Global Island Monitoring Scheme (GIMS): a proposal for the long-term coordinated survey and monitoring of native island forest biota

Species inventories which aim to collect data to assess the taxonomic, phylogenetic and functional facets of biodiversity often involve time-consuming sampling protocols. Species monitoring, in contrast, ideally builds on comprehensive inventories and focuses on a minimum set of indicators that involve less time-consuming protocols but deliver the minimum data necessary to infer temporal trends of community change. For example, inventorying may involve assessing alpha diversity by maximizing the sampled diversity with optimized effort (alpha-sampling sensu Cardoso et al. 2016), whereas monitoring may require assessment of temporal changes in species composition and/or abundances (i.e. beta-diversity) with minimum sampling effort (beta-sampling sensu Cardoso et al. 2016).

We therefore propose an integrated Global Island Monitoring Scheme (GIMS) that: (i) is accurate and sustainable in the long-term, (ii) includes protocols that can be applied to both ‘short-term’ (as a minimum, two comparisons in time within maximum of a 10 year interval) and ‘long-term’, continuous, monitoring programs, (iii) assesses long-term effects of key global change drivers, and (iv) generates data to inform biodiversity conservation and ecosystem management. Efficiency should be attained by minimizing spatial and seasonal variability (sampling the same sites through time, during the same season), maximizing the environmental variability within the sampled locations, and minimizing sampling bias (through sampling standardization) to reveal true trends in diversity descriptors. In the case of seasonal organisms (e.g. arthropods), sampling each season in a year over several years may be possible if resources are available (see e.g. Borges et al. 2017), but it is costly and in most cases the pragmatic choice is to sample in the best season in terms of number of species available.

Proxies of ecosystem change should consider effects of climate, land-use and exotic species. Based on the aforementioned objectives, and on the specificities of each island/archipelago, we advocate the implementation of a general experimental design that is optimized to cover the entire range of ecosystem change (assessed as an index—e.g. LDI—Landscape Disturbance Index [Cardoso et al. 2013]). Sampling along an elevational transect allows an evaluation of the impact of climate change at a mesoclimatic scale by investigating e.g. shifts in species abundances and distributional ranges in relation to key environmental gradients such as temperature and precipitation (Alexander et al. 2011). To cover the entire range of ecosystem change, we suggest first placing plots in native forest habitats, and, second, depending on the available resources, placing additional plots in modified habitats (e.g. production forests, secondary forest, and/or invaded forests). One key challenge is to combine environmental gradients (e.g. elevation) with gradients of forest disturbance, disentangling their relative influences on biodiversity. This could be achieved by statistical approaches based on variation partitioning techniques (Legendre 2008) or direct measurements using indicators of habitat quality, as described in GEO BON (2015) (see e.g. “Biodiversity Habitat Index” or “Local Biodiversity Intactness Index”). For instance, paired elevation/disturbance gradients are increasingly popular study systems in mainland areas (e.g. Ensslin et al. 2015; Gomez-Diaz et al. 2017).

We suggest the use of permanent plots of 50 m × 50 m, based on our previous experience (see NETBIOME-ISLANDBIODIV, Emerson et al. 2017). Nevertheless, there are islands (e.g. Tahiti) where the rugged terrain does not allow for the use of plots of that size, and there are taxa for which a plot-based approach is not adequate (e.g. vertebrates) and other methods may be more appropriate (e.g. transects; whole island surveys) (see Appendix S2). Moreover, subsampling can also be made inside these plots, for instance, for small organism such as a bryophytes. The frequency of sampling will differ depending on the objectives, protocols, taxa and available resources (see Appendix S2). Plots can be placed every 200 m along the elevational gradient, as in the NETBIOME-MOVECLIM project (Gabriel et al. 2014; Henriques et al. 2016), which may be sufficient to reflect climatic changes affecting species distribution in all selected taxonomic groups (see e.g. Ferreira et al. 2016; Patiño et al. 2016).

Within the GEO BON scheme, Pereira et al. (2010) advocate the use of plants and birds for global terrestrial species monitoring programs. However, a broader mix of taxa is needed for understanding ecosystem functions and the consequences of global changes. Based on our experience with native island forests, we suggest the monitoring of the following groups.

Bryophytes are a sensitive indicator group that respond rapidly to environmental change and also provide important ecosystem functions and services (e.g. water supply, nutrient cycles, erosion control). In the absence of roots and a protective cuticle, bryophytes are unable to regulate water exchange, and hence, depend more directly on precipitation and fog than higher plants, making them highly sensitive indicators of variation in precipitation regimes (Tuba et al. 2011) and human disturbance (e.g. Bates and Farmer 1992). Bryophytes are highly dispersive species, usually present in all elevation belts. As they are rarely consumed by herbivores, their distributions can be expected to be at equilibrium with climate (Mateo et al. 2013). Moreover, bryophytes are an important group of pioneer organisms and contribute to humus accumulation and regulation of microclimate (e.g. Van Tooren et al. 1985), which is particularly important for the maintenance of forest ecosystems. Together, this makes bryophytes an important group for the evaluation of the impacts of global changes on native island forests.

Vascular plants are the principal primary producers and structural elements of most island ecosystems, providing usually the largest share of biomass, as well as crucial ecosystem functions and services (e.g. water supply, nutrient cycles, erosion control). Vascular plants have been frequently used as model systems to study the impact of global change on ecosystem functioning (Naeem et al. 1999).

Vertebrates: birds/mammals are the best-known vertebrates inhabiting island forests. Birds are particularly sensitive to forest loss, degradation and alteration (Pereira et al. 2010) and this is more acutely true of the sub-set of forest-specialist species (Matthews et al. 2014). Insular mammal assemblages often contain a large proportion of non-native species (e.g. rabbits, rats, pigs, sheep, goats, cats), which can have strong negative effects on island ecosystems, particularly by suppressing plant cover and regeneration, leading to increased erosion and habitat degradation (Courchamp et al. 2003). However, large and less isolated islands often have native (often endemic) mammals that may be impacted by non-native species, and in these cases monitoring and eradication efforts should be designed to benefit native species (Harris 2009; Rickart et al. 2011). The monitoring of these vertebrate taxa is thus extremely important for the conservation and management of native island forests.

We acknowledge that other taxonomic groups are also of tremendous importance for forest ecosystems, e.g. fungi (including lichens), molluscs, bees, butterflies, parasitic wasps, moths, amphibians, reptiles, and could potentially be included in a future expansion of the current GIMS proposal, which is already available at IslandLab (http://​islandlab.​uac.​pt/​software/​ver.​php?​id=​27).

Several protocols have been proposed for the target taxonomic groups described above (Table 1 and Appendix S2). Often it is necessary to perform a detailed initial inventory for the selected permanent plots (or for the whole island in the case of vertebrates), particularly for less-studied groups (bryophytes, spiders, beetles). A combination of several methods is proposed to this end (Table 1). The BRYOLAT protocol (Ah-Peng et al. 2012; Gabriel et al. 2014) is proposed for bryophytes and ferns. Three protocols are proposed for vascular plants: one based on remote sensing, which can provide cost-efficient data; another for the vegetation as a whole, targeting both alpha and beta diversity; and a third one for monitoring rare species of vascular plants, which can also be applied to bryophytes. The COBRA protocol (Cardoso 2009) is proposed to monitor spiders and beetles. Protocols for birds and different groups of mammals are also suggested (see Appendix S2). The GIMS protocols should be conducted by trained field teams in order to generate reliable data and results, although non-scientists can be involved if properly familiar with the taxa and sampling techniques.

