Digital cameras are widely used for documenting phenological observations, and numerous images have been collected. However, intelligent approaches are required to extract valuable phenological information from time‐series images. In this study, we used machine learning (ML) algorithms, including convolutional neural network (CNN)‐based You Only Look Once (YOLO) object detection and semantic segmentation methods to identify flowers in images, establish curves of flower count and flower cover, and extract the phenophases of first, peak and end flowering. Random forests (RF) was performed to recognize flower pixels to calculate the flower cover, construct the flower cover curve and extract the same phenophases as those of the YOLO methods. Furthermore, flowering phenophases were also extracted through manual visual identification. We used a generalized additive model (GAM) to fit curves for flower count and flower cover, and extracted flowering phenophases by calculating the inflection points of the fitted curves. We found that (1) YOLO‐based methods could effectively identify flowers, and the variation in flower count and flower cover obtained from the YOLO object detection and semantic segmentation models reflected the trend of flowering phenology. The flower count and flower cover curves effectively supported the extraction of first and peak flowering. The difference between the YOLO‐identified and manually identified flowering phenophases ranged from 1 day to 3 days using the optimal thresholds. For end flowering, except for the end flowering identified based on flower count derived from YOLO object detection, the date difference in phenophases between the YOLO‐identified and manually identified ranged from 1 day to 8 days. (2) There are apparent outliers in the RF‐calculated flower cover values, particularly during the post‐peak‐flowering period. However, the identified flowering phenophases based on the RF‐derived flower cover curve after omitting outliers were consistent with those of manual visual identification and YOLO‐based methods (except end flowering identified based on flower count derived from YOLO object detection), with the date difference in phenophases ranging from 0 to 8 days. (3) The GAM performed well in fitting the trends of the normalized cumulative flower count and flower cover. Using the threshold generated by second derivate method, the identified end flowering was close to that of “late flowering” stage identified by manual visual identification, and the date difference ranged from 0 to 6 days. (4) Due to the variation in flowering rhythm and progression across different plant species, fixed thresholds are not fully optimal for all plants, and the thresholds used to extract flowering phenology require targeted adjustments based on specific observed species. Our study showed that a time‐lapse digital camera combined with ML algorithms can help improve the objectivity of phenology observations, indicating the possibility
数码相机被广泛用于记录物候观察，并收集了许多图像。然而，需要智能的方法从时间序列图像中提取有价值的物候信息。在这项研究中，我们使用机器学习（ML）算法，包括基于卷积神经网络（CNN）的You Only Look Once （YOLO）对象检测和语义分割方法来识别图像中的花卉，建立花数和花覆盖曲线，并提取开花的首、峰和末物候期。采用随机森林（Random forests， RF）方法识别花像素点，计算花覆盖，构建花覆盖曲线，提取与YOLO方法相同的物候期。此外，还通过人工视觉识别方法提取了开花物候。采用广义加性模型（GAM）拟合花数和花盖度曲线，通过计算拟合曲线的拐点提取开花物候。研究发现：(1)基于YOLO的方法可以有效识别花卉，YOLO对象检测和语义分割模型得到的花数和花盖度变化反映了开花物候变化趋势。花数和花盖曲线有效地支持了花期和花期的提取。使用最佳阈值，YOLO鉴定和人工鉴定的开花物候期之间的差异为1天至3天。对于终花期，除了根据YOLO目标检测得出的花数来鉴定的终花期外，YOLO鉴定的物候期与人工鉴定的物候期差异在1天到8天之间。(2) RF计算的花盖度值存在明显的异常值，特别是在花期高峰后。然而，剔除异常值后，基于RF提取的花盖曲线识别的开花物候期与人工视觉识别和基于YOLO的方法一致（基于YOLO对象检测提取的花数识别的终末花期除外），物候期的日期差异在0 ~ 8天之间。(3) GAM能较好地拟合归一化累计花数和花盖度的变化趋势。利用二阶导数法产生的阈值，鉴定的末花期与人工目测鉴定的“晚花期”接近，日期差异在0 ~ 6 d之间。(4)由于开花节律和开花进程在不同植物物种间的差异，固定的阈值并非对所有植物都是完全最优的，用于提取开花物候的阈值需要根据特定的观测物种进行有针对性的调整。我们的研究表明，将延时数码相机与ML算法相结合可以帮助提高物候观测的客观性，这表明使用ML算法识别开花物候的可能性。
{"title":"Time‐series digital camera photos combined with machine learning algorithms can realize accurate observation of flowering phenology","authors":"Chuangye Song, Yuan Jia, Lin Zhang, Dongxiu Wu","doi":"10.1002/rse2.70069","DOIUrl":"https://doi.org/10.1002/rse2.70069","url":null,"abstract":"Digital cameras are widely used for documenting phenological observations, and numerous images have been collected. However, intelligent approaches are required to extract valuable phenological information from time‐series images. In this study, we used machine learning (ML) algorithms, including convolutional neural network (CNN)‐based You Only Look Once (YOLO) object detection and semantic segmentation methods to identify flowers in images, establish curves of flower count and flower cover, and extract the phenophases