Ecological Informatics最新文献_第5页

DCMF: Deep Counterfactual Metric Framework for limited data plant disease recognition 有限数据植物病害识别的深度反事实度量框架

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-14 DOI: 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.

深度学习方法在植物病害识别方面取得了显著的成功。然而，这些方法依赖于大规模标记数据集进行训练，以确保经验风险最小化的可靠性。在现实世界中，获得如此广泛的疾病数据仍然具有挑战性。在数据有限的情况下，传统的基于相关性的学习框架可能会在疾病数据和疾病类别之间建立虚假的相关性，严重损害其泛化能力。我们通过提出深度反事实度量框架（DCMF）从因果角度解决了这个问题。具体而言，DCMF使用反事实推理模块（CRM）构建一个反事实世界，其中每个疾病图像仅包含健康特征，从而能够估计健康区域对疾病识别的直接影响。通过从对类别的总影响中减去这种直接影响，我们有效地消除了虚假相关性，使模型能够在有限的数据场景中学习稳健的疾病特定特征，从而实现可靠的泛化。在PlantVillage和PlantLeaves数据集上进行的5-shot和10-shot设置的大量实验表明，DCMF比最佳基线方法的平均性能提高了7.2%。这些改进验证了DCMF在有限数据植物病害识别中的有效性。

{"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":"2026-01-14","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}

引用次数: 0

Mapping alpine grassland species with the VI-MDACvT model based on UAV hyperspectral imagery 基于无人机高光谱影像的VI-MDACvT模型高寒草地物种制图

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-13 DOI: 10.1016/j.ecoinf.2026.103607

Hui Zhao , Jundi Wang , Huaan Jin , Da Wei , Zhengrong Yuan , Zhaoyi Zhang , Yaohua Luo , Xiaodan Wang

Hyperspectral imagery acquired by Unmanned Aerial Vehicles (UAVs) provides high spatial and spectral resolution, making it essential for accurate vegetation analysis. However, the reliance on a single deep learning model often limits the performance of vegetation classification in complex alpine grassland ecosystems. In this study, a UAV platform equipped with a hyperspectral imager was used to acquire detailed imagery of alpine grasslands. We developed a vegetation index–enhanced mobile three-dimensional atrous convolution vision transformer (VI-MDACvT), which integrates vegetation indices into a hybrid CNN–ViT framework to improve the representation of biologically meaningful features. The VI-MDACvT model combines a hybrid CNN–ViT architecture with vegetation indices (VIs) to strengthen feature extraction and expand the receptive field, enabling more precise classification of vegetation species. Experimental results demonstrated that integrating the VI module enhanced the ecological separability of vegetation features, increasing overall accuracy (OA), average accuracy (AA), and the Kappa coefficient by 0.25%, 0.2%, and 0.22%, respectively, compared with the benchmark MDvT model. By introducing 3D atrous convolution, the model effectively enlarged receptive field sizes across spectral–spatial dimensions, enabling better aggregation of contextual information for small and similar vegetation patches. The 3D atrous convolution module further improved OA, AA, and Kappa by 0.42%, 0.87%, and 0.96%, respectively. In comparisons with representative models (3D-CNN, ViT, MDvT, VI-MDvT and MDACvT), the VI-MDACvT model achieved the highest accuracy, with an OA of 98.48%, AA of 95.93%, and a Kappa coefficient of 97.87%. Moreover, the spatial distribution map of vegetation species generated by this method closely matched the actual distribution observed in the UAV imagery. These findings confirm the effectiveness of the proposed approach for alpine grassland vegetation classification and highlight its ecological application.

无人机获取的高光谱图像提供了高空间和光谱分辨率，对准确的植被分析至关重要。然而，在复杂的高寒草地生态系统中，对单一深度学习模型的依赖往往限制了植被分类的性能。本研究采用搭载高光谱成像仪的无人机平台，获取高寒草原的详细影像。我们开发了一个植被指数增强的移动三维自然卷积视觉转换器（VI-MDACvT），它将植被指数集成到一个混合CNN-ViT框架中，以提高对生物意义特征的表示。VI-MDACvT模型将CNN-ViT混合架构与植被指数（VIs）相结合，加强特征提取，扩大接受野，使植被种类分类更加精确。实验结果表明，与基准MDvT模型相比，集成VI模块增强了植被特征的生态可分性，总体精度（OA）、平均精度（AA）和Kappa系数分别提高了0.25%、0.2%和0.22%。通过引入三维光卷积，该模型有效地扩大了光谱空间维度上的接受野大小，从而能够更好地聚集小而相似的植被斑块的上下文信息。3D亚光卷积模块进一步提高OA、AA和Kappa分别为0.42%、0.87%和0.96%。与代表性模型（3D-CNN、ViT、MDvT、VI-MDvT和MDACvT）相比，VI-MDACvT模型的准确率最高，OA为98.48%，AA为95.93%，Kappa系数为97.87%。此外，该方法生成的植被种类空间分布图与无人机影像中观测到的实际分布非常吻合。这些结果证实了该方法在高寒草地植被分类中的有效性，并突出了其生态应用价值。

{"title":"Mapping alpine grassland species with the VI-MDACvT model based on UAV hyperspectral imagery","authors":"Hui Zhao , Jundi Wang , Huaan Jin , Da Wei , Zhengrong Yuan , Zhaoyi Zhang , Yaohua Luo , Xiaodan Wang","doi":"10.1016/j.ecoinf.2026.103607","DOIUrl":"10.1016/j.ecoinf.2026.103607","url":null,"abstract":"<div><div>Hyperspectral imagery acquired by Unmanned Aerial Vehicles (UAVs) provides high spatial and spectral resolution, making it essential for accurate vegetation analysis. However, the reliance on a single deep learning model often limits the performance of vegetation classification in complex alpine grassland ecosystems. In this study, a UAV platform equipped with a hyperspectral imager was used to acquire detailed imagery of alpine grasslands. We developed a vegetation index–enhanced mobile three-dimensional atrous convolution vision transformer (VI-MDACvT), which integrates vegetation indices into a hybrid CNN–ViT framework to improve the representation of biologically meaningful features. The VI-MDACvT model combines a hybrid CNN–ViT architecture with vegetation indices (VIs) to strengthen feature extraction and expand the receptive field, enabling more precise classification of vegetation species. Experimental results demonstrated that integrating the VI module enhanced the ecological separability of vegetation features, increasing overall accuracy (OA), average accuracy (AA), and the Kappa coefficient by 0.25%, 0.2%, and 0.22%, respectively, compared with the benchmark MDvT model. By introducing 3D atrous convolution, the model effectively enlarged receptive field sizes across spectral–spatial dimensions, enabling better aggregation of contextual information for small and similar vegetation patches. The 3D atrous convolution module further improved OA, AA, and Kappa by 0.42%, 0.87%, and 0.96%, respectively. In comparisons with representative models (3D-CNN, ViT, MDvT, VI-MDvT and MDACvT), the VI-MDACvT model achieved the highest accuracy, with an OA of 98.48%, AA of 95.93%, and a Kappa coefficient of 97.87%. Moreover, the spatial distribution map of vegetation species generated by this method closely matched the actual distribution observed in the UAV imagery. These findings confirm the effectiveness of the proposed approach for alpine grassland vegetation classification and highlight its ecological application.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103607"},"PeriodicalIF":7.3,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977425","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}

引用次数: 0

Explainable Hybrid Physics-Guided Neural Network (HPGNN) for diverse Inland Lakes water quality inversion 可解释混合物理引导神经网络（HPGNN）在不同内陆湖水质反演中的应用

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-13 DOI: 10.1016/j.ecoinf.2026.103611

Aamir Ali , Guanhua Zhou , Franz Pablo Antezana Lopez , Cheng Jiang , Guifei Jing , Yumin Tan

Accurate remote sensing of optically active water quality parameters (OAWQPs) in optically complex lakes remains challenging due to similar reflectance signatures, spectral overlap, and adjacency effects, leading to non-uniqueness and sensitivity to sparse, noisy in situ observations. While Machine Learning (ML) models capture complex relationships, they often lack physical consistency, resulting in limited generalization. Whereas conventional Physics-Guide Neutral Networks (PGNNs), integrating physics-based governing equations and auxiliary information, are effective but difficult to scale to satellite-based retrieval of OAWQPs, lacking sufficient observables and governing equations. This study proposes a Hybrid PGNN (HPGNN) that integrates reflectance-native physics-guided and statistical regularization constraints with data fidelity to simultaneously enforce physical plausibility, statistical robustness, and empirical fidelity. Using Sentinel-2 remote sensing reflectance and a broadly distributed in situ dataset, HPGNN models are developed with a hybrid loss function integrating the above constraints, achieving superior performance over conventional ML and empirical algorithms across multiple OAWQPs with R² (0.81–0.98). Using HPGNN models, quantitative and qualitative assessment results are produced across different continents, showing multiregional generalization and practical applicability. Finally, models' interpretability is ensured through SHapley Additive exPlanations (SHAP), which elucidates the influence of spectral bands on each parameter, providing hydrological insights. The proposed HPGNN framework provides a scalable and interpretable solution, enabling multiregional OAWQPs' spatio-temporal assessment to understand lake hydrology and take management decisions.

由于相似的反射特征、光谱重叠和邻接效应，光学复杂湖泊的光学活性水质参数（OAWQPs）的精确遥感仍然具有挑战性，导致对稀疏、嘈杂的原位观测的非唯一性和敏感性。虽然机器学习（ML）模型捕获复杂的关系，但它们通常缺乏物理一致性，导致泛化有限。而传统的物理引导中性网络（pgnn），集成了基于物理的控制方程和辅助信息，是有效的，但难以扩展到基于卫星的oawqp检索，缺乏足够的观测值和控制方程。本研究提出了一种混合PGNN (HPGNN)，它将反射原生物理指导和统计正则化约束与数据保真度相结合，同时增强物理可信性、统计稳健性和经验保真度。利用Sentinel-2遥感反射率和广泛分布的原位数据集，建立了包含上述约束的混合损失函数的HPGNN模型，在多个oawqp上取得了优于传统ML和经验算法的性能，R2（0.81-0.98）。利用HPGNN模型，得出了跨大洲的定量和定性评价结果，具有多区域通用性和实用性。最后，通过SHapley加性解释（SHAP）确保模型的可解释性，该解释阐明了光谱带对每个参数的影响，提供了水文见解。提出的HPGNN框架提供了一个可扩展和可解释的解决方案，使多区域oawqp的时空评估能够了解湖泊水文并做出管理决策。

{"title":"Explainable Hybrid Physics-Guided Neural Network (HPGNN) for diverse Inland Lakes water quality inversion","authors":"Aamir Ali , Guanhua Zhou , Franz Pablo Antezana Lopez , Cheng Jiang , Guifei Jing , Yumin Tan","doi":"10.1016/j.ecoinf.2026.103611","DOIUrl":"10.1016/j.ecoinf.2026.103611","url":null,"abstract":"<div><div>Accurate remote sensing of optically active water quality parameters (OAWQPs) in optically complex lakes remains challenging due to similar reflectance signatures, spectral overlap, and adjacency effects, leading to non-uniqueness and sensitivity to sparse, noisy in situ observations. While Machine Learning (ML) models capture complex relationships, they often lack physical consistency, resulting in limited generalization. Whereas conventional Physics-Guide Neutral Networks (PGNNs), integrating physics-based governing equations and auxiliary information, are effective but difficult to scale to satellite-based retrieval of OAWQPs, lacking sufficient observables and governing equations. This study proposes a Hybrid PGNN (HPGNN) that integrates reflectance-native physics-guided and statistical regularization constraints with data fidelity to simultaneously enforce physical plausibility, statistical robustness, and empirical fidelity. Using Sentinel-2 remote sensing reflectance and a broadly distributed in situ dataset, HPGNN models are developed with a hybrid loss function integrating the above constraints, achieving superior performance over conventional ML and empirical algorithms across multiple OAWQPs with R<sup>2</sup> (0.81–0.98). Using HPGNN models, quantitative and qualitative assessment results are produced across different continents, showing multiregional generalization and practical applicability. Finally, models' interpretability is ensured through SHapley Additive exPlanations (SHAP), which elucidates the influence of spectral bands on each parameter, providing hydrological insights. The proposed HPGNN framework provides a scalable and interpretable solution, enabling multiregional OAWQPs' spatio-temporal assessment to understand lake hydrology and take management decisions.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103611"},"PeriodicalIF":7.3,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146022544","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}

