Ecological Informatics最新文献_第2页

A novel model for mapping soil organic matter: Integrating temporal and spatial characteristics

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-12-01 DOI: 10.1016/j.ecoinf.2024.102923

Xinle Zhang , Guowei Zhang , Shengqi Zhang , Hongfu Ai , Yongqi Han , Chong Luo , Huanjun Liu

Mapping the spatial distribution of soil organic matter (SOM) content is crucial for land management decisions, yet its accurate mapping faces challenges due to complex soil-environment relationships and temporal feature capture limitations in machine learning models. This study focuses on the typical black soil region in Northeast China, specifically using Youyi Farm as the main research area and Heshan Farm as the transfer research area. A novel approach is proposed that combines the CNN-LSTM model with a Cosine Annealing Warm Restarts learning rate (CNN-LSTM-CAWR) to enhance the accuracy of SOM mapping. In this model, the Convolutional Neural Network (CNN) extracts spatial context features from static variables (e.g., climate and terrain variables), while the Long Short-Term Memory (LSTM) network captures temporal features from dynamic variables (e.g., Sentinel-2 time series from April to October). The incorporation of the CAWR learning rate helps alleviate overfitting issues. Comparing the CNN-LSTM model, CNN model, and traditional RF model, the results show that the CNN-LSTM-CAWR model achieves the highest accuracy within research Area 1 (R² = 0.64, RMSE = 0.54 %) and maintains strong performance in the transfer research area (R² = 0.60, RMSE = 0.57 %). CNN-LSTM-CAWR demonstrates faster convergence, thereby improving mapping precision and effectively utilizing temporal information from features to enhance overall model performance. This study underscores the significant potential of the hybrid CNN-LSTM with CAWR model, highlighting the valuable information for SOM mapping contained within Sentinel-2 time series data.

{"title":"A novel model for mapping soil organic matter: Integrating temporal and spatial characteristics","authors":"Xinle Zhang , Guowei Zhang , Shengqi Zhang , Hongfu Ai , Yongqi Han , Chong Luo , Huanjun Liu","doi":"10.1016/j.ecoinf.2024.102923","DOIUrl":"10.1016/j.ecoinf.2024.102923","url":null,"abstract":"<div><div>Mapping the spatial distribution of soil organic matter (SOM) content is crucial for land management decisions, yet its accurate mapping faces challenges due to complex soil-environment relationships and temporal feature capture limitations in machine learning models. This study focuses on the typical black soil region in Northeast China, specifically using Youyi Farm as the main research area and Heshan Farm as the transfer research area. A novel approach is proposed that combines the CNN-LSTM model with a Cosine Annealing Warm Restarts learning rate (CNN-LSTM-CAWR) to enhance the accuracy of SOM mapping. In this model, the Convolutional Neural Network (CNN) extracts spatial context features from static variables (e.g., climate and terrain variables), while the Long Short-Term Memory (LSTM) network captures temporal features from dynamic variables (e.g., Sentinel-2 time series from April to October). The incorporation of the CAWR learning rate helps alleviate overfitting issues. Comparing the CNN-LSTM model, CNN model, and traditional RF model, the results show that the CNN-LSTM-CAWR model achieves the highest accuracy within research Area 1 (R<sup>2</sup> = 0.64, RMSE = 0.54 %) and maintains strong performance in the transfer research area (R<sup>2</sup> = 0.60, RMSE = 0.57 %). CNN-LSTM-CAWR demonstrates faster convergence, thereby improving mapping precision and effectively utilizing temporal information from features to enhance overall model performance. This study underscores the significant potential of the hybrid CNN-LSTM with CAWR model, highlighting the valuable information for SOM mapping contained within Sentinel-2 time series data.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102923"},"PeriodicalIF":5.8,"publicationDate":"2024-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142746139","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Beyond observation: Deep learning for animal behavior and ecological conservation 超越观察：用于动物行为和生态保护的深度学习

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-26 DOI: 10.1016/j.ecoinf.2024.102893

Lyes Saad Saoud, Atif Sultan, Mahmoud Elmezain, Mohamed Heshmat, Lakmal Seneviratne, Irfan Hussain

Recent advancements in deep learning have profoundly impacted the field of animal behavioral research, offering researchers powerful tools for understanding the complexities of animal movements and cognition. This comprehensive review is dedicated to an in-depth examination of the latest techniques, tools, and applications of deep learning in this domain. This study examines the principles of deep-learning-based tracking, pose estimation, and behavioral analysis, emphasizing their respective strengths, limitations, and practical implementation. From markerless pose tracking to multi-animal behavior classification, we present a variety of methodologies that facilitate high-throughput and precise behavioral quantification across diverse species and settings. Furthermore, emerging trends, such as the integration of drones and computer vision for the study of group dynamics in natural environments, as well as advancements in semi-supervised and unsupervised learning for robust behavioral segmentation and classification, were also examined. Given the pivotal role of responsible research, we address the pivotal challenges of scalability, robustness, and ethical considerations, paving the way for future research. By synthesizing insights from seminal works in neuroscience, computer vision, and artificial intelligence, this study provides researchers with a comprehensive understanding of the powerful tools and methodologies available to unlock the secrets of animal behavior and make promising discoveries across the vast animal kingdom.

