Practice makes the expert: The importance of training volunteers in the generation of phenological data from photographs of biodiversity observation platforms

validation when the information obtained is doubtful, for example, when phenological events are identified outside the expected dates or in geographic regions that do not correspond to the known distribution area of the species, or when volunteers have doubts when making a phenological annotation. The Dk category in the classification of phenophases that was fol- lowed in this study was used to indicate photographs or phenophases that were blurred, that did not correspond to the species, or that showed evidence of being cultured. However, this scoring can also allow experts to identify if a photograph requires extra peer validation, if phe- nophase represents a particular difficulty in scoring, or if a more detailed training session is necessary. This category also allows doubtful data to be excluded automatically in a subsequent phenological analysis. Two of key components of monitoring phenology monitoring are the phenological state and the intensity of the phenophase. The phenological state, defined as presence or absence of a phenophase, provides information on its onset and duration, while the intensity provides information on the temporal distribution of the structures that characterize it [35]. We do not recommend using the method proposed here to calculate population-level phenophase inten- sity measures, such as detailed phenological peaks, as there may be biases in the availability of photographs that do not allow analysis of phenology at fine temporal resolution [66], as well as

limitations in the reliability of volunteers when performing more complex measurements, such as abundance or frequency of species or structures [51,52]. In addition, the low propor- tion of available observations with whole plants could generate an erroneous estimate of the start or end date of the phenophases, since structures present in a given individual have not been photographed. These incomplete records could be generate false negatives when they report the absence of structures [41]. Our experience suggests that these types of records, for example, incomplete plants without reproductive structures, are a small proportion of total observations, since collectors, expert or not, tend to focus on the most conspicuous structures, such as fruits, flowers, or buds [67–69]. We suggest that the methodology proposed here would be helpful when used to obtain species-level phenological estimates, including mean dates of each phenophase, or comparisons between the temporal distributions of phenophases between species or different geographic areas [51,52]. Ideally, phenological annotations should be conducted by citizen scientists at the time of recording their photographic observation. However, due to the large amount of information without phenological annotations to date, a posteriori classification of photographs was per- formed (such as the one in this study carried out on photographs taken by citizen scientists), can help expand the use of primary data. Approaching citizen science with a big data approach

would entail taking advantage of emerging data that have not been generated under the target of interest, but can nonetheless be useful in phenological or climate change studies [70]. Our study found that this approach must be underpinned by training provided to volunteers or cit- izen scientists assessing phenological data in order to have reliable data. Several training tools can be used to increase the levels of accuracy as part of the training protocols, including infor- mative talks, interactive forums, short videos, image collections or illustrated guides. The large amounts of phenological information generated through this approach can also be matched against herbarium records [67], and would represent, over time, a temporal con- tinuum of data on the presence of species and the corresponding phenology. Finally, the appli- cability of data generated through this approach ranges from use in environmental suitability analyses [37–40], documenting large-scale phenological changes (such as in the phenological response of invasive species to new environments and climates) [6], possible changes in the nutrient cycles of invaded ecosystems due to changes in leaf phenology [7], and the phenologi- cal response of plant species to GCC [12]. PLOS ONE The training of volunteers as a promoter of phenological studies PLOS ONE | https://doi.org/10.1371/journal.pone.0282750 March 7, 2023 12 / 17