Benthic ecological surveys yield a massive volume of seabed imagery, yet analyzing the abundance of organisms remains a time-consuming task for experts. This bottleneck hinders the analysis of all collected data. Convolutional Neural Networks (CNNs) offer a promising solution for automating image analysis. However, training CNNs requires images with bounding boxes drawn around the target organisms. Such datasets are often unavailable, as prior research primarily relied on manual point annotations for organism locations. This study presents a novel workflow for training CNN to identify benthic organisms using existing point annotations. We demonstrate that legacy point annotations from previous surveys can be used to annotate new images collected within the same study area. Our results show that the CNN's predictions were comparable to discrepancies found in inter-expert variability. While the accuracy may not surpass models trained with dedicated bounding box datasets, our approach proves that historical point annotations can effectively generate training data for object detection CNNs, particularly when dedicated bounding box datasets are scarce. Given the vast number of past and ongoing benthic surveys utilizing point annotations, this approach unlocks new avenues for machine learning in marine ecology.
{"title":"Training neural network for benthic image analysis using legacy point annotations: A case study in HAUSGARTEN LTER","authors":"Caroline Johansen , Yann Marcon , Lilian Böhringer , Autun Purser","doi":"10.1016/j.ecoinf.2025.103556","DOIUrl":"10.1016/j.ecoinf.2025.103556","url":null,"abstract":"<div><div>Benthic ecological surveys yield a massive volume of seabed imagery, yet analyzing the abundance of organisms remains a time-consuming task for experts. This bottleneck hinders the analysis of all collected data. Convolutional Neural Networks (CNNs) offer a promising solution for automating image analysis. However, training CNNs requires images with bounding boxes drawn around the target organisms. Such datasets are often unavailable, as prior research primarily relied on manual point annotations for organism locations. This study presents a novel workflow for training CNN to identify benthic organisms using existing point annotations. We demonstrate that legacy point annotations from previous surveys can be used to annotate new images collected within the same study area. Our results show that the CNN's predictions were comparable to discrepancies found in inter-expert variability. While the accuracy may not surpass models trained with dedicated bounding box datasets, our approach proves that historical point annotations can effectively generate training data for object detection CNNs, particularly when dedicated bounding box datasets are scarce. Given the vast number of past and ongoing benthic surveys utilizing point annotations, this approach unlocks new avenues for machine learning in marine ecology.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103556"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925479","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2025-11-22DOI: 10.1016/j.ecoinf.2025.103488
Javiera Boada-Campos , Felipe Lobos-Roco , Francisca Aguirre-Correa , Francisco Suárez
{"title":"Corrigendum to “A machine learning-based analysis of actual evaporation predictors across different land covers in high-elevation drylands” [Ecological Informatics, Volume 92 (2025), 103471, https://doi.org/10.1016/j.ecoinf.2025.103471]","authors":"Javiera Boada-Campos , Felipe Lobos-Roco , Francisca Aguirre-Correa , Francisco Suárez","doi":"10.1016/j.ecoinf.2025.103488","DOIUrl":"10.1016/j.ecoinf.2025.103488","url":null,"abstract":"","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103488"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147385302","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2025-12-27DOI: 10.1016/j.ecoinf.2025.103586
Rebecca Millington, Dale Partridge, Helen R. Powley, Gennadi Lessin, David Moffat, Jerry Blackford
Rapid advancement of machine learning and artificial intelligence is enabling new analysis techniques to be applied across all fields of scientific research. To aid analysis of the physical or biogeochemical characteristics of the ocean, marine systems have been subdivided into spatial regions where properties exhibit similar distributions or behaviour, such as the Longhurst provinces. Machine learning techniques enable the identification of spatial regions in a robust and transferable way. In this paper we drive clustering algorithms with a variety of input datasets to assess the consistency of resulting clusters. We compare the results of clustering analyses applied separately to physical, biogeochemical and ecological variables at different depths, using model output from a 3D hydrodynamical-biogeochemical model (NEMO-ERSEM) on the Northwest European shelf. Clustering outcomes depended on both the variables and depths input into the algorithm, although some similarities still existed in spatial patterns between each clustering analysis, e.g. clusters were smaller near the coast and relatively extensive in the open ocean. Clusters based on physical properties showed latitudinal distribution, while biogeochemical and ecological inputs resulted in a higher concentration of clusters near the coast. Results from depth-averaged and near-bottom inputs were similar and followed the limits of the shelf-edge, unlike clusters based on surface inputs. Overall, clustering algorithms offer a useful method to define spatial regions with similar characteristics, however, our results emphasise that input data choices should be carefully considered. Our results provide a knowledge foundation which can help future researchers make informed decisions when applying clustering to complex datasets.