As mentioned previously, an optimal inventory strategy requires the maximization of sampled alpha diversity while minimizing sampling effort (alpha-sampling). In contrast, optimal monitoring requires minimizing the difference between sampled and true beta diversity (i.e. beta-sampling; Cardoso et al. 2016), while still minimizing sampling effort. The protocols listed in Table 1 for vascular plants, birds and mammals are comparatively less time consuming than those for bryophytes and arthropods (BRYOLAT and COBRA protocols), which were originally designed for the inventory of biodiversity, and thus need to be optimized for monitoring. To optimize the protocols for the purpose of monitoring, we took several steps building on previous work by Cardoso et al. (2016), and developed a version of BRYOLAT-Monitoring and COBRA-Monitoring protocols (see Appendix S2- Protocols) using the R package BAT (Cardoso et al. 2015).

Our proposed protocols allow the calculation of key diversity metrics, including alpha and beta diversity, the latter involving two processes, replacement of species and loss or gain of species between sites or dates (Carvalho et al. 2012, 2013). This framework allows the measurement of directional turnover in assemblage structure (e.g. through time or along environmental gradients) and the quantification of changes in community composition (or of a single species population of particular interest) in space and time (Table 1 and Appendix S2). The changes through time can be compared with an initial baseline inventory for each plot, and true beta replacement and species loss or gain can then be distinguished from random changes with minimal effort. To monitor ecological and anthropogenic drivers and pressures, several indicators can be used, such as the relative abundance or percentage coverage of non-native species and invasive species, detailed microclimatic variables (e.g. Giambelluca et al. 2013) or indicators such as the “Biodiversity Habitat Index” or “Local Biodiversity Intactness Index”, which imply the use of remote-sensing with modelling of local species diversity metrics (GEO BON 2015).

The collection of standardized time-series data across regional and global site networks is also relevant from a basic science perspective and enables an evaluation of how different macroecological patterns vary in space and time. For example, as noted by Magurran (2007), temporal variation in the form of Species Abundance Distributions (SADs) has been largely overlooked. Analysis of changes in the shape of the SAD, and the position of individual species within it, should enable more rigorous tests of community assembly models (e.g. niche and neutral models), and facilitate a more comprehensive understanding of commonness and rarity in communities (Magurran 2007; Matthews and Whittaker 2015). The collection of standardized temporal abundance data also allows for a better understanding of sampling effects on different macroecological patterns (e.g. species–area relationships and SADs), and will be critical to advance ecological theory (Vellend 2016). For example, it has been shown that the SADs of samples collected over short durations approximate log-series distributions, whereas the SADs of samples collected over longer durations approximate log-normal distributions (Hubbell 2001; Magurran 2007). In addition, if ecological information regarding the sampled species is available (e.g. a species’ functional role) it is possible to explore how different species (e.g. the core–satellite dichotomy) affect different patterns, for example the SAD, and how these effects change through time (Magurran and Henderson 2003; Matthews and Whittaker 2015). Depending on the time interval between surveys, the collection of long-term monitoring data enables researchers to address questions related to temporal turnover in ecological communities, for example, how rates of temporal turnover vary across archipelagos, and how anthropogenic factors affect turnover rates (see Magurran et al. 2010). Such data may be a value input for initiatives such as the Intergovernmental Platform on Biodiversity and Ecosystem Services (IPBES).

The indicators that can be derived from our monitoring native island forests proposal are aligned with the Essential Biodiversity Variables (EBV) proposed by Pereira et al. (2013). In fact, the EBV classes “abundances and distributions”, “phenology”, and “taxonomic diversity” can be generated in straightforward manner from the implementation of the protocols proposed here (Appendix S2). Such data will be also crucial for validating model predictions (Honrado et al. 2016) and demonstrating the impacts of climate change on species distributions on islands (e.g. Ferreira et al. 2016; Patiño et al. 2016) and to develop proper programs for controlling and eradicating invasive species. This type of data will also be relevant for many island nations to identify threatened species through IUCN Red List assessments and to monitor the delivery of the Convention on Biological Diversity (CBD) Aichi Targets and the United Nations Sustainable Development Goals (SDGs). Key SDG indicators supported by this work will include the Red List Index and the proportion of countries resourcing work on invasive species.

