We selected 34 plant species and five threatened habitats as conservation features for the analysis. The 34 plant species originated from a list of 35 Species of Conservation Concern as defined by the UKOTs Team (2021), based on their restricted range and endemism (The BVI TIPAs National Team 2019a; Dani Sanchez et al. 2021) which were used for the identification of TIPAs in the BVI. One plant species, Picrasma excelsa (Sw.) Planch., had no recorded locations, thus the analysis did not consider this species. For the 34 Species of Conservation Concern, a total of 5,248 high accuracy (+/- 10 m) georeferenced location records were included in the analysis (Appendix 1: Table A1). Data were available for terrestrial seed plants only. All plant records were treated as presence/absence data, as the population size was not available for all records. For full details on data collection, see Dani Sanchez et al. (2021).

In addition to geographical plant records, we examined five threatened habitats. The BVI national list of threatened habitats consists of coastal shrubland, dry salt flats, mangrove, semi-deciduous gallery forest and upland evergreen forest (The BVI TIPAs National Team 2019a). These habitats meet the TIPAs criteria of being either natural or semi-natural habitats that support higher vascular plants and each currently cover less than 10% of BVI terrestrial land or is present in three or fewer islands (Darbyshire et al. 2017). Further, to be considered a threatened habitat, a continuous decline of the habitat has to have been observed, estimated, inferred or projected for the BVI (The BVI TIPAs National Team 2019a). Threatened habitats were mapped using QGIS (QGIS.org 2021) from two satellite-derived land-cover datasets: (1) The BVI Habitat map 2020 (Scarth and Pike 2020) derived from Sentinel-2 imagery was used primarily, due to its high spatial resolution of 10 × 10 m; (2) the Landsat-7 dataset in Kennaway et al. (2008) with a 30 × 30 m spatial resolution was used for the semi-deciduous gallery forest class as it was not present in The BVI Habitat map 2020.

Authors from the British Virgin Islands TIPAs National Team and University of Copenhagen held workshops to determine conservation targets for all conservation features (a percentage to protect for each plant population or habitat extent), making use of their collective practical knowledge on the conservation features, their spatial distributions and levels of threat. The threatened habitats were assigned a conservation target of 30% each in accordance with the 2030 Global Biodiversity Framework (CBD 2022). Targets for each conservation feature are shown in Appendix 1: Tables A1, A2.

We investigated where to strategically expand the current protected areas of the BVI within the boundaries of the TIPAs (The BVI TIPAs National Team 2019b; Dani Sanchez et al. 2021), using the decision-support system MARXAN (Ball et al. 2009). MARXAN seeks to identify near optimal solutions through a simulated annealing algorithm, by selecting areas (portfolios of planning units) that efficiently achieve conservation targets, while minimising costs (Ball et al. 2009). The cost parameters were provided by an overall boundary length and an Environmental Risk Surface (ERS; see below). The boundary length is controlled by setting a Boundary Length Modifier (BLM), which allows the user to change the compactness of the final solutions and thereby address the SLOSS approach (Ardron et al. 2010). Lower BLM values will result in the BLM to have less influence and, therefore, select portfolios with increased fragmentation (Schill and Raber 2011). It is important to note that since MARXAN applies a simulated annealing algorithm, it produces several near-optimal solutions (portfolios of planning units to protect) and a summed solution, rather than one single best solution. In terms of practical conservation planning, these portfolios would then have to be evaluated with respect to practical and/or political perspectives (Ball et al. 2009).

A grid of planning units with a size of 30 × 30 m was used, based on the fine scale of available input data (accuracy of +/- 10 m) and the relatively small land areas of the BVI and the TIPAs, making a compromise between data scale and practicality of implication and management (Schill and Raber 2011; Mo et al. 2019). We used the borders of the BVI Habitat map 2020 (Scarth and Pike 2020) to create the grid surface of planning units, removing only the Open water class to cover all land areas of BVI, as well as mangrove habitats. No planning units were clipped to the coastlines nor political boundaries within the BVI, as this would result in small, fragmented planning units. The final area of planning units covering the BVI (~ 166 km²) was, therefore, slightly larger than the official land area of the BVI (~ 153 km²) (The BVI TIPAs National Team 2019b).