of first, peak and end flowering. Random forests (RF) was performed to recognize flower pixels to calculate the flower cover, construct the flower cover curve and extract the same phenophases as those of the YOLO methods. Furthermore, flowering phenophases were also extracted through manual visual identification. We used a generalized additive model (GAM) to fit curves for flower count and flower cover, and extracted flowering phenophases by calculating the inflection points of the fitted curves. We found that (1) YOLO‐based methods could effectively identify flowers, and the variation in flower count and flower cover obtained from the YOLO object detection and semantic segmentation models reflected the trend of flowering phenology. The flower count and flower cover curves effectively supported the extraction of first and peak flowering. The difference between the YOLO‐identified and manually identified flowering phenophases ranged from 1 day to 3 days using the optimal thresholds. For end flowering, except for the end flowering identified based on flower count derived from YOLO object detection, the date difference in phenophases between the YOLO‐identified and manually identified ranged from 1 day to 8 days. (2) There are apparent outliers in the RF‐calculated flower cover values, particularly during the post‐peak‐flowering period. However, the identified flowering phenophases based on the RF‐derived flower cover curve after omitting outliers were consistent with those of manual visual identification and YOLO‐based methods (except end flowering identified based on flower count derived from YOLO object detection), with the date difference in phenophases ranging from 0 to 8 days. (3) The GAM performed well in fitting the trends of the normalized cumulative flower count and flower cover. Using the threshold generated by second derivate method, the identified end flowering was close to that of “late flowering” stage identified by manual visual identification, and the date difference ranged from 0 to 6 days. (4) Due to the variation in flowering rhythm and progression across different plant species, fixed thresholds are not fully optimal for all plants, and the thresholds used to extract flowering phenology require targeted adjustments based on specific observed species. Our study showed that a time‐lapse digital camera combined with ML algorithms can help improve the objectivity of phenology observations, indicating the possibility ","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"313 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147492810","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}
Meghan T. Hayden, Matthew W. Rossi, Laura E. Dee, Kyle Kovach, Cibele H. Amaral, Jacob Nesslage, Madeline Slimp, Rachel S. Meyer, E. Natasha Stavros
Biodiversity is under threat globally, with significant implications for the ecosystem processes that underpin human well‐being. Effective conservation efforts require scalable, replicable metrics to detect and monitor changes in biodiversity. However, a persistent challenge is deciding on the spatial scale over which to quantify biodiversity—including when using metrics derived from remote sensing—which is inherently scale‐dependent. Understanding the scaling properties of remote sensing metrics is thus important for biodiversity change detection and assessment. We address this challenge by investigating the scale dependence of two remotely sensed vegetation diversity metrics, spectral richness and divergence, across 15 diverse ecosystems that are part of the United States National Ecological Observatory Network (NEON). Our continental‐scale analysis builds on the success of similar studies that have shown scale dependence of spectral richness in select forest ecosystems. Our results corroborate prior findings that show that spectral richness follows well‐established ecological scaling laws by adhering to the sub‐linear scaling expected for species–area and functional diversity area relationships. We compare these scaling relationships to the null expectation of randomly distributed pixel values, demonstrating that empirical scaling relationships are non‐random. Comparing diverse ecosystems using the same data and methods, we show how scaling parameters encode important information on the relative roles of climate, geomorphology, and ecosystem structure on vegetation‐based biodiversity metrics. By advancing our understanding of the scale dependence of remotely sensed biodiversity metrics, this study lays a foundation for leveraging remote sensing data in global biodiversity monitoring and conservation.