引用次数: 0

BT-YOLO: An attention-enhanced framework for monitoring bioluminescent algal blooms in coastal tourism zones BT-YOLO：沿海旅游区生物发光藻华监测的关注增强框架

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-10 DOI: 10.1016/j.ecoinf.2026.103595

Naseeb Abbas , Kaijian Zheng , Zhenping Li , Wenqi Chi , Chun Chen , Heshan Lin , Degang Jiang , Haifeng Gu , Jianping Li

Monitoring of bioluminescent algae blooms, referred to in China as “Blue Tears”, is essential for managing ecological risks, ensuring public safety, and accommodating the growing tourist attention they attract in coastal regions. However, this task remains challenging due to the nocturnal nature of these events, limited visibility, the lack of real-time monitoring infrastructure, and the absence of publicly available training datasets. To address this, we present BT3.8k, a novel dataset comprising 3827 images extracted from 241 user-uploaded videos shared on social media platforms. These samples cover diverse lighting, sea states, and viewpoints. Annotations were generated using Roboflow and SAM2, and subsequently refined using a YOLOv11m-seg model. We applied augmentations including Gaussian blur, salt-and-pepper noise, flipping, rotation, hue, brightness, and contrast to enhance generalizability. Using this dataset, we trained BT-YOLO, a YOLOv11 framework enhanced with the Convolutional Block Attention Module (CBAM), designed for coastal Blue Tear monitoring. Among multiple models evaluated, BT-YOLO achieved the highest performance with a Dice score of 85%, IoU of 77%, AP@50 of 80%, and 90 FPS on GPU. This approach offers a scalable, real-time solution for BT monitoring, with potential impact across marine ecology, eco-tourism management, and environmental governance.

监测在中国被称为“蓝泪”的生物发光藻华，对于管理生态风险、确保公共安全以及适应沿海地区日益增长的游客关注度至关重要。然而，由于这些事件的夜间性质、有限的可见性、缺乏实时监控基础设施以及缺乏公开可用的训练数据集，这项任务仍然具有挑战性。为了解决这个问题，我们提出了BT3.8k，这是一个新的数据集，包括从社交媒体平台上分享的241个用户上传的视频中提取的3827张图像。这些样本涵盖了不同的照明、海况和视点。使用Roboflow和SAM2生成注释，随后使用YOLOv11m-seg模型进行细化。我们应用了增强功能，包括高斯模糊、椒盐噪声、翻转、旋转、色调、亮度和对比度，以增强可泛化性。使用该数据集，我们训练了BT-YOLO，这是一个经过卷积块注意力模块（CBAM）增强的YOLOv11框架，专为沿海蓝泪监测而设计。在评估的多个模型中，BT-YOLO获得了最高的性能，Dice得分为85%，IoU为77%，AP@50为80%，GPU上的FPS为90。这种方法为BT监测提供了可扩展的实时解决方案，对海洋生态、生态旅游管理和环境治理具有潜在影响。

{"title":"BT-YOLO: An attention-enhanced framework for monitoring bioluminescent algal blooms in coastal tourism zones","authors":"Naseeb Abbas , Kaijian Zheng , Zhenping Li , Wenqi Chi , Chun Chen , Heshan Lin , Degang Jiang , Haifeng Gu , Jianping Li","doi":"10.1016/j.ecoinf.2026.103595","DOIUrl":"10.1016/j.ecoinf.2026.103595","url":null,"abstract":"<div><div>Monitoring of bioluminescent algae blooms, referred to in China as “Blue Tears”, is essential for managing ecological risks, ensuring public safety, and accommodating the growing tourist attention they attract in coastal regions. However, this task remains challenging due to the nocturnal nature of these events, limited visibility, the lack of real-time monitoring infrastructure, and the absence of publicly available training datasets. To address this, we present BT3.8k, a novel dataset comprising 3827 images extracted from 241 user-uploaded videos shared on social media platforms. These samples cover diverse lighting, sea states, and viewpoints. Annotations were generated using Roboflow and SAM2, and subsequently refined using a YOLOv11m-seg model. We applied augmentations including Gaussian blur, salt-and-pepper noise, flipping, rotation, hue, brightness, and contrast to enhance generalizability. Using this dataset, we trained BT-YOLO, a YOLOv11 framework enhanced with the Convolutional Block Attention Module (CBAM), designed for coastal Blue Tear monitoring. Among multiple models evaluated, BT-YOLO achieved the highest performance with a Dice score of 85%, IoU of 77%, AP@50 of 80%, and 90 FPS on GPU. This approach offers a scalable, real-time solution for BT monitoring, with potential impact across marine ecology, eco-tourism management, and environmental governance.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103595"},"PeriodicalIF":7.3,"publicationDate":"2026-01-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976640","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}

引用次数: 0

CNN-based wheat yield prediction using multi-source and multi-stage data integration from UAV imagery and sensors 基于cnn的多源多级无人机图像和传感器数据集成小麦产量预测

