Identifying conservation priorities for gorgonian forests in Italian coastal waters with multiple methods including citizen science and social media content analysis

Gorgonian forests are among the most complex of subtidal habitats in the Mediterranean Sea, supporting high biodiversity and providing diverse ecosystem services. Despite their iconic status, the geographical distribution and condition of gorgonian species is poorly known. Using multiple online data sources, our primary aims were to compile, map and analyse observations of gorgonian forests in Italian coastal waters to assess the biological complexity of gorgonian forests, evaluate impacts and vulnerable species, and identify areas of special interest inside and outside of marine protected areas (MPAs) to help prioritize conservation strategies and actions.


| INTRODUC TI ON
Densely branching assemblages of gorgonians, referred to here as gorgonian forests, are among the most ecologically and economically important biotopes of Mediterranean subtidal seascapes (Kipson et al., 2011;Ponti et al., 2016). By increasing the complexity of the environment, gorgonian forests support high biodiversity, deviate currents, modify sedimentation rates, sequester carbon, delay the spread of invasive algae (Casas-Güell et al., 2015;Cerrano et al., 2010) and provide a wide range of ecosystem services (Ballesteros, 2003;de Ville d'Avray et al., 2019). For instance, through complex biophysical structures, gorgonian forests provide a valuable shelter, feeding and nursery habitat for many commercially important benthic invertebrates and fish (Ballesteros, 2003;Cerrano et al., 2010;Ponti et al., 2016Ponti et al., , 2018Valisano et al., 2016).
These architecturally complex marine animal forests are popular with recreational SCUBA divers generating socio-economic benefits and increasing the digital observational data on seascape conditions Ponti et al., 2011;Rodrigues et al., 2016).
The ecological integrity of gorgonian forests and thus the flow of interlinked ecosystem services is increasingly threatened by exposure to multiple stressors including thermal stress linked to accelerated global warming (Cerrano et al., 2000;Coma et al., 2009;Huete-Stauffer et al., 2011;Verdura et al., 2019), the spread of non-indigenous species (NIS) (Cebrian et al., 2018;Galil, 2019), and local anthropogenic activities such as anchoring, fishing, land-based sources of pollution and recreational SCUBA diving (Bavestrello et al., 1997;Milazzo et al., 2002;Mistri & Ceccherelli, 1995). Mass mortality events among gorgonian forests have been increasing in frequency and intensity over the past two decades in the Mediterranean Sea (Linares & Doak, 2010;Santangelo et al., 2015), with thermal stress and subsequent water stratification and disease associated with climate change being the most likely cause (Coma et al., 2009;Vezzulli et al., 2013). The loss of gorgonians and associated crustose coralline algae results in a transition towards a lower complexity autotrophic regime dominated by filamentous algae (Ponti et al., 2014). In 2017, the International Union for the Conservation of Nature (IUCN) estimated that at least 25% of anthozoans were threatened in the Mediterranean Sea, when accounting for similar conservation status in the 50% of species that are still data deficient (Otero et al., 2017). Although regarded as a 'Priority Habitat Type' in the EU Habitats Directive (Council Directive 92/43, 1992), and with some species listed in the IUCN Red List of Threatened Species (appendix II), habitat-forming gorgonian species, except for Corallium rubrum and Savalia savaglia in the Mediterranean Sea, still lack any significant protection under national laws and legal tools (Otero et al., 2017). Marine protected areas (MPAs) have been considered a potentially powerful tool for the protection of gorgonian forests in the Mediterranean Sea, contingent on relevant geographical placement, sufficient resourcing, effective regulations, and high compliance (Coma et al., 2004).
With increasing impact from marine heatwaves (Darmaraki et al., 2019;Smale et al., 2019), the spread of NIS (Cebrian et al., 2018) and other cumulative human impacts including fishing both inside and outside of MPAs (Betti et al., 2020), there is an urgent need for reliable baseline information on the geographical distribution, health and impacts to major habitat-forming gorgonian species in the Mediterranean region (Otero et al., 2017). Current gorgonian species distributions are poorly known in the coastal waters of Italy and long-term and spatially extended monitoring datasets are rare, with most studies focussing on specific sites usually within MPAs (Cupido et al., 2009;Coma et al., 2009;Gambi et al., 2010;Crisci et al., 2011).
Over the past few years, marine citizen science has emerged as an effective and low-cost tool to enhance information gathering across broader spatial and temporal scales than most conventional scientific projects, and at the same time foster increased ocean literacy and shared responsibility for ocean health (Di Minin et al., 2015;Earp & Liconti, 2020;Thiel et al., 2014;Vieira et al., 2020). In addition, an increasing number of recreational SCUBA divers are posting digital underwater videos and photographs on social media and uploading to crowdsourced web-based applications, of which some also have reliable metadata revealing the observer location, depth and the time of observation. Enabled by the growth of both underwater imaging and image sharing digital platforms, the SCUBA diving collective has created a vast ecological data repository of underwater videos and photographs that are publicly available on the World Wide Web (WWW) (Di Camillo et al., 2018). Collection and synthesis of crowdsourced information in ecological informatics is a rapidly emerging approach sometimes referred to as Web Ecological Knowledge (WEK) (sensu Di Camillo et al., 2018) or iEcology (sensu Jarić et al., 2020), defined as the study of ecological patterns and processes using online data usually generated for other purposes and stored digitally (e.g. Internet search activity, social media interactions and uploaded data and media) (Jarić et al., 2020). This accessible online information enables the investigation of historical trends and filling data gaps across broad geographical regions, complementing existing scientific studies to support decision making in marine conservation (Toivonen et al., 2019).
In Italy, growing concern over the mortality of gorgonians at local dive sites is being expressed by the SCUBA diving community and offers a more inclusive participatory model for crowdsourcing and repurposing under-utilized observations while also increasing ocean literacy.

K E Y W O R D S
biogeography, citizen science, gorgonian forests, marine policy, web ecological knowledge through frequent posts on social media that often remain unnoticed by the scientific community. To support effective conservation strategies through prioritization of sites for action, we created a new database of the geographical distribution, abundance, health and impacts to the main habitat-forming gorgonian species in Italian coastal waters. We applied a multi-source data mining approach that included WEK, in combination with data from citizen science, scientific publications, and questionnaires. Our objectives were to compile, map and analyse observations in Italian coastal waters to: (1) assess the complexity of gorgonian forests based on species composition and abundance; (2) assess the type and magnitude of impacts to gorgonian forests and species to identify vulnerability; (3) compare the complexity and magnitude of impact of gorgonian forests inside and outside of MPA boundaries and (4) identify areas of special interest both inside and outside of existing MPAs to help prioritize conservation strategies and actions.

| Multi-sourcedatasynthesis
Data on the distribution, abundance and condition of seven major habitat-forming species of gorgonians in Italian coastal waters were acquired, collated and evaluated using a multi-source data synthesis

| Literature review
Scientific data were gathered using a systematic literature review based on a specific list of search terms (English and Italian language) ( Table 1) and entered into the Internet search engine Google Scholar.
Data were extracted from a total of 50 scientific publications. For C. rubrum, a large amount of data was extracted from the comprehensive CorMedNet dataset (http://corme dnet.medre cover.org/).

| Citizen science
Data from citizen science activities were provided by Reef Check Italia (RCI) (https://www.reefc heckm ed.org/) and extracted from iNaturalist (https://www.inatu ralist.org/). These data originated from SCUBA diver surveys (Appendix S2) and underwater photographs conducted by trained observers. All records were quality checked with expert knowledge for RCI , and validated by researchers in iNaturalist.

| SCUBA diver questionnaires
A questionnaire was created to further investigate the distribution, abundance and health of the focal species in Italian coastal waters (Appendix S2). Following ethical approval, the questionnaire was  to impact classes following a standardized method, and additional visual information was also recorded.

| Dataprocessingandevaluation
The focal gorgonian species were chosen because they are found in Italian coastal waters, have known sensitivity to stressors, including thermal stress and harvesting in the case of C. rubrum, and are easily distinguished from one another by colour and shape which minimizes misidentification. For all data sources, the abundance of the gorgonian forest was recorded into abundance classes following RCI standardized methods. Abundances of E. cavolini, E. singularis, E. verrucosa, S. savaglia, L. sarmentosa and P. calavata were recorded according to the following classes: 1 = 1 individual, 2 = 2 individuals, 5 = 3-5 individuals, 10 = 5-10 individuals, 50 = 11-50 individuals and 100 = >51. For C. rubrum, the recorded classes were: 1 = an isolated specimen, 2 = some scattered specimens, 5 = several scattered specimens, 10 = one crowded area, 50 = some crowded areas and 100 = several crowded areas. The extent of damage of the colonies was logged in percentage classes: 0% damage for a healthy colony, 10%, 25%, 50%, 75%, 99% damage in order of gravity, and 100% for a dead colony (Appendix S2). Where available the water depth and the presence or absence of epibionts, mucilage assemblages, abandoned, lost or discarded fishing lines, and eggs/juveniles were recorded ( Table 2). The presence or absence of fishing lines was used to calculate the average occurrence of fishing lines inside and outside of MPAs.
All records were quality checked through a data validation process based on literature cross-checking (e.g. consistency among reported species) and manual procedures (e.g. matching dive site names to geographic coordinates). The origin of all inconsistent data was further investigated to attempt its correction. All data failing the quality control after further inspection were permanently deleted from the database. The resulting data set comprises separate records for each single species found by each observer/scientific publication in a diving site. Diving sites (n = 61) were georeferenced using geographical coordinates (Datum WGS84) with a minimum accuracy of ±15″ (=±0.00417°) and both diving sites and MPAs were mapped (Appendix S3). Three scores were assigned to each metric depending on extent:

| Creation of indices
Low (1), Medium (2), High (3) ( Table 3). The final values of the indices were obtained by summing the metric scores according to the following formulas: where S sp is the number of species; S ab the score of abundance mode of dominant species; S d the score of damage mode of dominant species; S i the score of abandoned fishing lines occurrence, given by the average occurrence of abandoned fishing lines recorded (0-1).
Resulting scores were then divided into the three classes to define the complexity (I complexity ) and impact (I impact ) extents of gorgonian forests (Low = 2 ≤ Index ≤3; Mid = 4; and High = 5 ≤ Index ≤6). Grid cells with both high complexity and high impact scores (as defined in Table 3) were defined as Areas of Special Interest (ASI) and mapped to inform biodiversity conservation. (1)

| Statistical analysis
All statistical analyses were carried out in R v3.2.1 (R Core Team, 2019). All data were tested for homogeneity using Levene's test and for normality of residuals with a Shapiro-Wilk test and found to be heterogeneous and non-normal. Differences between abundance and damage classes (dependent variables) inside and outside MPAs boundaries and between the seven focal species (fixed variables) were tested using a Generalized Least Squares fitting model  Poisson error distribution fit for the impact index (I impact ). In both models, factor p-value was calculated using a chi-squared maximumlikelihood ratio test. A GLM with binomial error structure was used to investigate the differences in occurrence of abandoned fishing lines inside and outside MPA boundaries, and p-value was calculated using a chi-squared maximum-likelihood ratio test. TA B L E 2 Data on 12 variables collected from four different data sources: World wide web (WWW) (823 records), diving questionnaires (DQ) (145 records), citizen science (CS) (3370 records), scientific publications (SCI) (348 records)

| Speciesdistribution,abundanceand gorgonian forest complexity
The database compiled for this study contained information on

| Arehabitat-forminggorgoniansmore abundant and more complex inside or outside MPAs?
Abundance was significantly higher outside of MPAs compared with inside (L-ratio = 4.75; df = 1; p < .05) and was significantly different

F I G U R E 2
The number of 15″ grid cells inside and outside of marine protected areas with focal gorgonian species presence recorded between species (L-ratio = 90.90; df = 6; p < .01) (Figure 4a and Appendix S6). A significant interaction between protection and species was also discovered (L-ratio = 20.92; df = 6; p < .01). The complexity index was also significantly higher outside MPA boundaries (I complexity F 1 = 35.421, p < .001); however, no significant difference in the index was found when compared across the seven focal species of habitat-forming gorgonians (I complexity F 6 = 8.609, p > .1). Despite not being statistically significant, C. rubrum, P. clavata, E. cavolini and E. singularis supported a higher complexity score than other species (Figure 4b).

| DogorgonianforestsinsideMPAshave significantly less damage than outside areas?
Protection did not have a significant effect on damage (F 1 = 0.02; the Italian coastline were low overall, with 76.3% of the 54′-grid cells exhibiting low impact scores (I impact <3) and only 4.4% displaying high impact (I impact >5) (Figure 3d and Appendix S7). The three locations with high gorgonian forest impact were as follows: Catania, Ventimiglia and Aeolian Islands. Of these, only one 54′grid cell (1.1%) in proximity to the location of Catania, showed very high gorgonian forest impact (Figure 3d). Analysis revealed that impact index was significantly higher outside MPA boundaries (I impact F 1 = 51.499; p < .001); however, no significant difference in the index was found when compared across the seven focal species of habitat-forming gorgonians (I impact F 6 = 2.799; p > .5)

| Identifyingareasofspecialinterestfor gorgonian forest conservation
Grid cells with both high complexity and high impact scores were defined as Areas of Special Interest (ASI) for further investigation and potential for conservation action. About 167 ASIs were identified in the study area (Figure 6a). Of these areas, 80 (47.9%) were found within MPA boundaries, and 87 (52.1%) were found outside MPAs (Figure 6b). Only two 54′-grid cells of ASI showed both high complexity and impact and they were found to be Ventimiglia and Catania, which were found respectively inside and outside MPA boundaries.

| GorgonianforestspatternsinItaliancoastal waters
Results from this study represent the most complete and up-todate baseline data of the distribution, abundance and ecological state of major habitat-forming gorgonian species in Italian coastal waters. The mapped geographical distributions of the seven focal habitat-building species provide: (1) evidence to support decision making in marine spatial planning and for designing place-based actions for addressing marine biodiversity conservation targets; and (2) additional data to enhance the reliability of species and habitat distribution modelling including development of forecasts (Bensoussan et al., 2010;Boavida et al., 2016), and (3) critical data for the evaluation of damaged gorgonian forests and the magnitude of threat from fishing gear (Sini et al., 2015). Differences in the mapped species distribution patterns are linked to environmental conditions (e.g. substrate type, depth and current regimes) and consistent with species habitat preferences (Gori et al., 2011;Linares et al., 2008;Rossi & Gili, 2009  high rates of harvest especially of shallow water colonies (Santangelo & Abbiati, 2001). The range of depths at which observation of red coral were collected in this study, suggested that abundant colonies of C. rubrum may be present below the conventional depth limits for most recreational SCUBA diving (Torrents et al., 2008), highlighting a need for survey techniques (towed video, ROV) that will increase the spatial knowledge of this keystone species at increased depth.
The purple gorgonian, P. clavata, exhibited the highest abundance class, the broadest geographical distribution and highest forest complexity in Italian coastal waters. Paramuricea clavata is a major characterizing and structurally important species of Mediterranean coralligenous habitats; its canopies reduce the range of environmental variability, supporting a wide range of key associated biota and therefore increasing biodiversity (Ponti et al., 2016;Rossi, 2013;Valisano et al., 2016). However, P. clavata was also the only species to experience significantly higher damage than most of the other focal species. This observation is further highlighting an already documented trend of increased damage from human activity for this structurally important species (Ponti et al., 2016).
Mortalities are usually size dependant, with larger colonies being more affected leading to a decrease in the structural complexity of habitat (Linares et al., 2013;Linares & Doak, 2010). Results from this study support the role of P. clavata as a major habitat-forming species and reinforce the case for implementation of targeted conservation strategies to protect this ecologically important species.
Paramuricea clavata is currently listed as 'vulnerable' under the IUCN

Mediterranean
Red List yet is not listed under any other legal instrument (IUCN, 2016;Otero et al., 2017). Italian legislation has no specific national laws for the protection of habitat-forming gorgonian species (Otero et al., 2017).
Alongside international and national legal tools, marine policies of reducing local stressors (e.g. fishing and agricultural run-offs) have been shown to lessen the effects of global stressors (e.g. marine heatwaves) (Micheli et al., 2013), and could play a vital role in limiting further damage to gorgonian forests. Actions to reduce the incidence of entanglement of gorgonians with fishing gear which cause tissue necrosis and subsequent aggregation of fouling organisms thus inhibiting the recovery of the colonies (Bavestrello et al., 1997;Ruitton et al., 2019) is a feasible management action.
The index of impact highlighted a spatial trend of geographically patchy patterns of damage, likely reflecting local cumulative human impacts; however, further research is needed to examine the exposure to threats and stressors. The historical patterns of recorded damage to gorgonian forests as revealed by records of mucilage events and mortality uncovered in our study could be modelled to examine co-occurrence with past marine heatwaves using satellite timeseries data (Bianchi et al., 2019). Similarly, patterns of impacts to gorgonian forests from fishing gear could be investigated with information of patterns of fishing activity using fishing vessel tracks or proximity to highly populated urban areas.
Alongside prioritizing high-low complexity and impact sites for F I G U R E 6 (a) Identified areas of special interest (ASI) for gorgonian forests in Italian coastal waters, summarized in 54′-grid cells. (b) Locations inside and outside of the existing MPA network conservation action, the index of impacts and complexity allows us to also consider gorgonian forests characterized by high resistance or high resilience to stressors. These areas could be used as study sites for understanding genetic traits expressed by different gorgonian populations, which could be useful evidence for restoration programs (Fava et al., 2010;Mokhtar-Jamaï et al., 2013). The mapping of indices of complexity and impact provide a tool to inform the design of adaptive monitoring strategies and create a set of sentinel sites to track condition and threats to give early warning of ecological changes.

| Evaluationofmulti-sourcemethodology
The multi-source approach proved to be an effective method to investigate patterns of species distribution, abundance and state over broad temporal and spatial scales, highlighting the power of public involvement in marine research Schläppy et al., 2017;Turicchia et al., 2021). All four data sources when integrated displayed a high complementarity, providing compelling evidence on the value of multi-source-approaches in marine subtidal monitoring. The proposed multi-source approach has potential to provide a structured, repeatable and inclusive monitoring tool for gorgonian forests, and other marine organisms within and outside MPAs.
This study highlighted the great potential of WEK in gathering large data sets of species distribution, without the costs and issues of fieldwork (Di Camillo et al., 2018). WEK also pro-  et al., 2020). The data set produced as part of this study could be used to train a neural network for initial recognition of gorgonians in videos/photographs, laying the first stepping-stones for AI to be implemented in subtidal ecosystems in Italian coastal waters.
In addition, videos collected in this study could be used to trial Structure from Motion photogrammetry methods, such as the one developed by Palma et al. (2018), with the potential to derive morphometrics and biomass of gorgonian forests from amateur recorded media.

| CON CLUS IONS
In times of rapid global change and financial constraints, the multisource data integration framework proved to be an effective tool for the collection and repurposing of under-utilized observations across a broad geographical and temporal scale to support decision making in marine conservation. The approach offers a more inclusive participatory model for crowdsourcing marine data while also increasing ocean literacy. The findings suggest that several habitat-forming gorgonian species receive insufficient protection under both international and local conservation measures, with special concern for the structurally important P. clavata, with widespread damage occurring across Italian coastal waters.
Understanding the efficacy of the existing MPAs for gorgonians and the potential for greater conservation measures to ensure future resilience of gorgonian forests presents a high applied research priority for the region. Our nation-wide data set, maps and indices hold great potential for informing targeted conservation actions, marine spatial planning, habitat-modelling, and to provide a baseline dataset for applied research projects, monitoring, and future comparative studies.

ACK N OWLED G EM ENT
We thank the volunteer SCUBA divers, scientists and coordinators of Reef Check Italia who provided the citizen science data that formed part of the multi-source data in this study.

CO N FLI C TO FI NTE R E S T
The authors have no conflict of interest to declare.

DATAAVA I L A B I L I T YS TAT E M E N T
The data that support the findings of this study are openly available in Dryad at https://doi.org/10.5061/dryad.kh189 326t (https:// datad ryad.org/stash/ share/ ghYIx sXK-CF00P wDa0G 7W85n ITEdf PxxqCYOji_wVQQ).

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13553.