{"title":"Scale dependence in remotely sensed biodiversity: Leveraging continental‐scale imaging spectroscopy from the National Ecological Observatory Network","authors":"Meghan T. Hayden, Matthew W. Rossi, Laura E. Dee, Kyle Kovach, Cibele H. Amaral, Jacob Nesslage, Madeline Slimp, Rachel S. Meyer, E. Natasha Stavros","doi":"10.1002/rse2.70068","DOIUrl":"https://doi.org/10.1002/rse2.70068","url":null,"abstract":"Biodiversity is under threat globally, with significant implications for the ecosystem processes that underpin human well‐being. Effective conservation efforts require scalable, replicable metrics to detect and monitor changes in biodiversity. However, a persistent challenge is deciding on the spatial scale over which to quantify biodiversity—including when using metrics derived from remote sensing—which is inherently scale‐dependent. Understanding the scaling properties of remote sensing metrics is thus important for biodiversity change detection and assessment. We address this challenge by investigating the scale dependence of two remotely sensed vegetation diversity metrics, spectral richness and divergence, across 15 diverse ecosystems that are part of the United States National Ecological Observatory Network (NEON). Our continental‐scale analysis builds on the success of similar studies that have shown scale dependence of spectral richness in select forest ecosystems. Our results corroborate prior findings that show that spectral richness follows well‐established ecological scaling laws by adhering to the sub‐linear scaling expected for species–area and functional diversity area relationships. We compare these scaling relationships to the null expectation of randomly distributed pixel values, demonstrating that empirical scaling relationships are non‐random. Comparing diverse ecosystems using the same data and methods, we show how scaling parameters encode important information on the relative roles of climate, geomorphology, and ecosystem structure on vegetation‐based biodiversity metrics. By advancing our understanding of the scale dependence of remotely sensed biodiversity metrics, this study lays a foundation for leveraging remote sensing data in global biodiversity monitoring and conservation.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"12 11 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147470950","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}
Anton Kuzmin, Lauri Korhonen, Topi Tanhuanpää, Mikko Kukkonen, Matti Maltamo, Timo Kumpula
In boreal forests, old deciduous trees, particularly European Aspen ( Populus tremula L.), play a crucial role in supporting biodiversity by providing unique habitats for cavity‐nesting birds, insects, and mammals. Despite their ecological importance, the low economic value and sparse distribution of aspen limit knowledge of their spatial and temporal distribution, hindering effective forest management and conservation. Similarly, standing dead trees are vital for biodiversity, offering habitats for numerous species. Accurate identification of tree species and standing dead trees is essential for forest mapping and biodiversity monitoring. Unmanned aerial vehicles (UAVs) have proven effective for detailed forest assessments, offering imagery with ultra‐high spatial resolution at relatively low costs. Their flexibility and customizable sensor payloads enable rapid data acquisition in challenging forest regions, making them a cost‐efficient alternative to manned aircraft. This study assessed the accuracy of different UAV‐based sensors and their combinations in classifying Scots pine ( Pinus sylvestris L.), Norway spruce ( Picea abies (L.) Karst.), birches ( Betula pendula Roth and Betula pubescens Ehrh.), European aspen, and standing dead trees. Spectral and structural features from true‐color (RGB) and multispectral (MSP) photogrammetric point clouds, as well as LiDAR data, were used as predictors. A total of 1,205 field‐measured trees (approx. 250 per class) were analyzed, with 70% used for training and 30% for validation. Our results showed that the LiDAR + MSP approach achieved the highest accuracy (78%) and kappa value (0.72), effectively leveraging LiDAR's structural detail and MSP's spectral richness. Among single sensors, MSP performed best (75% accuracy), while RGB and LiDAR achieved 71% and 60%, respectively. These findings highlight that while single‐sensor datasets can perform well, fusing spectral and structural data is essential for maximizing classification accuracy. UAV‐based multi‐sensor approaches offer significant potential for advancing assessments of biodiversity indicators and sustainable forest management.
{"title":"Classification of tree species and standing dead trees in Boreal forests using UAV‐based RGB, multispectral, and LiDAR point clouds","authors":"Anton Kuzmin, Lauri Korhonen, Topi Tanhuanpää, Mikko Kukkonen, Matti Maltamo, Timo Kumpula","doi":"10.1002/rse2.70070","DOIUrl":"https://doi.org/10.1002/rse2.70070","url":null,"abstract":"In boreal forests, old deciduous trees, particularly European Aspen ( <jats:italic>Populus tremula</jats:italic> L.), play a crucial role in supporting biodiversity by providing unique habitats for cavity‐nesting birds, insects, and mammals. Despite their ecological importance, the low economic value and sparse distribution of aspen limit knowledge of their spatial and temporal distribution, hindering effective forest management and conservation. Similarly, standing dead trees are vital for biodiversity, offering habitats for numerous species. Accurate identification of tree species and standing dead trees is essential for forest mapping and biodiversity monitoring. Unmanned aerial vehicles (UAVs) have proven effective for detailed forest assessments, offering imagery with ultra‐high spatial resolution at relatively low costs. Their flexibility and customizable sensor payloads enable rapid data acquisition in challenging forest regions, making them a cost‐efficient alternative to manned aircraft. This study assessed the accuracy of different UAV‐based sensors and their combinations in classifying Scots pine ( <jats:italic>Pinus sylvestris</jats:italic> L.), Norway spruce ( <jats:italic>Picea abies</jats:italic> (L.) Karst.), birches ( <jats:italic>Betula pendula</jats:italic> Roth and <jats:italic>Betula pubescens</jats:italic> Ehrh.), European aspen, and standing dead trees. Spectral and structural features from true‐color (RGB) and multispectral (MSP) photogrammetric point clouds, as well as LiDAR data, were used as predictors. A total of 1,205 field‐measured trees (approx. 250 per class) were analyzed, with 70% used for training and 30% for validation. Our results showed that the LiDAR + MSP approach achieved the highest accuracy (78%) and kappa value (0.72), effectively leveraging LiDAR's structural detail and MSP's spectral richness. Among single sensors, MSP performed best (75% accuracy), while RGB and LiDAR achieved 71% and 60%, respectively. These findings highlight that while single‐sensor datasets can perform well, fusing spectral and structural data is essential for maximizing classification accuracy. UAV‐based multi‐sensor approaches offer significant potential for advancing assessments of biodiversity indicators and sustainable forest management.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"6 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147470949","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}
The development of remote sensing and object detection technologies has advanced benthos surveys. However, challenges remain in accuracy and cost‐efficiency due to environmental interference. A practical method combining drone‐based image acquisition and deep learning techniques for benthos monitoring is presented. Field experiments objecting hermit crabs were conducted at Lake Hamana using drones at altitudes of 2 m, 5 m and 10 m. Super‐resolution reconstruction (SRR) was applied to enhance image quality, followed by small‐object detection using the self‐built V9‐BENTHOS. With a magnification factor × 4, Residual Dense Network (RDN) achieved optimal SRR performance (PSNR: 38.15 dB, SSIM: 88.51%) and V9‐BENTHOS reached a mean average precision of 95.5%. The effects of SRR algorithms and magnification factors on hermit crab detection were discussed. This case study provides a new approach to support benthos ecological monitoring.
{"title":"Deep learning‐based super‐resolution reconstruction and improved YOLOv9 for efficient benthos detection: a case study at Lake Hamana, Japan","authors":"Fan Zhao, Bangzhang Ma, Dianhan Xi, Jiaqi Wang, Yijia Chen, Yongying Liu, Xinlei Shao, Mowen Zhang, Guocheng Zhang, Jundong Chen, Katsunori Mizuno","doi":"10.1002/rse2.70066","DOIUrl":"https://doi.org/10.1002/rse2.70066","url":null,"abstract":"The development of remote sensing and object detection technologies has advanced benthos surveys. However, challenges remain in accuracy and cost‐efficiency due to environmental interference. A practical method combining drone‐based image acquisition and deep learning techniques for benthos monitoring is presented. Field experiments objecting hermit crabs were conducted at Lake Hamana using drones at altitudes of 2 m, 5 m and 10 m. Super‐resolution reconstruction (SRR) was applied to enhance image quality, followed by small‐object detection using the self‐built V9‐BENTHOS. With a magnification factor × 4, Residual Dense Network (RDN) achieved optimal SRR performance (PSNR: 38.15 dB, SSIM: 88.51%) and V9‐BENTHOS reached a mean average precision of 95.5%. The effects of SRR algorithms and magnification factors on hermit crab detection were discussed. This case study provides a new approach to support benthos ecological monitoring.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"9 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147448046","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}
Fannie W. Shabangu, Grant van der Heever, Charles von der Meden, Hannah Truter, Stephen J. Lamberth, Ofer Gon
Acoustic ecology of Southern Ocean fishes is currently unknown due to lack of dedicated fish acoustic research from those remote/inaccessible areas. The objective of this study was to investigate the monthly and diel acoustic occurrence pattern of benthic fishes relative to environmental conditions at the sub‐Antarctic Prince Edward Islands (PEIs) in the Southern Ocean. To collect our passive acoustic data, we used an autonomous recorder deployed at ~167 m water depth on an oceanographic mooring over 21 months (April 2021 to December 2022). Benthic Ski‐Monkey III towed camera was deployed around the PEIs to identify potential sources of recorded underwater fish sounds. Three types of sounds (pops, grunts and drum sounds) were detected and validated using random forest models based on their characteristics. Pops and grunts were produced in series and as singlets. Pops were the most frequently detected sounds and were detected in December 2021 through May 2022, whereas grunts were detected in January through March 2022. Drum sounds were rare and were detected as singlets on a few occasions in December 2021 through March 2022. These monthly fish occurrences correspond to the breeding season of fishes in the Southern Ocean, suggesting the use of acoustic cues during breeding. From camera footage, Nototheniops larseni (painted notothen) was the only fish species found around the acoustic recorder location, and pops were putatively attributed to this abundant species, whereas other sounds were attributed to other observed species. Fish sound occurrence increased around sunrise and sunset. Sea surface temperatures between 5.2°C and 8°C were the primary predictor of fish acoustic occurrence, underscoring the potential vulnerability of these fish to environmental change. This study provides the first evidence of monthly and diel acoustic occurrence of soniferous fishes and demonstrates that bioacoustics can monitor fish biodiversity and breeding phenology in the Southern Ocean.
