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

generation [23,53–55]. Our results indicate that reliable phenological data can also be gener- ated from photographic sources. Although not all the records reviewed by the volunteers were reviewed by the reviewers, a situation that entails a potential weakness, we believe that this could be partially corrected due to the large number of reviewers, the large number of observa- tions, and the random distribution of the observations that we have. Similar results have been found in other platforms [31,56] such as CrowdCurio, a crowdsourcing-type image annotation tool that involves the participation of financially rewarded, but not necessarily self-interested, non-expert workers who classify photographs for the purposes of some particular project [56]. The use of CrowdCurio also had comparable results between expert and non-experts when classifying photographic inputs for the identification of phenophases in digitized images from herbarium records. The Naturalista platform has the means to upload non mandatory phenological data at the time of recording an observation, but only about 30% of the photographs of both species reviewed here had a phenological annotation (data from www.naturalista.mx). The high pro- portion of observations classified as uncertain (Dk) in the group of untrained volunteers sug- gests that this may be due to the difficulties faced by citizen scientists without training in phenology and plant structures when identifying phenophases, rather than lack of interest,

since previous papers have reported that intrinsic motivations, such as interest to generate and share knowledge, are the main motivations for the participation of citizen scientists [57]. Feed- back training processes, such as the two-way conversations used in this research [58,59], help citizen science volunteers generate quality information, and professional scientists learned from the needs of the raters to expand the strategies for proper communication between spe- cialized an unspecialized audiences [60]. Our analysis of highlights the benefits of this approach and provides more evidence to sup- port the use of a promising alternative to the limitations posed by specific systems to track phe- nological data, while also reducing the time and effort required to obtain data [61]. Since this approach does not allow volunteers to corroborate the presence/absence of a particular pheno- phase, as would happen when rating a live individual, it is paramount to have high-quality images in which the phenophases can be clearly distinguished. Ideally, increased accuracy would be obtained from a series of high- resolution images depicting several portions of the individual, especially with species that have easily identifiable phenophases such as colorful shaped reproductive structures. The application of this methodology to species with less visible structures, as is the case of many grasses whose phenological study is equally important, would be cumbersome to the untrained eye, and would require longer and more detailed training.

Furthermore, we want to highlight the fact that photographs of species with inconspicuous structures may become impossible to classify with poor-resolution images, by people without extensive and detailed knowledge about the species, which is the main limitation of using this method followed for all phenophases. In addition to the obvious taxonomic bias that this entails, it is important to consider other biases into which one could fall, which are also typical of scientific collections. These include biases in taxonomic (such as dominant species in landscapes or preference for certain taxa), morphological (biases towards conspicuous individuals, etc.), geographic [62–64], and tempo- ral preferences [65]. On the other hand, the use of photographs from biodiversity observation platforms presents important advantages. These photographs are a verification tool that phe- nology platforms do not normally provide, and their opportunistic nature gives observers ease and freedom in data capture, leading to a huge amount of data with geographic and taxonomic coverage that can exceed those of the phenological monitoring platforms [41]. In addition, the presence of photographs makes it easier for researchers to corroborate the identity of the spe- cies if they consider it necessary [24,41]. The photographs also allow researchers to carry out a 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 11 / 17