-perfect performance despite substantial variation in footprint orientation, completeness, and source-image quality. Residual errors were concentrated among a small subset of morphologically similar species. Explainable AI analyses indicated that the model organized footprints according to morphological similarity and relied on biologically meaningful features, including toe placement, pad geometry, and overall footprint shape, rather than plate-specific artifacts. These findings demonstrate that individual small mammal footprints contain sufficient morphological information for robust species-level discrimination. This approach enables small mammal species-richness data to serve as a scalable, standardized, and non-invasive indicator, with strong potential to extend small mammal monitoring across protected areas, production landscapes, and community-based schemes aligned with the Kunming–Montreal Global Biodiversity Framework.

Authors

Institutions

Publication Details

Journal
Frontiers in Ecology and Evolution
Published
2026-11-19
DOI
https://doi.org/10.3389/fevo.2026.1983895
Primary Topic
Wildlife Ecology and Conservation
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An emerging tool for ecological surveillance: machine learning to identify small mammals from their footprints

Maria Kathleen Oosthuizen, Nico L. Avenant, Sky K. Alibhai, Zoë C. Jewell et al.
Frontiers in Ecology and Evolution
Wildlife Ecology and Conservation
article

An emerging tool for ecological surveillance: machine learning to identify small mammals from their footprints

Maria Kathleen Oosthuizen, Nico L. Avenant, Sky K. Alibhai, Zoë C. Jewell, Duncan N. MacFadyen, Debbie Liske, Lynn Carlson
article en

Abstract

Small mammals are important ecological indicators but remain under-represented in biodiversity monitoring because traditional survey methods can be labor-intensive, invasive, or difficult to scale. We introduce a non-invasive ecological surveillance approach that combines trackplates and computer vision to identify species from individual footprint images. By converting footprints into a standardized digital data stream, the method addresses a critical gap in biodiversity monitoring and ecological surveillance. Between October 2023 and March 2025, footprints from identified small mammal species were collected at five sites in South Africa. These data were curated into a reference library and processed into a benchmark dataset comprising 3,355 cropped footprint images from 122 usable trackplates representing 11 species. Each footprint received a primary species label and an auxiliary front-versus-hind label within a multi-task learning framework designed to promote anatomically meaningful feature learning. The final DINOv3-based Vision Transformer was evaluated using five-fold cross-validation grouped at the trackplate level to ensure assessment on previously unseen trackplates and prevent information leakage. The model achieved a mean cross-validation accuracy of 89.2% (SD = 2.4%) across five folds and a best-fold accuracy of 92.5%. Balanced species-level performance was also strong, with a mean macro F1 score of 88.7% (SD = 3.0%). Several species were classified with near-perfect performance despite substantial variation in footprint orientation, completeness, and source-image quality. Residual errors were concentrated among a small subset of morphologically similar species. Explainable AI analyses indicated that the model organized footprints according to morphological similarity and relied on biologically meaningful features, including toe placement, pad geometry, and overall footprint shape, rather than plate-specific artifacts. These findings demonstrate that individual small mammal footprints contain sufficient morphological information for robust species-level discrimination. This approach enables small mammal species-richness data to serve as a scalable, standardized, and non-invasive indicator, with strong potential to extend small mammal monitoring across protected areas, production landscapes, and community-based schemes aligned with the Kunming–Montreal Global Biodiversity Framework.

Frontiers in Ecology and EvolutionVol. 14
University of South Africa (ZA), Duke University (US), University of the Free State (ZA), National Museum (ZA), University of Pretoria (ZA)
Wildlife Ecology and Conservation
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon