Dataset Information

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

ABSTRACT: 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.

SUBMITTER: Blair J

PROVIDER: S-EPMC7713910 | biostudies-literature | 2020 Dec

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

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

Blair Jarrett J Weiser Michael D MD Kaspari Michael M Miller Matthew M Siler Cameron C Marshall Katie E KE

Ecology and evolution 20201111 23

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 a ...[more]

PMID: 33304524

Similar Datasets

Project description:Movement trajectories are usually recorded as a sequence of discrete movement events described by two parameters: step length (distance) and turning angle (bearing). One of the most widespread methods to record the geocoordinates of each step is by a GPS device. Such devices have limited suitability for recording fine movements of species with low dispersal ability including flightless carabid beetles at small spatio-temporal scales. As an alternative, the distance-bearing approach can avoid the measurement error of GPS units since it uses directly measured distances and compass azimuths. As no quantification of measurement error between distance-bearing and GPS approaches exists so far, we generated artificial fine-scale trajectories and in addition radio-tracked living carabids in a temperate forest and recorded each movement step by both methods. Trajectories obtained from distance-bearing were compared to those obtained by a GPS device in terms of movement parameters. Consequently, both types of trajectories were segmented by state-switching modeling into two distinct movement stages typical for carabids: random walk and directed movement. We found that the measurement error of GPS compared to distance-bearing was 1.878 m (SEM = 0.181 m) for distances and 31.330° (SEM = 2.066°) for bearings. Moreover, these errors increased under dense forest canopy and rainy weather. Distance error did not change with increasing distance recorded by distance-bearing but bearings were significantly more sensitive to error at short distances. State-switching models showed only slight, not significant, differences in movement states between the two methods in favor of the random walk in the distance-bearing approach. However, the shape of the GPS-measured trajectories considerably differed from those recorded by distance-bearing caused especially by bearing error at short distances. Our study showed that distance-bearing could be more appropriate for recording movement steps not only of ground-dwelling beetles but also other small animals at fine spatio-temporal scales.

Project description:The accurate sampling of communities is vital to any investigation of ecological processes and biodiversity. Dung beetles have emerged as a widely used focal taxon in environmental studies and can be sampled quickly and inexpensively using baited pitfalls. Although there is now a wealth of available data on dung beetle communities from around the world, there is a lack of standardisation between sampling protocols for accurately sampling dung beetle communities. In particular, bait choice is often led by the idiosyncrasies of the researcher, logistic problems and the dung sources available, which leads to difficulties for inter-study comparisons. In general, human dung is the preferred choice, however, it is often in short supply, which can severely limit sampling effort. By contrast, pigs may produce up to 20 times the volume. We tested the ability of human and pig dung to attract a primary forest dung beetle assemblage, as well as three mixes of the two baits in different proportions. Analyses focussed on the comparability of sampling with pig or human-pig dung mixes with studies that have sampled using human dung. There were no significant differences between richness and abundance sampled by each bait. The assemblages sampled were remarkably consistent across baits, and ordination analyses showed that the assemblages sampled by mixed dung baits were not significantly different from that captured by pure human dung, with the assemblages sampled by 10% and 90% pig mixes structurally most similar to assemblages sampled by human dung. We suggest that a 10:90 human:pig ratio, or similar, is an ideal compromise between sampling efficiency, inter-study comparability and the availability of large quantities of bait for sampling Amazonian dung beetles. Assessing the comparability of assemblage samples collected using different baits represents an important step to facilitating large-scale meta-analyses of dung beetle assemblages collected using non-standard methodology.

Project description:Background and aimsCarnivorous plants of the genus Nepenthes possess modified leaves that form pitfall traps in order to capture prey, mainly arthropods, to make additional nutrients available for the plant. These pitchers contain a digestive fluid due to the presence of hydrolytic enzymes. In this study, the composition of the digestive fluid was further analysed with regard to mineral nutrients and low molecular-weight compounds. A potential contribution of microbes to the composition of pitcher fluid was investigated.MethodsFluids from closed pitchers were harvested and analysed for mineral nutrients using analytical techniques based on ion-chromatography and inductively coupled plasma-optical emission spectroscopy. Secondary metabolites were identified by a combination of LC-MS and NMR. The presence of bacteria in the pitcher fluid was investigated by PCR of 16S-rRNA genes. Growth analyses of bacteria and yeast were performed in vitro with harvested pitcher fluid and in vivo within pitchers with injected microbes.Key resultsThe pitcher fluid from closed pitchers was found to be primarily an approx. 25-mm KCl solution, which is free of bacteria and unsuitable for microbial growth probably due to the lack of essential mineral nutrients such as phosphate and inorganic nitrogen. The fluid also contained antimicrobial naphthoquinones, plumbagin and 7-methyl-juglone, and defensive proteins such as the thaumatin-like protein. Challenging with bacteria or yeast caused bactericide as well as fungistatic properties in the fluid. Our results reveal that Nepenthes pitcher fluids represent a dynamic system that is able to react to the presence of microbes.ConclusionsThe secreted liquid of closed and freshly opened Nepenthes pitchers is exclusively plant-derived. It is unsuitable to serve as an environment for microbial growth. Thus, Nepenthes plants can avoid and control, at least to some extent, the microbial colonization of their pitfall traps and, thereby, reduce the need to vie with microbes for the prey-derived nutrients.

Dataset Information

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

Publications

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

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets