Automatic image‐based identification and biomass estimation of invertebrates

Low-cost, high-volume imaging for entomological digitization

Large-scale digitization of natural history collections requires automation of image acquisition and processing. Reflecting this fact, various approaches, some highly sophisticated, have been developed to support imaging of museum specimens. However, most of these systems are complex and expensive, restricting their deployment. Here we describe a simple, inexpensive technique for imaging arthropods larger than 5 mm. By mounting a digital SLR camera on a CNC (computer numerical control) motor-drive rig, we created a system that captures high-resolution z-axis stacked images (6960 × 4640 pixels) of 95 specimens in 30 minutes. This system can be assembled inexpensively ($1000 USD without a camera) and it is easy to set-up and maintain. By coupling low cost with high production capacity, it represents a solution for digitizing any natural history collection.

Towards a standardized framework for AI-assisted, image-based monitoring of nocturnal insects

Automated sensors have potential to standardize and expand the monitoring of insects across the globe. As one of the most scalable and fastest developing sensor technologies, we describe a framework for automated, image-based monitoring of nocturnal insects-from sensor development and field deployment to workflows for data processing and publishing. Sensors comprise a light to attract insects, a camera for collecting images and a computer for scheduling, data storage and processing. Metadata is important to describe sampling schedules that balance the capture of relevant ecological information against power and data storage limitations. Large data volumes of images from automated systems necessitate scalable and effective data processing. We describe computer vision approaches for the detection, tracking and classification of insects, including models built from existing aggregations of labelled insect images. Data from automated camera systems necessitate approaches that account for inherent biases. We advocate models that explicitly correct for bias in species occurrence or abundance estimates resulting from the imperfect detection of species or individuals present during sampling occasions. We propose ten priorities towards a step-change in automated monitoring of nocturnal insects, a vital task in the face of rapid biodiversity loss from global threats. This article is part of the theme issue 'Towards a toolkit for global insect biodiversity monitoring'.

Towards holistic insect monitoring: species discovery, description, identification and traits for all insects

Holistic insect monitoring needs scalable techniques to overcome taxon biases, determine species abundances, and gather functional traits for all species. This requires that we address taxonomic impediments and the paucity of data on abundance, biomass and functional traits. We here outline how these data deficiencies could be addressed at scale. The workflow starts with large-scale barcoding (megabarcoding) of all specimens from mass samples obtained at biomonitoring sites. The barcodes are then used to group the specimens into molecular operational taxonomic units that are subsequently tested/validated as species with a second data source (e.g. morphology). New species are described using barcodes, images and short diagnoses, and abundance data are collected for both new and described species. The specimen images used for species discovery then become the raw material for training artificial intelligence identification algorithms and collecting trait data such as body size, biomass and feeding modes. Additional trait data can be obtained from vouchers by using genomic tools developed by molecular ecologists. Applying this pipeline to a few samples per site will lead to greatly improved insect monitoring regardless of whether the species composition of a sample is determined with images, metabarcoding or megabarcoding. This article is part of the theme issue 'Towards a toolkit for global insect biodiversity monitoring'.

Arthropod diversity in the alpine tundra using metabarcoding: Spatial and temporal differences in alpha- and beta-diversity

All ecosystems face ecological challenges in this century. Therefore, it is becoming increasingly important to understand the ecology and degree of local adaptation of functionally important Arctic-alpine biomes by looking at the most diverse taxon of metazoans: the Arthropoda. This is the first study to utilize metabarcoding in the Alpine tundra, providing insights into the effects of micro-environmental parameters on alpha- and beta-diversity of arthropods in such unique environments. To characterize arthropod diversity, pitfall traps were set at three middle-alpine sampling sites in the Scandinavian mountain range in Norway during the snow-free season in 2015. A metabarcoding approach was then used to determine the small-scale biodiversity patterns of arthropods in the Alpine tundra. All DNA was extracted directly from the preservative EtOH from 27 pitfall traps. In order to identify the controlling environmental conditions, all sampling locations were equipped with automatic data loggers for permanent measurement of the microenvironmental conditions. The variables measured were: air temperature [°C] at 15 cm height, soil temperature [°C] at 15 cm depth, and soil moisture [vol.%] at 15 cm depth. A total of 233 Arthropoda OTUs were identified. The number of unique OTUs found per sampling location (ridge, south-facing slope, and depression) was generally higher than the OTUs shared between the sampling locations, demonstrating that niche features greatly impact arthropod community structure. Our findings emphasize the fine-scale heterogeneity of arctic-alpine ecosystems and provide evidence for trait-based and niche-driven adaptation. The spatial and temporal differences in arthropod diversity were best explained by soil moisture and soil temperature at the respective locations. Furthermore, our results show that arthropod diversity is underestimated in alpine-tundra ecosystems using classical approaches and highlight the importance of integrating long-term functional environmental data and modern taxonomic techniques into biodiversity research to expand our ecological understanding of fine- and meso-scale biogeographical patterns.

Automated software for counting and measuring Hyalella genus using artificial intelligence

Amphipods belonging to the Hyalella genus are macroinvertebrates that inhabit aquatic environments. They are of particular interest in areas such as limnology and ecotoxicology, where data on the number of Hyalella individuals and their allometric measurements are used to assess the environmental dynamics of aquatic ecosystems. In this study, we introduce HyACS, a software tool that uses a model developed with the YOLOv3's architecture to detect individuals, and digital image processing techniques to extract morphological metrics of the Hyalella genus. The software detects body metrics of length, arc length, maximum width, eccentricity, perimeter, and area of Hyalella individuals, using basic imaging capture equipment. The performance metrics indicate that the model developed can achieve high prediction levels, with an accuracy above 90% for the correct identification of individuals. It can perform up to four times faster than traditional visual counting methods and provide precise morphological measurements of Hyalella individuals, which may improve further studies of the species populations and enhance their use as bioindicators of water quality.

It Is Not Only Data-Freshwater Invertebrates Misused in Biological Monitoring

The article presents and discusses the issues of the use of free-living invertebrates to assess the ecological status of freshwater environments with different methods of biological monitoring. Invertebrates are excluded from ethical consideration in the procedures of environmental protection, which results in the killing of many more individuals during sampling than necessary. Biomonitoring is used as a routine method for environmental protection that results in the cruel death of even millions of aquatic animals annually. In many cases, the mortality of animals used in such types of activities has been shown as excessive, e.g., because the vast majority die due to unnecessary subsampling procedures. Improperly planned and conducted procedures which result in excessive mortality have or may have a negative impact on the environment and biodiversity. Their existence as sensitive beings is reduced to an information function; they become only data useful for biomonitoring purposes. The main problem when trying to determine the mortality of invertebrates due to biomonitoring activities and its impact on natural populations seems to be the lack of access to raw data presenting how many animals were killed during sampling.

Getting the bugs out of AI: Advancing ecological research on arthropods through computer vision

Deep learning for computer vision has shown promising results in the field of entomology, however, there still remains untapped potential. Deep learning performance is enabled primarily by large quantities of annotated data which, outside of rare circumstances, are limited in ecological studies. Currently, to utilize deep learning systems, ecologists undergo extensive data collection efforts, or limit their problem to niche tasks. These solutions do not scale to region agnostic models. However, there are solutions that employ data augmentation, simulators, generative models, and self-supervised learning that can supplement limited labelled data. Here, we highlight the success of deep learning for computer vision within entomology, discuss data collection efforts, provide methodologies for optimizing learning from limited annotations, and conclude with practical guidelines for how to achieve a foundation model for entomology capable of accessible automated ecological monitoring on a global scale.

