Leaves are the most abundant and visible plant organ, both in the modern world and the fossil record. Identifying foliage to the correct plant family based on leaf architecture is a fundamental botanical skill that is also critical for isolated fossil leaves, which often, especially in the Cenozoic, represent extinct genera and species from extant families. Resources focused on leaf identification are remarkably scarce; however, the situation has improved due to the recent proliferation of digitized herbarium material, live-plant identification applications, and online collections of cleared and fossil leaf images. Nevertheless, the need remains for a specialized image dataset for comparative leaf architecture. We address this gap by assembling an open-access database of 30,252 images of vouchered leaf specimens vetted to family level, primarily of angiosperms, including 26,176 images of cleared and x-rayed leaves representing 354 families and 4,076 of fossil leaves from 48 families. The images maintain original resolution, have user-friendly filenames, and are vetted using APG and modern paleobotanical standards. The cleared and x-rayed leaves include the Jack A. Wolfe and Leo J. Hickey contributions to the National Cleared Leaf Collection and a collection of high-resolution scanned x-ray negatives, housed in the Division of Paleobotany, Department of Paleobiology, Smithsonian National Museum of Natural History, Washington D.C.; and the Daniel I. Axelrod Cleared Leaf Collection, housed at the University of California Museum of Paleontology, Berkeley. The fossil images include a sampling of Late Cretaceous to Eocene paleobotanical sites from the Western Hemisphere held at numerous institutions, especially from Florissant Fossil Beds National Monument (late Eocene, Colorado), as well as several other localities from the Late Cretaceous to Eocene of the Western USA and the early Paleogene of Colombia and southern Argentina. The dataset facilitates new research and education opportunities in paleobotany, comparative leaf architecture, systematics, and machine learning.

Angiosperms, cleared leaves, data science, fossil leaves, leaf architecture, paleobotany

General patterns of angiosperm leaf architecture, the shape and venation characters of leaves, are well known for very few of the more than 400 angiosperm families. The development of a standard descriptive terminology (von Ettingshausen 1861; Hickey 1973, 1979; Ellis et al. 2009) has catalyzed increased detail and reproducibility in species descriptions of both living and fossil leaves. However, despite the use of numerous visual examples (e.g., Ellis et al. 2009), publications to date do not inform the reader how to accomplish the fundamental task of identifying leaves that, as for the great majority of leaf fossils, are isolated from the rest of the plant and missing diagnostic information from stipules, leaf organization, and reproductive and other organs.

To build their knowledge of leaf architecture, researchers still rely primarily on “oral tradition” from a dwindling number of knowledgeable colleagues and a handful of survey papers and field guides that emphasize purportedly diagnostic leaf features (Hickey and Wolfe 1975; Gentry 1993; da Ribeiro et al. 1999; Keller 2004). There is significant literature on the leaf architecture and leaf-fossil records of various taxa (among many others, von Ettingshausen 1858; Hill 1982; Jones 1986; Manchester 1987; Todzia and Keating 1991; Gandolfo and Romero 1992; Premoli 1996; Fuller and Hickey 2005; Martínez-Millán and Cevallos-Ferriz 2005; Doyle 2007; Kellner et al. 2012). However, many of the most diverse and ecologically significant groups of angiosperms have virtually no documentation of diagnostic leaf-blade features (e.g., Asteraceae, Rubiaceae), and thus their leaf fossils remain largely unrecognized, though probably hidden in plain sight in museum collections (see Wilf 2008; Wilf et al. 2016). More than half of fossil-leaf species in many older monographs are thought to have been misclassified (see Dilcher 1971), and most of the millions of leaf fossils in the general stratigraphic collections of the world’s museums are not yet identified.

Machine-vision algorithms, as seen in popular applications such as LeafSnap (Kumar et al. 2012), Pl@ntNet (Bonnet et al. 2018), and iNaturalist (Van Horn et al. 2018), are making spectacular breakthroughs in automated species identification of live plants; however, they provide little, if any, feedback about the diagnostic features they detect. Few algorithms have attempted to generalize above the species level (Wilf et al. 2016; Carranza-Rojas et al. 2018), and so far the methods do not work on leaf fossils, which mostly represent extinct species and often extinct genera.

Increasing general knowledge of leaf architecture for both human and machine learners depends on the development of customized, accessible, vetted visual libraries that allow rapid morphological comparisons of a high phylogenetic diversity of extant and fossil leaves. The recent proliferation of digitized plant-image resources comprises an invaluable reference for plant morphology, already including tens of millions of digitized herbarium sheets on portals and aggregator sites such as JStor Global Plants (https://plants.jstor.org), iDigBio (https://www.idigbio.org), RecolNat (https://www.recolnat.org), and many others, as well as servers located at numerous individual herbaria worldwide (e.g., Bakker et al. 2020). However, studying leaf comparative morphology is not simple because leaves only represent part of the visual field of a herbarium sheet and appear, with overlaps, at many different angles and sizes. Computer-vision algorithms that blur text or segment leaves from background or from other plant material are likely to help solve this issue (Hussein et al. 2020, 2021; Weaver et al. 2020; de Lutio et al. 2021). However, many visual distractors remain, and critical details of higher-order venation are often not visible in digitized herbarium sheets. Assessing leaf architecture at family level from digital herbaria also requires examination of extremely large numbers of specimens for all but the most species-poor families. In this regard, JStor Global Plants stands out for prioritizing type specimens collated digitally from across the world’s herbaria, thus allowing rapid surveys of the taxa in a family based on protologue voucher material. Finally, digitization efforts are far more advanced in resource-rich countries, whereas many significant collections are located in developing nations where herbarium digitization is occurring at a slower pace.

Cleared or x-rayed leaves from phylogenetically diverse taxa, selectively sampled from vouchered herbarium sheets, remain the most valuable visual reference for comparative study of leaf architecture because they have a similar visual presentation, with high capture of venation detail and comparatively few distractors. Existing collections of this type are fragile, mostly made decades ago as references for fossil leaf identification by selecting leaves from herbarium sheets, then either chemically clearing the specimens of most tissues other than veins and mounting them on glass slides or x-ray imaging them, in either case with extreme care and effort. Most cleared-leaf collections suffer from deterioration of the mounting media, which obscures large areas of the leaves; thus, photographic archiving offers a form of visual preservation before further degradation occurs. The largest and best-known cleared-leaf collections are those of the late Drs. Jack A. Wolfe and Leo J. Hickey, together now forming the National Cleared Leaf Collection (NCLC; NCLC-W and NCLC-H, respectively), housed in the Division of Paleobotany of the Smithsonian Institution National Museum of Natural History (NMNH, repository acronym USNM, Washington, D.C.).

For the many users who may find it challenging to visit these collections in person for suitable lengths of time, many cleared and x-rayed leaf collections are already accessible from various websites or in print. These valuable resources include the NCLC-W and other collections in the Cleared Leaf Image Database (http://clearedleavesdb.org; Das et al. 2014); the NCLC-H served from the Yale Peabody Museum (https://collections.peabody.yale.edu/pb/nclc); the Daniel I. Axelrod cleared-leaf collections of the University of California Museum of Paleontology (UCMP; https://UCMP.berkeley.edu/collections/paleobotany-collection/UCMP-cleared-leaf-collection); the National Museum of Nature and Science (NMNS, Ibaraki, Japan) Cleared Leaf Database by Drs. Toshimasa Tanai and Kazuhiko Uemura (https://www.kahaku.go.jp/research/db/geology-paleontology/cleared_leaf/database/?lg=en); leaf x-ray images of Australian rainforest plants by the late Dr. David C. Christophel and colleagues (Christophel and Hyland 1993; Christophel and Rowett 1996), some of which are maintained in the online Australian Tropical Rainforest Plants identification system (https://apps.lucidcentral.org/rainforest/text/intro/index.html); and the late Dr. Edward P. Klucking’s book series illustrating cleared leaves from selected families (Klucking 1986–2003). We also note an open-access image dataset of cleared leaves from Borneo, consisting of small (1 cm²) lamina samples (Blonder et al. 2019; Xu et al. 2021). In most of the online image sets, bulk downloads are not easily done, images are downsampled to low resolution, and the filenames are not standardized, requiring significant manual effort to re-organize and collate them for a particular project. Adding further complications to data modularity, taxonomic data have often become partially obsolete.