{"title":"Consistency of clustering analysis of complex 3D ocean datasets","authors":"Rebecca Millington, Dale Partridge, Helen R. Powley, Gennadi Lessin, David Moffat, Jerry Blackford","doi":"10.1016/j.ecoinf.2025.103586","DOIUrl":"10.1016/j.ecoinf.2025.103586","url":null,"abstract":"<div><div>Rapid advancement of machine learning and artificial intelligence is enabling new analysis techniques to be applied across all fields of scientific research. To aid analysis of the physical or biogeochemical characteristics of the ocean, marine systems have been subdivided into spatial regions where properties exhibit similar distributions or behaviour, such as the Longhurst provinces. Machine learning techniques enable the identification of spatial regions in a robust and transferable way. In this paper we drive clustering algorithms with a variety of input datasets to assess the consistency of resulting clusters. We compare the results of clustering analyses applied separately to physical, biogeochemical and ecological variables at different depths, using model output from a 3D hydrodynamical-biogeochemical model (NEMO-ERSEM) on the Northwest European shelf. Clustering outcomes depended on both the variables and depths input into the algorithm, although some similarities still existed in spatial patterns between each clustering analysis, e.g. clusters were smaller near the coast and relatively extensive in the open ocean. Clusters based on physical properties showed latitudinal distribution, while biogeochemical and ecological inputs resulted in a higher concentration of clusters near the coast. Results from depth-averaged and near-bottom inputs were similar and followed the limits of the shelf-edge, unlike clusters based on surface inputs. Overall, clustering algorithms offer a useful method to define spatial regions with similar characteristics, however, our results emphasise that input data choices should be carefully considered. Our results provide a knowledge foundation which can help future researchers make informed decisions when applying clustering to complex datasets.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103586"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925405","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2025-12-25DOI: 10.1016/j.ecoinf.2025.103573
Yufan Zhang, Fouad Jaber
Flood risk management in rapidly urbanizing areas requires frameworks that can simultaneously address flood susceptibility and vulnerability while accommodating future landscape and climate dynamics. This study proposes an upgraded GIS-based multi-criteria decision-making framework to develop flood control prioritization maps by integrating both susceptibility and vulnerability indices. The framework considers a wide range of conditioning factors across environmental, infrastructural, and socio-economic domains, and introduces four novel factors—dam density, road traffic density, bridge traffic density, and bridge vulnerability—to better capture the two-way interactions between infrastructure and flood risk. Factor weights were derived using the Analytic Hierarchy Process (AHP). Methodological innovations include the application of Kernel Density functions instead of traditional Euclidean distance to represent spatial influence, a new normalization approach for factor rating to reduce subjectivity, and a detailed GIS-based procedure for generating Curve Numbers (CN). Beyond current conditions, the framework also evaluates dynamic flood risk patterns under a 2045 scenario by incorporating projected changes in extreme rainfall depth, impervious surfaces, and traffic patterns. The approach was tested in the rapidly developing west Dallas–Fort Worth metroplex and validated against flood inventory data using Receiver Operating Characteristic (ROC) curves and, the Area Under the Curve (AUC) shows an acceptable value of 0.659. Results demonstrate the framework's ability to streamline flood mitigation planning and support decision-making in urban development under climate change and urban sprawl pressures. The proposed framework is transferable to other metropolitan regions seeking to enhance resilience through integrated flood risk management.