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-09 DOI: 10.1016/j.ecoinf.2026.103608

Süreyya Betül Rufaioglu , Ali Volkan Bilgili , Sibel Ipekesen , Amjed Mohamed Ismael , Yunus Kaya , João P. Matos-Carvalho

In this study, a convolutional neural network (CNN)-based deep learning model was developed to predict durum wheat yield by integrating RGB and multispectral UAV imagery, ground-based sensor measurements (SPAD, NDVI, LAI, and plant height), and climatic parameters under semi-arid conditions. The dataset was designed as a multi-source, multi-stage, and multi-year structure, comprising measurements collected across six phenological growth stages during the 2023 and 2024 growing seasons. Principal component analysis (PCA) indicated that approximately 69% of the total variance was explained by the first two components, with SPAD, NDVI, LAI, and plant height identified as the most influential variables in explaining yield variability. The CNN model achieved high predictive accuracy in both stage-based and year-based evaluations, with R² values ranging from 0.982 to 0.994, RMSE between 0.15 and 0.24 kg ha⁻¹, and MAE between 0.11 and 0.19 kg ha⁻¹. The highest performance was obtained during the heading and grain-filling stages. Overall, the results demonstrate that integrating UAV imagery, physiological sensor indicators, and climatic variables within a multi-source, multi-stage, multi-year deep learning framework substantially improves yield prediction accuracy compared with single-source approaches. This study presents a high-performance CNN architecture for yield forecasting and provides a robust foundation for generalizable and effective decision-support systems in precision agriculture.

本研究基于卷积神经网络（CNN）的深度学习模型，通过整合RGB和多光谱无人机图像、地面传感器测量数据（SPAD、NDVI、LAI和植物高度）以及半干旱条件下的气候参数，建立了预测硬粒小麦产量的模型。该数据集被设计成一个多来源、多阶段、多年份的结构，包括在2023年和2024年生长季节的六个物候生长阶段收集的测量数据。主成分分析（PCA）表明，约69%的总方差由前两个分量解释，其中SPAD、NDVI、LAI和株高被确定为解释产量变异的最具影响力的变量。CNN模型在阶段和年份评价中均具有较高的预测精度，R2值在0.982 ~ 0.994之间，RMSE在0.15 ~ 0.24 kg ha - 1之间，MAE在0.11 ~ 0.19 kg ha - 1之间。抽穗期和灌浆期产量最高。总体而言，研究结果表明，与单源方法相比，在多源、多阶段、多年的深度学习框架中集成无人机图像、生理传感器指标和气候变量可显著提高产量预测精度。该研究提出了一种用于产量预测的高性能CNN架构，为精准农业中可推广和有效的决策支持系统提供了坚实的基础。

{"title":"CNN-based wheat yield prediction using multi-source and multi-stage data integration from UAV imagery and sensors","authors":"Süreyya Betül Rufaioglu , Ali Volkan Bilgili , Sibel Ipekesen , Amjed Mohamed Ismael , Yunus Kaya , João P. Matos-Carvalho","doi":"10.1016/j.ecoinf.2026.103608","DOIUrl":"10.1016/j.ecoinf.2026.103608","url":null,"abstract":"<div><div>In this study, a convolutional neural network (CNN)-based deep learning model was developed to predict durum wheat yield by integrating RGB and multispectral UAV imagery, ground-based sensor measurements (SPAD, NDVI, LAI, and plant height), and climatic parameters under semi-arid conditions. The dataset was designed as a multi-source, multi-stage, and multi-year structure, comprising measurements collected across six phenological growth stages during the 2023 and 2024 growing seasons. Principal component analysis (PCA) indicated that approximately 69% of the total variance was explained by the first two components, with SPAD, NDVI, LAI, and plant height identified as the most influential variables in explaining yield variability. The CNN model achieved high predictive accuracy in both stage-based and year-based evaluations, with R<sup>2</sup> values ranging from 0.982 to 0.994, RMSE between 0.15 and 0.24 kg ha<sup>−1</sup>, and MAE between 0.11 and 0.19 kg ha<sup>−1</sup>. The highest performance was obtained during the heading and grain-filling stages. Overall, the results demonstrate that integrating UAV imagery, physiological sensor indicators, and climatic variables within a multi-source, multi-stage, multi-year deep learning framework substantially improves yield prediction accuracy compared with single-source approaches. This study presents a high-performance CNN architecture for yield forecasting and provides a robust foundation for generalizable and effective decision-support systems in precision agriculture.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103608"},"PeriodicalIF":7.3,"publicationDate":"2026-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977417","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}

引用次数: 0

Spatiotemporal statistical evaluation of recent active and passive satellite-derived soil moisture products across Central Asia under multiple scenarios 多情景下中亚地区近期主动式和被动式卫星土壤水分产品的时空统计评价

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-08 DOI: 10.1016/j.ecoinf.2026.103602

B.G. Mousa , Alim Samat , Peijun Du , Jilili Abuduwaili , Xiangzhuo Liu , Yousef A. Al-Masnay , Adel Nasri , Marzouk Mohamed Aly Abdelhamid