Isolated fossil leaves present an additional set of challenging problems (e.g., Wilf 2008), including incomplete preservation, morphological convergence, and the well-known legacy of innumerable taxonomic misidentifications in older publications (see Dilcher 1971, 1974; Hill 1982). Numerous high-quality systematic treatments have become available for many leaf-fossil taxa, especially over the last few decades, but the images are dispersed across publications and are usually of low resolution. An increasing number of images of vouchered fossil-leaf collections is available online from natural history museums. Examples include aggregator sites such as GBIF (gbif.org) and individual institutions such as the Yale Peabody Museum, (https://peabody.yale.edu/collections/paleobotany), the Burke Museum (www.burkemuseum.org/collections-and-research/geology-and-paleontology/collections-database/images.php), the University of Colorado Boulder Museum of Natural History (https://www.colorado.edu/cumuseum/research-collections/paleontology/invertebrates-plants), and the UCMP (https://UCMPdb.berkeley.edu). Nevertheless, museum servers and project sites (e.g., Traiser et al. 2018) usually retain the taxonomy as published, which is vital for the nomenclatural stability of type specimens but well known to be problematic, especially for the many older collections that have not been revised under modern standards. All these issues make it very difficult for researchers, students, and non-specialists to form a reliable base of knowledge about fossil-leaf identification and have perhaps engendered an overreliance on methods that do not require taxonomy at all, such as leaf morphotyping (see Wilf 2008).

Here, we meet the community need for a specialized dataset of leaf images by consolidating a set of original-resolution photographs of vouchered extant and fossil specimens (Fig. 1, Table 1), primarily of angiosperms, vetted to family level and relabeled to user-friendly filenames, into an open-access archive in a single, standard file format (jpeg, at minimum possible compression). A principal goal, based on many years of practical experience using leaf-image datasets in our research, is maximum and sustained ease of use with rapid access to the entire library. Thus, instead of creating an interactive database that may become obsolete and limit resolution or user flexibility, we simply provide the image files in labeled folders that can easily be downloaded, then viewed and searched using any visual browser (e.g., Adobe Bridge, Adobe Lightroom, Windows Explorer) on any suitable device, such as a personal computer.

Figure 1. 

Selected image pairs of confamilial extant and fossil (see Appendix 1) leaves from the dataset A Batesia floribunda Spruce ex. Benth. (Fabaceae), NCLC-W 6417, showing typical layout of a cleared-leaf slide with original annotations (other examples are cropped in this figure); source voucher Froes 12074, DS 291771 (at CAS), Amazonas, Brazil B Fabaceae sp. CJ1, SGC-ICP-10173; Cerrejón mine, middle-late Paleocene of Guajira Peninsula, Colombia C Crataegus viridis L. (Rosaceae), NCLC-W 11951b; H. Meyer s/n (collected 1974, no other voucher), cultivated, California, USA D Crataegus copeana (Rosaceae), UCMP 3610; Florissant, late Eocene of Colorado, USA; H. Meyer photograph number 0420 E Tetracentron sinense Oliv. (Trochodendraceae), S. Wing negative 71-002; E.H. Wilson 659, US 599036, Szechuan, China F Ziziphoides flabellum (Trochodendraceae), USNM 560134; Mexican Hat, early Paleocene of Montana, USA G Quercus prinus L. (Fagaceae), NCLC-W 6137; H. Foster 8223, US 1730249, Florida, USA H Fagopsis longifolia (Fagaceae), FLFO 003432A; Florissant, late Eocene of Colorado, USA I Eucalyptus astringens (Maiden) Maiden (Myrtaceae), NCLC-W 10489; J.H. Maiden (9 November 1909), Western Australia, UC 437518 J Eucalyptus frenguelliana (Myrtaceae), MPEF-Pb 2344; Laguna del Hunco, early Eocene of Chubut, Argentina K Cercidiphyllum obtritum (Cercidiphyllaceae), DMNH 25061; Republic, early Eocene of Washington, USA L Cercidiphyllum japonicum Siebold & Zucc. ex J.J.Hoffm. & J.H.Schult.bis (Cercidiphyllaceae), Axelrod cleared leaf 166; UCMP (no other voucher) M Platanus racemosa Nutt. (Platanaceae), NCLC-H 6631; Handel s/n (collected 1985, no other voucher), California, USA N Erlingdorfia montana (compound-leaved Platanaceae), DMNH 7642; Hell Creek Formation, Late Cretaceous of North Dakota, USA. Scale bars: centimeters as labeled (A, B, L, M); 1 cm when not labeled (C–K, N).