{"title":"An upgraded GIS-based multi-criteria decision-making approach for flood control prioritization mapping: Case study of West Dallas-Fort Worth metroplex","authors":"Yufan Zhang, Fouad Jaber","doi":"10.1016/j.ecoinf.2025.103573","DOIUrl":"10.1016/j.ecoinf.2025.103573","url":null,"abstract":"<div><div>Flood risk management in rapidly urbanizing areas requires frameworks that can simultaneously address flood susceptibility and vulnerability while accommodating future landscape and climate dynamics. This study proposes an upgraded GIS-based multi-criteria decision-making framework to develop flood control prioritization maps by integrating both susceptibility and vulnerability indices. The framework considers a wide range of conditioning factors across environmental, infrastructural, and socio-economic domains, and introduces four novel factors—dam density, road traffic density, bridge traffic density, and bridge vulnerability—to better capture the two-way interactions between infrastructure and flood risk. Factor weights were derived using the Analytic Hierarchy Process (AHP). Methodological innovations include the application of Kernel Density functions instead of traditional Euclidean distance to represent spatial influence, a new normalization approach for factor rating to reduce subjectivity, and a detailed GIS-based procedure for generating Curve Numbers (CN). Beyond current conditions, the framework also evaluates dynamic flood risk patterns under a 2045 scenario by incorporating projected changes in extreme rainfall depth, impervious surfaces, and traffic patterns. The approach was tested in the rapidly developing west Dallas–Fort Worth metroplex and validated against flood inventory data using Receiver Operating Characteristic (ROC) curves and, the Area Under the Curve (AUC) shows an acceptable value of 0.659. Results demonstrate the framework's ability to streamline flood mitigation planning and support decision-making in urban development under climate change and urban sprawl pressures. The proposed framework is transferable to other metropolitan regions seeking to enhance resilience through integrated flood risk management.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103573"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925406","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2026-01-14DOI: 10.1016/j.ecoinf.2026.103609
Richen Huang , Li Li , Lingrong Xu , Lloyd Hasson , Shuhua Peng , Jiali Luo
Deep learning methods have achieved remarkable success in plant disease recognition. However, these methods rely on large-scale labeled datasets for training to ensure the reliability of empirical risk minimization. In the real world, obtaining such extensive disease data remains challenging. With limited data, traditional correlation-based learning frameworks could establish spurious correlations between disease data and disease classes, which severely harms their generalization ability. We address this issue from a causal perspective by proposing the Deep Counterfactual Metric Framework (DCMF). Specifically, DCMF employs a Counterfactual Reasoning Module (CRM) to construct a counterfactual world where each disease image contains only healthy features, enabling estimation of the direct effect of healthy regions on disease recognition. By subtracting this direct effect from the total effect on classes, we effectively eliminate spurious correlations, allowing the model to learn robust disease-specific features for reliable generalization in limited data scenarios. Extensive experiments on PlantVillage and PlantLeaves datasets under 5-shot and 10-shot settings demonstrate that DCMF achieves an average performance improvement of 7.2% over the best baseline methods. These improvements validate the effectiveness of DCMF in limited data plant disease recognition.
