Ecological Informatics最新文献

{"title":"A digital-twin strategy using robots for marine ecosystem monitoring","authors":"Jacopo Aguzzi ,&nbsp;Elias Chatzidouros ,&nbsp;Damianos Chatzievangelou ,&nbsp;Morane Clavel-Henry ,&nbsp;Sascha Flögel ,&nbsp;Nixon Bahamon ,&nbsp;Michael Tangerlini ,&nbsp;Laurenz Thomsen ,&nbsp;Giacomo Picardi ,&nbsp;Joan Navarro ,&nbsp;Ivan Masmitja ,&nbsp;Nathan J. Robinson ,&nbsp;Tim Nattkemper ,&nbsp;Sergio Stefanni ,&nbsp;José Quintana ,&nbsp;Ricard Campos ,&nbsp;Rafael García ,&nbsp;Emanuela Fanelli ,&nbsp;Marco Francescangeli ,&nbsp;Luca Mirimin ,&nbsp;Jennifer Doyle","doi":"10.1016/j.ecoinf.2025.103409","DOIUrl":"10.1016/j.ecoinf.2025.103409","url":null,"abstract":"<div><div>Effective marine conservation and management require ecological monitoring in the form of intensive real-time data collection over large spatial scales. The combined use of fixed platforms (e.g., cabled observatories) and research vessels with platforms of different levels of teleoperated autonomy (e.g., remotely operated vehicles (ROVs) and autonomous underwater vehicles (AUVs) can contribute to the acquisition of large multiparametric biological and environmental data. If those data are spatially combined, sufficient spatial coverage can be achieved for ecological monitoring. A digital twin of the ocean (DTO) approach can then be used as a virtual representation of that monitored space, enabling multiparametric analyses of environmental patterns and processes affecting biodiversity and species distributions, as well as socioeconomic activities. Here, we propose a general architecture for a DTO centred on real-time data collection from local networks on fixed and mobile platforms, such as the physical twin observers (PTO), which is synergistically merged with platforms operating at large geographic scales. We describe a roadmap to achieve this DTO via 4 key steps: (1) acquisition of in situ data with a robotic network of platforms; (2) the application of AI in image processing for extracting biological data; (3) big data management with data bubbles; and (4) development of the resulting DTO framework for providing ecosystem monitoring via the computation of ecological indicators and socioecological modelling.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103409"},"PeriodicalIF":7.3,"publicationDate":"2025-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144920124","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}

{"title":"River basin urban flood resilience: A multi-dimensional framework for risk mitigation to adaptive management and ecosystem protection under changing climate","authors":"Shan-e-hyder Soomro ,&nbsp;Huaibin Wei ,&nbsp;Muhammad Waseem Boota ,&nbsp;Nishan-E-hyder Soomro ,&nbsp;Muhammad Faisal ,&nbsp;Sana Nazli ,&nbsp;Soraya sarwari ,&nbsp;Xiaotao Shi ,&nbsp;Caihong Hu ,&nbsp;Jiali Guo ,&nbsp;Yinghai Li","doi":"10.1016/j.ecoinf.2025.103412","DOIUrl":"10.1016/j.ecoinf.2025.103412","url":null,"abstract":"<div><h3>Study region</h3><div>This study aims at the Kunhar River Basin, Pakistan, that has been facing repeated flood occurrences on a recurring basis. As the flood susceptibility of this area is high, its topographic complexity demands correct predictive modeling for strategic flood planning.</div></div><div><h3>Study focus</h3><div>We developed a system of flood susceptibility mapping based on Geographic Information Systems (GIS), Principal Component Analysis (PCA), and Support Vector Machine (SVM) classification. Four kernel functions were applied, and the highest-performing was the Radial Basis Function (SVM-RBF). The model was validated and trained using historical flood inventories, morphometric parameters, and hydrologic variables, and feature dimensionality was reduced via PCA for increased efficiency.</div></div><div><h3>New hydrological insights</h3><div>The SVM-RBF model recorded an AUC of 0.8341, 88.02% success, 84.97% predictability, 0.89 Kappa value, and F1-score of 0.86, all of which indicated high predictability. Error analysis yielded a PBIAS of +2.14%, indicating negligible overestimation bias but within limits acceptable in hydrological modeling. The results support the superiority of the SVM-RBF approach compared to conventional bivariate methods in modeling flood susceptibility over the complex terrain of mountains. The results can be applied in guiding evidence-based flood mitigation, land-use planning, and adaptive management in the Kunhar River Basin.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103412"},"PeriodicalIF":7.3,"publicationDate":"2025-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144916487","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}