In order to design a biologically resilient network of protected areas, the planning units were divided into three geographic strata, named after the largest islands in each specific strata: Anegada, Tortola and Virgin Gorda (Fig. 1). The introduction of these strata, as opposed to running MARXAN on all of the BVI at once, were used to drive the optimal portfolio selection across the archipelago rather than being centred in a single area of the BVI. This design follows the previous marine protected area analysis within the BVI, which also employed three strata to force the selection of conservation features beyond Anegada’s large reef system (Woodfield-Pascoe et al 2013). Similarly, the use of strata in a terrestrial planning context would influence portfolio selection beyond Anegada’s rich terrestrial biodiversity. The use of strata also allowed different application of adaptive targets to plants which were more abundant in one stratum than another. MARXAN was run individually on each stratum and results were combined into collective portfolios for all of the BVI. A total of 184,791 planning units were analysed and assigned a status of “available” for MARXAN selection if they fell within the TIPAs boundaries and a status of “unavailable” for those outside the TIPAs boundaries. This allowed the prioritisation of areas within TIPAs for future protection. Planning units already part of the Protected Areas System Plan were assigned a “locked” status to ensure inclusion in the portfolios. Each planning unit was assigned a cost provided by an ERS (see below). For each planning unit, we calculated the total amount of each conservation feature within, either in number of occurrences (plant taxa) or area (habitats).

Figure 1.

Study area. Reference map showing the location of the BVI in the Caribbean Sea, as well as the strata division, named after the largest island in each specific area: Anegada; Tortola; and Virgin Gorda. Tropical Important Plant Areas (TIPAs) are shown, as well as areas within the current Protected Areas System Plan. A dashed line is added to show the distinction between islands allocated to the Tortola and Virgin Gorda strata.

Protected area solutions that are too fragmented can be difficult to implement and are ecologically less functional. Following a BLM sensitivity analysis, we settled on a BLM value of 0.1. This achieved an optimal degree of planning unit clustering and delivered a smaller total area with accepted initial coverage and connectivity. For each of the three strata of the BVI, 100 runs were executed using 1,000,000 iterations per run. MARXAN generates two standard outputs: the best solution, as well as a summed solution. The best solution portrays the portfolio of planning units with the lowest cost score, found in all the good combinations of planning units (portfolios) selected by MARXAN (Ball et al 2009). As we executed 100 runs, the best solution was found within one of these. The latter portrays the selection frequency, i.e. the number of times each planning unit was selected in each of the 100 runs, which Ardron et al (2010) suggests being one of the most useful outputs. We therefore examined two main portfolios of planning units; (1) the best solution output (hereafter ‘Best Solution’); (2) the planning units with a selection frequency of a 100, thus a portfolio of planning units that were selected in each of the 100 MARXAN runs (hereafter ‘SF100’). From those portfolios, we extracted two Crown land portfolios: (3) the best solution output that fell within Crown lands, thus not private land (hereafter ‘Best Solution CL’); and (4) the planning units with a selection frequency of 100 that fell within Crown lands (hereafter ‘SF100 CL’). The extraction of overlapping areas within the main portfolios and Crown land allowed us to examine portfolio areas that would not require land purchase. Planning units within the current Protected Areas System Plan were included in all four portfolios.

To estimate the level of threat to the conservation features and to provide MARXAN with a cost parameter beside the BLM, an Environmental Risk Surface (ERS) was produced to assign a cost value to each planning unit under consideration (Schill and Raber 2011). We assessed the locations of possible threats to biodiversity (e.g. an airport). The threats were drawn as polygons in GIS using a combination of OpenStreetMap layers (OpenStreetMap contributors 2021), the BVI Habitat map 2020 (Scarth and Pike 2020) and ground reference information acquired from Google Maps satellite imagery (Google 2021a, b), to capture the outlines of the threat features. Each threat was assigned an intensity score and an influence distance (Appendix 2: Table A3, Fig. A1). The latter expresses the intensity of the threat outside the immediate area, with the intensity decreasing the further away from the threat, until it no longer poses a threat to the conservation features. We examined two published assessments that applied the ERS methodology in Jamaica (McPherson et al. 2008) and across the Caribbean (Huggins et al. 2007) to understand the relationships between threat intensity scores and decay distances. In MARXAN, all costs are seen in relation to the other costs, so intensity scores in the modelled ERS are within a relative scale (Schill and Raber 2011). We selected four levels of intensity scores within the range of 0–99: 99 (high), 66 (medium), 33 (low) and 10 (very low). The scale was chosen to reflect fractions of thirds, followed by a lower intensity score of 10 to quantify the lower intensity for agriculture (Huggins et al. 2007; McPherson et al. 2008). To ensure MARXAN only selected planning units that were crucial for the portfolios, the minimum cost was set to a very low, non-zero, value (e.g. 0.01) to all planning units that did not overlap with the ERS modelled values. For each individual threat feature, multiple buffers were drawn outside the borders, until they reached the length of the defined decay distance. For each threat feature with multiple buffers, each buffer was assigned intensity scores from the aggregation of intensity scores and decay distances of the threat and the specific buffer’s distance to the threat, with the outermost buffer having the intensity score closest to zero. The ERS layer was turned into vector points in GIS, reflecting the grid surface of 30 × 30 m, with the maximum intensity score assigned for overlapping buffer values. The vector points were ultimately joined to the planning units as cost values.