{"title":"DCMF: Deep Counterfactual Metric Framework for limited data plant disease recognition","authors":"Richen Huang , Li Li , Lingrong Xu , Lloyd Hasson , Shuhua Peng , Jiali Luo","doi":"10.1016/j.ecoinf.2026.103609","DOIUrl":"10.1016/j.ecoinf.2026.103609","url":null,"abstract":"<div><div>Deep learning methods have achieved remarkable success in plant disease recognition. However, these methods rely on large-scale labeled datasets for training to ensure the reliability of empirical risk minimization. In the real world, obtaining such extensive disease data remains challenging. With limited data, traditional correlation-based learning frameworks could establish spurious correlations between disease data and disease classes, which severely harms their generalization ability. We address this issue from a causal perspective by proposing the Deep Counterfactual Metric Framework (DCMF). Specifically, DCMF employs a Counterfactual Reasoning Module (CRM) to construct a counterfactual world where each disease image contains only healthy features, enabling estimation of the direct effect of healthy regions on disease recognition. By subtracting this direct effect from the total effect on classes, we effectively eliminate spurious correlations, allowing the model to learn robust disease-specific features for reliable generalization in limited data scenarios. Extensive experiments on PlantVillage and PlantLeaves datasets under 5-shot and 10-shot settings demonstrate that DCMF achieves an average performance improvement of 7.2% over the best baseline methods. These improvements validate the effectiveness of DCMF in limited data plant disease recognition.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103609"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977423","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2025-11-30DOI: 10.1016/j.ecoinf.2025.103544
Antonio Pita, Juan M. Navarro
Monitoring acoustic environments in urban ecosystems poses a major challenge due to the temporal and spatial variability of soundscapes. Long-term data collection, often extending over a year, is recommended by regulations to establish reliable acoustic profiles, but such efforts are resource-intensive. In this study, we introduce a computational ecology approach to predict long-term acoustic patterns in cities using optimized combinations of time intervals as input for artificial neural networks. Unlike conventional methods relying on a single temporal window, our framework evaluates paired time intervals to enhance predictive performance and capture the dynamics of complex urban soundscapes. Multiple neural network architectures were designed and comparatively assessed, demonstrating that 2-slot datasets consistently improved classification accuracy and Balanced Accuracy Micro-Averaging across all categories. On average, temporal pairing increased Balanced Accuracy from 0.576 to 0.763 in the most variable category, reflecting a 32.4% improvement. These results highlight the importance of temporal diversity in ecological data modeling and underscore the potential of computational techniques to optimize temporary monitoring stations. The proposed method supports more efficient, data-driven strategies for environmental noise prediction, with direct implications for sustainable urban ecosystem management and decision-making in the context of global environmental change.
{"title":"Predicting long-term environmental acoustic urban patterns using 2-slot short-term measurement and feed-forward artificial neural networks","authors":"Antonio Pita, Juan M. Navarro","doi":"10.1016/j.ecoinf.2025.103544","DOIUrl":"10.1016/j.ecoinf.2025.103544","url":null,"abstract":"<div><div>Monitoring acoustic environments in urban ecosystems poses a major challenge due to the temporal and spatial variability of soundscapes. Long-term data collection, often extending over a year, is recommended by regulations to establish reliable acoustic profiles, but such efforts are resource-intensive. In this study, we introduce a computational ecology approach to predict long-term acoustic patterns in cities using optimized combinations of time intervals as input for artificial neural networks. Unlike conventional methods relying on a single temporal window, our framework evaluates paired time intervals to enhance predictive performance and capture the dynamics of complex urban soundscapes. Multiple neural network architectures were designed and comparatively assessed, demonstrating that 2-slot datasets consistently improved classification accuracy and Balanced Accuracy Micro-Averaging across all categories. On average, temporal pairing increased Balanced Accuracy from 0.576 to 0.763 in the most variable category, reflecting a 32.4% improvement. These results highlight the importance of temporal diversity in ecological data modeling and underscore the potential of computational techniques to optimize temporary monitoring stations. The proposed method supports more efficient, data-driven strategies for environmental noise prediction, with direct implications for sustainable urban ecosystem management and decision-making in the context of global environmental change.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103544"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145645758","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2025-12-01DOI: 10.1016/j.ecoinf.2025.103541