Using Machine Learning to make nanomaterials sustainable

Sustainable development is a key challenge for contemporary human societies; failure to achieve sustainability could threaten human survival. In this review article, we illustrate how Machine Learning (ML) could support more sustainable development, covering the basics of data gathering through each step of the Environmental Risk Assessment (ERA). The literature provides several examples showing how ML can be employed in most steps of a typical ERA.A key observation is that there are currently no clear guidance for using such autonomous technologies in ERAs or which standards/checks are required. Steering thus seems to be the most important task for supporting the use of ML in the ERA of nano- and smart-materials. Resources should be devoted to developing a strategy for implementing ML in ERA with a strong emphasis on data foundations, methodologies, and the related sensitivities/uncertainties. We should recognise historical errors and biases (e.g., in data) to avoid embedding them during ML programming.

Accurate image-based identification of macroinvertebrate specimens using deep learning-How much training data is needed?

Image-based methods for species identification offer cost-efficient solutions for biomonitoring. This is particularly relevant for invertebrate studies, where bulk samples often represent insurmountable workloads for sorting, identifying, and counting individual specimens. On the other hand, image-based classification using deep learning tools have strict requirements for the amount of training data, which is often a limiting factor. Here, we examine how classification accuracy increases with the amount of training data using the BIODISCOVER imaging system constructed for image-based classification and biomass estimation of invertebrate specimens. We use a balanced dataset of 60 specimens of each of 16 taxa of freshwater macroinvertebrates to systematically quantify how classification performance of a convolutional neural network (CNN) increases for individual taxa and the overall community as the number of specimens used for training is increased. We show a striking 99.2% classification accuracy when the CNN (EfficientNet-B6) is trained on 50 specimens of each taxon, and also how the lower classification accuracy of models trained on less data is particularly evident for morphologically similar species placed within the same taxonomic order. Even with as little as 15 specimens used for training, classification accuracy reached 97%. Our results add to a recent body of literature showing the huge potential of image-based methods and deep learning for specimen-based research, and furthermore offers a perspective to future automatized approaches for deriving ecological data from bulk arthropod samples.

Toward global integration of biodiversity big data: a harmonized metabarcode data generation module for terrestrial arthropods

Metazoan metabarcoding is emerging as an essential strategy for inventorying biodiversity, with diverse projects currently generating massive quantities of community-level data. The potential for integrating across such data sets offers new opportunities to better understand biodiversity and how it might respond to global change. However, large-scale syntheses may be compromised if metabarcoding workflows differ from each other. There are ongoing efforts to improve standardization for the reporting of inventory data. However, harmonization at the stage of generating metabarcode data has yet to be addressed. A modular framework for harmonized data generation offers a pathway to navigate the complex structure of terrestrial metazoan biodiversity. Here, through our collective expertise as practitioners, method developers, and researchers leading metabarcoding initiatives to inventory terrestrial biodiversity, we seek to initiate a harmonized framework for metabarcode data generation, with a terrestrial arthropod module. We develop an initial set of submodules covering the 5 main steps of metabarcode data generation: (i) sample acquisition; (ii) sample processing; (iii) DNA extraction; (iv) polymerase chain reaction amplification, library preparation, and sequencing; and (v) DNA sequence and metadata deposition, providing a backbone for a terrestrial arthropod module. To achieve this, we (i) identified key points for harmonization, (ii) reviewed the current state of the art, and (iii) distilled existing knowledge within submodules, thus promoting best practice by providing guidelines and recommendations to reduce the universe of methodological options. We advocate the adoption and further development of the terrestrial arthropod module. We further encourage the development of modules for other biodiversity fractions as an essential step toward large-scale biodiversity synthesis through harmonization.

Emerging technologies revolutionise insect ecology and monitoring

Insects are the most diverse group of animals on Earth, but their small size and high diversity have always made them challenging to study. Recent technological advances have the potential to revolutionise insect ecology and monitoring. We describe the state of the art of four technologies (computer vision, acoustic monitoring, radar, and molecular methods), and assess their advantages, current limitations, and future potential. We discuss how these technologies can adhere to modern standards of data curation and transparency, their implications for citizen science, and their potential for integration among different monitoring programmes and technologies. We argue that they provide unprecedented possibilities for insect ecology and monitoring, but it will be important to foster international standards via collaboration.