First, it is advisable to acquire both the most recent and the oldest aerial photographs, satellite imagery, historical maps and palaeoecological data (see e.g. Nogué et al. 2017), for comparative analyses. However, some caution should be taken when using only the most recent and oldest aerial photos, since these may lead to misleading conclusions if land cover change did not follow a linear pattern over time. Information on biomass evolution, growth form and chemical composition can be obtained by remote sensing, which appears to be the only method for efficiently collecting data over large spatial extents at high spatial resolution (Strand et al. 2007). At a high/medium spatial and medium spectral resolution, remote sensing multispectral imagery has been successfully used to map target plant species and communities (e.g. Mitchell and Glenn 2009). Hyperspectral images are currently the most widely used imaging source for vegetation studies. The main advantages of using hyperspectral data is that detailed spectral profiles can be developed for native and alien plants, and that specific spectral regions can be analyzed and that those data are most sensitive to the abundance of the species of interest (Underwood et al. 2003). If spectral information is not available, Synthetic Aperture Radar (SAR) data can be useful for vegetation management decision-support when combined with good field knowledge (Huang and Asner 2009). Airborne Laser Scanning (ALS, also referred to as LIDAR—Light Detection And Ranging) may also constitute a suitable tool for mapping key ecosystem properties such as canopy height, biomass and stand structural complexity (Maltamo et al. 2014). The connection between the differences in vegetation structure observed in the field and the corresponding ALS-derived parameters can be established in a straightforward way (Zlinszky et al. 2012). In fact, some of the most successful and detailed insular remote sensing-based vegetation mapping studies have been undertaken on the Hawaiian Islands using a combination of both airborne hyperspectral and LIDAR technologies (e.g. Asner et al. 2011). To date, due to climate constraints and the general lack of free-of-charge satellite multispectral remote sensing data (e.g. Landsat) covering most oceanic islands, few successful examples of vegetation mapping are available for islands. Promising case studies exist, focusing on alien invasive species, and have been based on high/medium-spatial resolution multispectral satellite remote sensing data (e.g. Gil et al. 2011; Massetti et al. 2016).

We propose to sample every 200 m along the elevational gradients from sea level (when native forest habitats are available) to the maximum elevation. The main goal is to study native island forests, selecting at least one 50 m × 50 m plot every 200-m of elevation (in the case of BRYOLAT two smaller 10 m × 10 m plots are set up; Ah-Peng et al. 2012). However, another possibility it to select at each elevation a set of plots that cover the entire range of disturbance to forests, from pristine (e.g. NETBIOME-MOVECLIM project; Gabriel et al. 2014) to anthropogenic habitats (e.g. Florencio et al. 2016). After selecting a 50 m × 50 m plot at each 200-m elevation, an index of disturbance/transformation based on the distance to the nearest anthropogenic habitats should be calculated for each plot (see e.g. Cardoso et al. 2013). This will allow the monitoring of the impact of the pool of alien species available in the surrounding landscape. Variation partitioning techniques can be used to disentangle interdependency between level of ecosystem change and the environmental gradients (Legendre 2008). Along most elevational gradients on islands, the lower elevations are usually dominated by modified habitats. This might make it difficult to find native habitats at all elevation for monitoring. We suggest starting the lower elevation plot at the minimum elevation with adequate native or near-native habitat. In addition, ecological processes and key ecosystem variables (e.g. microclimate variables, soil pH, C:N ratios, net primary productivity, etc.) should be measured in each site/plot using standardized protocols (see e.g. Herrick et al. 2006).