Chao Sun , Ming Hu , Shu Zhang , Xingru Shen , Chenwei Zhao , Penghui Jiang
Understanding how cropland morphology influences ecological quality is essential for sustaining agricultural development, but these two dimensions are often studied independently. This study develops an integrated framework that links cropland morphological evolution with ecological quality by extending the landscape ecology “pattern–process–function” theory to agricultural systems. Using time-series Landsat data, we applied a continuous change detection model to generate annual cropland maps, classify morphological structures (core, perforated, edge, scattered), and encode pixel-level transitions to track evolutionary pathways. A Comprehensive Ecological Evaluation Index (CEEI) was constructed from multiple remote sensing indicators to quantify ecological quality, and Cohen's d was used to compare differences among morphological types. Applied to the Hangzhou Bay Area (1990–2020), the framework demonstrated that: (i) 25.48 % of cropland underwent morphological transitions, with over 91 % shifting from core to scattered configurations; (ii) Two dominant stepwise pathways—core→edge→scattered (51 %) and core→perforated→edge (25 %)—together with one direct pathway (core→scattered, 12 %) characterized these transitions; (iii) Regional ecological quality continuously declined, with the mean CEEI decreasing from 0.648 to 0.600, accompanied by intensified heat stress (LST relative change: −0.037); (iv) Early-stage transitions from core cropland (core→edge and core→perforated) contributed most to ecological degradation, corresponding to mean CEEI reductions of 0.055 and 0.041, respectively. These findings indicate that preventing the loss of core cropland, especially in the plains of Shanghai and Jiaxing, is key to preserving agricultural ecosystem health. Our proposed framework is flexible in the selection of remote sensing indicators and is broadly applicable to other ecosystems, providing actionable insights for ecological restoration based on morphological configuration.
{"title":"A novel framework for bridging cropland morphological structure and ecological quality using a remote sensing-based continuous change detection model","authors":"Chao Sun , Ming Hu , Shu Zhang , Xingru Shen , Chenwei Zhao , Penghui Jiang","doi":"10.1016/j.ecoinf.2025.103541","DOIUrl":"10.1016/j.ecoinf.2025.103541","url":null,"abstract":"<div><div>Understanding how cropland morphology influences ecological quality is essential for sustaining agricultural development, but these two dimensions are often studied independently. This study develops an integrated framework that links cropland morphological evolution with ecological quality by extending the landscape ecology “pattern–process–function” theory to agricultural systems. Using time-series Landsat data, we applied a continuous change detection model to generate annual cropland maps, classify morphological structures (core, perforated, edge, scattered), and encode pixel-level transitions to track evolutionary pathways. A Comprehensive Ecological Evaluation Index (CEEI) was constructed from multiple remote sensing indicators to quantify ecological quality, and Cohen's d was used to compare differences among morphological types. Applied to the Hangzhou Bay Area (1990–2020), the framework demonstrated that: (i) 25.48 % of cropland underwent morphological transitions, with over 91 % shifting from core to scattered configurations; (ii) Two dominant stepwise pathways—core→edge→scattered (51 %) and core→perforated→edge (25 %)—together with one direct pathway (core→scattered, 12 %) characterized these transitions; (iii) Regional ecological quality continuously declined, with the mean CEEI decreasing from 0.648 to 0.600, accompanied by intensified heat stress (LST relative change: −0.037); (iv) Early-stage transitions from core cropland (core→edge and core→perforated) contributed most to ecological degradation, corresponding to mean CEEI reductions of 0.055 and 0.041, respectively. These findings indicate that preventing the loss of core cropland, especially in the plains of Shanghai and Jiaxing, is key to preserving agricultural ecosystem health. Our proposed framework is flexible in the selection of remote sensing indicators and is broadly applicable to other ecosystems, providing actionable insights for ecological restoration based on morphological configuration.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103541"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685095","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2025-12-21DOI: 10.1016/j.ecoinf.2025.103567
Agnethe S. Olsen , Paul L. Rosin , Christopher B. Jones , Jo Cable , Sarah E. Perkins
Effective disease surveillance in wild fish populations is essential for food security and biodiversity conservation, but data acquisition can be limited by ad hoc reporting and resource-intensive laboratory diagnostics. We developed and evaluated a computer vision pipeline to detect saprolegniasis-like infections, a devastating disease in salmonids that manifests as visible signs.