Multifunctionality of belowground food webs: resource, size and spatial energy channels

The belowground compartment of terrestrial ecosystems drives nutrient cycling, the decomposition and stabilisation of organic matter, and supports aboveground life. Belowground consumers create complex food webs that regulate functioning, ensure stability and support biodiversity both below and above ground. However, existing soil food-web reconstructions do not match recently accumulated empirical evidence and there is no comprehensive reproducible approach that accounts for the complex resource, size and spatial structure of food webs in soil. Here I build on generic food-web organisation principles and use multifunctional classification of soil protists, invertebrates and vertebrates, to reconstruct a 'multichannel' food web across size classes of soil-associated consumers. I infer weighted trophic interactions among trophic guilds using feeding preferences and prey protection traits (evolutionarily inherited traits), size and spatial distributions (niche overlaps), and biomass-dependent feeding. I then use food-web reconstruction, together with assimilation efficiencies, to calculate energy fluxes assuming a steady-state energetic system. Based on energy fluxes, I propose a number of indicators, related to stability, biodiversity and multiple ecosystem-level functions such as herbivory, top-down control, translocation and transformation of organic matter. I illustrate this approach with an empirical example, comparing it with traditional resource-focused soil food-web reconstruction. The multichannel reconstruction can be used to assess 'trophic multifunctionality' (analogous to ecosystem multifunctionality), i.e. simultaneous support of multiple trophic functions by the food web, and compare it across communities and ecosystems spanning beyond the soil. With further empirical validation of the proposed functional indicators, this multichannel reconstruction approach could provide an effective tool for understanding animal diversity-ecosystem functioning relationships in soil. This tool hopefully will inspire more researchers to describe soil communities and belowground-aboveground interactions comprehensively. Such studies will provide informative indicators for including consumers as active agents in biogeochemical models, not only locally but also on regional and global scales.

Automating insect monitoring using unsupervised near-infrared sensors

Insect monitoring is critical to improve our understanding and ability to preserve and restore biodiversity, sustainably produce crops, and reduce vectors of human and livestock disease. Conventional monitoring methods of trapping and identification are time consuming and thus expensive. Automation would significantly improve the state of the art. Here, we present a network of distributed wireless sensors that moves the field towards automation by recording backscattered near-infrared modulation signatures from insects. The instrument is a compact sensor based on dual-wavelength infrared light emitting diodes and is capable of unsupervised, autonomous long-term insect monitoring over weather and seasons. The sensor records the backscattered light at kHz pace from each insect transiting the measurement volume. Insect observations are automatically extracted and transmitted with environmental metadata over cellular connection to a cloud-based database. The recorded features include wing beat harmonics, melanisation and flight direction. To validate the sensor’s capabilities, we tested the correlation between daily insect counts from an oil seed rape field measured with six yellow water traps and six sensors during a 4-week period. A comparison of the methods found a Spearman’s rank correlation coefficient of 0.61 and a p-value = 0.0065, with the sensors recording approximately 19 times more insect observations and demonstrating a larger temporal dynamic than conventional yellow water trap monitoring.

Seasonal variations of serotonin in the visual system of an ant revealed by immunofluorescence and a machine learning approach

Hibernation, as an adaptation to seasonal environmental changes in temperate or boreal regions, has profound effects on mammalian brains. Social insects of temperate regions hibernate as well, but despite abundant knowledge on structural and functional plasticity in insect brains, the question of how seasonal activity variations affect insect central nervous systems has not yet been thoroughly addressed. Here, we studied potential variations of serotonin-immunoreactivity in visual information processing centres in the brain of the long-lived ant species Lasius niger. Quantitative immunofluorescence analysis revealed stronger serotonergic signals in the lamina and medulla of the optic lobes of wild or active laboratory workers than in hibernating animals. Instead of statistical inference by testing, differentiability of seasonal serotonin-immunoreactivity was confirmed by a machine learning analysis using convolutional artificial neuronal networks (ANNs) with the digital immunofluorescence images as input information. Machine learning models revealed additional differences in the third visual processing centre, the lobula. We further investigated these results by gradient-weighted class activation mapping. We conclude that seasonal activity variations are represented in the ant brain, and that machine learning by ANNs can contribute to the discovery of such variations.

