Species identification by experts and non-experts: comparing images from field guides

Resurrecting Historical Observations to Characterize Species-Specific Nesting Traits of Bumblebees

AbstractIn recent years, ecological research has become increasingly synthetic, relying on revolutionary changes in data availability and accessibility. In spite of their strengths, these approaches may cause us to overlook natural history knowledge that is not part of the digitized English-language scientific record. Here, we combine historic and modern documents to quantify species-specific nesting habitat associations of bumblebees (Bombus spp. Latreille, 1802 Apidae). We compiled nest location data from 316 documents, of which 81 were non-English and 93 were published before 1950. We tested whether nesting traits show phylogenetic signal, examined relationships between habitat associations at different scales, and compared methodologies used to locate nests. We found no clear phylogenetic signals, but we found that nesting habitat associations were somewhat generalizable within subgenera. Landcover associations were related to nesting substrate associations; for example, surface-nesting species also tended to be associated with grasslands. Methodology was associated with nest locations; community scientists were most likely and researchers using nest boxes were least likely to report nests in human-dominated environments. These patterns were not apparent in past syntheses based only on the modern digital record. Our findings highlight the tremendous value of historic accounts for quantifying species' traits and other basic biological knowledge needed to interpret global-scale patterns.

Assessing deep convolutional neural network models and their comparative performance for automated medicinal plant identification from leaf images

Medicinal plants have got notable attention in recent years in the field of pharmaceutical and drug research. The high demand of herbal medicine in the rural areas of developing countries and drug industries necessitates correct identification of the medicinal plant species which is challenging in absence of expert taxonomic knowledge. Against this backdrop, we attempted to assess the performance of seven advanced deep learning algorithms in the automated identification of the plants from their leaf images and to suggest the best model from a comparative study of the models. We meticulously trained VGG16, VGG19, DenseNet201, ResNet50V2, Xception, InceptionResNetV2, and InceptionV3 deep neural network models. This training utilized a dataset comprising 5878 images encompassing 30 medicinal species distributed among 20 families. Our approach involved two avenues: the utilization of public data (PI) and a blend of public and field data (PFI), the latter featuring intricate backgrounds. Our study elucidates the robustness of these models in accurately identifying and classifying both interfamily and interspecies variations. Despite variations in accuracy across diverse families and species, the models demonstrated adeptness in these classifications. Comparing the models, we unearthed a crucial insight: the Normalized leverage factor (γ ω ) for DenseNet201 stands at 0.19, elevating it to the pinnacle position for PI with a remarkable 99.64 % accuracy and 98.31 % precision. In the PFI scenario, the same model achieves a γ ω of 0.15 with a commendable 97 % accuracy. These findings serve as a guiding beacon for shaping future application tools designed to automate medicinal plant identification at the user level.

Quantifying effectiveness and best practices for bumblebee identification from photographs

Understanding pollinator networks requires species level data on pollinators. New photographic approaches to identification provide avenues to data collection that reduce impacts on declining bumblebee species, but limited research has addressed their accuracy. Using blind identification of 1418 photographed bees, of which 561 had paired specimens, we assessed identification and agreement across 20 bumblebee species netted in Montana, North Dakota, and South Dakota by people with minimal training. An expert identified 92.4% of bees from photographs, whereas 98.2% of bees were identified from specimens. Photograph identifiability decreased for bees that were wet or matted; bees without clear pictures of the abdomen, side of thorax, or top of thorax; bees photographed with a tablet, and for species with more color morphs. Across paired specimens, the identification matched for 95.1% of bees. When combined with a second opinion of specimens without matching identifications, data suggested a similar misidentification rate (2.7% for photographs and 2.5% specimens). We suggest approaches to maximize accuracy, including development of rulesets for collection of a subset of specimens based on difficulty of identification and to address cryptic variation, and focused training on identification that highlights detection of species of concern and species frequently confused in a study area.

Environmental DNA (eDNA) removal rates in streams differ by particle size under varying substrate and light conditions

The use of environmental DNA (eDNA) as a sampling tool offers insights into the detection of invasive and/or rare aquatic species and enables biodiversity assessment without traditional sampling approaches, which are often labor-intensive. However, our understanding of the environmental factors that impact eDNA removal (i.e., how rapidly eDNA is removed from the water column by the combination of decay and physical removal) in flowing waters is limited. This limitation constrains predictions about the location and density of target organisms after positive detection. To address this question, we spiked Common Carp (Cyprinus carpio) eDNA into recirculating mesocosms (n = 24) under varying light (shaded versus open) and benthic substrate conditions (no substrate, bare substrate, and biofilm-colonized substrate). We then collected water samples from each mesocosm at four time points (40 min, 6 h, 18 h, and 48 h), and sequentially filtered the samples through 10, 1.0, and 0.2 μm filters to quantify removal rates for different eDNA particle sizes under varying light and substrate conditions. Combining all size classes, total eDNA removal rates were higher for mesocosms with biofilm-colonized substrate compared to those with no substrate or bare (i.e., no biofilm) substrate, which is consistent with previous findings linking biofilm colonization with increased eDNA removal and degradation. Additionally, when biofilm was present, light availability increased eDNA removal; eDNA levels fell below detection after 6-18 h for open mesocosms versus 18-48 h for shaded mesocosms. Among size classes, larger particles (>10 μm) were removed faster than small particles (1.0-0.2 μm). These results suggest that changes in the distribution of eDNA size classes over time (e.g., with downstream transport) and with differing environmental conditions could be used to predict the location of target organisms in flowing waters, which will advance the use of eDNA as a tool for species monitoring and management.