The full image dataset and supporting data files are available open-access for download in a single Figshare Plus data collection at https://doi.org/10.25452/figshare.plus.14980698 (hereafter, “the Figshare archive”). The components described below are summarized in Table 1, along with relevant online resources where many of the specimens can already be searched, usually at lower resolution. Although individual linkage of each specimen with online resources would be desirable, it is highly impractical at present because the necessary tags and lookup tables have never been compiled and vetted for most of the collections used here. For readability, we use “leaves” to refer to all specimens discussed here, whether they are leaves, leaflets, or other plant organs that are included in small numbers.

The cleared and x-rayed leaf-image collections included here were chosen for availability of a large number of botanically diverse, high-quality images, accessible voucher data, and open-access re-use permissions. The collections primarily represent non-monocot (“dicot”) angiosperm leaves, with minor representation of monocots, other vascular plant groups, and non-foliar plant organs. Several other large cleared and x-rayed leaf collections exist (see Introduction) but were not used in the dataset presented here for various reasons. For example, the significant cleared-leaf atlas series by Klucking (1986–2003) was manually scanned, cropped, and made into a dataset as part of a machine-learning study (Wilf et al. 2016); however, that dataset is not retained here due to comparatively low resolution, moiré patterns, and other artifacts of printing (Wilf et al. 2016). In addition, the University of California Berkeley collection of over 800 vouchered cleared leaves (distinct from the Axelrod collection) has not been included here because it has not yet been digitized.

A master inventory of the 26,176 images of cleared and x-rayed specimens from >4,500 genera and >17,300 extant species in 354 plant families (Table 1) is provided in the accompanying Figshare archive. A small number of specimens are represented by multiple images, such as close-ups or lighting variants. Taxonomic fields include the family, genus, and species as provided in the respective collection catalog, with additional fields for updated Angiosperm Phylogeny Group (APG) family and order (APG IV 2016). Taxonomic and geographic coverage are uneven, constrained by the general availability of herbarium materials to the creators of the collections (similar issues occur even in recent, large herbarium-image datasets; e.g., de Lutio et al. 2021). Eight families are represented by more than 1,000 images each (Fabaceae, Sapindaceae, Rosaceae, Fagaceae, Annonaceae, Rubiaceae, Ulmaceae, and Malvaceae), whereas 173 families have fewer than ten images apiece. Photographs were taken by many people (see Acknowledgments) over an extended period of time and at different institutions, with a wide variety of cameras and methods that we do not attempt to detail; however, the original camera or scanner EXIF (Exchangeable Image File Format) metadata remain embedded in most of the images and are viewable in standard file browsers. We have also maintained the original pixel resolution and image dimensions in all photographs.

Catalog numbers of cleared or x-rayed leaves in the master inventory (available in the accompanying Figshare archive) refer to a unique glass slide (for the cleared leaves) or a film-negative number (for the x-rays) used to organize the respective collection, as designated by the creator of the collection. The catalog numbers of the cleared and x-rayed leaf collections are usually secondary, i.e., specific to the collection but linked in museum records (as legacy data and thus without hyperlinks) to a primary source voucher at a herbarium (Table 1). Thus, the collection-specific secondary numbers are usually the information needed to search the specimens online (using resources listed in Table 1 and further described below) or in paper catalogs to locate the primary source-voucher data. In some cases there is no herbarium or other voucher besides the mounted slide, and then the curated cleared-leaf specimen, usually specially collected for the purpose, is a primary collection. We also provide in the accompanying Figshare archive a catalog file containing the voucher data for the Wing X-Ray Collection, for which data are not otherwise available online. In publications, specimens should be formally cited by primary voucher as well as secondary catalog number if possible (see Fig. 1).