{"title":"Rhyming in the cold: first evidence of soniferous fishes in the Southern Ocean","authors":"Fannie W. Shabangu, Grant van der Heever, Charles von der Meden, Hannah Truter, Stephen J. Lamberth, Ofer Gon","doi":"10.1002/rse2.70065","DOIUrl":"https://doi.org/10.1002/rse2.70065","url":null,"abstract":"Acoustic ecology of Southern Ocean fishes is currently unknown due to lack of dedicated fish acoustic research from those remote/inaccessible areas. The objective of this study was to investigate the monthly and diel acoustic occurrence pattern of benthic fishes relative to environmental conditions at the sub‐Antarctic Prince Edward Islands (PEIs) in the Southern Ocean. To collect our passive acoustic data, we used an autonomous recorder deployed at ~167 m water depth on an oceanographic mooring over 21 months (April 2021 to December 2022). Benthic Ski‐Monkey III towed camera was deployed around the PEIs to identify potential sources of recorded underwater fish sounds. Three types of sounds (pops, grunts and drum sounds) were detected and validated using random forest models based on their characteristics. Pops and grunts were produced in series and as singlets. Pops were the most frequently detected sounds and were detected in December 2021 through May 2022, whereas grunts were detected in January through March 2022. Drum sounds were rare and were detected as singlets on a few occasions in December 2021 through March 2022. These monthly fish occurrences correspond to the breeding season of fishes in the Southern Ocean, suggesting the use of acoustic cues during breeding. From camera footage, <jats:italic>Nototheniops larseni</jats:italic> (painted notothen) was the only fish species found around the acoustic recorder location, and pops were putatively attributed to this abundant species, whereas other sounds were attributed to other observed species. Fish sound occurrence increased around sunrise and sunset. Sea surface temperatures between 5.2°C and 8°C were the primary predictor of fish acoustic occurrence, underscoring the potential vulnerability of these fish to environmental change. This study provides the first evidence of monthly and diel acoustic occurrence of soniferous fishes and demonstrates that bioacoustics can monitor fish biodiversity and breeding phenology in the Southern Ocean.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"412 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147447947","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}
Lubomír Tichý, Patricia Singh, Petra Hájková, Anna Müllerová, Tomáš Peterka, Zuzana Plesková, Karel Prach, Adéla Široká, Kamila Vítovcová, Michal Hájek
Small temperate fens rank among the most endangered habitats in temperate Europe. In agricultural landscapes, they are highly vulnerable to eutrophication and desiccation, which accelerate biodiversity loss and shifts in the carbon balance due to peat mineralization. The initial signs of habitat change are commonly manifested by shifts in vegetation structure and dominance, accompanied by increasing productivity, which precede major qualitative changes in species composition. The in‐time monitoring of vegetation productivity and site wetness at large areas is essential for guiding conservation management strategies for fens to slow down or reverse undesired changes. Here, we evaluated the ability of satellite (Sentinel‐2) and high‐resolution aerial imagery to detect early, structure‐ and productivity‐related signals of fen deterioration. We compared multispectral and optical imagery with ground‐based data, including both direct measurements and indicators derived from the species composition of the vegetation plots. At the landscape scale where both the acidic poor fens and the base‐rich fens occurred, MSAVI and NGRDI indices performed best, indicating primarily the vascular plant cover, species richness and representation of nutrient‐demanding species. At the within‐site scale, where the differences among plots were largely driven by habitat deterioration, NDVI, NDWI and RENDVI well captured differences in vascular plant productivity estimates and moss biomass measurements. Our results indicate that remote sensing is applicable for the identification of individual fen habitats and their nutrient status at the landscape scale and is even effective in detecting incipient habitat deterioration associated with increasing productivity. We demonstrate that remote sensing also performs well for small, island‐like fen patches. Its wider integration into the mire research would improve monitoring and enhance the amount of available ecological data.
{"title":"Identification of initial vegetation and habitat changes in small temperate fens using remote sensing","authors":"Lubomír Tichý, Patricia Singh, Petra Hájková, Anna Müllerová, Tomáš Peterka, Zuzana Plesková, Karel Prach, Adéla Široká, Kamila Vítovcová, Michal Hájek","doi":"10.1002/rse2.70067","DOIUrl":"https://doi.org/10.1002/rse2.70067","url":null,"abstract":"Small temperate fens rank among the most endangered habitats in temperate Europe. In agricultural landscapes, they are highly vulnerable to eutrophication and desiccation, which accelerate biodiversity loss and shifts in the carbon balance due to peat mineralization. The initial signs of habitat change are commonly manifested by shifts in vegetation structure and dominance, accompanied by increasing productivity, which precede major qualitative changes in species composition. The in‐time monitoring of vegetation productivity and site wetness at large areas is essential for guiding conservation management strategies for fens to slow down or reverse undesired changes. Here, we evaluated the ability of satellite (Sentinel‐2) and high‐resolution aerial imagery to detect early, structure‐ and productivity‐related signals of fen deterioration. We compared multispectral and optical imagery with ground‐based data, including both direct measurements and indicators derived from the