Algorithms for retrieving Surface Soil Moisture (SSM) from microwave remote sensing are continually refined, necessitating the evaluation of newly released products to identify their optimal applications and potential for improvement. This study evaluates ASCAT H119, SMAP DCA, SMAP MTDCA, SMAP-IB, AMSR2, and SMOS-IC products across Central Asia (CA) using a multi-scenario approach. First, the products were validated against three independent reference datasets: ERA5, ESA CCI (combined), and GLDAS-Noah. Second, Triple Collocation Analysis (TCA) was employed to estimate the Fractional Mean Squared Error (fMSE) and the error variance of each product. Third, Hovmöller diagrams were used to identify regional spatiotemporal trends and anomalies. Additionally, the quality of SSM products was examined in relation to key ecological variables across CA. All evaluations were conducted during the SSM growing season (April–October) from 2016 to 2019. Results showed that SMAP products, particularly SMAP DCA and SMAP-IB, delivered the most accurate SSM estimates, achieving the highest average correlation (0.63 to 0.85), low average bias (−0.059 to −0.077 m³ m⁻³), and the lowest average ubRMSE (0.030 to 0.049 m³ m⁻³) across the three reference datasets. ASCAT and AMSR2 exhibited moderate performance, while SMOS-IC performed the weakest overall. All products performed best against ESA CCI, followed by ERA5 and then GLDAS-Noah. The performance ranking derived from TCA was generally consistent with the reference-based validation, except for ASCAT and AMSR2. The proposed evaluation framework offers a reliable alternative for assessing SSM products in regions with sparse ground measurements. This study provides valuable insights for enhancing SSM monitoring, hydrological modeling, and agricultural management in the CA ecosystem.

从微波遥感中检索地表土壤湿度（SSM）的算法不断完善，需要对新发布的产品进行评估，以确定其最佳应用和改进潜力。本研究使用多场景方法评估了中亚地区（CA）的ASCAT H119、SMAP DCA、SMAP MTDCA、SMAP- ib、AMSR2和SMOS-IC产品。首先，对三个独立的参考数据集进行验证：ERA5、ESA CCI（组合）和GLDAS-Noah。其次，采用三重搭配分析法（Triple matching Analysis， TCA）估计各产品的分数均方误差（fMSE）和误差方差；第三，利用Hovmöller图识别区域时空趋势和异常。此外，还研究了与全CA关键生态变量相关的SSM产品质量。所有评估都是在2016年至2019年SSM生长季节（4月至10月）进行的。结果表明，SMAP产品，特别是SMAP DCA和SMAP- ib，提供了最准确的SSM估计，在三个参考数据集中实现了最高的平均相关性（0.63至0.85），低平均偏差（- 0.059至- 0.077 m3 m−3）和最低的平均ubRMSE（0.030至0.049 m3 m−3）。ASCAT和AMSR2表现中等，而SMOS-IC表现最差。所有产品在ESA CCI测试中表现最佳，其次是ERA5，然后是GLDAS-Noah。除ASCAT和AMSR2外，TCA得出的性能排名与基于参考的验证基本一致。提出的评估框架为在地面测量稀疏的地区评估SSM产品提供了可靠的替代方案。该研究为加强加州生态系统的SSM监测、水文建模和农业管理提供了有价值的见解。

{"title":"Spatiotemporal statistical evaluation of recent active and passive satellite-derived soil moisture products across Central Asia under multiple scenarios","authors":"B.G. Mousa , Alim Samat , Peijun Du , Jilili Abuduwaili , Xiangzhuo Liu , Yousef A. Al-Masnay , Adel Nasri , Marzouk Mohamed Aly Abdelhamid","doi":"10.1016/j.ecoinf.2026.103602","DOIUrl":"10.1016/j.ecoinf.2026.103602","url":null,"abstract":"<div><div>Algorithms for retrieving Surface Soil Moisture (SSM) from microwave remote sensing are continually refined, necessitating the evaluation of newly released products to identify their optimal applications and potential for improvement. This study evaluates ASCAT H119, SMAP DCA, SMAP MTDCA, SMAP-IB, AMSR2, and SMOS-IC products across Central Asia (CA) using a multi-scenario approach. First, the products were validated against three independent reference datasets: ERA5, ESA CCI (combined), and GLDAS-Noah. Second, Triple Collocation Analysis (TCA) was employed to estimate the Fractional Mean Squared Error (fMSE) and the error variance of each product. Third, Hovmöller diagrams were used to identify regional spatiotemporal trends and anomalies. Additionally, the quality of SSM products was examined in relation to key ecological variables across CA. All evaluations were conducted during the SSM growing season (April–October) from 2016 to 2019. Results showed that SMAP products, particularly SMAP DCA and SMAP-IB, delivered the most accurate SSM estimates, achieving the highest average correlation (0.63 to 0.85), low average bias (−0.059 to −0.077 m<sup>3</sup> m<sup>−3</sup>), and the lowest average ubRMSE (0.030 to 0.049 m<sup>3</sup> m<sup>−3</sup>) across the three reference datasets. ASCAT and AMSR2 exhibited moderate performance, while SMOS-IC performed the weakest overall. All products performed best against ESA CCI, followed by ERA5 and then GLDAS-Noah. The performance ranking derived from TCA was generally consistent with the reference-based validation, except for ASCAT and AMSR2. The proposed evaluation framework offers a reliable alternative for assessing SSM products in regions with sparse ground measurements. This study provides valuable insights for enhancing SSM monitoring, hydrological modeling, and agricultural management in the CA ecosystem.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103602"},"PeriodicalIF":7.3,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925325","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}

引用次数: 0

Calibrated general-purpose smartphone augmented reality for urban street-tree structural data acquisition 校准通用智能手机增强现实，用于城市街道树结构数据采集

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-08 DOI: 10.1016/j.ecoinf.2026.103599

Ya Zou , Feng Shi , Junsong Wang , Nan Mo , Lan Pan , Qinglin Meng

Urban street trees are key to urban climate regulation, carbon storage and habitat provision, yet many cities still lack fine-scale, up-to-date field data on their structure and site conditions. Conventional tape and laser measurements are accurate but slow, labour-intensive and often impractical along narrow, obstacle-rich sidewalks. This study develops and tests a smartphone-based augmented-reality (AR) framework for measuring three ecologically critical geometric parameters of street trees—diameter at breast height (DBH), planting-pit area and tree spacing. Using general-purpose AR ruler apps on three consumer smartphones, we implemented a standardized field protocol and a generic geometric-plus-statistical calibration workflow that converts basal rectangular measurements into calibrated DBH. Field validation along a 1.5-km street corridor in Guangzhou, China (125 trees) shows that, after calibration, mean absolute percentage errors are 6–8% for DBH and below 3% for planting-pit area and spacing, satisfying Class B precision under the Chinese Regulations on Main Technical Specifications for Forestry Professional Surveys. AR workflows reduced survey time by 35–40% and required only one operator. The framework thus provides an operationally efficient, low-cost and transferable pathway to generate standardized structural data for urban-forest inventories and ecological-informatics applications.

城市行道树是城市气候调节、碳储存和栖息地提供的关键，但许多城市仍然缺乏关于其结构和场地条件的精细尺度、最新的实地数据。传统的卷尺和激光测量虽然准确，但速度慢、劳动密集，而且在狭窄、障碍物多的人行道上往往不切实际。本研究开发并测试了一个基于智能手机的增强现实（AR）框架，用于测量行道树的三个生态关键几何参数——胸径（DBH）、种植坑面积和树间距。在三台消费者智能手机上使用通用AR尺应用程序，我们实现了标准化的现场协议和通用的几何加统计校准工作流程，将基本矩形测量值转换为校准的胸径。在中国广州1.5 km的街道走廊（125棵树）上进行的实地验证表明，经校正后，胸径平均绝对百分比误差在6-8%，栽坑面积和间距平均绝对百分比误差在3%以下，满足中国林业专业调查主要技术规范规定的B级精度。AR工作流程减少了35-40%的调查时间，只需要一名操作员。因此，该框架为城市森林清单和生态信息学应用产生标准化结构数据提供了一个操作效率高、成本低和可转移的途径。

{"title":"Calibrated general-purpose smartphone augmented reality for urban street-tree structural data acquisition","authors":"Ya Zou , Feng Shi , Junsong Wang , Nan Mo , Lan Pan , Qinglin Meng","doi":"10.1016/j.ecoinf.2026.103599","DOIUrl":"10.1016/j.ecoinf.2026.103599","url":null,"abstract":"<div><div>Urban street trees are key to urban climate regulation, carbon storage and habitat provision, yet many cities still lack fine-scale, up-to-date field data on their structure and site conditions. Conventional tape and laser measurements are accurate but slow, labour-intensive and often impractical along narrow, obstacle-rich sidewalks. This study develops and tests a smartphone-based augmented-reality (AR) framework for measuring three ecologically critical geometric parameters of street trees—diameter at breast height (DBH), planting-pit area and tree spacing. Using general-purpose AR ruler apps on three consumer smartphones, we implemented a standardized field protocol and a generic geometric-plus-statistical calibration workflow that converts basal rectangular measurements into calibrated DBH. Field validation along a 1.5-km street corridor in Guangzhou, China (125 trees) shows that, after calibration, mean absolute percentage errors are 6–8% for DBH and below 3% for planting-pit area and spacing, satisfying Class B precision under the Chinese <em>Regulations on Main Technical Specifications for Forestry Professional Surveys</em>. AR workflows reduced survey time by 35–40% and required only one operator. The framework thus provides an operationally efficient, low-cost and transferable pathway to generate standardized structural data for urban-forest inventories and ecological-informatics applications.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103599"},"PeriodicalIF":7.3,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977420","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}

引用次数: 0

YOLO-HARVEST: A hybrid ViT architecture with locality-enhanced attention for automated wildlife species classification YOLO-HARVEST：一种混合ViT架构，具有位置增强的关注，用于自动野生动物物种分类

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-08 DOI: 10.1016/j.ecoinf.2026.103605

Anuruddha Paul , Rishi Raj , Mahendra Kumar Gourisaria , Amitkumar V. Jha , Nicu Bizon

Wildlife conservation efforts increasingly depend on automated species classification for processing large-scale camera trap data, yet existing approaches struggle with accuracy and computational efficiency in resource-constrained environments. This paper introduces HARVEST (Hierarchical Attention for Robust Vision Enhancement with Shifted Tokenization), a novel hybrid architecture integrating YOLOv8 object detection with transformer-based classification. The architecture incorporates three key innovations: Shifted Patch Tokenization (SPT) for boundary information preservation, Local Information Enhancer (LIFE) for spatial feature extraction, and Locality-Enhanced Attention (LEA) for adaptive feature integration. The model is evaluated on two comprehensive datasets: a challenging 45-species Ohio State University (OSU) Small Animals dataset exhibiting an extreme class imbalance (6320:1 ratio) and a balanced 6-species African wildlife dataset. The HARVEST demonstrates excellent performance and achieves 85.27% accuracy on the OSU dataset and 94.74% accuracy on the Wildlife dataset with only 13.0M parameters, representing an 85% reduction compared to standard Vision Transformers while maintaining superior performance. The OSU evaluation demonstrates robust performance across highly imbalanced real-world conditions with species sample sizes ranging from 1 to 6320 images, validating practical applicability for conservation scenarios. Qualitative analysis reveals biologically meaningful attention patterns focusing on taxonomically relevant features. The efficient architecture enables real-world deployment in conservation applications, providing a practical solution for automated wildlife monitoring and biodiversity surveillance.