Identifying, reducing, and communicating uncertainty in community science: a focus on alien species

Community science (also often referred to as citizen science) provides a unique opportunity to address questions beyond the scope of other research methods whilst simultaneously engaging communities in the scientific process. This leads to broad educational benefits, empowers people, and can increase public awareness of societally relevant issues such as the biodiversity crisis. As such, community science has become a favourable framework for researching alien species where data on the presence, absence, abundance, phenology, and impact of species is important in informing management decisions. However, uncertainties arising at different stages can limit the interpretation of data and lead to projects failing to achieve their intended outcomes. Focusing on alien species centered community science projects, we identified key research questions and the relevant uncertainties that arise during the process of developing the study design, for example, when collecting the data and during the statistical analyses. Additionally, we assessed uncertainties from a linguistic perspective, and how the communication stages among project coordinators, participants and other stakeholders can alter the way in which information may be interpreted. We discuss existing methods for reducing uncertainty and suggest further solutions to improve data reliability. Further, we make suggestions to reduce the uncertainties that emerge at each project step and provide guidance and recommendations that can be readily applied in practice. Reducing uncertainties is essential and necessary to strengthen the scientific and community outcomes of community science, which is of particular importance to ensure the success of projects aimed at detecting novel alien species and monitoring their dynamics across space and time.

Deep learning increases the availability of organism photographs taken by citizens in citizen science programs

Citizen science programs using organism photographs have become popular, but there are two problems related to photographs. One problem is the low quality of photographs. It is laborious to identify species in photographs taken outdoors because they are out of focus, partially invisible, or under different lighting conditions. The other is difficulty for non-experts to identify species. Organisms usually have interspecific similarity and intraspecific variation, which hinder species identification by non-experts. Deep learning solves these problems and increases the availability of organism photographs. We trained a deep convolutional neural network, Xception, to identify bee species using various quality of bee photographs that were taken by citizens. These bees belonged to two honey bee species and 10 bumble bee species with interspecific similarity and intraspecific variation. We investigated the accuracy of species identification by biologists and deep learning. The accuracy of species identification by Xception (83.4%) was much higher than that of biologists (53.7%). When we grouped bee photographs by different colors resulting from intraspecific variation in addition to species, the accuracy of species identification by Xception increased to 84.7%. The collaboration with deep learning and experts will increase the reliability of species identification and their use for scientific researches.

The Native Bees of Texas: Evaluating the Benefits of a Public Engagement Course

Declines in native bee communities due to forces of global change have become an increasing public concern. Despite this heightened interest, there are few publicly available courses on native bees, and little understanding of how participants might benefit from such courses. In October of 2018 and 2019, we taught the 'Native Bees of Texas' course to the public at The University of Texas at Austin Lady Bird Johnson Wildflower Center botanical gardens in an active learning environment with slide-based presentations, printed photo-illustrated resources, and direct insect observations. In this study, we evaluated course efficacy and learning outcomes with a pre/post-course test, a survey, and open-ended feedback, focused on quality improvement findings. Overall, participants' test scores increased significantly, from 60% to 87% correct answers in 2018 and from 64% to 87% in 2019, with greater post-course differences in ecological knowledge than in identification skills. Post-course, the mean of participants' bee knowledge self-ratings was 4.56 on a five-point scale. The mean of participants' ratings of the degree to which they attained the course learning objectives was 4.43 on a five-point scale. Assessment results provided evidence that the course enriched participants' knowledge of native bee ecology and conservation and gave participants a basic foundation in bee identification. This highlights the utility of systematic course evaluations in public engagement efforts related to biodiversity conservation.

Linking camera-trap data to taxonomy: Identifying photographs of morphologically similar chipmunks

Remote cameras are a common method for surveying wildlife and recently have been promoted for implementing large-scale regional biodiversity monitoring programs. The use of camera-trap data depends on the correct identification of animals captured in the photographs, yet misidentification rates can be high, especially when morphologically similar species co-occur, and this can lead to faulty inferences and hinder conservation efforts. Correct identification is dependent on diagnosable taxonomic characters, photograph quality, and the experience and training of the observer. However, keys rooted in taxonomy are rarely used for the identification of camera-trap images and error rates are rarely assessed, even when morphologically similar species are present in the study area. We tested a method for ensuring high identification accuracy using two sympatric and morphologically similar chipmunk (Neotamias) species as a case study. We hypothesized that the identification accuracy would improve with use of the identification key and with observer training, resulting in higher levels of observer confidence and higher levels of agreement among observers. We developed an identification key and tested identification accuracy based on photographs of verified museum specimens. Our results supported predictions for each of these hypotheses. In addition, we validated the method in the field by comparing remote-camera data with live-trapping data. We recommend use of these methods to evaluate error rates and to exclude ambiguous records in camera-trap datasets. We urge that ensuring correct and scientifically defensible species identifications is incumbent on researchers and should be incorporated into the camera-trap workflow.