Compiling a dataset of 4526 images (494 infected, 4032 healthy) from citizen science platforms and stakeholders, we used data augmentation to address the significant class imbalance. We then fine-tuned and compared four pre-trained convolutional neural network architectures (EfficientNetV2S, EfficientNetV2B0, ResNet50, and MobileNetV3S), chosen to represent a range of standard and efficient models, to classify healthy versus infected fish across datasets of varying host taxonomic specificity.
The EfficientNetV2S model achieved the highest performance on a Salmo spp. specific dataset, with a mean recall (proportion of infected fish images correctly identified) of 0.898 ( 0.043) and precision (proportion of correctly identified infected fish among all fish identified as infected) of 0.858 ( 0.067). Performance varied with host taxonomic scope, with models achieving lower metrics on broader host taxa datasets. Despite challenges including variable image quality, water surface reflections, and inherent class imbalance, these results show computer vision can support large-scale disease surveillance in wild fish populations.
Computer vision-based surveillance could enable earlier outbreak detection and targeted diagnostics, improving freshwater ecosystem health management. While successful implementation hinges on acquiring sufficient high-quality imagery, this study highlights the potential of applying tailored Artificial Intelligence tools for monitoring visually detectable diseases across diverse wildlife species.
{"title":"Computer vision for infectious disease surveillance; Saprolegnia spp. in salmonids","authors":"Agnethe S. Olsen , Paul L. Rosin , Christopher B. Jones , Jo Cable , Sarah E. Perkins","doi":"10.1016/j.ecoinf.2025.103567","DOIUrl":"10.1016/j.ecoinf.2025.103567","url":null,"abstract":"<div><div>Effective disease surveillance in wild fish populations is essential for food security and biodiversity conservation, but data acquisition can be limited by ad hoc reporting and resource-intensive laboratory diagnostics. We developed and evaluated a computer vision pipeline to detect saprolegniasis-like infections, a devastating disease in salmonids that manifests as visible signs.</div><div>Compiling a dataset of 4526 images (494 infected, 4032 healthy) from citizen science platforms and stakeholders, we used data augmentation to address the significant class imbalance. We then fine-tuned and compared four pre-trained convolutional neural network architectures (EfficientNetV2S, EfficientNetV2B0, ResNet50, and MobileNetV3S), chosen to represent a range of standard and efficient models, to classify healthy versus infected fish across datasets of varying host taxonomic specificity.</div><div>The EfficientNetV2S model achieved the highest performance on a <em>Salmo</em> spp. specific dataset, with a mean recall (proportion of infected fish images correctly identified) of 0.898 (<span><math><mo>±</mo></math></span> 0.043) and precision (proportion of correctly identified infected fish among all fish identified as infected) of 0.858 (<span><math><mo>±</mo></math></span> 0.067). Performance varied with host taxonomic scope, with models achieving lower metrics on broader host taxa datasets. Despite challenges including variable image quality, water surface reflections, and inherent class imbalance, these results show computer vision can support large-scale disease surveillance in wild fish populations.</div><div>Computer vision-based surveillance could enable earlier outbreak detection and targeted diagnostics, improving freshwater ecosystem health management. While successful implementation hinges on acquiring sufficient high-quality imagery, this study highlights the potential of applying tailored Artificial Intelligence tools for monitoring visually detectable diseases across diverse wildlife species.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103567"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925319","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-01Epub Date: 2025-12-25DOI: 10.1016/j.ecoinf.2025.103584
Li-Dunn Chen , Stephen Dodds , Molly McGuire , Maria Franke , Gabriela Mastromonaco