Family and order updates were done iteratively by first doing automatic lookups to family of the catalog genera and species, using the tables provided in The Plant List (www.theplantlist.org) and its successor, World Flora Online (WFO; www.worldfloraonline.org; Borsch et al. 2020). These resources include standardized lookups to family and order for most generic and species names, modified slightly to include a few taxa not listed in WFO. Failed lookups were flagged and corrected manually. Most lookup failures resulted from typographical errors of generic and species names in the catalog data, and these were manually corrected. Others resulted from genera not being listed or having ambiguous or unverified family status in WFO; these taxa were then manually vetted using other standard resources such as Tropicos (www.tropicos.org), the International Plant Names Index (www.ipni.org), and the Angiosperm Phylogeny Website (www.mobot.org/MOBOT/research/APweb). For consistency, the WFO was designated as the priority lookup for conflicting results among taxonomic databases.

For reference, we note other online resources for batch-vetting plant names that we did not use, including Taxonomic Names Resolution Service (tnrs.iplantcollaborative.org), taxize (github.com/ropensci/taxize), and the Kew Vascular Plant Families and Genera database (data.kew.org/vpfg1992/vascplnt.html). In addition, an automated tool, the WORLDFLORA R package, is now available for batch lookups from the WFO taxonomic backbone file (Kindt 2020), although this would not have resolved the large number of taxa with uncertain status in WFO that required manual vetting.

Due to the intensive labor that would be required to update the large number of names below family level, even with the aid of batch services, and the emphasis here on family-level vetting, generic and species names were for the most part not updated except to correct misspellings that would hinder future lookups. A full vetting below family level would also require manually consulting and hyperlinking all the primary herbarium records to check for new determinations, a process of several years. However, any user can easily find taxa of interest using the specimen list provided (accompanying Figshare archive) and access updated nomenclature and voucher data using the resources listed.

The resulting master inventory of cleared and x-rayed leaves was manually inspected repeatedly to eliminate variant spellings and other inconsistencies, until no more were found. Even after this stage, many issues remained from duplicate and corrupt files, invalid paths, labeling errors, ghost folders of problem images, and other common legacy database errors. Automated and reproducible data analysis and cleaning was done (by J. Rose and R. Saha) largely in Jupyter Notebooks and scripted in Python. In an iterative process, we used the Pandas library to load, sort, and filter the dataset in the form of a table, mapping metadata values in each column to unique specimens in each row. From there, we verified each file path’s full compliance with a pair of requirements, namely that it be both (a) a unique absolute path, and (b) a valid path specifying an existing, uncorrupted image file that can be successfully opened and closed. Rows that failed this test were flagged and taken out for manual review.

Further file path cleaning included the use of a fuzzy matching algorithm, through which all possible matches between a flagged query file path q and a possible near-duplicate reference path f_R were compared by calculating the Levenshtein Distance (e.g., https://xlinux.nist.gov/dads/HTML/Levenshtein.html). This distance serves as a measure of the character-level similarity between two strings, from which all pairs are sorted in order of decreasing similarity to the flagged file q. Several duplicated source files that had evaded detection in previous stages were identified in this way, by manually scanning the top few most similar matches and searching for signs of typos. This procedure for automating the identification of the most likely near-duplicate strings allowed us to automatically verify that none of the tens of thousands of species in thousands of genera, hundreds of families, and dozens of orders included any artificial categories created by a misspelling. An example could be two samples from the same family, where one’s family was spelled “Fabaceae” (correct), whereas the other was accidentally entered as “Fabeceae.” This is an easy typo to miss, but it can skew downstream analyses.