深度学习的最新进展对动物行为研究领域产生了深远影响，为研究人员了解动物运动和认知的复杂性提供了强大的工具。这篇综合评论致力于深入研究深度学习在这一领域的最新技术、工具和应用。本研究探讨了基于深度学习的跟踪、姿势估计和行为分析的原理，强调了它们各自的优势、局限性和实际应用。从无标记姿势跟踪到多动物行为分类，我们介绍了各种方法，这些方法有助于在不同物种和环境中进行高通量和精确的行为量化。此外，我们还探讨了新出现的趋势，例如将无人机与计算机视觉结合起来研究自然环境中的群体动态，以及在半监督和无监督学习方面取得的进展，以实现稳健的行为细分和分类。鉴于负责任研究的关键作用，我们探讨了可扩展性、稳健性和伦理考虑等关键挑战，为未来研究铺平了道路。本研究综合了神经科学、计算机视觉和人工智能领域开创性著作中的见解，让研究人员全面了解可用来揭开动物行为秘密的强大工具和方法，并在广袤的动物王国中取得有希望的发现。

{"title":"Beyond observation: Deep learning for animal behavior and ecological conservation","authors":"Lyes Saad Saoud, Atif Sultan, Mahmoud Elmezain, Mohamed Heshmat, Lakmal Seneviratne, Irfan Hussain","doi":"10.1016/j.ecoinf.2024.102893","DOIUrl":"10.1016/j.ecoinf.2024.102893","url":null,"abstract":"<div><div>Recent advancements in deep learning have profoundly impacted the field of animal behavioral research, offering researchers powerful tools for understanding the complexities of animal movements and cognition. This comprehensive review is dedicated to an in-depth examination of the latest techniques, tools, and applications of deep learning in this domain. This study examines the principles of deep-learning-based tracking, pose estimation, and behavioral analysis, emphasizing their respective strengths, limitations, and practical implementation. From markerless pose tracking to multi-animal behavior classification, we present a variety of methodologies that facilitate high-throughput and precise behavioral quantification across diverse species and settings. Furthermore, emerging trends, such as the integration of drones and computer vision for the study of group dynamics in natural environments, as well as advancements in semi-supervised and unsupervised learning for robust behavioral segmentation and classification, were also examined. Given the pivotal role of responsible research, we address the pivotal challenges of scalability, robustness, and ethical considerations, paving the way for future research. By synthesizing insights from seminal works in neuroscience, computer vision, and artificial intelligence, this study provides researchers with a comprehensive understanding of the powerful tools and methodologies available to unlock the secrets of animal behavior and make promising discoveries across the vast animal kingdom.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102893"},"PeriodicalIF":5.8,"publicationDate":"2024-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142720558","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Deep learning approaches for bias correction in WRF model outputs for enhanced solar and wind energy estimation: A case study in East and West Malaysia 用于 WRF 模型输出偏差校正的深度学习方法，以增强太阳能和风能估算：东西马来西亚案例研究

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-23 DOI: 10.1016/j.ecoinf.2024.102898

Abigail Birago Adomako , Ehsan Jolous Jamshidi , Yusri Yusup , Emad Elsebakhi , Mohd Hafiidz Jaafar , Muhammad Izzuddin Syakir Ishak , Hwee San Lim , Mardiana Idayu Ahmad

Accurate estimation of wind and solar energy potentials is crucial for successfully integrating renewable energy into power grids. Traditional numerical weather prediction models, such as the Weather Research and Forecasting (WRF) model, often suffer from biases that lead to inaccurate energy forecasts. This study employs advanced deep learning (DL) techniques to correct these biases in WRF model outputs, specifically to enhance wind and solar energy estimations in East and West Malaysia. Unlike previous studies, this research integrates a diverse array of DL models: Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Feedforward Neural Networks (FNN), to address both temporal and spatial prediction challenges. The models were trained and tested using historical weather data and ground-based measurements to improve the accuracy of wind speed and solar radiation predictions. Evaluation metrics, root mean square error (RMSE), mean bias error (MBE), and mean absolute error (MAE), demonstrate the better performance of CNN and FNN models over the sole WRF approach. The findings reveal that CNN achieved the lowest RMSE in wind speed estimation (0.91 in CEMACS and 0.97 in Kuching compared to WRF RMSEs of 1.92 and 1.39). At the same time, FNN significantly improved solar radiation prediction (RMSE of 86.86 in Kuching and 99.23 in CEMACS compared to WRF RMSEs of 154.44 and 370.66). Given the low wind speeds, the corrected data from CNN was used to estimate wind energy at 536 kWh at Kuching and 0 kWh at CEMACS. FNN-corrected data was also used to estimate solar energy at 19 kWh and 18 kWh at Kuching and CEMACS, respectively. This research not only shows the effectiveness of DL in mitigating biases in numerical weather prediction models but also contributes a novel methodology for reliable renewable energy assessments.

准确估计风能和太阳能的潜力对于成功将可再生能源并入电网至关重要。传统的数值天气预报模型，如天气研究与预测（WRF）模型，往往存在偏差，导致能源预测不准确。本研究采用了先进的深度学习（DL）技术来纠正 WRF 模型输出中的这些偏差，特别是为了增强马来西亚东部和西部的风能和太阳能估算。与以往的研究不同，本研究整合了多种深度学习模型：循环神经网络 (RNN)、长短期记忆 (LSTM)、卷积神经网络 (CNN) 和前馈神经网络 (FNN)，以解决时间和空间预测难题。利用历史气象数据和地面测量数据对这些模型进行了训练和测试，以提高风速和太阳辐射预测的准确性。评估指标（均方根误差 (RMSE)、平均偏差误差 (MBE) 和平均绝对误差 (MAE)）表明，与单一的 WRF 方法相比，CNN 和 FNN 模型具有更好的性能。研究结果表明，CNN 在风速估计方面的 RMSE 最低（CEMACS 为 0.91，古晋为 0.97，而 WRF 的 RMSE 分别为 1.92 和 1.39）。同时，FNN 显著改善了太阳辐射预测（古晋和 CEMACS 的 RMSE 分别为 86.86 和 99.23，而 WRF 的 RMSE 分别为 154.44 和 370.66）。鉴于风速较低，使用 CNN 修正数据估算的风能在古晋为 536 千瓦时，在 CEMACS 为 0 千瓦时。经 FNN 修正的数据还用于估算古晋和 CEMACS 的太阳能，分别为 19 千瓦时和 18 千瓦时。这项研究不仅显示了 DL 在减少数值天气预报模型偏差方面的有效性，还为可靠的可再生能源评估提供了一种新方法。

{"title":"Deep learning approaches for bias correction in WRF model outputs for enhanced solar and wind energy estimation: A case study in East and West Malaysia","authors":"Abigail Birago Adomako , Ehsan Jolous Jamshidi , Yusri Yusup , Emad Elsebakhi , Mohd Hafiidz Jaafar , Muhammad Izzuddin Syakir Ishak , Hwee San Lim , Mardiana Idayu Ahmad","doi":"10.1016/j.ecoinf.2024.102898","DOIUrl":"10.1016/j.ecoinf.2024.102898","url":null,"abstract":"<div><div>Accurate estimation of wind and solar energy potentials is crucial for successfully integrating renewable energy into power grids. Traditional numerical weather prediction models, such as the Weather Research and Forecasting (WRF) model, often suffer from biases that lead to inaccurate energy forecasts. This study employs advanced deep learning (DL) techniques to correct these biases in WRF model outputs, specifically to enhance wind and solar energy estimations in East and West Malaysia. Unlike previous studies, this research integrates a diverse array of DL models: Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Feedforward Neural Networks (FNN), to address both temporal and spatial prediction challenges. The models were trained and tested using historical weather data and ground-based measurements to improve the accuracy of wind speed and solar radiation predictions. Evaluation metrics, root mean square error (RMSE), mean bias error (MBE), and mean absolute error (MAE), demonstrate the better performance of CNN and FNN models over the sole WRF approach. The findings reveal that CNN achieved the lowest RMSE in wind speed estimation (0.91 in CEMACS and 0.97 in Kuching compared to WRF RMSEs of 1.92 and 1.39). At the same time, FNN significantly improved solar radiation prediction (RMSE of 86.86 in Kuching and 99.23 in CEMACS compared to WRF RMSEs of 154.44 and 370.66). Given the low wind speeds, the corrected data from CNN was used to estimate wind energy at 536 kWh at Kuching and 0 kWh at CEMACS. FNN-corrected data was also used to estimate solar energy at 19 kWh and 18 kWh at Kuching and CEMACS, respectively. This research not only shows the effectiveness of DL in mitigating biases in numerical weather prediction models but also contributes a novel methodology for reliable renewable energy assessments.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102898"},"PeriodicalIF":5.8,"publicationDate":"2024-11-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142720556","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Shrinking horizons: Climate-induced range shifts and conservation status of hickory trees (Carya Nutt.) 缩小的地平线：气候引起的山核桃树（Carya Nutt.）

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-22 DOI: 10.1016/j.ecoinf.2024.102910

Winnie W. Mambo , Guang-Fu Zhu , Richard I. Milne , Moses C. Wambulwa , Oyetola O. Oyebanji , Boniface K. Ngarega , Daniel Carver , Jie Liu

Understanding the intricate interplay between the geographic distributions of species and the dynamics of environmental factors is crucial for effective biodiversity management. Crop wild relatives are important resources for the improvement of cultivated plants. However, our understanding of how these species might respond to future climatic changes and their implications for conservation remains incomplete. In this study, we focus on the ecologically and economically significant hickory trees to address this knowledge gap. We employed the Biomod2 ensemble model to predict the potential distributions of 12 North American and five East Asian Carya species based on 13,643 occurrence points and 26 environmental variables. We analyzed the distribution range dynamics of hickory trees across the past, present, and future emission scenarios (2090; SSP126 and SSP585), assessed their conservation status, and conducted a preliminary threat assessment. Our results indicate that most Carya species expanded their habitat range from the Last Glacial Maximum to the present, with substantial contraction projected under both future scenarios. A northward migration shift to high elevations was observed for most species from the LGM to the future. Sixteen species were categorized as “medium priority” for further conservation action, and only one (C. tonkinensis) as “high priority”. Preliminary threat assessment classified one species (C. luana) as critically endangered, eight endangered, four vulnerable, and five as least concern. This study underscores the potential effects of climate change on the distribution of Carya species, providing crucial insights for their conservation and highlighting the broader impacts of climate change on crop wild relatives.

了解物种地理分布与环境因素动态之间错综复杂的相互作用，对于有效管理生物多样性至关重要。作物野生近缘种是改良栽培植物的重要资源。然而，我们对这些物种如何应对未来气候变化及其对保护的影响的了解仍然不全面。在本研究中，我们重点研究了具有重要生态和经济意义的山核桃树，以填补这一知识空白。我们利用 Biomod2 组合模型，基于 13643 个发生点和 26 个环境变量，预测了 12 个北美山核桃树种和 5 个东亚山核桃树种的潜在分布。我们分析了山核桃树在过去、现在和未来排放情景（2090 年；SSP126 和 SSP585）下的分布范围动态，评估了它们的保护状况，并进行了初步的威胁评估。我们的研究结果表明，从末次冰川极盛时期到现在，大多数山杉树物种的栖息地范围都在扩大，而在两种未来情景下，它们的栖息地范围预计都会大幅缩小。从末次冰期到未来，大多数物种都向高海拔地区北移。有 16 个物种被列为 "中度优先"，需要采取进一步保护措施，只有一个物种（C. tonkinensis）被列为 "高度优先"。初步威胁评估将一个物种（C. luana）列为极度濒危，八个濒危，四个易危，五个最不值得关注。这项研究强调了气候变化对 Carya 物种分布的潜在影响，为保护这些物种提供了重要的启示，并突出了气候变化对作物野生近缘种的广泛影响。