DiversityScanner: Robotic handling of small invertebrates with machine learning methods

Invertebrate biodiversity remains poorly understood although it comprises much of the terrestrial animal biomass, most species and supplies many ecosystem services. The main obstacle is specimen-rich samples obtained with quantitative sampling techniques (e.g., Malaise trapping). Traditional sorting requires manual handling, while molecular techniques based on metabarcoding lose the association between individual specimens and sequences and thus struggle with obtaining precise abundance information. Here we present a sorting robot that prepares specimens from bulk samples for barcoding. It detects, images and measures individual specimens from a sample and then moves them into the wells of a 96-well microplate. We show that the images can be used to train convolutional neural networks (CNNs) that are capable of assigning the specimens to 14 insect taxa (usually families) that are particularly common in Malaise trap samples. The average assignment precision for all taxa is 91.4% (75%-100%). This ability of the robot to identify common taxa then allows for taxon-specific subsampling, because the robot can be instructed to only pick a prespecified number of specimens for abundant taxa. To obtain biomass information, the images are also used to measure specimen length and estimate body volume. We outline how the DiversityScanner can be a key component for tackling and monitoring invertebrate diversity by combining molecular and morphological tools: the images generated by the robot become training images for machine learning once they are labelled with taxonomic information from DNA barcodes. We suggest that a combination of automation, machine learning and DNA barcoding has the potential to tackle invertebrate diversity at an unprecedented scale.

For flux's sake: General considerations for energy-flux calculations in ecological communities

Global change alters ecological communities with consequences for ecosystem processes. Such processes and functions are a central aspect of ecological research and vital to understanding and mitigating the consequences of global change, but also those of other drivers of change in organism communities. In this context, the concept of energy flux through trophic networks integrates food-web theory and biodiversity-ecosystem functioning theory and connects biodiversity to multitrophic ecosystem functioning. As such, the energy-flux approach is a strikingly effective tool to answer central questions in ecology and global-change research. This might seem straight forward, given that the theoretical background and software to efficiently calculate energy flux are readily available. However, the implementation of such calculations is not always straight forward, especially for those who are new to the topic and not familiar with concepts central to this line of research, such as food-web theory or metabolic theory. To facilitate wider use of energy flux in ecological research, we thus provide a guide to adopting energy-flux calculations for people new to the method, struggling with its implementation, or simply looking for background reading, important resources, and standard solutions to the problems everyone faces when starting to quantify energy fluxes for their community data. First, we introduce energy flux and its use in community and ecosystem ecology. Then, we provide a comprehensive explanation of the single steps towards calculating energy flux for community data. Finally, we discuss remaining challenges and exciting research frontiers for future energy-flux research.

Identification of Species by Combining Molecular and Morphological Data Using Convolutional Neural Networks

Integrative taxonomy is central to modern taxonomy and systematic biology, including behaviour, niche preference, distribution, morphological analysis and DNA barcoding. However, decades of use demonstrate that these methods can face challenges when used in isolation, for instance, potential misidentifications due to phenotypic plasticity for morphological methods, and incorrect identifications because of introgression, incomplete lineage sorting and horizontal gene transfer for DNA barcoding. Although researchers have advocated the use of integrative taxonomy, few detailed algorithms have been proposed. Here, we develop a convolutional neural network method (morphology-molecule network (MMNet)) that integrates morphological and molecular data for species identification. The newly proposed method (MMNet) worked better than four currently-available alternative methods when tested with 10 independent datasets representing varying genetic diversity from different taxa. High accuracies were achieved for all groups, including beetles (98.1% of 123 species), butterflies (98.8% of 24 species), fishes (96.3% of 214 species) and moths (96.4% of 150 total species). Further, MMNet demonstrated a high degree of accuracy (>98%) in four datasets including closely related species from the same genus. The average accuracy of two modest sub-genomic (single nucleotide polymorphism) datasets, comprising eight putative subspecies respectively, is 90%. Additional tests show that the success rate of species identification under this method most strongly depends on the amount of training data, and is robust to sequence length and image size. Analyses on the contribution of different data types (image versus gene) indicate that both morphological and genetic data are important to the model, and that genetic data contribute slightly more. The approaches developed here serve as a foundation for the future integration of multi-modal information for integrative taxonomy, such as image, audio, video, 3D scanning and biosensor data, to characterize organisms more comprehensively as a basis for improved investigation, monitoring and conservation of biodiversity.