野生动物保护工作越来越依赖于自动物种分类来处理大规模相机陷阱数据，然而现有的方法在资源有限的环境中存在准确性和计算效率的问题。本文介绍了一种将YOLOv8目标检测与基于变换的分类相结合的新型混合架构——HARVEST （Hierarchical Attention for Robust Vision Enhancement with shifting Tokenization）。该体系结构包含三个关键创新：用于边界信息保存的移位补丁标记化（SPT），用于空间特征提取的局部信息增强器（LIFE）和用于自适应特征集成的位置增强注意（LEA）。该模型在两个综合数据集上进行了评估：一个具有挑战性的45种俄亥俄州立大学（OSU）小动物数据集，显示出极端的类不平衡（6380:1的比例）和一个平衡的6种非洲野生动物数据集。HARVEST表现出优异的性能，在OSU数据集上实现了85.27%的准确率，在野生动物数据集上实现了94.74%的准确率，仅使用13.0M参数，与标准视觉变形器相比降低了85%，同时保持了优异的性能。俄勒冈州立大学的评估表明，在高度不平衡的现实世界条件下，物种样本量从1到6320张不等，验证了保护场景的实际适用性。定性分析揭示了生物学上有意义的注意力模式，集中在分类学上相关的特征上。这种高效的架构能够在保护应用中进行实际部署，为自动野生动物监测和生物多样性监测提供实用的解决方案。

{"title":"YOLO-HARVEST: A hybrid ViT architecture with locality-enhanced attention for automated wildlife species classification","authors":"Anuruddha Paul , Rishi Raj , Mahendra Kumar Gourisaria , Amitkumar V. Jha , Nicu Bizon","doi":"10.1016/j.ecoinf.2026.103605","DOIUrl":"10.1016/j.ecoinf.2026.103605","url":null,"abstract":"<div><div>Wildlife conservation efforts increasingly depend on automated species classification for processing large-scale camera trap data, yet existing approaches struggle with accuracy and computational efficiency in resource-constrained environments. This paper introduces HARVEST (Hierarchical Attention for Robust Vision Enhancement with Shifted Tokenization), a novel hybrid architecture integrating YOLOv8 object detection with transformer-based classification. The architecture incorporates three key innovations: Shifted Patch Tokenization (SPT) for boundary information preservation, Local Information Enhancer (LIFE) for spatial feature extraction, and Locality-Enhanced Attention (LEA) for adaptive feature integration. The model is evaluated on two comprehensive datasets: a challenging 45-species Ohio State University (OSU) Small Animals dataset exhibiting an extreme class imbalance (6320:1 ratio) and a balanced 6-species African wildlife dataset. The HARVEST demonstrates excellent performance and achieves 85.27% accuracy on the OSU dataset and 94.74% accuracy on the Wildlife dataset with only 13.0M parameters, representing an 85% reduction compared to standard Vision Transformers while maintaining superior performance. The OSU evaluation demonstrates robust performance across highly imbalanced real-world conditions with species sample sizes ranging from 1 to 6320 images, validating practical applicability for conservation scenarios. Qualitative analysis reveals biologically meaningful attention patterns focusing on taxonomically relevant features. The efficient architecture enables real-world deployment in conservation applications, providing a practical solution for automated wildlife monitoring and biodiversity surveillance.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103605"},"PeriodicalIF":7.3,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080600","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}

引用次数: 0

Dual-branch time–frequency network with channel masking for automated insect sound monitoring toward field environments 基于信道掩蔽的双支路时频网络野外环境昆虫声自动监测

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-08 DOI: 10.1016/j.ecoinf.2026.103600

Chunjie Guo , Nanbo Xu , Jinfeng Li , Xuefei Feng , Jun Li

Insect acoustic signals convey vital ecological information, supporting behaviors such as mating and territory defense. Monitoring these sounds toward field environments offers a non-invasive way to assess biodiversity and ecosystem health, yet automatic classification is hindered by short signal durations, spectral variability, and noise. We propose DB-CAFNet, a lightweight time–frequency model combining a 1D CNN and spectrogram encoder (WaveSpecNet) with a novel ChannelMask classifier that introduces channel-wise perturbations to enhance robustness. On InsectSet32, our method achieves a best accuracy of 89.47%, surpassing the previously reported model (80.21%) (He et al., 2024). Consistently strong results across five additional public datasets further demonstrate its generalization ability, with only 4.09M parameters and 2.55 GFLOPs under simulated field conditions. Its compact design facilitates deployment on resource-limited hardware, making the model a promising candidate for future edge-based monitoring workflows in field-oriented settings. Future work will incorporate real outdoor recordings to complete ecological validation in field environments.

昆虫的声音信号传递重要的生态信息，支持交配和领土防御等行为。在野外环境中监测这些声音提供了一种非侵入性的评估生物多样性和生态系统健康的方法，但由于信号持续时间短、频谱变化和噪声，自动分类受到阻碍。我们提出了DB-CAFNet，这是一种轻量级的时频模型，结合了1D CNN和频谱图编码器（WaveSpecNet），并采用了一种新的ChannelMask分类器，该分类器引入了信道扰动以增强鲁棒性。在昆虫set32上，我们的方法达到了89.47%的最佳准确率，超过了之前报道的模型（80.21%）（He et al., 2024）。在另外5个公共数据集上的一致结果进一步证明了它的泛化能力，在模拟的现场条件下，只有4.09M个参数和2.55个GFLOPs。其紧凑的设计便于在资源有限的硬件上部署，使该模型成为面向现场设置的未来基于边缘的监测工作流程的有希望的候选者。未来的工作将包括真实的户外记录，以完成野外环境的生态验证。

{"title":"Dual-branch time–frequency network with channel masking for automated insect sound monitoring toward field environments","authors":"Chunjie Guo , Nanbo Xu , Jinfeng Li , Xuefei Feng , Jun Li","doi":"10.1016/j.ecoinf.2026.103600","DOIUrl":"10.1016/j.ecoinf.2026.103600","url":null,"abstract":"<div><div>Insect acoustic signals convey vital ecological information, supporting behaviors such as mating and territory defense. Monitoring these sounds toward field environments offers a non-invasive way to assess biodiversity and ecosystem health, yet automatic classification is hindered by short signal durations, spectral variability, and noise. We propose DB-CAFNet, a lightweight time–frequency model combining a 1D CNN and spectrogram encoder (WaveSpecNet) with a novel ChannelMask classifier that introduces channel-wise perturbations to enhance robustness. On InsectSet32, our method achieves a best accuracy of 89.47%, surpassing the previously reported model (80.21%) (He et al., 2024). Consistently strong results across five additional public datasets further demonstrate its generalization ability, with only 4.09M parameters and 2.55 GFLOPs under simulated field conditions. Its compact design facilitates deployment on resource-limited hardware, making the model a promising candidate for future edge-based monitoring workflows in field-oriented settings. Future work will incorporate real outdoor recordings to complete ecological validation in field environments.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103600"},"PeriodicalIF":7.3,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977424","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}