species composition of the vegetation plots. At the landscape scale where both the acidic poor fens and the base‐rich fens occurred, MSAVI and NGRDI indices performed best, indicating primarily the vascular plant cover, species richness and representation of nutrient‐demanding species. At the within‐site scale, where the differences among plots were largely driven by habitat deterioration, NDVI, NDWI and RENDVI well captured differences in vascular plant productivity estimates and moss biomass measurements. Our results indicate that remote sensing is applicable for the identification of individual fen habitats and their nutrient status at the landscape scale and is even effective in detecting incipient habitat deterioration associated with increasing productivity. We demonstrate that remote sensing also performs well for small, island‐like fen patches. Its wider integration into the mire research would improve monitoring and enhance the amount of available ecological data.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"1143 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147447951","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}
Martynas Bielinis, Michelle LaRue, Benjamin M. Kraemer, Catalina Munteanu
Satellite imagery extending as far back as the 1960's has the potential to inform Antarctic conservation by providing insights into habitat and population dynamics that are otherwise difficult to observe. Here we demonstrate the detection of Emperor Penguin ( Aptenodytes forsteri ) guano stains on sea ice using Keyhole, Landsat, and Sentinel‐2 imagery from the 1960s to 2024. For 18 of the 66 known emperor penguin colonies, we confirmed colony presence in images that predate their earliest published records. Beyond presence detection, we examined the colony with the densest available imagery (Cape Washington) to quantify change in guano area over time. The guano area detected with satellites was correlated with observed chick counts from ground surveys (Spearman's ρ = 0.59, P ‐value = 0.017), and showed no strong evidence for a long‐term trend ( P = 0.61). Taken together, our results indicate substantial interannual and intra‐annual variability in colony size, but no evidence for a consistent long‐term directional trend and highlight that the use of remote sensing imagery across the Antarctic could inform conservation efforts and benefit the ongoing historical studies of penguin colony dynamics.
{"title":"Historical remote sensing highlights long‐term persistence of Emperor Penguin ( Aptenodytes forsteri ) colonies","authors":"Martynas Bielinis, Michelle LaRue, Benjamin M. Kraemer, Catalina Munteanu","doi":"10.1002/rse2.70064","DOIUrl":"https://doi.org/10.1002/rse2.70064","url":null,"abstract":"Satellite imagery extending as far back as the 1960's has the potential to inform Antarctic conservation by providing insights into habitat and population dynamics that are otherwise difficult to observe. Here we demonstrate the detection of Emperor Penguin ( <jats:italic>Aptenodytes forsteri</jats:italic> ) guano stains on sea ice using Keyhole, Landsat, and Sentinel‐2 imagery from the 1960s to 2024. For 18 of the 66 known emperor penguin colonies, we confirmed colony presence in images that predate their earliest published records. Beyond presence detection, we examined the colony with the densest available imagery (Cape Washington) to quantify change in guano area over time. The guano area detected with satellites was correlated with observed chick counts from ground surveys (Spearman's ρ = 0.59, <jats:italic>P</jats:italic> ‐value = 0.017), and showed no strong evidence for a long‐term trend ( <jats:italic>P</jats:italic> = 0.61). Taken together, our results indicate substantial interannual and intra‐annual variability in colony size, but no evidence for a consistent long‐term directional trend and highlight that the use of remote sensing imagery across the Antarctic could inform conservation efforts and benefit the ongoing historical studies of penguin colony dynamics.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"12 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147447954","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}
Claire McGinnity, Connor C.G. Bamford, Nathan Fenney, Andrew Fleming, Jaume Forcada, Michael S. Tift, Luis A. Hückstädt, Daniel P. Costa, Peter T. Fretwell
Over the past 25 years, the Western Antarctic Peninsula (WAP) has experienced dramatic shifts in sea ice extent. This change has coincided with rapid alterations in ice‐dependent ecosystems, including those supporting crabeater seals—the most abundant Antarctic seal and one of the largest mammalian consumers of krill. Despite their ecological importance, population estimates for ice seals remain scarce due to the difficulty of surveying large‐scale, remote, ice‐covered habitats. In 2023, during an abnormally low sea ice year, we conducted aerial surveys over Crystal Sound and Marguerite Bay during the end of the breeding season, flying over 1000 km of transects. Seals were extremely sparse in the resulting imagery—occupying less than 1% of the surveyed area. This posed a significant challenge for both manual annotation and automated detection. Here, we present a semi‐automated, rule‐based image analysis pipeline to substantially reduce human annotation time. Our method leverages hierarchical clustering with just two tuneable parameters, avoiding the computational burden and opacity of deep learning models. Using this method, we identified 758 seals within an ~350 km 2 survey subset, achieving a test recall of 79% ± 9.1%. In the absence of concurrent tagging data to estimate haul‐out corrections, we refrain from extrapolating to a population estimate. However, the low observed densities highlight the urgent need for continued monitoring. Our improved data processing pipeline is a key step in facilitating the large‐scale analysis required to inform conservation strategies for this key species.
