The recent growth in herbarium-based phenological research is arguably a product of the growing interest in climate change and phenology around the turn of 21st century . Researchers realized that herbarium specimens could potentially be used to detect and quantify long-term phenological shifts in response to climate change. This, in turn, led to the use of specimens to estimate phenological sensitivity to various environmental factors, including temperature, day length, and precipitation . To date, specimens have been used to estimate the onset of several phenophases, including first flowering, peak flowering, leaf-out, and fruit set, as well as the duration of entire growth phases. These phenophase estimates have been used to study long-term shifts in phenology and phenological sensitivity to interannual climate variation . A literature review focused on the modern use of herbarium specimens to study phenological responses to climate reveals interesting generalities and insights. First, drainage planter pot studies that have investigated longterm shifts in phenology have generally found that flowering and leaf-out times have advanced, in some cases dramatically, over the past century .
These long-term trends are often in agreement with studies that have used alternative sources, such as observational data, to study phenological shifts. Second, for most of the studies we reviewed the onset dates of spring flowering and leaf-out tended to be negatively associated with winter or spring temperatures ; that is, plants tended to flower and leaf-out earlier in warmer years. However, some species and regions exhibit delayed or mixed phenological responses under warmer temperatures, potentially because they did not experience sufficient winter chilling requirements or because the imprint of past climate conditions has resulted in a response lag. Third, given the span of time and geographic area that specimens encompass, they almost always capture a greater range of climatic variation experienced by a species than traditional long-term observational data, and thus can provide a more complete estimate of phenological shifts over time as well as phenological sensitivity to interannual or spatial variation in climate . Most studies that have used herbarium specimens have focused on a single phenological event, most commonly the date of onset for a single phenophase .
The most frequently studied phenophase in relation to climate change is flowering , with a specific focus on either mean flowering date or peak flowering date . Only a handful of studies have attempted to quantify different events within a phenophase, such as the onset, peak, and end of flowering date. Thus, the opportunities for expanded application of comparable and new techniques are abundant. For example, specimens can be used to assess multiple phenological characters at different stages of development , allowing researchers to estimate the sensitivity of different points in a given phenophase as well as determine how different phenophases are related. Additionally, most herbarium-based studies have been limited to northern, temperate biomes , mirroring geographic biases in long-term observational data. The potential to expand phenological investigation into nontemperate biomes using specimens, however, is considerable, as illustrated by the density of tropical and subtropical specimen records in the Integrated Digitized Biocollections database alone . Several recent studies have validated herbarium phenological estimates by comparing them with independent estimates of similar phenological phenomena . Generally, comparisons with independent phenological data – using photographs and field observations show that herbarium-based estimates of both phenological timing and phenological sensitivity to climate are reliable . At a broader scale, additional validation of herbarium-based phenological data has come from comparisons with satellite observations of ‘green up’ .
While these studies provide important validation of herbarium-based phenological data, they are nonetheless limited in their phylogenetic scope and number of regional comparisons. As the use of herbarium-based phenological data grows, so too should efforts to independently validate these data.Herbarium-based data, like all sources of data, are subject to potential biases and limitations of which researchers must be aware. Such limitations are present from the specimen collection phase, to the digitization and processing of specimens, to the analysis and interpretation of specimen data. By understanding and addressing these challenges, researchers can make full and appropriate use of specimens for phenological research. Some limitations of using herbarium data for phenology are common to other observational datasets and originate at the time of specimen collection, including accurate species identification and phenological event and phase discrimination. While specimens are often correctly identified by experienced botanists, they may still be misidentified or labeled according to outdated taxonomy. Unlike with observational datasets, however, species and phenophase identifications for herbarium data can easily be confirmed by revisiting anomalous specimens.Herbarium data are known to contain additional, unique biases that stem from the opportunistic nature of their collection. Botanists often collect samples depending on their interests, schedule, and location and not to capture the phenological status of the plant perse. Collection biases relating to plant habit, morphology, and nativity may also occur in herbarium datasets; for example, Schmidt-Lebuhn et al. discovered strong biases against very small plants, plants with brown or green inflorescences, and introduced species in a sample of Australian Asteraceae. Rich and Woodruff noted that collections are biased towards common, showy plants that grow in clumps. Additionally, broader taxonomic, spatial, and temporal biases have been identified with Global Biodiversity Information Facility occurrence records, which include herbarium records . Specific to phenology, plants may be less likely to be collected at the very beginning or end of a reproductive season, especially if a species is difficult to identify during these stages or is inconspicuous. For example, Davis et al. found that first-flowering date estimates from specimens were, on average, 3 days later than first-flowering date estimates from field observations. Botanists may also collect only those individuals exhibiting a certain phenological stage to facilitate identification. However, it is also true that botanists may deliberately collect plants that are flowering or fruiting out of season and are therefore not representative of the overall phenology of the species. Another source of collection bias is the tendency for large numbers of specimens to be collected during single collecting trips, which can result in oversampling and the generation of duplicate specimens distributed to multiple institutions that are subsequently treated as independent samples. Duplication of records is a well-known problem, however, and efforts are currently underway to better account for duplicate records across databases and data portals. Finally, herbarium specimens often represent only a fragment of an entire plant , plant pot with drainage which makes it important to consider how accurately specimens represent the phenology of the whole plant or local population from which they are sampled.Data quality issues in herbarium data may also arise after collection, during label transcription, or due to digitization. For example, ambiguous handwriting or descriptions can lead to the incorrect transcription of a specimen’s location or collection date. In addition to transcriptionerrors, discriminating among phenophases can be even more difficult if observers are assessing digital images rather than the physical specimens themselves.
While these problems can often be resolved from other contextual clues , each of these aspects of data quality must be assessed and managed when studying phenology. Moreover, different countries and individuals have developed separate methods for recording specimen information, which presents a challenge for data aggregation. This topic has recently received renewed attention and methods to improve the standardization and integration of these data are currently being developed .Clearly, herbarium records are subject to error, as are all sources of data, and may contain geographic, phylogenetic, temporal, or other biases because they were not assembled to answer phenological questions. Nevertheless, one of the strengths of herbarium data is that their biases can be minimized by careful selection of species and phenological phases for assessment, rigorous training of observers, high-quality imaging, and the continued development of statistical methods to test and correct for biases.In the phenological studies that have been completed to date , researchers often evaluated phenological stages differently according to their research priorities and rarely madedata publicly available, thus limiting the utility of those data beyond the life of the individual projects. The most serious challenge for the future of herbarium-based phenological research is the standardization of phenological terms and methods for scoring phenophases and phenological events. Such standardization is important not only to ensure that herbarium-based studies are comparable but also to facilitate effective integration with other types of phenological data such as citizen science observations, satellite imagery, and stationary camera images. Biodiversity data standards for the biocollections community have already been established in the Darwin Core Data Standards. Most digitizing institutions generate data conforming to the Darwin Core, which comprises defined metadata properties and a small set of classes; however, phenological terms are not currently defined by the Darwin Core and instead are captured in unrelated fields such as ‘occurrenceRemarks’, ‘organismRemarks’, ‘dynamicProperties’, or ‘fieldNotes’. Many institutions capture flowering information in the ‘reproductiveCondition’ field, but this field lacks a standardized vocabulary. For example, we discovered 3900 unique terms to describe reproductive status in a search of the ‘reproductive Condition’ field of 5.7 million specimens in SEINet, a portal of digitized specimens for Arizona and New Mexico, USA. Lack of standardization complicates data integration and presents a huge obstacle to the mobilization and consolidation of herbarium data from multiple institutions for phenological research. The development of standards and ontologies is a vital step toward unlocking the research potential of digitized specimens. Standardization of herbarium specimen data, in combination with the availability of new data management tools, will facilitate the large-scale collection and use of phenological data from specimens. The task of scoring phenological data from millions of digitized specimens, however, is a monumental task. As noted above, herbarium-based phenological studies to date have typically focused on only a single phenophase and classified specimens in binary terms . This limited approach is due in no small part to the challenge of scoring phenology for a large number of specimens. Standardization can facilitate the collection of these data in two ways: by providing a template for scoring phenology that can be easily incorporated into the digitization or post-digitization workflow; and by providing guidelines for the conversion of raw count data collected via citizen science crowd sourcing into predefined phenophases.Efforts to scale up the collection of phenological data using new tools are already under way and would only benefit from the incorporation of a standardized ontology and data structure. The New England Vascular Plant project, for instance, has developed an extension of the specimen management system Symbiota that provides an interactive online platform to score a range of predefined phenophases based on coarse estimates of different phenological characteristics . This approach has the advantage of speed and efficiency and can be easily incorporated into an existing digitization pipeline where, along with transcribing the label information, technicians input phenological scores. Another tool, similarly meant to be implemented within an existing collection database, is the Phenological Predictability Index module in the Botanical Research and Herbarium Management System. The PPI module, however, is geared more toward standardizing estimates of phenological activity as opposed to scaling the collection of the data itself. Another avenue for scaling phenological data collection is the use of citizen science crowdsourcing. The popular citizen science platform Zooniverse has utilized crowdsourcing in the collection of data from digital specimens including label transcription [Notes from Nature ] and even phenological data [Orchid Observers ]. Another crowd sourcing tool that has been developed to collect phenological data from specimens is CrowdCurio. Preliminary results from CrowdCurio have demonstrated that phenological data collected from non-expert users are comparable to those compiled by expert users, suggesting that it has the potential to be a powerful tool for the collection of detailed, accurate phenological data. In addition to crowd sourcing, machine learning – the ability of computers to learn a task without being specifically programmed – offers an exciting new tool for the collection of large amounts of phenological data from specimens. Several recent studies have demonstrated that machine learning can be used to identify species with a high degree of accuracy based on leaf shape and venation. In either case data collected with these new and powerful tools should be made to conform to standardization efforts so that they can be easily incorporated into existing herbarium databases.