{"title":"Linking groundwater variability to ecosystem carbon and water use efficiencies across India","authors":"Abhishek Chakraborty ,&nbsp;M. Sekhar ,&nbsp;Soumendra N. Bhanja ,&nbsp;Lakshminarayana Rao","doi":"10.1016/j.ecoinf.2025.103411","DOIUrl":"10.1016/j.ecoinf.2025.103411","url":null,"abstract":"<div><div>Carbon use efficiency (CUE) and water use efficiency (WUE) are important indicators of ecosystem health, reflecting the balance between carbon uptake and allocation, and the relationship between carbon assimilation and water loss. Although India shows large spatiotemporal variation in water table depth (WTD), and increasing groundwater stress, the influence of WTD on ecosystem functioning remains underexplored. This study uses satellite-based, modeled, and in-situ datasets to (1) quantify variations in CUE and WUE under shallow (SWTD) and deep (DWTD) WTD conditions across six homogeneous meteorological regions (HMRs), (2) evaluate temporal roles of gross primary productivity (GPP), net primary productivity (NPP), and evapotranspiration (ET) in driving these efficiencies, and (3) examine temporal responses to WTD shifts. SWTD regions generally showed 12 to 18 % higher GPP, 10 to 15 % higher NPP, and 10 to 20 % higher ET than DWTD zones, especially in semi-arid croplands and forests, leading to 8 to 12 % higher CUE and WUE. However, in humid and heavily irrigated regions, CUE was up to 10 % higher in DWTD zones, possibly due to reduced respiration and better soil aeration compared to SWTD areas affected by waterlogging. During the <em>Kharif</em> (wet) season, DWTD croplands in humid zones had higher efficiencies, while in the <em>Rabi</em> (dry) season, SWTD croplands in northern India benefited from irrigation and cooler temperatures. These results highlight strong influence of WTD on carbon and water use processes and support the need for region-specific groundwater strategies.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103411"},"PeriodicalIF":7.3,"publicationDate":"2025-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145018931","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}

{"title":"BiFusionNet: A lightweight model for detecting Red Turpentine Beetle infestation in pine trees","authors":"Xiaorong Zhang ,&nbsp;Yong Xu ,&nbsp;Han Liao","doi":"10.1016/j.ecoinf.2025.103403","DOIUrl":"10.1016/j.ecoinf.2025.103403","url":null,"abstract":"<div><div>The Red Turpentine Beetle (RTB) poses a substantial threat to global pine resources. While remote sensing technologies like unmanned aerial vehicles (UAVs) offer an alternative to inefficient manual inspections, their practical application is often constrained by high computational demands and the scattered distribution of infected trees. To address these challenges, this study introduces BiFusionNet, a novel lightweight deep learning framework for RTB detection. BiFusionNet integrates several architectural innovations to achieve a superior balance between accuracy and efficiency, including a dual-path fusion backbone (DPF-Backbone) for adaptive feature extraction, a LightScale feature pyramid network (LSFPN) for efficient multi-scale fusion, and a lightweight detection head (LWDHead) to minimize computational overhead. The model’s localization accuracy is further enhanced by the Wise-CIoU loss function. On the Pests and Diseases Tree (PDT) dataset, BiFusionNet-n achieves a mean Average Precision (mAP) of 91.53 ± 0.21% at IoU 0.5. Critically, its model size of 0.8 MB and computational load of 0.9 GFLOPs are 84.6% and 85.7% lower, respectively, than those of the You Only Look Once Version 11 (YOLOv11) baseline. By effectively balancing detection accuracy with computational efficiency, BiFusionNet demonstrates significant potential as a practical solution for large-scale, UAV-based forest pest monitoring.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103403"},"PeriodicalIF":7.3,"publicationDate":"2025-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144907889","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}