在过去的25年里，南极半岛西部（WAP）经历了海冰范围的巨大变化。这种变化与依赖冰的生态系统的快速变化相吻合，包括那些支持食蟹海豹的生态系统——数量最多的南极海豹和磷虾最大的哺乳动物之一。尽管它们具有重要的生态意义，但由于难以对大范围、偏远、冰雪覆盖的栖息地进行调查，对冰海豹的数量估计仍然很少。2023年，在海冰异常稀少的年份，我们在繁殖季节结束时对水晶湾和玛格丽特湾进行了空中调查，飞越了1000多公里的样带。在最终的图像中，海豹极其稀少，只占调查面积的不到1%。这对手动注释和自动检测都提出了重大挑战。在这里，我们提出了一个半自动化的，基于规则的图像分析管道，以大大减少人工注释时间。我们的方法利用只有两个可调参数的分层聚类，避免了深度学习模型的计算负担和不透明性。使用该方法，我们在约350 km 2的调查子集内识别了758个密封件，测试召回率为79%±9.1%。在没有并发标记数据来估计拖出更正的情况下，我们避免外推到总体估计。然而，观测到的低密度突出表明迫切需要继续进行监测。我们改进的数据处理管道是促进大规模分析所需的关键步骤，为这一关键物种的保护策略提供信息。
{"title":"Semi‐automated seal detection on the Western Antarctic Peninsula: an unsupervised machine learning approach for detecting ice seals in aerial survey data","authors":"Claire McGinnity, Connor C.G. Bamford, Nathan Fenney, Andrew Fleming, Jaume Forcada, Michael S. Tift, Luis A. Hückstädt, Daniel P. Costa, Peter T. Fretwell","doi":"10.1002/rse2.70060","DOIUrl":"https://doi.org/10.1002/rse2.70060","url":null,"abstract":"Over the past 25 years, the Western Antarctic Peninsula (WAP) has experienced dramatic shifts in sea ice extent. This change has coincided with rapid alterations in ice‐dependent ecosystems, including those supporting crabeater seals—the most abundant Antarctic seal and one of the largest mammalian consumers of krill. Despite their ecological importance, population estimates for ice seals remain scarce due to the difficulty of surveying large‐scale, remote, ice‐covered habitats. In 2023, during an abnormally low sea ice year, we conducted aerial surveys over Crystal Sound and Marguerite Bay during the end of the breeding season, flying over 1000 km of transects. Seals were extremely sparse in the resulting imagery—occupying less than 1% of the surveyed area. This posed a significant challenge for both manual annotation and automated detection. Here, we present a semi‐automated, rule‐based image analysis pipeline to substantially reduce human annotation time. Our method leverages hierarchical clustering with just two tuneable parameters, avoiding the computational burden and opacity of deep learning models. Using this method, we identified 758 seals within an ~350 km <jats:sup>2</jats:sup> survey subset, achieving a test recall of 79% ± 9.1%. In the absence of concurrent tagging data to estimate haul‐out corrections, we refrain from extrapolating to a population estimate. However, the low observed densities highlight the urgent need for continued monitoring. Our improved data processing pipeline is a key step in facilitating the large‐scale analysis required to inform conservation strategies for this key species.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"253 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147373924","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}
Marina D. A. Scarpelli, Stewart Macdonald, Maryam Golchin, Simon Linke, Jens G. Froese
Invasive alien species are a major threat to biodiversity, with significant impacts on threatened species and priority sites. Monitoring is essential to inform appropriate management strategies, and autonomous sensors are increasingly used to address data collection at large spatio‐temporal scales. Feral pigs ( Sus scrofa ) are a major threat to native fauna in Australia. Here, the utility of passive acoustic monitoring for detecting feral pigs and its complementarity to camera trap detection was tested. A custom‐built deep‐learning BirdNET recogniser was used to automatically scan sound for pig presence; image data was manually scanned. Detection probabilities and effects of covariates were compared for detections of each method, separately and combined, using multi‐season occupancy models. There was little spatio‐temporal overlap between image and sound detections. Modelled detection probability was the highest when sound and image detections were combined, followed by sound and, lastly, images. Seasonality affected detectability: camera traps were most successful in the Late Wet, when sound detection was poor. Sound detection was more successful in all other seasons, with the highest detection probability in the Late Dry. The intrinsic variation across survey methods along with the effects of environmental factors in species behaviour can be accounted for by combining methods, improving overall detections and providing complementary information on the same species. Autonomous sensors can provide comprehensive data to inform land management decisions, including population control and impact mitigation of invasive species. However, the utility of different sensors is context‐dependent. Combining multiple technologies can harness the strengths of each and mitigate against weaknesses. Increasing technology accessibility and decreasing costs is key to facilitate uptake by land managers.
