2.1 Selection of species, populations and habitats of interest

Three groups of plants of interest were defined. We first identified conservation priority plants following regional, national or international official lists. This group includes threatened species (critically endangered, in extinction risk, endangered, vulnerable and sensitive to habitat perturbation), as well as those listed in Annexes II and IV of the Habitats Directive of the European Union (full list of plants in Appendix S1 in Supporting Information). Since our Institution holds an important herbarium, we have extensive and detailed information on plant distributions in the region. Common plants co-occurring with priority ones were also monitored when possible, to control for possible habitat deterioration affecting all species in the study area irrespective of conservation status, and because they contribute greatly to overall richness. A second group of target plants was selected according to their singularity or regional rarity. These included endemics of the Pyrenean range or of a small part of the Iberian Peninsula, and species occurring at their distribution range limit within the study area. All these can be considered of interest but not of conservation concern. Finally, a third group of species included indicator plants of a variety of non-forested habitats of the Habitats Directive (Annex I) that cannot be monitored through remote sensing, such as endorheic lakes, bogs, cliffs, shrublands in the subalpine belt or on gypsum soil in the lowlands, and highly diverse (sub)alpine grasslands. Because of their ecological specificity, many of them can be considered good indicators of climatic change (e.g. when very dependent of humidity) or land-use change (e.g. when occurring in habitats that are under slow transformation).

Specific monitoring sites (‘monitoring units’, hereafter MUs) were selected according to expert knowledge, the conservation status of particular plant populations (e.g. under some specific threat), terrain accessibility, and in some cases, participants’ preferences (e.g. within a protected area, or proximity to their residence). MUs (N = 213 in total), range from 2 to 4 km², typically contain one or two species of interest (range: 1–12 plants), are distributed along a very large altitudinal gradient (139–2,837 m a.s.l.) and are often located within the European Natura 2000 network (69%; Figure 1).

2.2 Monitoring protocol

Table 1 shows the general rules to select appropriate plant abundance estimation methods (counts, presence or plant cover) and type of permanent sampling units. Such units go from macro-plots of hundreds of m² to micro-plots of 20 × 20 cm evenly distributed along transects (usually 10–50 meters long), or 10 × 10 cm miniquadrats in grids. For cliff-dwelling species, we use ‘virtual plots’ where individual plants are counted, by drawing fixed areas on good-quality digital photos that are surveyed every year by eye or aided by binoculars (see the bottom left picture in Figure 1). For very small and uncountable plants forming very small population, we estimate the gain or loss of occupancy area by comparing pictures through time. For count-based surveys in medium to large populations (~74% of monitored populations, Table 1), we add sampling units until a minimum of 300 plant units are counted. More than 1,000 records are usually used for presence surveys (~21% of monitored populations), and hundreds for plant cover surveys (4%) in highly replicated micro-plots along transects. To minimize observation error, we adjust the size of sampling units to the plant size and density, to include less than 30 plant units per plot if possible. For presence surveys, the number and distance between micro-plots along transects are adjusted for total records to fall within the range (20%–80%) of presences so that small changes can be detected between years. Sampling design (number of plots or transects and their distribution in space, size and distance of micro-plots along transects, etc.) is usually conducted in situ by an experienced scientist in agreement with the team of volunteers or rangers. This is because the design must be guided by two principles: (a) being as simple and effective as possible so that participants will be able to complete surveys on their own in just one working day, for a minimum period of 10 years and (b) being scientifically robust, to produce trustable indices based on the data gathered. Nevertheless, new MUs are established sometimes by participants, who must decide the sampling design on the go following suggestions of Table 1, or with our remote assistance. Anyway, the final sampling method can be adjusted (increasing or reducing sampling effort) from the second year. Each sampling design is thus unique and specific to each MU in terms or number of sampling units, micro-plots along transects, etc., because of such adjustment. A variety of examples and suggestions to set the sampling method are provided in García, Sanz, et al. (2019).

2.3 Statistical analyses

Data gathered were analysed to look for three signals of weakness or bad performance: small populations, and negative or highly fluctuating trends. Given that species differ in their conservation status, our primary hypotheses were that priority species are more prone to show them, or suffer more threats.

Since population dynamics might vary across years and we are interested in overall and comparative results, we cleaned the dataset prior to analysis. We first removed MUs with less than 3 years of surveys and/or extremely low population sizes (<20 individuals), as these might be severely affected by demographic stochasticity. We also excluded species monitored by counting only flowering individuals (flowering does not necessarily correlates with the dynamics of the whole population) and annual plants (their dynamics are extremely dependent on stochastic climatic factors and therefore require very long-term series, not yet available). Our final dataset for the analysis of trends and fluctuations contained 242 plant populations of 150 taxa, and a total of 942 transitions (annual changes between consecutive years; range: 3–10 surveys per MU; median: 4).

We first analysed the association between population sizes and priority status with a contingency table and Chi-squared test (see also Appendix S4 for the same analysis but using random samples of same size). To test whether priority species are more prone to decline or fluctuate than non-priority ones, we fitted linear models with PAch and PAch_sd as response variables, separately. Residuals of both linear models showed non-constant spread of variance, thus indicating violation of the homogeneity assumption (Appendix S5 and Figures S1 and S2). By plotting residuals against each explanatory variable of the models, we found differences in the spread of residuals between levels of priority, as well as across levels of population sizes (Figures S1 and S2). As data transformation was not sufficient to overcome heterogeneity problems, we fitted a generalized least squares (GLS) model with a residual variance structure (Zuur et al., 2009). Along with the two categorical variables (priority status and population size) and their interaction, we included in our initial model three other covariates: data type (counts/presence/plant cover; included to account for the possible effect of the method used to estimate abundance), number of transitions over which the response variables were averaged (to account for the robustness of the mean trends and temporal fluctuations, and the possible ‘learning’ effect of participants if they became more skilful the longer the temporal series was) and the elevation of MUs (to account for the fact that species tend to show more stable dynamics at higher elevations; Morris et al., 2008). In preliminary analyses, we also explored the possible dependency among observations of the same habitat using habitat type as a random intercept in a linear mixed model with the same fixed components. We selected a GLS model over a mixed model because dependency between observations was negligible between different habitats and sites (intra-class correlations were both close to 0). Following the protocol outlined by Zuur et al. (2009), we started with a saturated GLS model without variance structure. This model was then compared to models with different variance structures using Akaike's information criterion (AIC; Table S1). For covariate selection, we compared nested models fitted by Maximum Likelihood (ML) estimation using log-likelihood ratio tests (Zuur et al., 2009). The final model was refitted using Restricted Maximum Likelihood (REML) to reduce bias in the estimation of the variance components. The same procedure was repeated with temporal fluctuations or PAch_sd (Figure S2). Models were fitted using package nlme (Pinheiro et al., 2020) in R v.3.6.3 (R Core Team, 2020). Pairwise comparisons between levels of population size were done using the post-hoc Tukey correction with R package multicomp (Hothorn et al., 2008). Finally, given that eight MUs of priority species also contained co-occurring common plants, we tested whether they differed in PAch using a paired t-test after checking that normality and homogeneity of variance assumptions were met.

The response variables were not phylogenetically structured (Appendix S6), and hence there was no need for controlling for the residuals being potentially non-independent due to the shared evolutionary history of monitored taxa (Felsenstein, 1985).

4.1 What makes the ‘Adopt a plant’ a unique Citizen Science programme?

Citizen Science programmes dealing with plants are frequent, but they often collect unstructured data such as local records, systematic monitoring of presence/absence (e.g. Martin et al., 2019; Pescott et al., 2015) or overall abundance for populations of rare and threatened plants (e.g. Barnard et al., 2017). Our AP programme collects structured data instead, following rigorous sampling protocols designed ad hoc for each target plant species. It is also unique in focusing on the dynamics of local populations, which we combine with population sizes and threats. Since the same team of participants is responsible for tracking the abundance of target plants year after year in each MU, the potential bias in data collection is minimized.

The two indicators of population extinction risk estimated from such information (abundance trend and temporal fluctuations), along with population size and distribution ranges, provide baseline information for the IUCN evaluation process and population viability analyses (Morris et al., 2002). These variables have been selected by Pereira et al. (2013) to detect environmental changes (Haase et al., 2018) and defined as an Essential Biodiversity Variables (EBV) to prioritize, standardize and facilitate comparisons. Hence, our methodological proposal is enabling us to compile the first extensive and well-structured dataset of these EBVs, of high value for conservation challenges.

Another important feature of our AP programme is that obtained trends are disseminated to participants on a regular basis in annual meetings, and to the general public through the webpage of the project (https://biodiversidadipe.csic.es/ciencia_ciudadana.html). They are also transferred to conservation managers for them to know the dynamics of priority plants so that they take actions if necessary, to avoid declines or extinctions, and fulfil national and international mandatory requirements.