{"title":"TwinEco: A unified framework for dynamic data-driven digital twins in ecology","authors":"Taimur Khan ,&nbsp;Koen de Koning ,&nbsp;Dag Endresen ,&nbsp;Desalegn Chala ,&nbsp;Erik Kusch","doi":"10.1016/j.ecoinf.2025.103407","DOIUrl":"10.1016/j.ecoinf.2025.103407","url":null,"abstract":"<div><div>A Digital Twin (DT) is a virtual replica of a physical object or process that is continuously updated at a certain frequency and can steer change on the physical system, enabling a seamless integration of observation, understanding, and action. Although initially applied primarily in industry, DT is emerging as a powerful tool in ecology, offering new possibilities for dynamic simulations of change in the biosphere. However, since DTs are relatively new in this field, there is currently no standard framework to guide their conceptualisation and development. Thus, DTs in ecological applications are already experiencing fragmentation in software concepts and design philosophies, leading to incompatibilities across DT implementations. This fragmentation risks undermining the progress and potential of the DT concept in ecology. A unifying framework, such as TwinEco, can address these discrepancies and establish a cohesive foundation for the effective adoption and integration of DTs across ecological domains. TwinEco is a modular framework designed to aid and harmonise ecologists' efforts to build DTs. In doing so, TwinEco focuses on three major design goals:<ul><li><span>1.</span><span><div>Modularity, flexibility, and interoperability of DTs facilitated through distinct “components” nested within DT “layers”.</div></span></li><li><span>2.</span><span><div>Dynamic modelling of ecological processes and states that evolve over time.</div></span></li><li><span>3.</span><span><div>Linking ecological modelling to downstream actions or decisions made on the ecological object or process of study.</div></span></li></ul></div><div>TwinEco's architecture builds upon the feedback loops and state management strategies introduced in the Dynamic Data-Driven Application Systems (DDDAS) paradigm, which has already inspired many DTs across scientific domains. We also discuss the usefulness and ease-of-use of TwinEco by demonstrating its applicability to computational case studies and suggesting future recommendations to the community of data infrastructure builders and modellers in regards to open considerations. By introducing a shared terminology and emphasising model-data fusion, TwinEco highlights the importance of a unified framework to avoid fragmentation in the burgeoning field of ecological digital twinning.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103407"},"PeriodicalIF":7.3,"publicationDate":"2025-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144932453","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}

{"title":"Multi-features deep learning framework for aboveground biomass mapping in semi-arid forests by integrating LiDAR with Sentinel-1 and Sentinel-2 time series","authors":"Linjing Zhang ,&nbsp;Jing Chen ,&nbsp;Xinran Yin ,&nbsp;Yaru Wang","doi":"10.1016/j.ecoinf.2025.103408","DOIUrl":"10.1016/j.ecoinf.2025.103408","url":null,"abstract":"<div><div>Semi-arid forests harbour major carbon stocks. However, they have received little attention and are disappearing rapidly due to the global expansion of agriculture. The high-precision, large-scale estimation of semi-arid forest aboveground biomass (AGB) presents a major challenge to remote sensing because of the high spatiotemporal heterogeneity and structural complexity of semi-arid forests. This study developed a multi-features fusion transformer temporal-spatial model (MFF-TTSM) for mapping the AGB of semiarid forests by integrating time series Sentinel-1 (S-1) and Sentinel-2 (S-2) data with Global Ecosystem Dynamics Investigation (GEDI) data. An AGB reference map was constructed by combining the measured AGB and variables from airborne LiDAR data by utilising the Random Forest (RF) method, thus providing precise AGB estimates and high-accuracy sample data for subsequent modelling (R² = 0.833, RMSE = 21.926 Mg/ha and RMSE_r = 14.785 %). Various combinations of S-1, S-2 and GEDI data, including S-1 data (ASA), S-2 data (AOP), S-1 and S-2 data (AOP + ASA), S-2 and GEDI data (AOP + AGE), S-1 and S-2 and GEDI data (AOP + ASA + AGE) and the top 15 % of all data (the TOP15%), were compared. AOP + ASA + AGE exhibited the optimal estimation performance on the basis of the MFF-TTSM model, with an R² of 0.884, RMSE of 18.224 Mg/ha, and RMSE_r of 12.382 %. Compared with other advanced models, the MFF-TTSM model exhibited the highest accuracy under three different combinations (AOP + ASA + AGE, TOP15%, and AOP + ASA). The TOP15% experiment with the MFF-TTSM model also gave promising results (R² = 0.848, RMSE = 20.839 Mg/ha, RMSE_r = 14.159 %), yielding an AGB map for the research region. This study thus provides an advanced deep learning method for the high-precision mapping of AGB in semi-arid forests. Such an approach is critical for the large-scale sustainable management and carbon stock monitoring of semi-arid forests.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103408"},"PeriodicalIF":7.3,"publicationDate":"2025-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144925572","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}