{"title":"Monitoring feral pigs ( Sus scrofa ): Complementarity between autonomous sensing methods increases detection probability","authors":"Marina D. A. Scarpelli, Stewart Macdonald, Maryam Golchin, Simon Linke, Jens G. Froese","doi":"10.1002/rse2.70062","DOIUrl":"https://doi.org/10.1002/rse2.70062","url":null,"abstract":"Invasive alien species are a major threat to biodiversity, with significant impacts on threatened species and priority sites. Monitoring is essential to inform appropriate management strategies, and autonomous sensors are increasingly used to address data collection at large spatio‐temporal scales. Feral pigs ( <jats:italic>Sus scrofa</jats:italic> ) are a major threat to native fauna in Australia. Here, the utility of passive acoustic monitoring for detecting feral pigs and its complementarity to camera trap detection was tested. A custom‐built deep‐learning BirdNET recogniser was used to automatically scan sound for pig presence; image data was manually scanned. Detection probabilities and effects of covariates were compared for detections of each method, separately and combined, using multi‐season occupancy models. There was little spatio‐temporal overlap between image and sound detections. Modelled detection probability was the highest when sound and image detections were combined, followed by sound and, lastly, images. Seasonality affected detectability: camera traps were most successful in the Late Wet, when sound detection was poor. Sound detection was more successful in all other seasons, with the highest detection probability in the Late Dry. The intrinsic variation across survey methods along with the effects of environmental factors in species behaviour can be accounted for by combining methods, improving overall detections and providing complementary information on the same species. Autonomous sensors can provide comprehensive data to inform land management decisions, including population control and impact mitigation of invasive species. However, the utility of different sensors is context‐dependent. Combining multiple technologies can harness the strengths of each and mitigate against weaknesses. Increasing technology accessibility and decreasing costs is key to facilitate uptake by land managers.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"51 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-03-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147373925","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}
Jorge Sereno‐Cadierno, Tim R. Hofmeester, Marcus Becker, Alice Bernard, Lizette Moolman, Hervé Fritz, Pelayo Acevedo
Camera traps (CTs) are widely used in wildlife monitoring, but sampling design choices can introduce significant biases in trapping rates (TR) that, depending on the evaluated parameter, can be propagated to dependent estimates (e.g., density). This study evaluates the effect of camera height placement on TR across five experiments encompassing 172 paired sampling points (i.e., with a low and a high camera per point) in four biomes across Europe, North America and Africa. We analysed data of 49 vertebrate species, ranging from small mammals and birds to large ungulates and carnivores (0.013–461 kg), using generalised linear and multinomial models to assess how TR varies with body mass and camera height. Our results show that lower camera placements significantly increase TR for small (0–10 kg) and medium‐sized species (11–50 kg), while the opposite is found in larger animals. Simultaneous detections by both high‐ and low‐placed cameras increased with body mass, but small species were often missed by high cameras alone. Camera height introduces systematic biases in TR, affecting its comparability across time and space. For multispecies monitoring, lower cameras (30–50 cm above ground) offer better overall performance, though higher placements may be more suitable for large‐bodied focal species. We recommend consistent, standardised height measurements in long‐term monitoring to ensure reliable TR‐based inferences and validate the recommendation of using target species' shoulder height when monitoring single species. This study provides the most comprehensive cross‐continental evaluation of camera height effects to date and offers empirically grounded guidance for optimising sampling design in wildlife monitoring.
{"title":"Knee height is often right: evaluating device height effects on camera trapping rate","authors":"Jorge Sereno‐Cadierno, Tim R. Hofmeester, Marcus Becker, Alice Bernard, Lizette Moolman, Hervé Fritz, Pelayo Acevedo","doi":"10.1002/rse2.70053","DOIUrl":"https://doi.org/10.1002/rse2.70053","url":null,"abstract":"Camera traps (CTs) are widely used in wildlife monitoring, but sampling design choices can introduce significant biases in trapping rates (TR) that, depending on the evaluated parameter, can be propagated to dependent estimates (e.g., density). This study evaluates the effect of camera height placement on TR across five experiments encompassing 172 paired sampling points (i.e., with a low and a high camera per point) in four biomes across Europe, North America and Africa. We analysed data of 49 vertebrate species, ranging from small mammals and birds to large ungulates and carnivores (0.013–461 kg), using generalised linear and multinomial models to assess how TR varies with body mass and camera height. Our results show that lower camera placements significantly increase TR for small (0–10 kg) and medium‐sized species (11–50 kg), while the opposite is found in larger animals. Simultaneous detections by both high‐ and low‐placed cameras increased with body mass, but small species were often missed by high cameras alone. Camera height introduces systematic biases in TR, affecting its comparability across time and space. For multispecies monitoring, lower cameras (30–50 cm above ground) offer better overall performance, though higher placements may be more suitable for large‐bodied focal species. We recommend consistent, standardised height measurements in long‐term monitoring to ensure reliable TR‐based inferences and validate the recommendation of using target species' shoulder height when monitoring single species. This study provides the most comprehensive cross‐continental evaluation of camera height effects to date and offers empirically grounded guidance for optimising sampling design in wildlife monitoring.","PeriodicalId":21132,"journal":{"name":"Remote Sensing in Ecology and Conservation","volume":"41 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146215636","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：