{"title":"Shrinking horizons: Climate-induced range shifts and conservation status of hickory trees (Carya Nutt.)","authors":"Winnie W. Mambo , Guang-Fu Zhu , Richard I. Milne , Moses C. Wambulwa , Oyetola O. Oyebanji , Boniface K. Ngarega , Daniel Carver , Jie Liu","doi":"10.1016/j.ecoinf.2024.102910","DOIUrl":"10.1016/j.ecoinf.2024.102910","url":null,"abstract":"<div><div>Understanding the intricate interplay between the geographic distributions of species and the dynamics of environmental factors is crucial for effective biodiversity management. Crop wild relatives are important resources for the improvement of cultivated plants. However, our understanding of how these species might respond to future climatic changes and their implications for conservation remains incomplete. In this study, we focus on the ecologically and economically significant hickory trees to address this knowledge gap. We employed the <em>Biomod2</em> ensemble model to predict the potential distributions of 12 North American and five East Asian <em>Carya</em> species based on 13,643 occurrence points and 26 environmental variables. We analyzed the distribution range dynamics of hickory trees across the past, present, and future emission scenarios (2090; SSP126 and SSP585), assessed their conservation status, and conducted a preliminary threat assessment. Our results indicate that most <em>Carya</em> species expanded their habitat range from the Last Glacial Maximum to the present, with substantial contraction projected under both future scenarios. A northward migration shift to high elevations was observed for most species from the LGM to the future. Sixteen species were categorized as “medium priority” for further conservation action, and only one (<em>C. tonkinensis</em>) as “high priority”. Preliminary threat assessment classified one species (<em>C. luana</em>) as critically endangered, eight endangered, four vulnerable, and five as least concern. This study underscores the potential effects of climate change on the distribution of <em>Carya</em> species, providing crucial insights for their conservation and highlighting the broader impacts of climate change on crop wild relatives.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102910"},"PeriodicalIF":5.8,"publicationDate":"2024-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142720557","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

TreeLearn: A deep learning method for segmenting individual trees from ground-based LiDAR forest point clouds TreeLearn：从地面激光雷达森林点云中分割单棵树木的深度学习方法

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-22 DOI: 10.1016/j.ecoinf.2024.102888

Jonathan Henrich , Jan van Delden , Dominik Seidel , Thomas Kneib , Alexander S. Ecker

Laser-scanned point clouds of forests make it possible to extract valuable information for forest management. To consider single trees, a forest point cloud needs to be segmented into individual tree point clouds. Existing segmentation methods are usually based on hand-crafted algorithms, such as identifying trunks and growing trees from them, and face difficulties in dense forests with overlapping tree crowns. In this study, we propose TreeLearn, a deep learning-based approach for tree instance segmentation of forest point clouds. TreeLearn is trained on already segmented point clouds in a data-driven manner, making it less reliant on predefined features and algorithms. Furthermore, TreeLearn is implemented as a fully automatic pipeline and does not rely on extensive hyperparameter tuning, which makes it easy to use. Additionally, we introduce a new manually segmented benchmark forest dataset containing 156 full trees. The data is generated by mobile laser scanning and contributes to create a larger and more diverse data basis for model development and fine-grained instance segmentation evaluation. We trained TreeLearn on forest point clouds of 6665 trees, labeled using the Lidar360 software. An evaluation on the benchmark dataset shows that TreeLearn performs as well as the algorithm used to generate its training data. Furthermore, the performance can be vastly improved by fine-tuning the model using manually annotated datasets. We evaluate TreeLearn on our benchmark dataset and the Wytham Woods dataset, outperforming the recent SegmentAnyTree, ForAINet and TLS2Trees methods. The TreeLearn code and all datasets that were created in the course of this work are made publicly available.

激光扫描的森林点云可为森林管理提取有价值的信息。要考虑单棵树木，需要将森林点云分割为单棵树木点云。现有的分割方法通常基于手工制作的算法，例如识别树干并从中生长出树木，在树冠重叠的茂密森林中面临困难。在本研究中，我们提出了一种基于深度学习的森林点云树木实例分割方法--TreeLearn。TreeLearn 以数据驱动的方式在已分割的点云上进行训练，从而减少了对预定义特征和算法的依赖。此外，TreeLearn 是以全自动管道的形式实现的，不依赖于大量的超参数调整，因此易于使用。此外，我们还引入了一个新的人工分割基准森林数据集，其中包含 156 棵完整的树木。这些数据由移动激光扫描生成，有助于为模型开发和细粒度实例分割评估创建更大、更多样化的数据基础。我们在使用 Lidar360 软件标注的 6665 棵树的森林点云上训练 TreeLearn。在基准数据集上进行的评估表明，TreeLearn 的性能与用于生成其训练数据的算法不相上下。此外，通过使用人工标注的数据集对模型进行微调，还能大大提高性能。我们在基准数据集和 Wytham Woods 数据集上对 TreeLearn 进行了评估，结果表明 TreeLearn 的性能优于最新的 SegmentAnyTree、ForAINet 和 TLS2Trees 方法。TreeLearn 代码和在此工作过程中创建的所有数据集均可公开获取。

{"title":"TreeLearn: A deep learning method for segmenting individual trees from ground-based LiDAR forest point clouds","authors":"Jonathan Henrich , Jan van Delden , Dominik Seidel , Thomas Kneib , Alexander S. Ecker","doi":"10.1016/j.ecoinf.2024.102888","DOIUrl":"10.1016/j.ecoinf.2024.102888","url":null,"abstract":"<div><div>Laser-scanned point clouds of forests make it possible to extract valuable information for forest management. To consider single trees, a forest point cloud needs to be segmented into individual tree point clouds. Existing segmentation methods are usually based on hand-crafted algorithms, such as identifying trunks and growing trees from them, and face difficulties in dense forests with overlapping tree crowns. In this study, we propose TreeLearn, a deep learning-based approach for tree instance segmentation of forest point clouds. TreeLearn is trained on already segmented point clouds in a data-driven manner, making it less reliant on predefined features and algorithms. Furthermore, TreeLearn is implemented as a fully automatic pipeline and does not rely on extensive hyperparameter tuning, which makes it easy to use. Additionally, we introduce a new manually segmented benchmark forest dataset containing 156 full trees. The data is generated by mobile laser scanning and contributes to create a larger and more diverse data basis for model development and fine-grained instance segmentation evaluation. We trained TreeLearn on forest point clouds of 6665 trees, labeled using the Lidar360 software. An evaluation on the benchmark dataset shows that TreeLearn performs as well as the algorithm used to generate its training data. Furthermore, the performance can be vastly improved by fine-tuning the model using manually annotated datasets. We evaluate TreeLearn on our benchmark dataset and the Wytham Woods dataset, outperforming the recent SegmentAnyTree, ForAINet and TLS2Trees methods. The TreeLearn code and all datasets that were created in the course of this work are made publicly available.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102888"},"PeriodicalIF":5.8,"publicationDate":"2024-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142701323","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Changes in vegetation ecosystem carbon sinks and their response to drought in the karst concentration distribution area of Asia 亚洲岩溶集中分布区植被生态系统碳汇的变化及其对干旱的响应

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-22 DOI: 10.1016/j.ecoinf.2024.102907

Shunfu Yang , Yuan Li , Yuluan Zhao , Anjun Lan , Chunfang Zhou , Hongxing Lu , Luanyu Zhou

Changes in net ecosystem productivity (NEP) in karst areas can have a significant impact on terrestrial ecosystem carbon cycling, yet quantifying changes in vegetation NEP and its response to factors such as drought and hydroclimate remains a difficult challenge because of its special climatic and hydrological conditions. We used remote sensing data to estimate vegetation NEP in the Asian karst concentrated distribution area (AKC), analyzed its spatial and temporal variations annually (2000−2020) and during rainy season (May–November), established the drought fluorescence monitoring index (DFMI), and used a ridge regression model to explore the response mechanism of vegetation NEP to dry and wet conditions response mechanism. The results showed the following: (1) Compared with the annual changes, the vegetation NEP changes in the rainy season differed significantly on the karst geographic divisions, in which there was a significant increasing trend in Southwest China (SC) and its karst areas, while a significant decreasing trend in the Indochina Peninsula (IP) and its karst areas. (2) DFMI was the main driver of vegetation NEP change, of which the contributions were 38.05 % and 32.82 % at the annual scale and in the rainy season, respectively, which drove the increase in SC vegetation NEP, and the decrease in IP; note that the increase in vapor pressure deficit (VPD) was the key factor causing the decrease in NEP in the IP karst area during the rainy season. (3) In the lagged effect of drought on vegetation NEP, the time scale of the lag was found to be 1 month. The study revealed differences in the changes in the vegetation carbon sinks in different karst geographic divisions. We obtained a new finding: a significant trend of decreasing vegetation NEP in the IP and its karst area was influenced by the long-term effects of changes in DFMI and VPD. Therefore, the variability of different karst areas, as well as changes in drought and water resources, should be considered in carbon-cycle regulation and vegetation restoration efforts in karst areas.

岩溶地区生态系统净生产力（NEP）的变化会对陆地生态系统的碳循环产生重大影响，但由于其特殊的气候和水文条件，量化植被NEP的变化及其对干旱和水文气候等因素的响应仍是一项艰巨的挑战。我们利用遥感数据估算了亚洲喀斯特集中分布区（AKC）的植被NEP，分析了其每年（2000-2020年）和雨季（5-11月）的时空变化，建立了干旱荧光监测指数（DFMI），并利用山脊回归模型探讨了植被NEP对干湿条件响应机制。结果表明(1）与全年变化相比，雨季植被NEP的变化在喀斯特地理分区上存在显著差异，其中中国西南（SC）及其喀斯特地区呈显著上升趋势，而印度支那半岛（IP）及其喀斯特地区呈显著下降趋势。(2）DFMI 是植被 NEP 变化的主要驱动因子，其在年尺度和雨季的贡献率分别为 38.05% 和 32.82%，驱动了西南地区植被 NEP 的增加和中南半岛植被 NEP 的减少；值得注意的是，水汽压差（VPD）的增加是导致中南半岛岩溶地区雨季植被 NEP 减少的关键因素。(3) 在干旱对植被 NEP 的滞后效应中，发现滞后的时间尺度为 1 个月。该研究揭示了不同喀斯特地理分区植被碳汇变化的差异。我们得到了一个新的发现：受 DFMI 和 VPD 变化的长期影响，IP 及其岩溶地区的植被 NEP 呈显著下降趋势。因此，在岩溶地区的碳循环调节和植被恢复工作中，应考虑不同岩溶地区的差异性以及干旱和水资源的变化。

{"title":"Changes in vegetation ecosystem carbon sinks and their response to drought in the karst concentration distribution area of Asia","authors":"Shunfu Yang , Yuan Li , Yuluan Zhao , Anjun Lan , Chunfang Zhou , Hongxing Lu , Luanyu Zhou","doi":"10.1016/j.ecoinf.2024.102907","DOIUrl":"10.1016/j.ecoinf.2024.102907","url":null,"abstract":"<div><div>Changes in net ecosystem productivity (NEP) in karst areas can have a significant impact on terrestrial ecosystem carbon cycling, yet quantifying changes in vegetation NEP and its response to factors such as drought and hydroclimate remains a difficult challenge because of its special climatic and hydrological conditions. We used remote sensing data to estimate vegetation NEP in the Asian karst concentrated distribution area (AKC), analyzed its spatial and temporal variations annually (2000−2020) and during rainy season (May–November), established the drought fluorescence monitoring index (DFMI), and used a ridge regression model to explore the response mechanism of vegetation NEP to dry and wet conditions response mechanism. The results showed the following: (1) Compared with the annual changes, the vegetation NEP changes in the rainy season differed significantly on the karst geographic divisions, in which there was a significant increasing trend in Southwest China (SC) and its karst areas, while a significant decreasing trend in the Indochina Peninsula (IP) and its karst areas. (2) DFMI was the main driver of vegetation NEP change, of which the contributions were 38.05 % and 32.82 % at the annual scale and in the rainy season, respectively, which drove the increase in SC vegetation NEP, and the decrease in IP; note that the increase in vapor pressure deficit (VPD) was the key factor causing the decrease in NEP in the IP karst area during the rainy season. (3) In the lagged effect of drought on vegetation NEP, the time scale of the lag was found to be 1 month. The study revealed differences in the changes in the vegetation carbon sinks in different karst geographic divisions. We obtained a new finding: a significant trend of decreasing vegetation NEP in the IP and its karst area was influenced by the long-term effects of changes in DFMI and VPD. Therefore, the variability of different karst areas, as well as changes in drought and water resources, should be considered in carbon-cycle regulation and vegetation restoration efforts in karst areas.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102907"},"PeriodicalIF":5.8,"publicationDate":"2024-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142701321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Bottlenose dolphin identification using synthetic image-based transfer learning 利用基于合成图像的迁移学习识别瓶鼻海豚

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-20 DOI: 10.1016/j.ecoinf.2024.102909

Changsoo Kim , Byung-Yeob Kim , Dong-Guk Paeng

The Indo-Pacific bottlenose dolphin (IPBD) (Tursiops aduncus) is a key species in marine ecosystems. Photo-identification (photo-ID) is a fundamental method for studying dolphin populations by identifying individuals based on the unique features of their dorsal fins. Despite recent developments in learning-based photo-ID algorithms, the lack of training data for these models has become a bottleneck for improving the accuracy of these algorithms. In this study, we used synthetic image generation and deep learning to improve photography-based IPBD identification. We generated 7500 synthetic dorsal fin images of 30 dolphins and trained a custom triplet neural network using ResNet50 to distinguish individuals. The model achieved 84.8 % accuracy within the top 10-ranked positions and 72.2 % accuracy in the top 5-ranked positions, demonstrating the potential of these technologies to enhance IPBD monitoring and conservation efforts.

印度-太平洋宽吻海豚（IPBD）（Tursiops aduncus）是海洋生态系统中的重要物种。照片识别（photo-ID）是研究海豚种群的基本方法，它根据海豚背鳍的独特特征来识别海豚个体。尽管基于学习的照片识别算法最近有了新的发展，但这些模型缺乏训练数据已成为提高这些算法准确性的瓶颈。在这项研究中，我们利用合成图像生成和深度学习来改进基于摄影的 IPBD 识别。我们生成了 30 头海豚的 7500 张合成背鳍图像，并使用 ResNet50 训练了一个自定义三元神经网络来区分个体。该模型在排名前 10 位的准确率达到 84.8%，在排名前 5 位的准确率达到 72.2%，证明了这些技术在加强 IPBD 监测和保护工作方面的潜力。

引用次数: 0

Self-supervised feature learning for acoustic data analysis 用于声学数据分析的自监督特征学习

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-20 DOI: 10.1016/j.ecoinf.2024.102878

Ahmet Pala , Anna Oleynik , Ketil Malde , Nils Olav Handegard

Acoustic surveys play a pivotal role in fisheries management. During the surveys, acoustic signals are sent into the water and the strength of the reflection, so-called backscatter, is recorded. The collected data are typically annotated manually, a process that is both labor-intensive and time-consuming, to support acoustic target classification (ATC). The primary objective of this study is to develop an annotation-free deep learning model that extracts acoustic features and improves the representation of acoustic data. For this purpose, we adopt a self-supervised method inspired by the Self DIstillation with NO Labels (DINO) model. Extracting useful acoustic features is an intricate task due to the inherent variability and complexity in biological targets, as well as environmental and technical factors influencing sound interactions. The proposed model is trained with three sampling methods: random sampling, which ignores class imbalance present in the acoustic survey data; class-balanced sampling, which ensures equal representation of known categories; and intensity-based sampling, which selects data to capture backscatter variations. The quality of extracted features is then evaluated and compared. We show that the extracted features lead to improvement, in comparison to using the untreated data, in the discriminative power of several machine learning methods (k-nearest neighbor (kNN), linear regression, multinomial logistic regression) for ATC. The improvement was measured through higher accuracy in kNN (77.55% vs. 71.93%), Macro AUC in logistic regression (0.92 vs. 0.80), and

R^{2}

in linear regression (0.69 vs. 0.45) when comparing extracted features to the untreated data. Our findings highlight the advantage of applying emerging self-supervised techniques in fisheries acoustics. This study thus contributes to the ongoing efforts to improve the efficiency of acoustic surveys in fisheries management.

声学调查在渔业管理中发挥着举足轻重的作用。在勘测过程中，声学信号被送入水中，并记录反射的强度，即所谓的反向散射。为支持声学目标分类（ATC），收集到的数据通常需要人工标注，这一过程既耗费人力又耗费时间。本研究的主要目的是开发一种无需注释的深度学习模型，以提取声学特征并改进声学数据的表示。为此，我们采用了一种自监督方法，其灵感来自无标签自静音（DINO）模型。由于生物目标固有的多变性和复杂性，以及影响声音相互作用的环境和技术因素，提取有用的声音特征是一项复杂的任务。所提出的模型采用三种采样方法进行训练：随机采样，忽略声学调查数据中存在的类别不平衡；类别平衡采样，确保已知类别的平等代表性；基于强度的采样，选择数据以捕捉反向散射变化。然后对提取特征的质量进行评估和比较。我们发现，与使用未经处理的数据相比，提取的特征提高了几种机器学习方法（k-近邻（kNN）、线性回归、多项式逻辑回归）对 ATC 的判别能力。在将提取的特征与未经处理的数据进行比较时，KNN 的准确率（77.55% 对 71.93%）、逻辑回归的宏观 AUC（0.92 对 0.80）和线性回归的 R2（0.69 对 0.45）均有所提高。我们的研究结果凸显了在渔业声学中应用新兴自监督技术的优势。因此，这项研究有助于提高声学调查在渔业管理中的效率。