Machine learning (ML)-aided technologies can be applied to many of the existing wildlife science tools (e.g., camera traps) used to support conservation initiatives both in situ and ex situ. The automated nature of ML methods reduces manual labour, extends monitoring efforts past regular daylight/working hours, and improves the overall diagnostic capacity of tools routinely applied by wildlife biologists and animal care staff at zoological institutions. Though the conservation aims and expectations may differ among zoos and aquariums, simple monitoring tools that impose less demand on animal care staff should serve as an important aid for advancing management strategies for threatened species. We applied computer vision-based predictive models built on CCTV footage from a zoo-housed Panthera tigris individual to develop an automated behavioural monitoring tool (“PantherAI”) capable of rapidly assessing activity budget and space use across variable lighting and weather conditions. We applied YOLOv8 as the model backbone to detect and classify several tiger behaviours (e.g., stereotypical pacing, resting, enrichment interaction, feeding); the trained models were then applied with scripts to autonomously generate customized activity budgets and space use heatmaps from 24-h video samples. PantherAI yielded a mean average precision >75% on test data, where it detected and classified tiger behaviours with varying levels of accuracy (stereotypical pacing: 92.2%, resting: 72.2%, locomotion: 65.4%, feeding: 34.4%, object manipulation: 43.8%). Activity budgets varied (p < 0.05) across habitats and by time of day for several behaviours. PantherAI provided reliable estimates of behaviour and space usage, two important ecological metrics commonly used to establish baseline activity budgets and assess indicators of animal welfare. Overall, ML-coupled technologies can facilitate daily data collection and monitoring procedures, both of which are integral for objectively measuring behavioural outcomes as newly implemented husbandry practices (e.g., alterations to diet, environment, social group, enrichment) are enacted in zoological and other ex situ conservation settings.
{"title":"PantherAI: An autonomous behavioural monitoring tool for assessing activity budget and space use in a zoo-housed tiger","authors":"Li-Dunn Chen , Stephen Dodds , Molly McGuire , Maria Franke , Gabriela Mastromonaco","doi":"10.1016/j.ecoinf.2025.103584","DOIUrl":"10.1016/j.ecoinf.2025.103584","url":null,"abstract":"<div><div>Machine learning (ML)-aided technologies can be applied to many of the existing wildlife science tools (e.g., camera traps) used to support conservation initiatives both in situ and ex situ. The automated nature of ML methods reduces manual labour, extends monitoring efforts past regular daylight/working hours, and improves the overall diagnostic capacity of tools routinely applied by wildlife biologists and animal care staff at zoological institutions. Though the conservation aims and expectations may differ among zoos and aquariums, simple monitoring tools that impose less demand on animal care staff should serve as an important aid for advancing management strategies for threatened species. We applied computer vision-based predictive models built on CCTV footage from a zoo-housed <em>Panthera tigris</em> individual to develop an automated behavioural monitoring tool (“PantherAI”) capable of rapidly assessing activity budget and space use across variable lighting and weather conditions. We applied YOLOv8 as the model backbone to detect and classify several tiger behaviours (e.g., stereotypical pacing, resting, enrichment interaction, feeding); the trained models were then applied with scripts to autonomously generate customized activity budgets and space use heatmaps from 24-h video samples. PantherAI yielded a mean average precision >75% on test data, where it detected and classified tiger behaviours with varying levels of accuracy (stereotypical pacing: 92.2%, resting: 72.2%, locomotion: 65.4%, feeding: 34.4%, object manipulation: 43.8%). Activity budgets varied (<em>p</em> < 0.05) across habitats and by time of day for several behaviours. PantherAI provided reliable estimates of behaviour and space usage, two important ecological metrics commonly used to establish baseline activity budgets and assess indicators of animal welfare. Overall, ML-coupled technologies can facilitate daily data collection and monitoring procedures, both of which are integral for objectively measuring behavioural outcomes as newly implemented husbandry practices (e.g., alterations to diet, environment, social group, enrichment) are enacted in zoological and other ex situ conservation settings.