Central Attention and a Dual Path Convolutional Neural Network in Real-World Tree Species Recognition

Identifying plants is not only the job of professionals, but also useful or essential for the plant lover and the general public. Although deep learning approaches for plant recognition are promising, driven by the success of convolutional neural networks (CNN), their performances are still far from the requirements of an in-field scenario. First, we propose a central attention concept that helps focus on the target instead of backgrounds in the image for tree species recognition. It could prevent model training from confused vision by establishing a dual path CNN deep learning framework, in which the central attention model combined with the CNN model based on InceptionV3 were employed to automatically extract the features. These two models were then learned together with a shared classification layer. Experimental results assessed the effectiveness of our proposed approach which outperformed each uni-path alone, and existing methods in the whole plant recognition system. Additionally, we created our own tree image database where each photo contained a wealth of information on the entire tree instead of an individual plant organ. Lastly, we developed a prototype system of an online/offline available tree species identification working on a consumer mobile platform that can identify the tree species not only by image recognition, but also detection and classification in real-time remotely.

Using Bumble Bee Watch to investigate the accuracy and perception of bumble bee (Bombus spp.) identification by community scientists

Community science programs provide an opportunity to gather scientific data to inform conservation policy and management. This study examines the accuracy of community science identifications submitted to the North American Bumble Bee Watch program on a per species level and as compared to each species’ conservation status, as well as users (members of the public) and experts (those with expertise in the field of bumble bee biology) perceived ease of species identification. Photos of bumble bees (Hymenoptera: Apidae: Bombus) are submitted to the program by users and verified (species name corrected or assigned as necessary) by an expert. Over 22,000 records from over 4,900 users were used in the analyses. Accuracy was measured in two ways: percent agreement (percent of all records submitted correctly by users) and veracity (percent of all verified records submitted correctly by the users). Users generally perceived it harder to identify species than experts. User perceptions were not significantly different from the observed percent agreement or veracity, while expert perceptions were significantly different (overly optimistic) from the observed percent agreement but not the veracity. We compared user submitted names to final expert verified names and found that, for all species combined, the average percent agreement was 53.20% while the average veracity was 55.86%. There was a wide range in percent agreement values per species, although sample size and the role of chance did affect some species agreements. As the conservation status of species increased to higher levels of extinction risk, species were increasingly more likely to have a lower percent agreement but higher levels of veracity than species of least concern. For each species name submitted, the number of different species verified by experts varied from 1 to 32. Future research may investigate which factors relate to success in user identification through community science. These findings could play a role in informing the design of community science programs in the future, including for use in long-term and national-level monitoring of wild pollinators.

Not that kind of tree: Assessing the potential for decision tree-based plant identification using trait databases

PREMISE

Advancements in machine learning and the rise of accessible "big data" provide an important opportunity to improve trait-based plant identification. Here, we applied decision-tree induction to a subset of data from the TRY plant trait database to (1) assess the potential of decision trees for plant identification and (2) determine informative traits for distinguishing taxa.

METHODS

Decision trees were induced using 16 vegetative and floral traits (689 species, 20 genera). We assessed how well the algorithm classified species from test data and pinpointed those traits that were important for identification across diverse taxa.

RESULTS

The unpruned tree correctly placed 98% of the species in our data set into genera, indicating its promise for distinguishing among the species used to construct them. Furthermore, in the pruned tree, an average of 89% of the species from the test data sets were properly classified into their genera, demonstrating the flexibility of decision trees to also classify new species into genera within the tree. Closer inspection revealed that seven of the 16 traits were sufficient for the classification, and these traits yielded approximately two times more initial information gain than those not included.

DISCUSSION

Our findings demonstrate the potential for tree-based machine learning and big data in distinguishing among taxa and determining which traits are important for plant identification.

Citizen-Science and Participatory Research as a Means to Improve Stakeholder Engagement in Resource Management: A Case Study of Vietnamese American Fishers on the US Gulf Coast

This study examines the engagement of Vietnamese American commercial fisheries stakeholders in the US Gulf Coast with state and federal agencies and the role that citizen science and participatory research may play in improving this engagement. Using a mixed methods study including surveys, interviews, and focus groups, findings highlight language, lack of trust, and outreach misfit as key barriers to engaging Vietnamese American stakeholders as demanded for collaborative resource management or co-management. However, findings also demonstrate the potential role for citizen science and participatory research that collaboratively engages stakeholders in research to overcome some of these barriers to engaging diverse fishing stakeholders.

Capacity of United States federal government and its partners to rapidly and accurately report the identity (taxonomy) of non-native organisms intercepted in early detection programs

The early detection of and rapid response to invasive species (EDRR) depends on accurate and rapid identification of non-native species. The 2016–2018 National Invasive Species Council Management Plan called for an assessment of US government (federal) capacity to report on the identity of non-native organisms intercepted through early detection programs. This paper serves as the response to that action item. Here we summarize survey-based findings and make recommendations for improving the federal government’s capacity to identify non-native species authoritatively in a timely manner. We conclude with recommendations to improve accurate identification within the context of EDRR by increasing coordination, maintaining taxonomic expertise, creating an identification tools clearinghouse, developing and using taxonomic standards for naming and identification protocols, expanding the content of DNA and DNA Barcode libraries, ensuring long-term sustainability of biological collections, and engaging and empowering citizens and citizen science groups.

Utility of citizen science data: A case study in land-based shark fishing

Involving citizen scientists in research has become increasingly popular in natural resource management and allows for an increased research effort at low cost, distribution of scientific information to relevant audiences, and meaningful public engagement. Scientists engaging fishing tournament participants as citizen scientists represent ideal scenarios for testing citizen science initiatives. For example, the Texas Shark Rodeo has begun shifting to conservation-oriented catch-and-release practices, which provides a unique opportunity to collect data on a large scale for extended periods of time, particularly through tagging large numbers of sharks for very little cost compared to a directed scientific study. However, critics are somewhat skeptical of citizen science due to the potential for lack of rigor in data collection and validation. A major management concern for shark fisheries is the ability of anglers to identify species. We tested some of the assumptions and value of citizen-collected data by cross-verifying species identification. Specifically, the purpose of this study was to evaluate the accuracy of shark species identifications made by anglers fishing in the Texas Shark Rodeo using photographs that were submitted as a requirement for tournament participation. Using a confusion matrix, we determined that anglers correctly identified 97.2% of all shark catches submitted during the Texas Shark Rodeo from 2014-2018; however, smaller sharks and certain species, including blacknose and spinner sharks, were more difficult to identify than others. Most commonly confused with blacktip sharks, spinner sharks were most commonly identified incorrectly (76.1% true positive rate [TPR]) followed by blacknose (86.8% TPR), finetooth (88.0% TPR), and Atlantic sharpnose sharks (93.8% TPR). This study demonstrated that citizen scientists have the ability to identify sharks with relatively low error. This is important for science and management, as these long-term datasets with relatively wide geographic scope could potentially be incorporated into future assessments of sharks in the Gulf of Mexico.

Exploring snake occurrence records: Spatial biases and marginal gains from accessible social media

A species' distribution provides fundamental information on: climatic niche, biogeography, and conservation status. Species distribution models often use occurrence records from biodiversity databases, subject to spatial and taxonomic biases. Deficiencies in occurrence data can lead to incomplete species distribution estimates. We can incorporate other data sources to supplement occurrence datasets. The general public is creating (via GPS-enabled cameras to photograph wildlife) incidental occurrence records that may present an opportunity to improve species distribution models. We investigated (1) occurrence data of a cryptic group of animals: non-marine snakes, in a biodiversity database (Global Biodiversity Information Facility (GBIF)) and determined (2) whether incidental occurrence records extracted from geo-tagged social media images (Flickr) could improve distribution models for 18 tropical snake species. We provide R code to search for and extract data from images using Flickr's API. We show the biodiversity database's 302,386 records disproportionately originate from North America, Europe and Oceania (250,063, 82.7%), with substantial gaps in tropical areas that host the highest snake diversity. North America, Europe and Oceania averaged several hundred records per species; whereas Asia, Africa and South America averaged less than 35 per species. Occurrence density showed similar patterns; Asia, Africa and South America have roughly ten-fold fewer records per 100 km2than other regions. Social media provided 44,687 potential records. However, including them in distribution models only marginally impacted niche estimations; niche overlap indices were consistently over 0.9. Similarly, we show negligible differences in Maxent model performance between models trained using GBIF-only and Flickr-supplemented datasets. Model performance appeared dependent on species, rather than number of occurrences or training dataset. We suggest that for tropical snakes, accessible social media currently fails to deliver appreciable benefits for estimating species distributions; but due to the variation between species and the rapid growth in social media data, may still be worth considering in future contexts.

Automated Taxonomic Identification of Insects with Expert-Level Accuracy Using Effective Feature Transfer from Convolutional Networks

Rapid and reliable identification of insects is important in many contexts, from the detection of disease vectors and invasive species to the sorting of material from biodiversity inventories. Because of the shortage of adequate expertise, there has long been an interest in developing automated systems for this task. Previous attempts have been based on laborious and complex handcrafted extraction of image features, but in recent years it has been shown that sophisticated convolutional neural networks (CNNs) can learn to extract relevant features automatically, without human intervention. Unfortunately, reaching expert-level accuracy in CNN identifications requires substantial computational power and huge training data sets, which are often not available for taxonomic tasks. This can be addressed using feature transfer: a CNN that has been pretrained on a generic image classification task is exposed to the taxonomic images of interest, and information about its perception of those images is used in training a simpler, dedicated identification system. Here, we develop an effective method of CNN feature transfer, which achieves expert-level accuracy in taxonomic identification of insects with training sets of 100 images or less per category, depending on the nature of data set. Specifically, we extract rich representations of intermediate to high-level image features from the CNN architecture VGG16 pretrained on the ImageNet data set. This information is submitted to a linear support vector machine classifier, which is trained on the target problem. We tested the performance of our approach on two types of challenging taxonomic tasks: 1) identifying insects to higher groups when they are likely to belong to subgroups that have not been seen previously and 2) identifying visually similar species that are difficult to separate even for experts. For the first task, our approach reached $CDATA[$CDATA[$>$$92% accuracy on one data set (884 face images of 11 families of Diptera, all specimens representing unique species), and $CDATA[$CDATA[$>$$96% accuracy on another (2936 dorsal habitus images of 14 families of Coleoptera, over 90% of specimens belonging to unique species). For the second task, our approach outperformed a leading taxonomic expert on one data set (339 images of three species of the Coleoptera genus Oxythyrea; 97% accuracy), and both humans and traditional automated identification systems on another data set (3845 images of nine species of Plecoptera larvae; 98.6 % accuracy). Reanalyzing several biological image identification tasks studied in the recent literature, we show that our approach is broadly applicable and provides significant improvements over previous methods, whether based on dedicated CNNs, CNN feature transfer, or more traditional techniques. Thus, our method, which is easy to apply, can be highly successful in developing automated taxonomic identification systems even when training data sets are small and computational budgets limited. We conclude by briefly discussing some promising CNN-based research directions in morphological systematics opened up by the success of these techniques in providing accurate diagnostic tools.

Invasive plants negatively impact native, but not exotic, animals

Despite our growing understanding of the impacts of invasive plants on ecosystem structure and function, important gaps remain, including whether native and exotic species respond differently to plant invasion. This would elucidate basic ecological interactions and inform management. We performed a meta-analytic review of the effects of invasive plants on native and exotic resident animals. We found that invasive plants reduced the abundance of native, but not exotic, animals. This varied by animal phyla, with invasive plants reducing the abundance of native annelids and chordates, but not mollusks or arthropods. We found dissimilar impacts among "wet" and "dry" ecosystems, but not among animal trophic levels. Additionally, the impact of invasive plants increased over time, but this did not vary with animal nativity. Our review found that no studies considered resident nativity differences, and most did not identify animals to species. We call for more rigorous studies of invaded community impacts across taxa, and most importantly, explicit consideration of resident biogeographic origin. We provide an important first insight into how native and exotic species respond differently to invasion, the consequences of which may facilitate cascading trophic disruptions further exacerbating global change consequences to ecosystem structure and function.

Evaluating the ability of citizen scientists to identify bumblebee (Bombus) species

Citizen science is an increasingly popular way of engaging volunteers in the collection of scientific data. Despite this, data quality remains a concern and there is little published evidence about the accuracy of records generated by citizen scientists. Here we compare data generated by two British citizen science projects, Blooms for Bees and BeeWatch, to determine the ability of volunteer recorders to identify bumblebee (Bombus) species. We assessed recorders' identification ability in two ways-as recorder accuracy (the proportion of expert-verified records correctly identified by recorders) and recorder success (the proportion of recorder-submitted identifications confirmed correct by verifiers). Recorder identification ability was low (<50% accuracy; <60% success), despite access to project specific bumblebee identification materials. Identification ability varied significantly depending on bumblebee species, with recorders most able to correctly identify species with distinct appearances. Blooms for Bees recorders (largely recruited from the gardening community) were markedly less able to identify bumblebees than BeeWatch recorders (largely individuals with a more specific interest in bumblebees). Within both projects, recorders demonstrated an improvement in identification ability over time. Here we demonstrate and quantify the essential role of expert verification within citizen science projects, and highlight where resources could be strengthened to improve recorder ability.

Designing online species identification tools for biological recording: the impact on data quality and citizen science learning

In recent years, the number and scale of environmental citizen science programmes that involve lay people in scientific research have increased rapidly. Many of these initiatives are concerned with the recording and identification of species, processes which are increasingly mediated through digital interfaces. Here, we address the growing need to understand the particular role of digital identification tools, both in generating scientific data and in supporting learning by lay people engaged in citizen science activities pertaining to biological recording communities. Starting from two well-known identification tools, namely identification keys and field guides, this study focuses on the decision-making and quality of learning processes underlying species identification tasks, by comparing three digital interfaces designed to identify bumblebee species. The three interfaces varied with respect to whether species were directly compared or filtered by matching on visual features; and whether the order of filters was directed by the interface or a user-driven open choice. A concurrent mixed-methods approach was adopted to compare how these different interfaces affected the ability of participants to make correct and quick species identifications, and to better understand how participants learned through using these interfaces. We found that the accuracy of identification and quality of learning were dependent upon the interface type, the difficulty of the specimen on the image being identified and the interaction between interface type and ‘image difficulty’. Specifically, interfaces based on filtering outperformed those based on direct visual comparison across all metrics, and an open choice of filters led to higher accuracy than the interface that directed the filtering. Our results have direct implications for the design of online identification technologies for biological recording, irrespective of whether the goal is to collect higher quality citizen science data, or to support user learning and engagement in these communities of practice.

Measuring agreement among experts in classifying camera images of similar species

Camera trapping and solicitation of wildlife images through citizen science have become common tools in ecological research. Such studies collect many wildlife images for which correct species classification is crucial; even low misclassification rates can result in erroneous estimation of the geographic range or habitat use of a species, potentially hindering conservation or management efforts. However, some species are difficult to tell apart, making species classification challenging-but the literature on classification agreement rates among experts remains sparse. Here, we measure agreement among experts in distinguishing between images of two similar congeneric species, bobcats (Lynx rufus) and Canada lynx (Lynx canadensis). We asked experts to classify the species in selected images to test whether the season, background habitat, time of day, and the visible features of each animal (e.g., face, legs, tail) affected agreement among experts about the species in each image. Overall, experts had moderate agreement (Fleiss' kappa = 0.64), but experts had varying levels of agreement depending on these image characteristics. Most images (71%) had ≥1 expert classification of "unknown," and many images (39%) had some experts classify the image as "bobcat" while others classified it as "lynx." Further, experts were inconsistent even with themselves, changing their classifications of numerous images when they were asked to reclassify the same images months later. These results suggest that classification of images by a single expert is unreliable for similar-looking species. Most of the images did obtain a clear majority classification from the experts, although we emphasize that even majority classifications may be incorrect. We recommend that researchers using wildlife images consult multiple species experts to increase confidence in their image classifications of similar sympatric species. Still, when the presence of a species with similar sympatrics must be conclusive, physical or genetic evidence should be required.

New records of sharks (Elasmobranchii) from the Andaman and Nicobar Archipelago in India with notes on current checklists

The diversity of sharks occurring off the Andaman and Nicobar Archipelago in India has received increased attention in recent years. Yet, available checklists are out of date, often with inaccurate information and a number of commercially important species have not been documented through research and fish landing surveys. Here we report on shark species examined during fish landing surveys conducted from January 2017 to April 2018. Records of twelve previously unreported species from the archipelago are presented and include the bignose shark (Carcharhinusaltimus), pigeye shark (Carcharhinusamboinensis), bull shark (Carcharhinusleucas), snaggletooth shark (Hemipristiselongata), slender weasel shark (Paragaleusrandalli), Arabian smoothhound shark (Mustelusmosis), Indonesian houndshark (Hemitriakisindroyonoi), sand tiger shark (Carchariastaurus), Indonesian bambooshark (Chiloscylliumhasseltii), tawny nurse shark (Nebriusferrugineus), dwarf gulper shark (Centrophorusatromarginatus), and the Indonesian shortsnout spurdog (Squalushemipinnis). These records increase the reported shark species for the archipelago from 47 to 59 and for India from 114 to 116. Additionally, a size extension in the total length of C.hasseltii by 27 cm and of P.randalli by 8 cm is reported. Owing to the bio-geographical location of these islands, species diversity around the archipelago is unique and appears to overlap with that of southeast Asia. With increasing reports of over-exploitation and the operation of a targeted shark fishery by distant water fleets in these waters, the limited information on shark diversity from this region is of concern. Systematic and long-term monitoring of catches, combined with accurate species identification, is crucial to provide information on management measures.

A critical evaluation of the exotic bird collection of the Šariš Museum in Bardejov, Slovakia

A collection of exotic birds deposited in the Šariš Museum in Bardejov (SMB), Slovakia, has not been evaluated critically since their deposition. We assessed the accuracy of identification of 465 bird specimens deposited in SMB with native distributions outside of Slovakia. Specimens belonged to 322 species of 82 families and 26 orders. Of the specimen represented, 34 belonged to species considered as ‘near-threatened’ (7.3%), 16 as ‘vulnerable’ (3.4%) and one as ‘endangered’ (0.2%). The SMB collection holds 10 of 28 extant Cuban endemic species and another 11 species endemic to the Caribbean archipelago. Even among birds that are relatively easy to identify, many specimens were identified incorrectly or species identification was missing. Of 465 specimens evaluated, 95 (20.4%) were identified incorrectly or were missing species identification, and another 79 (17%) were identified correctly, but their names have changed over time due to taxonomic shift, thus they required correction.

Recommending plant taxa for supporting on-site species identification

BACKGROUND

Predicting a list of plant taxa most likely to be observed at a given geographical location and time is useful for many scenarios in biodiversity informatics. Since efficient plant species identification is impeded mainly by the large number of possible candidate species, providing a shortlist of likely candidates can help significantly expedite the task. Whereas species distribution models heavily rely on geo-referenced occurrence data, such information still remains largely unused for plant taxa identification tools.

RESULTS

In this paper, we conduct a study on the feasibility of computing a ranked shortlist of plant taxa likely to be encountered by an observer in the field. We use the territory of Germany as case study with a total of 7.62M records of freely available plant presence-absence data and occurrence records for 2.7k plant taxa. We systematically study achievable recommendation quality based on two types of source data: binary presence-absence data and individual occurrence records. Furthermore, we study strategies for aggregating records into a taxa recommendation based on location and date of an observation.

CONCLUSION

We evaluate recommendations using 28k geo-referenced and taxa-labeled plant images hosted on the Flickr website as an independent test dataset. Relying on location information from presence-absence data alone results in an average recall of 82%. However, we find that occurrence records are complementary to presence-absence data and using both in combination yields considerably higher recall of 96% along with improved ranking metrics. Ultimately, by reducing the list of candidate taxa by an average of 62%, a spatio-temporal prior can substantially expedite the overall identification problem.

Automated plant species identification-Trends and future directions

Current rates of species loss triggered numerous attempts to protect and conserve biodiversity. Species conservation, however, requires species identification skills, a competence obtained through intensive training and experience. Field researchers, land managers, educators, civil servants, and the interested public would greatly benefit from accessible, up-to-date tools automating the process of species identification. Currently, relevant technologies, such as digital cameras, mobile devices, and remote access to databases, are ubiquitously available, accompanied by significant advances in image processing and pattern recognition. The idea of automated species identification is approaching reality. We review the technical status quo on computer vision approaches for plant species identification, highlight the main research challenges to overcome in providing applicable tools, and conclude with a discussion of open and future research thrusts.

Plant identification is not exclusively the job of botanists and plant ecologists. It is required or useful for large parts of society, from professionals (such as landscape architects, foresters, farmers, conservationists, and biologists) to the general public (like ecotourists, hikers, and nature lovers). But the identification of plants by conventional means is difficult, time consuming, and (due to the use of specific botanical terms) frustrating for novices. This creates a hard-to-overcome hurdle for novices interested in acquiring species knowledge. In recent years, computer science research, especially image processing and pattern recognition techniques, have been introduced into plant taxonomy to eventually make up for the deficiency in people's identification abilities. We review the technical status quo on computer vision approaches for plant species identification, highlight the main research challenges to overcome in providing applicable tools, and conclude with a discussion of open and future research thrusts.

Species identification of juvenile fishes of the genus Pseudoblennius using mitochondrial DNA barcoding

Species identification is important in natural science and should be precise. Six specimens of juvenile Pseudoblennius were collected from the eastern coastal waters of the Korean Peninsula and Jeju Island in 2016-2017, and identified for the first time using DNA barcoding based on mitochondrial DNA cytochrome oxidase subunit I sequences. DNA barcoding analysis supported three adult species of genus Pseudoblennius (P. cottoides, P. marmoratus, and P. percoides) being quite distinct from each other. Six juvenile specimens were completely identified: two as P. cottoides; two more as P. marmoratus; and the final two as P. percoides. Mitochondrial DNA COI can be effective as a means of species identification method for the genus Pseudoblennius.

Fish DNA barcoding around large marine infrastructure for improved biodiversity assessment and monitoring

Accurate species-level identification is pivotal for environmental assessments and monitoring. The PERU LNG terminal is composed of large marine infrastructure located on the central coast of Peru. Since construction, taxonomically challenging species such as drum fishes (Sciaenidae) have been attracted to the new hard-bottom habitat. We conducted a DNA barcoding study to investigate fish diversity and constructed a DNA barcode reference library. We examined 56 vouchered specimens and identified 24 unique species. Intra- and interspecific divergence estimates ranged between 0 and 0.64% and 11 and 35.5%, respectively. We assessed the efficiency of the reference library to identify 29 non-vouchered specimens. We had 82.5% efficiency by using both our reference library (n = 17) and GenBank (n = 24). We highlight the importance of implementing molecular barcoding for complementing biodiversity assessments in marine environments. This study represents a first step towards generating a comprehensive DNA barcode reference library for marine fishes in Peru.