The goal of many biodiversity-oriented CS programmes is to record species occurrences. This approach is becoming very popular thanks to the increasing possibilities offered by technological devices, software and platforms for the collection and exchange of biodiversity data (smartphones, apps, GBIF). These data have contributed to the development of distribution models in the last decade. However, short-term predictions of local abundance with real data are at least as important as modelling the long-term future distribution of species in hypothetical environmental scenarios (Ehrlén & Morris, 2015). A robust estimation of local abundance trends requires sampling designs adapted to each particular population, systematic fieldwork, and mechanisms to reduce and account for observation error. However, these types of designs are very scarce even among studies of endangered species (Bayraktarov et al., 2019; Morris et al., 2002). In fact, there are many demographic studies of plants based on structured matrix models, but they span short periods (<5 years; Salguero-Gómez et al., 2015) probably because of the high effort required both in the field and subsequent model analysis, as well as the lack of long-term stable funding. Methods proposed for the AP programme are adapted to people with no previous expertise, and generate data for robust estimation of future trajectories and to evaluate extinction risk (Elderd et al., 2003). Our methodological proposal, nevertheless, needs an important scientific investment, particularly at the onset of each monitoring (choosing the population of the target plant, setting the specific sampling designs, personal training of participants, data validation and mentoring over years; García, Silva, et al., 2019). Long-term species-specific fieldwork surveys like ours, however, will continue being the only trustworthy way of documenting real changes in the abundance of many small organisms like plants, and thus in assessing the trajectory and vulnerability of an important part of biodiversity.

4.2 Assessing and comparing plant trends and vulnerability

In this study, we have shown how a Citizen Science programme can contribute to detecting population declines and associated potential threats, essential for effective biodiversity management. Contrary to expectations, our results highlight widespread stability of population trends for many different plant species in an environmentally heterogeneous European region, and very few threats, particularly for priority plants. This finding is in line with the increase in abundance of many plant species found in the European Alps over a longer period (more than 4 decades; Rumpf et al., 2018).

Another remarkable and novel result obtained from our AP programme is the evidence that trends of priority plants did not differ neither from co-occurring common plants nor from other non-priority plants in the region, casting doubt on the classical conservation biology tenet of worse performance of the former ones. In contrast, our results do support the general expectation that small populations perform worse irrespective of their conservation status (see also Matthies et al., 2004). We found that priority plants might not always be as vulnerable as thought at local scale when their populations are medium or large sized (the most common cases in our study). This finding suggests that estimating local population sizes should be the first and priority step for assessing potential vulnerability. In another large-scale analysis based on sporadic data gathered by volunteers, Lawson et al. (2008) found that small (<50 individuals) isolated populations were performing similar to large ones. These authors, however, acknowledged that a substantial proportion of unexplained variation was likely due to measurement error, a source of variation that we have minimized with our specific sampling designs.

The scarcity of threats and the overall stability found in monitored plants are probably associated with their frequent location in areas covered by the European 2000 Natura network. Whatever the reason, our results contrast with the overwhelming evidence of negative trends resulting from animal monitoring (e.g. Burns et al., 2018; Rosenberg et al., 2019; but see Wiens, 2016), and highlight the risk of accepting general rules and conclusions derived from studies with vertebrates to assess the conservation status and risk of other groups like plants (Knapp, 2011). Although our results are restricted to a single European region and are based on a relatively short temporal series (3–10 years), they result from analyses of more than 300 plant populations covering a broad variety of biological and ecological conditions: 3,000 m of altitudinal range, a wide habitat heterogeneity across Mediterranean and Eurosiberian biogeographical regions, and most life forms such as biannuals, perennial herbs, geophytes or shrubs.

In summary, contrary to classical surveillance monitoring, our AP programme meets the critical components of effective structured monitoring aimed at generating long-term, high-quality datasets that allow testing specific hypotheses (sensu Bayraktarov et al., 2019; Lindenmayer & Likens, 2010; Pescott et al., 2015). We tested here the presumed worse performance of threatened plants, and results demonstrated the benefit of moving away from a model solely focusing on them towards a more comprehensive approach for comparative purposes. Monitoring population dynamics constitutes the core of the adaptive management process, providing the necessary link between threats, objectives and management alternatives (Bakker & Doak, 2009; Bayraktarov et al., 2019; Lahoz-Monfort et al., 2014). Our AP programme also shows that mandatory evaluation requirements such as the Habitats Directive can be delivered with CS approaches. We therefore encourage conservation policymakers to implement this or similar approaches, and we advocate for its expansion and integration into larger national or international platforms. A larger-scale participative network would also allow conservation managers to share standardized information to better face current challenges in Biodiversity conservation. Long-term monitoring of non-priority species will also help to address question-driven large-scale analyses of trends and pace of biodiversity changes, considering different situations such as central and edge locations (the central–peripheral hypothesis), contrasted habitat types, different life forms or in protected versus unprotected areas. Comprehensive programmes involving citizen scientists and different kinds of plants will assist conservation policymakers to protect the most vulnerable species while managing other structurally important components of the communities within which they are integrated.