Based on the four resulting portfolios, we found that, by expanding the current Protected Areas System Plan to the main portfolios: ‘Best Solution’ and the ‘SF100’, it is possible to reach the conservation targets for all conservation features and meet the 2030 Global Biodiversity Framework commitments for the BVI terrestrial land. The ‘SF100’ portfolio resulted in a more fragmented and scattered selection of planning units; however, both portfolios covered the same conservation features, suggesting that the additional planning units selected within the ‘Best Solution’ may only improve connectivity. The additional two Crown land portfolios, which extracted overlapping areas between the main portfolios and Crown land, allowed us to evaluate whether it would be possible to meet conservation targets on areas that do not require land purchase, as approximately 80% of land in the BVI is privately owned. Our results show that we can achieve the 2030 Global Biodiversity Framework targets by solely expanding protected areas into Crown land, although private land would be needed to achieve the conservation targets for all plants, including three of the four BVI endemics (Tables 1, 2). All individual habitat types met their 30% protection targets within the main portfolios; however, this was not the case for the Crown land portfolios as the target was not reached for upland evergreen forest. This further suggests that, to protect 30% of all threatened habitats, protected areas need to expand beyond existing Crown lands. Regardless, the four portfolios offer a significant improvement in protection for all conservation features when compared to the current Protected Areas System Plan and provide evidence and a robust framework for guiding Protected Area expansion.

Many conservation features exceeded their conservation targets (14 plant species and two threatened habitats exceeded their targets by over 50 percentage points) within the main portfolios, as well as four plant species and one habitat within the Crown land portfolios (Appendix 3: Table A6). Although mangrove and salt-pond habitats had already met their targets within the current Protected Areas System Plan, their selected extent also increased within the four portfolios, as well as that for 10 plant species (Appendix 3: Table A6). Given that the primary objective of MARXAN’s selection algorithm is efficiency, there is a high likelihood that multiple rare occurrences of conservation features within a planning unit or across neighbouring planning units will be included in the final portfolio. When these conservation features are clustered together in close proximity, the selection of neighbouring planning units can result in targets being exceeded.

If a perfect solution based on perfect input data and variables existed, it might not be the best solution possible due to local politics, land ownership, funding for purchase, stakeholder involvement or current and future land use. MARXAN solutions provide decision support and should be vetted and combined with expert and local stakeholder knowledge to arrive at a solution to implement. This approach is endorsed by Cowling et al. (2003), who suggest the use of systematic conservation planning algorithms should be closely integrated with expert knowledge to combine the strengths of both approaches. The review of MARXAN portfolios by local experts can provide valuable specific insight into landscape context, political environment, biodiversity knowledge and management issues that are not likely captured in a MARXAN portfolio selection (Cowling et al. 2003). Ardron et al. (2010) stated that MARXAN should be used to model a suite of alternative protected area network designs for stakeholders to review and select from, as opposed to giving one definite solution. This is why it is important to model a variety of scenarios whose input parameters originate from stakeholder workshops. Portfolio results should then be presented and reviewed by experts and stakeholders, having the most appropriate portfolio selected or modified.

TIPAs are often used as a way to highlight protected area gaps and have been used in other areas of the world. Couch et al. (2019) mapped 22 TIPAs in Guinea, West Africa occupying 3.5% of Guinea’s land surfaces, highlighting areas of irreplaceable plant diversity and the need to protect against species extinction. Not all 22 areas were protected at the time, although it was stated that if all 22 TIPAs were protected, a vast amount of Guinea’s wild plant resources would be safeguarded, including over 60% of threatened plant species and a majority of species and representative areas for nine threatened habitats found across Guinea (Couch et al. 2019). As suggested by The BVI TIPAs National Team (2019a), the integration of the TIPAs into a revised Protected Areas System Plan would be beneficial for the conservation of the 34 Species of Conservation Concern and threatened habitats. Some TIPAs are entire islands or cover large parts of private land and further assessments on these areas within the TIPAs Network are needed to protect a critical percentage of conservation features. Addressing this, the four portfolios presented in this study highlight areas within the TIPAs, where conservation efforts should be prioritised to adequately manage and promote species preservation. Our work provides a science-based framework towards guiding the expansion and subsequent implementation of conservation efforts for the 34 plant species and the five threatened habitats, based on the areas selected for TIPAs (Dani Sanchez et al. 2021) and the Protected Areas System Plan 2007–2017 (Gardner et al. 2008).

