GEOREFERENCING OF THE MOSCOW UNIVERSITY HERBARIUM (MW) DATASET IS COMPLETED

Ten years ago we started scanning the Moscow University Herbarium collections (Seregin, 2016). Since 2016, we have been publishing them online on the Moscow Digital Herbarium website (https://plant.depo.msu.ru/). Later, we launched a GBIF-mediated dataset to promote large-scale access to the collections (https://doi.org/10.15468/cpnhcc). As of July 2025, physical collections of the Moscow University Herbarium include 1,137,501 specimens making it the second-largest herbarium in Russia after Komarov Institute in St. Petersburg and the 60th-largest herbarium in the world (Thiers, 2025). Specimen imaging in the MW Herbarium was completed at the end of 2018. All scans were published online with key metadata (i.e., “filed as” name, currently accepted name, one of 60 geographic area codes, country of origin, collection date, and primary collector). However, it was just the beginning of our long-term goal to digitize complete label data and georeference the collection, which was our most challenging task (Seregin, 2018). The MW Herbarium dataset is part of the Moscow Digital Herbarium online-consortium. This website is used by eleven Russian herbaria as a freely available product maintained by the Moscow State University. Besides MW (1,085,763 digitized specimens / 811,338 georeferences), these include MHA (183,479 / 131,959), IRKU (42,297 / 35,729), KUZ (20,136 / 20,129), TUL (10,098 / 10,089), TULGU (5054 / 5047), KULPOL (3657 / 3657), MAG (2604 / 2541), TKM (2967 / 2930), SMR (1796 / 638), KRSU (1268 / 31). Twenty herbaria confirmed their intention to join the consortium as new members in the near future. As of 25 July 2025, there are 1,085,763 digitized specimens in the Moscow University Herbarium of which 811,338 specimens, or 75%, are georeferenced. These georeferenced specimens are only from the MW Herbarium and do not include specimens from the other consortium members. At the moment, we have three main sources of geodata in the database including: (1) coordinates available from labels; (2) manual georeferences made by operators; (3) automatic georeferences based upon the ISTRA algorithm (Seregin & al., 2023); and also (4) specimens which do not have sufficient information to generate georeferenced data. (1) Coordinates available from labels (151,743 records, or 19%). Labels from specimens collected in the 21st century usually contain geographic coordinates of the collection site. They are entered in a tabular form by label data operators. Since December 2022, such input is carried out by volunteers in the “Pomogator” module (https://plant.depo.msu.ru/pomogator) with automatic error control based upon double independent data entry. The proportion of collections in the MW Herbarium that directly indicate the coordinates of the collection site is steadily increasing. In 2000, when portable GPS receivers were just beginning to appear, 39.5% of collected samples had coordinates included in the specimen labels; in 2005, this figure was already 55.2%; in 2010, this proportion had grown to 60.3%; and by 2015, it had reached 78.0%, remaining at approximately the same level until very recently. Further check of the coordinates taken directly from the labels to confirm they are accurate is carried out. (2) Manual georeferencing by operators (i.e., staff members and part time staff) for a specific specimen (or a group of specimens with the same label text) includes 303,973 records (37%). Manual georeferencing is the establishment of the collection site of a specimen using historical and modern cartographic and reference materials that are widely available in the public domain on the Internet. In addition, we establish georeferencing accuracy for each manual georeference. Our standards and protocols comply with general methodological recommendations (Chapman & Wieczorek, 2020). To carry out this stage, the key initial data are, as a rule, not derived from digital scans, but label transcriptions. Moscow Digital Herbarium uses a strict protocol for transcribing label data, which allows for unification of the entry procedures by different operators with minimal discrepancies. For example, a single text field “All geography and ecology (label text)” is used, while country name and administrative units are usually set later automatically based on the coordinates. Georeferencing operators work with spreadsheets of label transcriptions. This allows for quick identification of repeating toponyms and understanding of routes of specific collectors, which reduces input time and improves the accuracy of positioning of closely located points. If the work is carried out in a spreadsheet editor, then at different stages the array is sorted by various parameters. Finally, operators of georeferencing often perform final editing (cleaning) of the label transcriptions. (3) Georeferencing established automatically based upon the ISTRA algorithm (341,760 records, or 42%). The ISTRA algorithm (Intellectual System of Toponymic Reading and Attribution) was developed by Oleg Platko, programmer of the Skulachev Project and architect and chief developer of the Moscow Digital Herbarium. Today, ISTRA combines all specimens from the database into groups by the “collector – date” pair with additional normalization by geographical areas. One specimen from the group is sent for manual georeferencing, and all specimens collected by the collector during one day are linked to the same collection site. Automatically georeferenced specimens receive a special mark in the database, and their collection site can gradually be specified through manual georeferencing. To ensure the operation of the ISTRA algorithm, a commercial partner carried out a mass input of the “collector – date” pair for 80% of the specimens of the MW Herbarium in July 2018. This data was entered for all specimens in which the collection date and the collector's name were indicated unambiguously and completely. The next day, using the ISTRA algorithm resulted in the automatic addition of ca. 63,000 specimens and the number of georeferences suddenly raised up to 217,971 records. At this moment, approximately 25% of the total specimens present in the database received coordinates. The ISTRA system can handle all collections in the Moscow Digital Herbarium consortium. This data network includes the original collections of each herbarium and duplicates coming from other collections. Any georeferenced point entered into the system by any consortium participant updates the database of reference geodata required for the operation of the ISTRA system. Tools and protocols for automatic and semi-automatic error search and correction (both within the platform and in GBIF) help maintain data cleaning. As new data are added and the database grows, the ISTRA system refines its georeference coordinates for specimens that have already been georeferenced earlier by ISTRA. For example, label capturing enables more detailed grouping of specimens based upon precise locality descriptions. (4) In addition, there is a relatively small group of specimens for which it is impossible to establish a georeference. This is due to the lack of the locality description or only the most general geographical indications like “Caucasus”, “Siberia”, “Kamchatka”, etc. We encountered 23,174 specimens in this category. ISTRA inspects this group once a week and linking specimens them with georeferenced records with the same date and collector. We estimate that most of the 274,425 specimens which are not yet georeferenced in the MW Herbarium will not be easily georeferenced using ISTRA or manual input. For instance, many old specimens are lacking locality names, complete date, indication of collector or collector team, etc. In many cases, we failed to find coordinates for some expeditions into uninhabited areas due to lack of archives and itineraries. As of July 2025, the Moscow University Herbarium dataset includes 811,338 georeferenced specimens, making it the seventh-largest set of freely available e-geodata associated with plant specimens from a single herbarium collection after Missouri (MO, 2,701,965 georeferenced specimens), Naturalis (L and others, ca. 2,058,355), New York (NY, ca. 1,905,840), Lund (LD, ca. 1,038,651), Canberra (CANB, ca. 835,916), and Melbourne (MEL, ca. 832,738). Georeferences from the Beijing Herbarium (PE, ca. 1,049,845) are obscured and associated with county administrative centres due to limitations on geodata accuracy introduced in the Chinese national legislation (Fig. 1). The Botanical Database of Southern Africa (BODATSA consortium) published ca. 1,125,924 georeferences, but the Pretoria Herbarium (PRE) as the largest contributor gives ca. 801,246 records only. The current phase and bulk of georeferencing for the MW Herbarium has been completed. Further intensive georeferencing is going to be highly time-consuming, since collections with traceable exact localities are already done. Expansion of geodata will be mostly based upon new accessions with available GPS-coordinates with an average rate of new acquisitions ca. 15,000 specimens per year. Geographically, our data cover large areas of temperate Eurasia with a focus on Russia, the former Soviet Union, and Mongolia, that have long been poorly represented in GBIF (Fig. 1). This makes spatial data on the world's plants more uniform and will help to avoid bias in global biodiversity research and other areas that have been actively developing recently (Soltis, 2017). The reported project is currently supported by a grant from the Russian Science Foundation (project No. 21-77-20042, https://rscf.ru/en/project/21-77-20042/).