Based on Table 1 and Appendix S2, researchers may first need to perform an inventory of biodiversity using different protocols for bryophytes (1A), vascular plants, including ferns (3), spiders/beetles (5A), birds (6), and mammals (7–10). After the initial inventory, monitoring should proceed using the protocols 1B (bryophytes), 2–4 (vascular plants), 5B (spiders/beetles), 6 (birds), and 7–10 (mammals). Caution should be taken when sampling and manipulation rare and threatened species. Some island states have lists of threatened species and permits should be obtained to sample and monitor those species (see more details in Appendix S2).

The workload required for comprehensive taxonomic surveys remains high and is particularly challenging for under-surveyed islands and taxa. For taxa like vascular plants and vertebrates, species identification should be straightforward and based on vouchers for morphological identification and/or DNA barcoding (see e.g. Gonzalez et al. 2009). For groups such as spiders and beetles, and some critical bryophyte and vascular plants genera, the taxonomic impediment (Linnean shortfall) can be an issue in tropical islands (e.g. Cardoso et al. 2011; Hortal et al. 2015), and in addition to the need for taxonomic expertise for identifying the species or determining morphospecies, it is desirable to have access to DNA barcoding for the detection of cryptic species and help on assignment of species that may be affected by splitting and lumping processes in sorting morphospecies (see also Emerson et al. 2017). Furthermore, the recent advances in sequencing techniques now allow third-generation sequencing directly in the field (e.g. Parker et al. 2017), which opens very promising avenues for novel ways of DNA-based identification beyond the limits of predefined barcoding regions. In particular, given the importance of being able to assess identity and abundance of all individuals in an entire community, approaches are being developed that allow massive sequencing of entire communities at minimal cost (Krehenwinkel et al. 2018). While species richness can be routinely identified by sequencing bulk samples, new approaches also allows estimation of species abundance (Krehenwinkel et al. 2017). This in turn allows development of metrics of community stability and turnover in the context of food web structure and species abundance distributions. Additionally, by integrating data on identity and food webs with phylogenetic data, these approaches provide insights into changes in communities based on genetic data.

Data archiving and sharing strategies are critical for the successful implementation of long-term biodiversity studies (Michener 2015; Mills et al. 2015). First, specimen archiving should be guaranteed using public natural history collections (registered herbaria for vascular plants and bryophytes, zoological collections for arthropods and eventually some birds and mammals or their DNA). Collected specimens are the primary data arising from surveys and monitoring and have to be stored and made publicly available for cross-checking and future research (see Schilthuizen et al. 2015). We advise also that some ethical procedures should be follow avoiding sampling with removal of rare plants and vertebrates. Data papers should be promoted (e.g. Borges et al. 2016b) to foster synthesis and comparative studies and we propose the creation of a repository of raw data following common community standards such as the Darwin Core, which is available at Biodiversity Data Journal via the Data Paper format and connected to GBIF using the Pensoft IPT server or other IPT server available in other institutions. Alternatively, a more sophisticated option will be the Ecological Metadata Language (EML), which can also be used to document ecological data in modules (see https://​knb.​ecoinformatics.​org/​#external/​/​emlparser/​docs/​index.​html).

Obtained distribution data should be also made available in regional portals such as the Azores Bioportal, https://​azoresbioportal.​uac.​pt/​ (Borges et al. 2010) or BIOTA Canarias (http://​www.​biodiversidadcan​arias.​es/​atlantis/​common/​index.​jsf) as well as globally aggregated in GBIF and through the IUCN Red List of Threatened species, as appropriate. It is further necessary to ensure that the data can be noticed and used by decision makers, and allow national reporting to multilateral environment agreements such as the CBD. Intellectual property rights and authorship issues need to be considered when making the data available, but access restrictions often hinder scientific use and synthesis of data from many monitory programs and databases (e.g. GLORIA, MIREN, TRY, sPlot, etc.). Monitoring protocols should thus ensure that data become (at least in time) publicly available and easily accessible. All these actions should be complemented by activities that promote science communication for conservation managers and the general public, using participatory strategies and web technologies (e.g. Amorim et al. 2016; Arroz et al. 2016).