Our portfolio results need to be presented as options for experts to review, modify and implement (as per Cowling et al. (2003)). Stakeholders from the National Parks Trust of the Virgin Islands (NPTVI) and Kew need to carefully assess our results and integrate them into a strategic conservation planning design for resource management implementation at relevant scales (Huggins et al. 2007). The Crown land portfolios presented in this work demonstrate how a large number of conservation targets can be achieved without focusing on areas that require land purchase. A recommended first step would be to evaluate areas for conservation within the Crown land portfolios, followed by prioritisation outside Crown lands, paying special attention to conservation features that did not meet their targets, as well as the planning units selected in each MARXAN run; the ‘SF100’ portfolio. Notably, the protection of the BVI endemic plant species would require voluntary partnerships with BVI private landowners. Private protection is possible under the Virgin Islands National Parks Act (Virgin Islands National Parks Act 2006) conservation agreements or under the Virgin Islands Physical Planning Act (Virgin Islands Physical Planning Act 2004) that can protect species within declared Environmental Protection Areas. Moreover, Ball et al. (2009) recommended that planning units identified as having both high selection frequency and a high cost are under immediate threat and should be considered priority areas where closer examination is warranted. Further, in cases where land purchase is not possible, areas could still be recognised within the Protected Areas System Plan as OECMs (other effective area-based conservation measures; IUCN-WCPA Task Force on OECMs (2019)) by engaging with private landowners to promote conservation actions and focused management within their lands.

When expanding protected area networks, we note that management effectiveness guided by a management plan is critical for protected areas to be successful tools for biodiversity conservation (Geldmann et al. 2013). The first target of the 2030 Global Biodiversity Framework focuses on integrating areas in biodiversity-inclusive spatial planning and other effective management activities, to ensure that loss of areas of high biodiversity value is brought close to zero, with respect for local communities and indigenous people (CBD 2022). Watson et al. (2014) stated how a lack of resources, most notably in developing countries, was the primary driver behind weakened performance of protected areas. As protected areas continue to expand, these areas are likely to have increased contact with local communities, which may lead to conflicts (Gooden and ’t Sas-Rolfes 2020). Further, in 2003, CBD found that only 6% of involved countries reported to have adequate resources for carrying out protected area management (Watson et al. 2014). Therefore, we suggest that conservation efforts and any plans to expand protected areas should be followed by an evaluation of adequate resource availability to carry out recommended management actions once the protected areas have been established.

The authors have declared that no competing interests exist.

No ethical statement was reported.

We wish to acknowledge the following funding sources that have enabled this dataset to be compiled: The UK Government’s Darwin Initiative and Biodiversity Challenge Fund (DPLUS012, DPLUS039 and DPLUS084), HSBC’s 150^th Fund; The Mohamed bin Zayed Species Conservation Fund (project number 13257818). B.D. was supported by Independent Research Fund Denmark (grant/award number 0135-00333B).

Each of the authors has fulfilled the following: (1) Contributed significantly to the conception or design of the study; or data acquisition, analysis or interpretation; (2) Participated in drafting the manuscript or provided critical revisions to enhance its content; (3) Given approval for the publication of the final version; and (4) Committed to being responsible for all aspects of the work, ensuring that any enquiries regarding the accuracy or integrity of the work are thoroughly investigated and resolved.

Thomas Heller https://orcid.org/0000-0003-2004-2614

Michele Dani Sanchez https://orcid.org/0000-0001-6998-9433

Sara Bárrios https://orcid.org/0000-0002-6541-1295

Martin Allen Hamilton https://orcid.org/0000-0002-5127-8438

Colin Clubbe https://orcid.org/0000-0002-0532-1722

Data were obtained with approval from Virgins Islands Government via the National Parks Trust of the Virgin Islands and Royal Botanic Gardens, Kew. The plant occurrence data used in these analyses were collected over many years by staff of the National Parks Trust of the Virgin Islands and the Royal Botanic Gardens, Kew in collaboration with our regional partners. The complete datasets used are not publicly accessible, in order to safeguard the precise locations of threatened species. Vetted data for the TIPAs can be found on https://tipas.kew.org.