Species identification by conservation practitioners using online images: accuracy and agreement between experts

Emerging technologies have led to an increase in species observations being recorded via digital images. Such visual records are easily shared, and are often uploaded to online communities when help is required to identify or validate species. Although this is common practice, little is known about the accuracy of species identification from such images. Using online images of newts that are native and non-native to the UK, this study asked holders of great crested newt (Triturus cristatus) licences (issued by UK authorities to permit surveying for this species) to sort these images into groups, and to assign species names to those groups. All of these experts identified the native species, but agreement among these participants was low, with some being cautious in committing to definitive identifications. Individuals’ accuracy was also independent of both their experience and self-assessed ability. Furthermore, mean accuracy was not uniform across species (69–96%). These findings demonstrate the difficulty of accurate identification of newts from a single image, and that expert judgements are variable, even within the same knowledgeable community. We suggest that identification decisions should be made on multiple images and verified by more than one expert, which could improve the reliability of species data.

The power of monitoring: optimizing survey designs to detect occupancy changes in a rare amphibian population

Biodiversity conservation requires reliable species assessments and rigorously designed surveys. However, determining the survey effort required to reliably detect population change can be challenging for rare, cryptic and elusive species. We used a tropical bromeliad-dwelling frog as a model system to explore a cost-effective sampling design that optimizes the chances of detecting a population decline. Relatively few sampling visits were needed to estimate occupancy and detectability with good precision, and to detect a 30% change in occupancy with 80% power. Detectability was influenced by observer expertise, which therefore also had an effect on the sampling design - less experienced observers require more sampling visits to detect the species. Even when the sampling design provides precise parameter estimates, only moderate to large changes in occupancy will be detected with reliable power. Detecting a population change of 15% or less requires a large number of sites to be surveyed, which might be unachievable for range-restricted species occurring at relatively few sites. Unless there is high initial occupancy, rare and cryptic species will be particularly challenging when it comes to detecting small population changes. This may be a particular issue for long-term monitoring of amphibians which often display low detectability and wide natural fluctuations.

Utilization of photographs taken by citizens for estimating bumblebee distributions

Citizen science is a powerful tool for collecting large volumes of observational data on various species. These data are used to estimate distributions using environmental factors with Species Distribution Models (SDM). However, if citizens are inexperienced in recognizing organisms, they may report different species as the subject species. Here we show nation-wide bumblebee distributions using photographs taken by citizens in our project, and estimated distributions for six bumblebee species using land use, climate, and altitude data with SDM. We identified species from photographic images, and took their locations from GPS data of photographs or the text in e-mails. When we compared our data with conventional data for specimens in the Global Biodiversity Information Facility (GBIF), we found that the volume and the number of species were larger, and the bias of spatial range was lower, than those of GBIF. Our estimated distributions were more consistent with bumblebee distributions reported in previous studies than with those of GBIF. Our method was effective for collecting distribution data, and estimating distributions with SDM. The estimated SDM allows us to predict the previous and future species distributions, and to develop conservation policies taking account of future city planning and/or global climate changes.

A visual processing advantage for young-adolescent deaf observers: Evidence from face and object matching tasks

It is unresolved whether the permanent auditory deprivation that deaf people experience leads to the enhanced visual processing of faces. The current study explored this question with a matching task in which observers searched for a target face among a concurrent lineup of ten faces. This was compared with a control task in which the same stimuli were presented upside down, to disrupt typical face processing, and an object matching task. A sample of young-adolescent deaf observers performed with higher accuracy than hearing controls across all of these tasks. These results clarify previous findings and provide evidence for a general visual processing advantage in deaf observers rather than a face-specific effect.

Acquiring and preprocessing leaf images for automated plant identification: understanding the tradeoff between effort and information gain

BACKGROUND

Automated species identification is a long term research subject. Contrary to flowers and fruits, leaves are available throughout most of the year. Offering margin and texture to characterize a species, they are the most studied organ for automated identification. Substantially matured machine learning techniques generate the need for more training data (aka leaf images). Researchers as well as enthusiasts miss guidance on how to acquire suitable training images in an efficient way.

METHODS

In this paper, we systematically study nine image types and three preprocessing strategies. Image types vary in terms of in-situ image recording conditions: perspective, illumination, and background, while the preprocessing strategies compare non-preprocessed, cropped, and segmented images to each other. Per image type-preprocessing combination, we also quantify the manual effort required for their implementation. We extract image features using a convolutional neural network, classify species using the resulting feature vectors and discuss classification accuracy in relation to the required effort per combination.

RESULTS

The most effective, non-destructive way to record herbaceous leaves is to take an image of the leaf's top side. We yield the highest classification accuracy using destructive back light images, i.e., holding the plucked leaf against the sky for image acquisition. Cropping the image to the leaf's boundary substantially improves accuracy, while precise segmentation yields similar accuracy at a substantially higher effort. The permanent use or disuse of a flash light has negligible effects. Imaging the typically stronger textured backside of a leaf does not result in higher accuracy, but notably increases the acquisition cost.

CONCLUSIONS

In conclusion, the way in which leaf images are acquired and preprocessed does have a substantial effect on the accuracy of the classifier trained on them. For the first time, this study provides a systematic guideline allowing researchers to spend available acquisition resources wisely while yielding the optimal classification accuracy.