引用次数: 0

Advancing provenance assignment using machine learning and time series analysis of chemical chronologies in archival tissues 利用机器学习和时间序列分析档案组织中的化学年表，推进出处分配

IF 7.3 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2026-01-07 DOI: 10.1016/j.ecoinf.2026.103604

Kohma Arai , Malte Willmes , Rachel C. Johnson , Anna M. Sturrock

Accurate provenance assignment is critical for understanding ecological connectivity and guiding conservation and management. Natural chemical chronologies stored in metabolically inert, incrementally growing tissues (e.g., otoliths) provide a powerful tool for this purpose. However, traditional approaches face biological challenges, collapse chronological data into single values, and require subjective interpretation, limiting accuracy and scalability. We present a novel framework that integrates machine learning, time series analysis, and ensemble modeling to improve provenance assignment from archival tissue chemistry. Using otolith ⁸⁷Sr/⁸⁶Sr profiles from 17 distinct natal sources (n = 255) of California Central Valley Chinook salmon, we developed automated feature extraction, explicit time series classification (including dynamic time warping [DTW] with k-nearest neighbors [KNN]), and ensembles combining multiple classifiers. Furthermore, to address gaps in under-sampled life histories within natal sources, we added simulated chemical profiles to the reference baseline and tested whether they improved model performance. Time series approaches (mean accuracy: 0.56–0.62) consistently outperformed traditional methods (0.48), particularly for sources influenced by maternal signatures or early dispersal. Feature extraction approaches performed best when profiles followed predictable life-stage patterns, while explicit time series classification (DTW + KNN) excelled for distinct profile “shapes”. Ensembles leveraged complementary strengths and outperformed any single method. Our results highlight the advantages of treating archival chemical data as time series and applying machine learning and ensemble strategies to enhance the accuracy and scalability of provenance assignment. This framework is broadly applicable across taxa, tissue types, and chemical markers, offering a roadmap for advancing ecological inference and conservation.

准确的种源分配对于认识生态连通性和指导保护与管理至关重要。储存在代谢惰性、逐渐生长的组织（如耳石）中的天然化学年表为这一目的提供了有力的工具。然而，传统的方法面临着生物学的挑战，将时间顺序数据分解为单一值，并且需要主观解释，限制了准确性和可扩展性。我们提出了一个集成了机器学习、时间序列分析和集成建模的新框架，以改善档案组织化学的来源分配。利用加利福尼亚中央山谷支诺克鲑鱼17种不同出生来源（n = 255）的耳石87Sr/86Sr谱，我们开发了自动特征提取、显式时间序列分类（包括基于k近邻的动态时间整型[DTW]）和结合多个分类器的集合。此外，为了解决在出生来源中采样不足的生活史中的差距，我们在参考基线中添加了模拟化学剖面，并测试了它们是否提高了模型的性能。时间序列方法（平均精度：0.56-0.62）始终优于传统方法（0.48），特别是对于受母体特征或早期分散影响的来源。当轮廓遵循可预测的生命阶段模式时，特征提取方法表现最佳，而显式时间序列分类（DTW + KNN）在不同的轮廓“形状”上表现出色。集成利用互补优势，优于任何单一方法。我们的研究结果突出了将档案化学数据作为时间序列处理，并应用机器学习和集成策略来提高来源分配的准确性和可扩展性的优势。该框架广泛适用于分类群、组织类型和化学标记，为推进生态推断和保护提供了路线图。

{"title":"Advancing provenance assignment using machine learning and time series analysis of chemical chronologies in archival tissues","authors":"Kohma Arai , Malte Willmes , Rachel C. Johnson , Anna M. Sturrock","doi":"10.1016/j.ecoinf.2026.103604","DOIUrl":"10.1016/j.ecoinf.2026.103604","url":null,"abstract":"<div><div>Accurate provenance assignment is critical for understanding ecological connectivity and guiding conservation and management. Natural chemical chronologies stored in metabolically inert, incrementally growing tissues (e.g., otoliths) provide a powerful tool for this purpose. However, traditional approaches face biological challenges, collapse chronological data into single values, and require subjective interpretation, limiting accuracy and scalability. We present a novel framework that integrates machine learning, time series analysis, and ensemble modeling to improve provenance assignment from archival tissue chemistry. Using otolith <sup>87</sup>Sr/<sup>86</sup>Sr profiles from 17 distinct natal sources (<em>n</em> = 255) of California Central Valley Chinook salmon, we developed automated feature extraction, explicit time series classification (including dynamic time warping [DTW] with <em>k</em>-nearest neighbors [KNN]), and ensembles combining multiple classifiers. Furthermore, to address gaps in under-sampled life histories within natal sources, we added simulated chemical profiles to the reference baseline and tested whether they improved model performance. Time series approaches (mean accuracy: 0.56–0.62) consistently outperformed traditional methods (0.48), particularly for sources influenced by maternal signatures or early dispersal. Feature extraction approaches performed best when profiles followed predictable life-stage patterns, while explicit time series classification (DTW + KNN) excelled for distinct profile “shapes”. Ensembles leveraged complementary strengths and outperformed any single method. Our results highlight the advantages of treating archival chemical data as time series and applying machine learning and ensemble strategies to enhance the accuracy and scalability of provenance assignment. This framework is broadly applicable across taxa, tissue types, and chemical markers, offering a roadmap for advancing ecological inference and conservation.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"93 ","pages":"Article 103604"},"PeriodicalIF":7.3,"publicationDate":"2026-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976637","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}

引用次数: 0