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103584"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Deep learning has been extensively used in fisheries science, as it enables the acquisition of information regarding the body length and stock-abundance index of target fish from images, thereby facilitating stock assessment and management. However, generally, multiple species appear together in images obtained from fisheries, necessitating the classification of fish species before extracting relevant biological information. Improving the performance of fish detection and species classification is crucial as it affects the quality of biological information that could be inferred from images. Previous studies have reported that increasing the input-image size can affect the classification accuracy. Identification characteristics of fish are small in comparison with their body size, and increasing the image size can affect the classification accuracy; however, there are no reports on the effect of image size on fish species-classification accuracy. Herein, different input-image sizes were taken to evaluate the effect of input-image size on the performance of fish detection and species classification. Fish images (41,922 fish across 41 classes) were acquired from conveyor belts to sort set-net fish catches. Fish were detected and classified using a mask region-based convolutional neural network. The effect of input-image size on performance was examined using nine datasets in three image sizes of 1333 × 888, 2000 × 1333, and 2666 × 1777 pixels, obtaining an average mAP50–95 value of 0.586, 0.612, and 0.609, respectively. Larger image sizes offered improved performance compared with that of the smallest, averaging 0.026 and 0.023 improvements in mAP50–95 at two larger image sizes. However, when comparing the degree of improvement between image sizes of 2000 × 1333 pixels and 2666 × 1777 pixels under fine-tuning conditions, the former size resulted in higher performance. Performance was observed to improve for species with low performance at the smallest image size; therefore, we can say that increasing the input-image size is a simple and effective way for improving detection and classification performance for these species.
{"title":"Effects of input-image size on performance of fish detection and species classification using deep learning","authors":"Yuka Iwahara , Yasutoki Shibata , Masahiro Manano , Tomoya Nishino , Ryosuke Kariya , Hiroki Yaemori","doi":"10.1016/j.ecoinf.2025.103566","DOIUrl":"10.1016/j.ecoinf.2025.103566","url":null,"abstract":"<div><div>Deep learning has been extensively used in fisheries science, as it enables the acquisition of information regarding the body length and stock-abundance index of target fish from images, thereby facilitating stock assessment and management. However, generally, multiple species appear together in images obtained from fisheries, necessitating the classification of fish species before extracting relevant biological information. Improving the performance of fish detection and species classification is crucial as it affects the quality of biological information that could be inferred from images. Previous studies have reported that increasing the input-image size can affect the classification accuracy. Identification characteristics of fish are small in comparison with their body size, and increasing the image size can affect the classification accuracy; however, there are no reports on the effect of image size on fish species-classification accuracy. Herein, different input-image sizes were taken to evaluate the effect of input-image size on the performance of fish detection and species classification. Fish images (41,922 fish across 41 classes) were acquired from conveyor belts to sort set-net fish catches. Fish were detected and classified using a mask region-based convolutional neural network. The effect of input-image size on performance was examined using nine datasets in three image sizes of 1333 × 888, 2000 × 1333, and 2666 × 1777 pixels, obtaining an average mAP50–95 value of 0.586, 0.612, and 0.609, respectively. Larger image sizes offered improved performance compared with that of the smallest, averaging 0.026 and 0.023 improvements in mAP50–95 at two larger image sizes. However, when comparing the degree of improvement between image sizes of 2000 × 1333 pixels and 2666 × 1777 pixels under fine-tuning conditions, the former size resulted in higher performance. Performance was observed to improve for species with low performance at the smallest image size; therefore, we can say that increasing the input-image size is a simple and effective way for improving detection and classification performance for these species.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103566"},"PeriodicalIF":7.3,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925483","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号