{"title":"Self-supervised feature learning for acoustic data analysis","authors":"Ahmet Pala , Anna Oleynik , Ketil Malde , Nils Olav Handegard","doi":"10.1016/j.ecoinf.2024.102878","DOIUrl":"10.1016/j.ecoinf.2024.102878","url":null,"abstract":"<div><div>Acoustic surveys play a pivotal role in fisheries management. During the surveys, acoustic signals are sent into the water and the strength of the reflection, so-called backscatter, is recorded. The collected data are typically annotated manually, a process that is both labor-intensive and time-consuming, to support acoustic target classification (ATC). The primary objective of this study is to develop an annotation-free deep learning model that extracts acoustic features and improves the representation of acoustic data. For this purpose, we adopt a self-supervised method inspired by the Self DIstillation with NO Labels (DINO) model. Extracting useful acoustic features is an intricate task due to the inherent variability and complexity in biological targets, as well as environmental and technical factors influencing sound interactions. The proposed model is trained with three sampling methods: random sampling, which ignores class imbalance present in the acoustic survey data; class-balanced sampling, which ensures equal representation of known categories; and intensity-based sampling, which selects data to capture backscatter variations. The quality of extracted features is then evaluated and compared. We show that the extracted features lead to improvement, in comparison to using the untreated data, in the discriminative power of several machine learning methods (k-nearest neighbor (kNN), linear regression, multinomial logistic regression) for ATC. The improvement was measured through higher accuracy in kNN (77.55% vs. 71.93%), Macro AUC in logistic regression (0.92 vs. 0.80), and <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> in linear regression (0.69 vs. 0.45) when comparing extracted features to the untreated data. Our findings highlight the advantage of applying emerging self-supervised techniques in fisheries acoustics. This study thus contributes to the ongoing efforts to improve the efficiency of acoustic surveys in fisheries management.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102878"},"PeriodicalIF":5.8,"publicationDate":"2024-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142719997","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Tracing the range shifts of African tree ferns: Insights from the last glacial maximum and beyond 追踪非洲蕨类植物的分布范围变化：从末次冰川极盛时期及其后的洞察力

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-18 DOI: 10.1016/j.ecoinf.2024.102896

Mwihaki J. Karichu , Boniface K. Ngarega , Joyce M. Jefwa , Bette A. Loiselle , Emily B. Sessa

African tropical forests are experiencing rapid decline as a result of several factors, including increasing population pressure, recurrent wildfires, selective logging practices, land use changes, intensified agricultural activities, and other social and economic issues. Using MaxEnt, paleoclimatic data, and future climate scenarios, the present study seeks to explore the presence of tree ferns in tropical and Saharan Africa during the Last Glacial Maximum, African Holocene Humid Period (AHHP; ca. 14,500–5000 years ago) and to project the effects of climate change on the distribution of tree ferns in Africa under two future climatic scenarios, Representation Concentration Pathways (RCP) 4.5 and 8.5. Our study reveals that despite a significant increase in precipitation during the AHHP, precipitation distribution was variable and insufficient to support the five tree fern species examined in this study. While some tree fern species have experienced range shifts over time, we found that most of them have maintained their presence within refuge areas that probably endured the late Pleistocene extinction event. These refugia provided a haven for some tree ferns, allowing them to persist and survive amidst challenging and varying environmental conditions. This highlights tree ferns' remarkable adaptability in changing climate as well as the critical importance of these refugial areas in safeguarding their populations during climatic upheaval. Our study further demonstrates that different species respond to climate change differently, with some experiencing minimal range contractions of 2.0 %, up to more than 57.0 % range expansion in other species. Preserving refugia not only safeguards tree fern populations but also contributes to conserving overall forest biodiversity and ecosystem functioning. This knowledge is crucial for implementing targeted conservation actions that promote sustainable forest management and can mitigate the threats posed by climate change and anthropogenic activities in African closed wet forests.

由于人口压力不断增加、野火频发、选择性伐木、土地用途改变、农业活动加剧以及其他社会和经济问题等多种因素，非洲热带森林正在经历快速衰退。本研究利用MaxEnt、古气候数据和未来气候情景，试图探讨非洲热带和撒哈拉地区在末次冰川极盛时期、非洲全新世湿润时期（AHHP，约14500-5000年前）树蕨的存在情况，并预测在两种未来气候情景（代表浓度途径（RCP）4.5和8.5）下气候变化对非洲树蕨分布的影响。我们的研究表明，尽管在非洲高原水文计划期间降水量显著增加，但降水分布不均，不足以支持本研究中考察的五种树蕨物种。随着时间的推移，一些树蕨类物种的分布范围发生了变化，但我们发现，大多数树蕨类物种都留在了可能经历了晚更新世物种灭绝事件的避难区内。这些避难所为一些树蕨类植物提供了庇护所，使它们能够在充满挑战和多变的环境条件下存活下来。这凸显了树蕨类植物在不断变化的气候中的卓越适应能力，以及这些避难区在气候动荡期间保护其种群的极端重要性。我们的研究进一步表明，不同物种对气候变化的反应各不相同，有些物种的分布区收缩幅度很小，仅为2.0%，而另一些物种的分布区则扩大了57.0%。保护避难所不仅能保护蕨类植物种群，还有助于保护整个森林的生物多样性和生态系统功能。这些知识对于实施有针对性的保护行动至关重要，这些行动可促进可持续森林管理，并能减轻气候变化和人类活动对非洲封闭湿林造成的威胁。

{"title":"Tracing the range shifts of African tree ferns: Insights from the last glacial maximum and beyond","authors":"Mwihaki J. Karichu , Boniface K. Ngarega , Joyce M. Jefwa , Bette A. Loiselle , Emily B. Sessa","doi":"10.1016/j.ecoinf.2024.102896","DOIUrl":"10.1016/j.ecoinf.2024.102896","url":null,"abstract":"<div><div>African tropical forests are experiencing rapid decline as a result of several factors, including increasing population pressure, recurrent wildfires, selective logging practices, land use changes, intensified agricultural activities, and other social and economic issues. Using MaxEnt, paleoclimatic data, and future climate scenarios, the present study seeks to explore the presence of tree ferns in tropical and Saharan Africa during the Last Glacial Maximum, African Holocene Humid Period (AHHP; ca. 14,500–5000 years ago) and to project the effects of climate change on the distribution of tree ferns in Africa under two future climatic scenarios, Representation Concentration Pathways (RCP) 4.5 and 8.5. Our study reveals that despite a significant increase in precipitation during the AHHP, precipitation distribution was variable and insufficient to support the five tree fern species examined in this study. While some tree fern species have experienced range shifts over time, we found that most of them have maintained their presence within refuge areas that probably endured the late Pleistocene extinction event. These refugia provided a haven for some tree ferns, allowing them to persist and survive amidst challenging and varying environmental conditions. This highlights tree ferns' remarkable adaptability in changing climate as well as the critical importance of these refugial areas in safeguarding their populations during climatic upheaval. Our study further demonstrates that different species respond to climate change differently, with some experiencing minimal range contractions of 2.0 %, up to more than 57.0 % range expansion in other species. Preserving refugia not only safeguards tree fern populations but also contributes to conserving overall forest biodiversity and ecosystem functioning. This knowledge is crucial for implementing targeted conservation actions that promote sustainable forest management and can mitigate the threats posed by climate change and anthropogenic activities in African closed wet forests.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102896"},"PeriodicalIF":5.8,"publicationDate":"2024-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142701325","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Improving the explainability of CNN-LSTM-based flood prediction with integrating SHAP technique 利用 SHAP 技术提高基于 CNN-LSTM 的洪水预测的可解释性

IF 5.8 2区环境科学与生态学 Q1 ECOLOGY

Ecological Informatics

Pub Date : 2024-11-17 DOI: 10.1016/j.ecoinf.2024.102904

Hao Huang , Zhaoli Wang , Yaoxing Liao , Weizhi Gao , Chengguang Lai , Xushu Wu , Zhaoyang Zeng

Convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) are popular deep learning architectures currently used for rapid flood simulations. However, deep learning algorithms are difficult to explain, like a “black box” that lacks insight. In order to reveal the intrinsic mechanism of prediction by such architectures, we adopted a coupled CNN-LSTM model based on the explainability technique SHapley Additive exPlanations (SHAP) to predict the rainfall-runoff process and identify key input feature factors, and took the Beijiang River Basin in China as an example, so as to improve the explainability and credibility of this black-box model. The results show that the coupled CNN-LSTM model performs better than the flood predictions compared to the individual CNN or LSTM models under the longest foresight period of 25 h. In particular, the Nash-Sutcliffe Efficiency (NSE) of the former model reaches 0.838, while those of the latter two models are 0.737 and 0.745, respectively. The coupled CNN-LSTM model has a high-accuracy prediction capability, consistently exhibiting NSEs greater than 0.8 for different input time steps and foresight periods. The prediction accuracy is mainly influenced by the observed runoff at the downstream hydrological station from previous time points, while the effects of the input time step length and the foresight period are comparatively negligible. This study provides a new perspective for understanding the potential physical mechanism of black-box models for rainfall-runoff prediction and emphasizes the importance and prospect of the application of explainability techniques.

卷积神经网络（CNN）和长短期记忆网络（LSTM）是目前用于快速洪水模拟的流行深度学习架构。然而，深度学习算法很难解释，就像一个缺乏洞察力的 "黑盒子"。为了揭示这类架构预测的内在机理，我们采用了基于可解释性技术SHapley Additive exPlanations（SHAP）的耦合CNN-LSTM模型来预测降雨径流过程，识别关键输入特征因子，并以中国北江流域为例，以提高这一黑箱模型的可解释性和可信度。结果表明，与单独的 CNN 或 LSTM 模型相比，耦合 CNN-LSTM 模型在最长预报期 25 h 的洪水预报中表现更好，其中前者的 Nash-Sutcliffe 效率（NSE）达到 0.838，而后两者分别为 0.737 和 0.745。耦合 CNN-LSTM 模型具有高精度预测能力，在不同输入时间步长和预测周期下，NSE 均大于 0.8。预测精度主要受下游水文站前几个时间点观测到的径流量影响，而输入时间步长和预见期的影响相对较小。这项研究为理解降雨-径流预测黑箱模型的潜在物理机制提供了一个新的视角，并强调了可解释性技术应用的重要性和前景。

{"title":"Improving the explainability of CNN-LSTM-based flood prediction with integrating SHAP technique","authors":"Hao Huang , Zhaoli Wang , Yaoxing Liao , Weizhi Gao , Chengguang Lai , Xushu Wu , Zhaoyang Zeng","doi":"10.1016/j.ecoinf.2024.102904","DOIUrl":"10.1016/j.ecoinf.2024.102904","url":null,"abstract":"<div><div>Convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) are popular deep learning architectures currently used for rapid flood simulations. However, deep learning algorithms are difficult to explain, like a “black box” that lacks insight. In order to reveal the intrinsic mechanism of prediction by such architectures, we adopted a coupled CNN-LSTM model based on the explainability technique SHapley Additive exPlanations (SHAP) to predict the rainfall-runoff process and identify key input feature factors, and took the Beijiang River Basin in China as an example, so as to improve the explainability and credibility of this black-box model. The results show that the coupled CNN-LSTM model performs better than the flood predictions compared to the individual CNN or LSTM models under the longest foresight period of 25 h. In particular, the Nash-Sutcliffe Efficiency (NSE) of the former model reaches 0.838, while those of the latter two models are 0.737 and 0.745, respectively. The coupled CNN-LSTM model has a high-accuracy prediction capability, consistently exhibiting NSEs greater than 0.8 for different input time steps and foresight periods. The prediction accuracy is mainly influenced by the observed runoff at the downstream hydrological station from previous time points, while the effects of the input time step length and the foresight period are comparatively negligible. This study provides a new perspective for understanding the potential physical mechanism of black-box models for rainfall-runoff prediction and emphasizes the importance and prospect of the application of explainability techniques.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"84 ","pages":"Article 102904"},"PeriodicalIF":5.8,"publicationDate":"2024-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142701332","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0