{"title":"Assessing the effect of environmental factors and land use changes on benthic macroinvertebrate communities in stream ecosystems","authors":"Jong-Won Lee,&nbsp;Sang-Woo Lee,&nbsp;Se-Rin Park","doi":"10.1016/j.ecoinf.2025.103404","DOIUrl":"10.1016/j.ecoinf.2025.103404","url":null,"abstract":"<div><div>The ecological integrity of stream communities is shaped by the complex interplay of multiple environmental factors. This study employed a Bayesian Network (BN) approach based on Structural Equation Modeling (SEM) to evaluate the relative influence of environmental factors on the condition of benthic macroinvertebrates in streams. The analysis focused on 24 impaired streams within the Hangang River Basin in South Korea, using monitoring data collected from 2018 to 2022. Eleven input variables related to watershed land use (WLU), water quality (WQ), and physical habitat quality (PHQ) were incorporated into the model. The population density ratio of benthic macroinvertebrates and the benthic macroinvertebrate index were used as endpoints, serving as dependent variables. Sensitivity analysis identified WQ as the key factor influencing benthic macroinvertebrate communities in urban and agricultural watersheds, whereas PHQ was the most significant factor in forest watersheds. Scenario analyses further demonstrated that changes in PHQ and WQ could significantly alter the likelihood of benthic macroinvertebrate health degradation, depending on WLU changes. Notably, the probability of benthic macroinvertebrate impairment due to WQ degradation was higher in streams with good PHQ than in those with poor PHQ, suggesting that streams with better physical conditions are more vulnerable to the negative effects of poor WQ. This study provides a comprehensive understanding of how freshwater organisms respond to complex environmental dynamics within streams and watersheds. The findings underscore the importance of integrating both direct and indirect environmental impacts into watershed and stream management strategies to preserve the biological integrity of aquatic communities.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103404"},"PeriodicalIF":7.3,"publicationDate":"2025-08-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144904655","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}

{"title":"Modeling of mercury distribution in human body under conditions of chronic exposure: Development of a Biologically Based Dynamic (BBD) model with application to the Italian adult population","authors":"Giovanni Denaro ,&nbsp;Luciano Curcio ,&nbsp;Bartolomeo Cosenza ,&nbsp;Giuseppe Avellone","doi":"10.1016/j.ecoinf.2025.103405","DOIUrl":"10.1016/j.ecoinf.2025.103405","url":null,"abstract":"<div><div>The biologically based dynamic (BBD) model was used to study the concentration dynamics of methylmercury (MeHg) and its inorganic metabolites (IHg) in the human body. The study focused on the populations residing close to an industrial site characterized by mercury (Hg) pollution, with the main objective of supporting public health decision-making. The BBD model was modified to introduce some novelties compared to previous investigations. First, the BBD model considered the estimated weekly intake of methylmercury (EWI) and the body weight (BDW) as a function of age. Second, the calibration procedure of the BBD model parameters was based on the human biomonitoring data published in previous studies, and metabolism differences between the two genders was considered. Third, the theoretical Hg burdens in major organs and excreta were converted into Hg concentrations to compare the numerical results with the experimental data. The total mercury concentrations in biological matrices were theoretically reproduced for the Italian population, showing a reasonable agreement with values measured in blood (<span><math><msup><mover><mi>χ</mi><mo>∼</mo></mover><mn>2</mn></msup><mo>=</mo><mn>1.507</mn></math></span> for men and for <span><math><msup><mover><mi>χ</mi><mo>∼</mo></mover><mn>2</mn></msup><mo>=</mo><mn>0.778</mn></math></span> for women), urine (<span><math><msup><mover><mi>χ</mi><mo>∼</mo></mover><mn>2</mn></msup><mo>=</mo><mn>0.202</mn></math></span> for men and for <span><math><msup><mover><mi>χ</mi><mo>∼</mo></mover><mn>2</mn></msup><mo>=</mo><mn>0.364</mn></math></span> for women) and hair (<span><math><msup><mover><mi>χ</mi><mo>∼</mo></mover><mn>2</mn></msup><mo>=</mo><mn>0.425</mn></math></span> for men and for <span><math><msup><mover><mi>χ</mi><mo>∼</mo></mover><mn>2</mn></msup><mo>=</mo><mn>0.067</mn></math></span> for women) of the adult population residing in Augusta Bay between October 2012 and April 2013. Furthermore, the BBD model simulated the methylmercury and inorganic mercury concentrations in major organs, i.e. brain, kidney and liver, of local foodstuff consumers residing in highly polluted areas, showing an acceptable agreement with values measured in cadavers from Hyogo Prefecture (Japan) between November 1971 and May 1972. The model results depended strongly on the diet preferences of the investigated population and the mercury content in consumed foodstuffs. The BBD model can be improved by considering the variations of some biological parameters as a function of age, even if this would require a lot of experimental data on the main organs, which are difficult to obtain. By introducing these improvements, the BBD model could become a useful tool for assessing mercury chronic exposure risks near industrial areas and for improving policies aimed at preventing diseases associated with mercury pollution.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103405"},"PeriodicalIF":7.3,"publicationDate":"2025-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144904656","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}