Once all taxonomic and archival fields were validated, we assigned each sample a new filename that accomplishes both (a) directly encoding multiple levels of metadata into human-readable format within the filename, and (b) allowing easy sorting and searching of files on disk, without any additional alterations or struggling with a full relational database. The new filename format is constructed in the form: “Family_Genus_species_Collection_Catalog number”. This user-friendly format facilitates, for the first time, rapid alphabetic sorting, visual inspection, and searching of all the merged images from multiple sources in standard personal-computer windows and visual browsers. In the filenames, as just described, the family is updated to APG standard according to World Flora Online and other resources, whereas the genus and species fields are usually not updated except to correct spelling errors, especially those that could cause lookup failures.

We provide 4,076 vouchered leaf-fossil images of specimens that are assigned to family level, in total covering 44 angiosperm and four non-angiosperm families from a variety of sites in the AmeriCAS that are well known to the authors (Table 1; Appendix 1). Although far from comprehensive, this image set nevertheless covers at least a majority of angiosperm families that are reliably known in the fossil record from nearly-complete leaf remains; it provides a starter set both for comparative learning in angiosperm paleobotany and training machine-learning algorithms.

Unlike the images from the cleared and x-rayed collections, which were not adjusted except for cropping of the x-rays, the fossil-leaf images were all manually and reversibly rotated, close-cropped, and contrast- and temperature-adjusted (all whole-image adjustments, other than cropping) in Adobe Camera Raw so that they are approximately similar in relative frame alignment and overall contrast, with emphasis on making vein features visible (for some photographs taken on early-model digital cameras with barrel distortion in macro mode, the lens distortion was corrected manually using Adobe Camera Raw). This procedure minimizes strong distractors such as rock matrix for machine learning of fossil leaves, an interest of several of the authors (Wilf et al. 2016), and we found that it also enhances human learning for the same reason, by increasing visual comparability of the leaf features and eliminating distractors and variable orientation. In all CASes, we have maintained the full pixel resolution and (post-crop) dimensions of the original image and resaved processed images from Camera Raw to jpeg format only once (usually with tiff format as a lossless intermediate step), using the minimum compression ratio to maintain image quality. A cost of this approach was removal of most of the scale bars. However, nearly all scaling information can be accessed if needed from online (usually much lower resolution but suitable for scaling) versions of the images or sets of uncropped originals that we have included where necessary (see General Collection, below). In addition, all physical voucher specimens can be accessed at their respective repositories. As for the cleared and x-rayed leaves, original camera or scanner EXIF data remain embedded in the image metadata.

The fossil set of 4,076 images is comprised of two parts (Table 1, Appendix 1): first, a concentrated collection of 3,320 images from a single prolific site, the late Eocene Florissant fossil beds of Colorado; and second, a smaller general collection of 756 images from a variety of latest Cretaceous and Paleogene sites in North and South America (Appendix 1; accompanying Figshare archive). Appendix 1 annotates and lists authorities and taxonomic references for the ca. 222 species used in the fossil dataset, and individual catalog numbers are embedded in the filenames of the images. Appendix 1 also lists references for site-specific collections that pertain directly to the specimens used here, if the latter are different from the taxonomic references. Some specimens are represented by multiple images, such as close-ups or lighting variations (but not by duplicate images), and many images of counterparts are included. Although the major target for the collection was “dicot” leaves, images of a few species of monocots, ferns, and conifers that were readily available were included to help seed future expansions. Several generic names that may be botanically doubtful are left in as-published or as-cataloged form (Appendix 1), but all included material is considered reliably placed at family level. Informal morphotypes are included if they have reliable features at family level. Filenames, as for the cleared and x-rayed leaves, embed taxonomy to enable rapid auto-sorting and searching in standard PC windows, followed by collection data.

