Biology subjects

Deck, J.

Publications and source records attributed to Deck, J..

5 recordsLinked to original sources

bioRxiv2025

Petal to the metal: The slow road to automating large-scale phenology labeling for herbarium specimens

Herbarium specimens represent critical historical records of plant phenology, yet automating annotation of reproductive structures remains challenging given the diversity of floral morphologies, specimen age and quality, and image quality. Here, we present a machine learning pipeline that uses an ensemble modeling approach to detect flowers on herbarium specimens and deliver these data to the phenology research community. After testing multiple strategies for generating training data, we found in-house expert-curated annotations were essential for producing reliable results. Expert validation found relatively strong accuracy for detecting present floral structures, but still had moderately high false negative rates. Applying the ensemble to our filtered final image dataset of 22 million records resulted in 11.1 million records labeled with flowers present. However, only 2.9 million of these contained complete metadata necessary for downstream phenology research, highlighting the need for full label digitization efforts. Still, this dataset represents a large compilation of historical herbarium-derived phenology records available as a resource for the phenology community. We end by demonstrating how integrating these machine-labeled records into Phenobase, a publicly-available phenology database, expands taxonomic and temporal coverage for large-scale phenological analyses, and discuss remaining challenges and next steps.

ecology↗

bioRxiv2025

Turning a new leaf: PhenoVision provides leaf phenology data at the global scale

Plant phenology dictates many aspects of community function and ecosystem dynamics. Yet, global phenology data are still limited, especially in areas lacking monitoring programs. Here we present a new data resource, PhenoVision-Leaf, which extends a computer-vision pipeline utilizing iNaturalist digital image vouchers to produce global-scale leaf phenophase data for deciduous, woody genera. We first discuss our implementation of a new human annotation framework for leaf phenology on iNaturalist, aligning with phenophase definitions used by the larger phenology community. We then showcase the use of 165,988 crowdsourced annotated records to train a Vision Transformer model with a two-stage regime to maximize accuracy across single- and multi-image records. This approach extends Phenovision from scoring individual images to aggregating at the iNaturalist record level, better aligning with human annotation processes. Post-hoc validation showed high performance for detecting present green and colored leaves (>98% accuracy), and reasonable accuracy for breaking leaf buds (>87% accuracy). Applying PhenoVision-Leaf to over 26 million iNaturalist records yielded 5.6 million record-level phenology observations across 6,500 species and 57 families, filling geographic and taxonomic gaps. These data, now accessible through the Phenobase portal, establish a foundation for near real-time monitoring of leaf phenology, supporting global-scale synthesis analyses.

ecology↗

bioRxiv2024

PhenoVision: A framework for automating and delivering research-ready plant phenology data from field images

Plant phenology plays a fundamental role in shaping ecosystems, and global change-induced shifts in phenology have cascading impacts on species interactions and ecosystem structure and function. Detailed, high-quality observations of when plants undergo seasonal transitions such as leaf-out, flowering, and fruiting are critical for tracking causes and consequences of phenology shifts, but these data are often sparse and biased globally. These data gaps limit broader generalizations and forecasting improvements in the face of continuing disturbance. One solution to closing such gaps is to document phenology on field images taken by public participants. iNaturalist, in particular, provides global scale research-grade data and is expanding rapidly. Here we utilize over 53 million field images of plants and millions of human annotations from iNaturalist - data spanning all angiosperms and drawn from across the globe - to train a computer vision model (PhenoVision) to detect the presence of fruits and flowers. PhenoVision utilizes a vision transformer architecture pretrained with a masked autoencoder to improve classification success, and it achieves high accuracy for flower (98.5%) and fruit presence (95%). Key to producing research-ready phenology data is post-calibration tuning and validation focused on reducing noise inherent in field photographs, and maximizing the true positive rate. We also develop a standardized set of quality metrics and metadata so that results can be used effectively by the community. Finally, we showcase how this effort vastly increases phenology data coverage, including regions of the globe where data have been limited before. Our end products are tuned models, new data resources, and an application streamlining discovery and use of those data for the broader research and management community. We close by discussing next steps, including automating phenology annotations, adding new phenology targets, e.g., leaf phenology, and further integration with other resources to form a global central database integrating all in-situ plant phenology resources.

ecology↗

bioRxiv2022

Metadata preservation and stewardship for genomic data is possible, but must happen now

Genetic diversity within species represents a fundamental yet underappreciated level of biodiversity. Because genetic diversity can indicate species and population resilience to changing climate, its measurement is relevant to many national and global conservation policy targets. Many studies of evolutionary biology, molecular ecology and conservation genetics produce large amounts of genome-scale genetic diversity data for wild populations. While open data policies have ensured an abundance of freely available genomic data stored in the databases of the International Nucleotide Sequence Database Collaboration (INSDC), only about 13% of current accessions have the associated spatial and temporal metadata in INSDC necessary to be reused in monitoring programs, macrogenetic studies, or for acknowledging the sovereignty of nations or Indigenous Peoples. We undertook a "distributed datathon" to quantify the availability of these missing metadata in sources external to the INSDC and to test the hypothesis that these metadata decay with time. We also worked to remediate these missing metadata by extracting them, when present, from associated published papers, online repositories, and/or from direct communication with authors. Starting with 848 programmatically identified candidate datasets (INSDC BioProjects), we manually determined that 561 contained samples from wild populations. We successfully restored spatiotemporal metadata (locality name and/or geospatial coordinates and collection year) for 78% of these 561 datasets (N = 440 BioProjects comprising 45,105 individuals or BioSamples from 762 species in 17 phyla). We also quantified the availability of 33 additional categories of metadata in sources external to the INSDC. Information about associated publications and the type of habitat from which the samples were taken was the most easily found; information about sampling permits was the most challenging to locate. Looking at papers and online repositories was much more fruitful than contacting authors, who only replied to our email requests 45% of the time. Overall, 23% of our email queries to authors discovered useful metadata. Importantly, we found that the probability of retrieving spatiotemporal metadata declines significantly with the age of the dataset, with a 13.5% yearly decrease for metadata located in published papers or online repositories and up to a 22% yearly decrease for metadata that were only available from authors. This observable metadata decay, mirrored in studies of other types of biological data, should motivate swift updates to data sharing policies and researcher practices to ensure that the valuable context provided by metadata is not lost forever.

bioinformatics↗

bioRxiv2017

Identifiers for the 21st century:How to design, provision, and reuse persistent identifiers to maximize utility and impact of life science data

In many disciplines, data is highly decentralized across thousands of online databases (repositories, registries, and knowledgebases). Wringing value from such databases depends on the discipline of data science and on the humble bricks and mortar that make integration possible; identifiers are a core component of this integration infrastructure. Drawing on our experience and on work by other groups, we outline ten lessons we have learned about the identifier qualities and best practices that facilitate large-scale data integration. Specifically, we propose actions that identifier practitioners (database providers) should take in the design, provision and reuse of identifiers; we also outline important considerations for those referencing identifiers in various circumstances, including by authors and data generators. While the importance and relevance of each lesson will vary by context, there is a need for increased awareness about how to avoid and manage common identifier problems, especially those related to persistence and web-accessibility/resolvability. We focus strongly on web-based identifiers in the life sciences; however, the principles are broadly relevant to other disciplines.

bioinformatics↗