{"title":"Deep-CABPred: Deep learning model for predicting functional chlorophyll a-b binding proteins in trait-based plant ecology using hybrid embedding with semi-normalized temporal convolutional networks","authors":"Farman Ali ,&nbsp;Raed Alsini ,&nbsp;Tamim Alkhalifah ,&nbsp;Fahad Alturise ,&nbsp;Wajdi Alghamdi ,&nbsp;Majdi Khalid","doi":"10.1016/j.ecoinf.2025.103400","DOIUrl":"10.1016/j.ecoinf.2025.103400","url":null,"abstract":"<div><div>Chlorophyll a-b binding proteins (CABs) are crucial for photosynthesis, directly influencing plant efficiency and environmental adaptation. Identifying these proteins is vital for understanding ecological function and productivity, but traditional experimental methods are laborious. To overcome this, we developed Deep-CABPred, a novel deep learning model for efficient CAB prediction. Our approach leverages a combination of advanced feature embedding techniques. We extract multi-source protein information from primary sequences, employing FastText for discriminative subword-level patterns and ProtBERT (a protein large language model) for contextualized sequential features. These independently extracted features are then fused into a comprehensive representation. This fused feature set is subsequently integrated into a Semi-Normalized Temporal Convolutional Network (SN-TCN) for model training. Deep-CABPred's performance was rigorously validated using a five-fold cross-validation strategy, achieving impressive accuracies of 88.60 % on the training dataset and 83.68 % on the testing dataset. This model offers an effective computational solution for CAB prediction and holds significant potential for advancing our understanding of plant functional traits, ultimately supporting agricultural and conservation efforts in a changing climate.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103400"},"PeriodicalIF":7.3,"publicationDate":"2025-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144889193","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}

{"title":"Advances in deep learning-driven photo identification and meta analysis of cetaceans in large data repositories","authors":"Alexander Barnhill ,&nbsp;Jared R. Towers ,&nbsp;Tasli J.H. Shaw ,&nbsp;Magdalena Arias ,&nbsp;Adrián Bécares ,&nbsp;Thomas Doniol-Valcroze ,&nbsp;Lorenzo von Fersen ,&nbsp;Rodrigo Genoves ,&nbsp;Tim Rörup ,&nbsp;Gary J. Sutton ,&nbsp;Sheila Thornton ,&nbsp;Michael Weiss ,&nbsp;Andreas Maier ,&nbsp;Elmar Nöth ,&nbsp;Christian Bergler","doi":"10.1016/j.ecoinf.2025.103396","DOIUrl":"10.1016/j.ecoinf.2025.103396","url":null,"abstract":"<div><div>Photo-identification of cetaceans remains a labor-intensive task, requiring expert annotation of long-tailed image datasets in which most individuals are rarely encountered. We present a scalable, end-to-end framework that automates this process using lightweight deep learning models optimized for resource-constrained environments. Our modular pipeline integrates state-of-the-art detection (YOLOv8-small), individual identification via metric learning (EfficientNet-B0 with a contrastive head), and auxiliary modules for image quality scoring, side classification, and identifiability prediction. Unlike previous approaches limited to single-species applications or high-resource settings, our framework generalizes across five cetacean populations with diverse visual characteristics. We achieve top-1 identification accuracies of 0.92 for Bigg's killer whales (<em>Orcinus orca rectipinnus</em>), 0.96 for Southern resident killer whales (<em>Orcinus orca ater</em>), 0.96 for Lahille's bottlenose dolphins (<em>Tursiops truncatus gephyreus</em>), 0.82 for common minke whales (<em>Balaenoptera acutorostrata scammoni</em>), and 0.85 for humpback whales (<em>Megaptera novaeangliae</em>), yielding a cross-species accuracy of 0.90. To support image triage in large datasets, we include a quality scoring module that predicts image utility using learned embedding features. This module achieves an R² of 0.799, enabling intelligent prioritization of data. Runtime evaluations show processing speeds of 1.6–3.2 images/s on CPU and 9.6–23.3 FPS with GPU acceleration, making it suitable for archival and real-time applications. We also evaluate the impact of demographic metadata (age, sex) on identification performance and provide practical recommendations for future dataset design. The system is available via a web interface designed to support real-world conservation workflows with minimal computational overhead.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"91 ","pages":"Article 103396"},"PeriodicalIF":7.3,"publicationDate":"2025-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144907888","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}