The dataset presented here consolidates thousands of hours of labor by many people (see Acknowledgments) into a single accessible platform. Due to the extraordinary effort involved, it is unlikely that many new, large-scale cleared and x-rayed leaf collections will ever be assembled and digitally processed. Thus, the future prospects for significantly increasing the overall sample size and improving the coverage of taxonomy and geography in digital leaf-reference collections most likely lie elsewhere. The greatest potential appears to come from the advancing techniques for segmenting and enhancing leaf images from the enormous, widely available resource of digitized herbarium sheets (Hussein et al. 2020; Weaver et al. 2020), which have the significant additional advantage of direct linkage to the global data infrastructure for biodiversity (e.g., Bakker et al. 2020). To reach comparability with cleared leaves, segmented leaf images will require high pixel resolution, optimized contrast for the capture of venation details, and the careful retention of significant edge features such as the leaf margin. For leaf fossils, increasing the sample size of well-identified specimens is straightforward in principle but will require efforts far beyond the resources of a single collaboration. Thus, we plan a community initiative for this purpose.

We look forward to seeing the assembled image dataset catalyze advances in research, education, and outreach. The images and supporting data are available open-access on Figshare Plus at https://doi.org/10.25452/figshare.plus.14980698. Some mistakes are inevitable in a first-version database of this nature; please report any errors observed to the corresponding author. Corrections and updates may be applied to the Figshare archive under new version numbers; the version precisely corresponding to this article will remain preserved as version 1.0.

Funding for this work came from NSF grants EAR-1925755, EAR-1925481, and EAR-1925552 (PW, TS, MAG, and others); DEB-1556666 and DEB-1556136 (PW, MAG, and others); and the National Park Service (HWM).

Many researchers, staff, students, and volunteers contributed to the development of the collections aggregated here over many years. These include the investigators named in the manuscript, the collectors and field crews who made the primary collections around the world, and the collections staff, technicians, and volunteers at the numerous involved herbaria and fossil repositories. We further acknowledge the following, with apologies for any missing names. Assistance in the original assembly of the NCLC-W cleared-leaf collection, including performance of most specimen selection, registration, and leaf clearing and mounting: Sandy Wilson, Robyn Burnham, Russell O’Connell, H. Meyer, and others. Databasing and photography of NCLC-W at NMNH: Dane Miller, Ian Tom, Stephanie Bailey, and many volunteers at the Smithsonian Institution. Photography and curatorial support for NCLC-H at YPM: Whitney Barlow, Donna Beeson, Alyssa Cheung, Serra Vidinli Dedeoglu, Larry Gall, Michelle Garcia, Gabriela Gonzalez, William Guth, Zoe Kitchel, Linda Klise, Philip Kuchuk, Joanna Liu, Ivette Lopez, Steven Mordarski, Sally Palatto, Paul Pena, John Petrucelli, Ornella Rossi, Carl Russell, Jared Shayne, Harry Shyket, Robert Swerling, Cecilia Tenorio, Tim White. Sampling and imaging support for the Wing x-ray collection: Hazel Beehler, Keith Boi, Lynn Gillespie, Scott Krueger, David and Susan Rosen. Florissant fossil photography: Ashley Ferguson, Kelly Hattori, several other Geoscientist-in-the-Parks (GIP) interns, and Conni O’Connor. Additional fossil photography: Bárbara Cariglino, CASsi Knight, Lisa Merkhofer, Dana Royer. Cerrejón and Bogotá (SGC-ICP) support: Carlos Jaramillo and Mónica R. Carvalho. Additional database support: Sarah Allen, Sven Eberhardt, Ana Van Gulick, Jenny Kissell, Thao Nguyen, Edward Spagnuolo, Alysa Young. Museum curators and staff (other than the authors) who provided collections support and re-use permissions for images of fossils: Patricia Coorough Burke, Milwaukee Public Museum; Thomas Demere, San Diego Natural History Museum; Peta Hayes, Natural History Museum, London; Kathy Hollis and Jon Wingerath, NMNH; Ashley Klymiuk, Field Museum; Andrew Knoll and Michaela Schmull, Harvard University Herbaria; Kristen MacKenzie and Ian Miller, Denver Museum of Nature & Science; Steven Manchester and Hongshan Wang, University of Florida; Ruth O’Leary, American Museum of Natural History; Andrew Ross, National Museums Scotland. We thank Steven Manchester, Patrick Herendeen, Susanne Renner, one anonymous reviewer, and Editor Sandra Knapp for their helpful comments that improved the manuscript.