A re-analysis of the data in Sharkey et al.'s (2021) minimalist revision reveals that BINs do not deserve names, but BOLD Systems needs a stronger commitment to open science

Halting biodiversity decline is one of the most critical challenges for humanity, but monitoring biodiversity is hampered by taxonomic impediments. One impediment is the large number of undescribed species (here called "dark taxon impediment") whereas another is caused by the large number of superficial species descriptions, that can only be resolved by consulting type specimens ("superficial description impediment"). Recently, Sharkey et al. (2021) proposed to address the dark taxon impediment for Costa Rican braconid wasps by describing 403 species based on COI barcode clusters ("BINs") computed by BOLD Systems. More than 99% of the BINs (387 of 390) were converted into species by assigning binominal names (e.g. BIN "BOLD:ACM9419" becomes Bracon federicomatarritai) and adding a minimal diagnosis (consisting only of a consensus barcode for most species). We here show that many of Sharkey et al.'s species are unstable when the underlying data are analyzed using different species delimitation algorithms. Add the insufficiently informative diagnoses, and many of these species will become the next "superficial description impediment" for braconid taxonomy because they will have to be tested and redescribed after obtaining sufficient evidence for confidently delimiting species. We furthermore show that Sharkey et al.'s approach of using consensus barcodes as diagnoses is not functional because it cannot be applied consistently. Lastly, we reiterate that COI alone is not suitable for delimiting and describing species, and voice concerns over Sharkey et al.'s uncritical use of BINs because they are calculated by a proprietary algorithm (RESL) that uses a mixture of public and private data. We urge authors, reviewers and editors to maintain high standards in taxonomy by only publishing new species that are rigorously delimited with open-access tools and supported by publicly available evidence.

Do amateurs and citizen science fill the gaps left by scientists?

The diversity of insects is tremendous and so is the effort needed to assess it in order to better understand insect ecology as well as their role for the functioning of ecosystems. While the interest of academics and naturalists for these species has always existed, it is only recently that such interest started to reach society more generally. From insect taxonomy and distribution to the collection of large range and long scale monitoring data, the involvement of non-academics in research outputs is growing. Is this a sign of scientists not being able to meet expectations or of science getting more and more entrenched in society? We argue for the latter, highlighting the opportunities that such involvement of amateurs in insect science represent for insect conservation.

Deep learning and computer vision will transform entomology

Most animal species on Earth are insects, and recent reports suggest that their abundance is in drastic decline. Although these reports come from a wide range of insect taxa and regions, the evidence to assess the extent of the phenomenon is sparse. Insect populations are challenging to study, and most monitoring methods are labor intensive and inefficient. Advances in computer vision and deep learning provide potential new solutions to this global challenge. Cameras and other sensors can effectively, continuously, and noninvasively perform entomological observations throughout diurnal and seasonal cycles. The physical appearance of specimens can also be captured by automated imaging in the laboratory. When trained on these data, deep learning models can provide estimates of insect abundance, biomass, and diversity. Further, deep learning models can quantify variation in phenotypic traits, behavior, and interactions. Here, we connect recent developments in deep learning and computer vision to the urgent demand for more cost-efficient monitoring of insects and other invertebrates. We present examples of sensor-based monitoring of insects. We show how deep learning tools can be applied to exceptionally large datasets to derive ecological information and discuss the challenges that lie ahead for the implementation of such solutions in entomology. We identify four focal areas, which will facilitate this transformation: 1) validation of image-based taxonomic identification; 2) generation of sufficient training data; 3) development of public, curated reference databases; and 4) solutions to integrate deep learning and molecular tools.

Getting science priorities straight: how to increase the reliability of specimen identification?

'We advise the authors to find a native English speaker to proofread the manuscript'. This is a standard feedback journals give to non-native English speakers. Journals are justifiably concerned with grammar but do not show the same rigour about another step crucial to biological research: specimen identification. Surveying the author guidelines of 100 journals, we found that only 6% of them request explicitly citation of the literature used in specimen identification. Authors hamper readers from contesting specimen identification whenever vouchers, identification methods, and taxon concepts are not provided. However, unclear taxonomic procedures violate the basic scientific principle of reproducibility. The scientific community must continuously look for practical alternatives to improve taxonomic identification and taxonomic verification. We argue that voucher pictures are an accessible, cheap and time-effective alternative to mitigate (not abolish) bad taxonomy by exposing preventable misidentifications. Voucher pictures allow scientists to judge specimen identification actively, based on available data. The popularization of high-quality image devices, photo-identification technologies and computer vision algorithms yield accurate scientific photo-documentation, improving taxonomic procedures. Taxonomy is timeless, transversal and essential to most scientific disciplines in biological sciences. It is time to demand rigour in taxonomic identifications.

Assessing the potential for deep learning and computer vision to identify bumble bee species from images

Pollinators are undergoing a global decline. Although vital to pollinator conservation and ecological research, species-level identification is expensive, time consuming, and requires specialized taxonomic training. However, deep learning and computer vision are providing ways to open this methodological bottleneck through automated identification from images. Focusing on bumble bees, we compare four convolutional neural network classification models to evaluate prediction speed, accuracy, and the potential of this technology for automated bee identification. We gathered over 89,000 images of bumble bees, representing 36 species in North America, to train the ResNet, Wide ResNet, InceptionV3, and MnasNet models. Among these models, InceptionV3 presented a good balance of accuracy (91.6%) and average speed (3.34 ms). Species-level error rates were generally smaller for species represented by more training images. However, error rates also depended on the level of morphological variability among individuals within a species and similarity to other species. Continued development of this technology for automatic species identification and monitoring has the potential to be transformative for the fields of ecology and conservation. To this end, we present BeeMachine, a web application that allows anyone to use our classification model to identify bumble bees in their own images.

Nonlinear trends in abundance and diversity and complex responses to climate change in Arctic arthropods

Time series data on arthropod populations are critical for understanding the magnitude, direction, and drivers of change. However, most arthropod monitoring programs are short-lived and restricted in taxonomic resolution. Monitoring data from the Arctic are especially underrepresented, yet critical to uncovering and understanding some of the earliest biological responses to rapid environmental change. Clear imprints of climate on the behavior and life history of some Arctic arthropods have been demonstrated, but a synthesis of population-level abundance changes across taxa is lacking. We utilized 24 y of abundance data from Zackenberg in High-Arctic Greenland to assess trends in abundance and diversity and identify potential climatic drivers of abundance changes. Unlike findings from temperate systems, we found a nonlinear pattern, with total arthropod abundance gradually declining during 1996 to 2014, followed by a sharp increase. Family-level diversity showed the opposite pattern, suggesting increasing dominance of a small number of taxa. Total abundance masked more complicated trajectories of family-level abundance, which also frequently varied among habitats. Contrary to expectation in this extreme polar environment, winter and fall conditions and positive density-dependent feedbacks were more common determinants of arthropod dynamics than summer temperature. Together, these data highlight the complexity of characterizing climate change responses even in relatively simple Arctic food webs. Our results underscore the need for data reporting beyond overall trends in biomass or abundance and for including basic research on life history and ecology to achieve a more nuanced understanding of the sensitivity of Arctic and other arthropods to global changes.

DNA metabarcoding improves the detection of multiple stressor responses of stream invertebrates to increased salinity, fine sediment deposition and reduced flow velocity

Worldwide, multiple stressors affect stream ecosystems and frequently lead to complex and non-linear biological responses. These combined stressor effects on ecologically diverse and functionally important macroinvertebrate communities are often difficult to assess, in particular species-specific responses across many species and effects of different stressors and stressor levels in concert. A central limitation in many studies is the taxonomic resolution applied for specimen identification. DNA metabarcoding can resolve taxonomy and provide greater insights into multiple stressor effects. This was detailed by results of a recent multiple stressor mesocosm experiment, where only for the dipteran family Chironomidae 183 Operational Taxonomic Units (OTUs) could be distinguished. Numerous OTUs showed very different response patterns to multiple stressors. In this study, we applied DNA metabarcoding to assess multiple stressor effects on all non-chironomid invertebrates from the same experiment. In the experiment, we applied three stressors (increased salinity, deposited fine sediment, reduced flow velocity) in a full-factorial design. We compared stressor responses inferred through DNA metabarcoding of the mitochondrial COI gene to responses based on morphotaxonomic taxa lists. We identified 435 OTUs, of which 122 OTUs were assigned to EPT (Ephemeroptera, Plecoptera, Trichoptera) taxa. The most common 35 OTUs alone showed 15 different response patterns to the experimental manipulation, ranging from insensitivity to any applied stressor to sensitivity to single and multiple stressors. These response patterns even comprised differences within one family. The species-specific taxonomic resolution and the inferred response patterns to stressors highlights the potential of DNA metabarcoding in the context of multiple stressor research, even for well-known taxa such as EPT species.

Robust and simplified machine learning identification of pitfall trap-collected ground beetles at the continental scale

Insect populations are changing rapidly, and monitoring these changes is essential for understanding the causes and consequences of such shifts. However, large-scale insect identification projects are time-consuming and expensive when done solely by human identifiers. Machine learning offers a possible solution to help collect insect data quickly and efficiently.Here, we outline a methodology for training classification models to identify pitfall trap-collected insects from image data and then apply the method to identify ground beetles (Carabidae). All beetles were collected by the National Ecological Observatory Network (NEON), a continental scale ecological monitoring project with sites across the United States. We describe the procedures for image collection, image data extraction, data preparation, and model training, and compare the performance of five machine learning algorithms and two classification methods (hierarchical vs. single-level) identifying ground beetles from the species to subfamily level. All models were trained using pre-extracted feature vectors, not raw image data. Our methodology allows for data to be extracted from multiple individuals within the same image thus enhancing time efficiency, utilizes relatively simple models that allow for direct assessment of model performance, and can be performed on relatively small datasets.The best performing algorithm, linear discriminant analysis (LDA), reached an accuracy of 84.6% at the species level when naively identifying species, which was further increased to >95% when classifications were limited by known local species pools. Model performance was negatively correlated with taxonomic specificity, with the LDA model reaching an accuracy of ~99% at the subfamily level. When classifying carabid species not included in the training dataset at higher taxonomic levels species, the models performed significantly better than if classifications were made randomly. We also observed greater performance when classifications were made using the hierarchical classification method compared to the single-level classification method at higher taxonomic levels.The general methodology outlined here serves as a proof-of-concept for classifying pitfall trap-collected organisms using machine learning algorithms, and the image data extraction methodology may be used for nonmachine learning uses. We propose that integration of machine learning in large-scale identification pipelines will increase efficiency and lead to a greater flow of insect macroecological data, with the potential to be expanded for use with other noninsect taxa.

Arthropods and climate change - arctic challenges and opportunities

The harsh climate, limited human infrastructures, and basic autecological knowledge gaps represent substantial challenges for studying arthropods in the Arctic. At the same time, rapid climate change, low species diversity, and strong collaborative networks provide unique and underexploited Arctic opportunities for understanding species responses to environmental change and testing ecological theory. Here, I provide an overview of individual, population, and ecosystem level responses to climate change in Arctic arthropods. I focus on thermal performance, life history variation, population dynamics, community composition, diversity, and biotic interactions. The species-poor Arctic represents a unique opportunity for testing novel, automated arthropod monitoring methods. The Arctic can also potentially provide insights to further understand and mitigate the effects of climate change on arthropods worldwide.