The spatial extent of this study encompasses four sites within Taiwan and the Philippines: one in the northernmost part of the Philippines (Batanes) and three in the southernmost part of Taiwan (Kenting, Orchid Island, and Green Island). Since Batanes is in the upstream of the Kuroshio current, it was assumed that this area seeds larvae to Taiwan. Therefore, a strategic approach would be to protect this area to benefit the countries in the downstream of the Kuroshio current. For this reason, Batanes was chosen to be the main planning region where the selection of MPAs occur (Fig. 1a). This planning region of 300,000 ha is composed of 3000 cells (100 rows and 30 columns) of 100 ha area each. Random starting points for virtual larval dispersal were generated throughout the study area, and their trajectories were simulated using 27 years of acoustic doppler current profiler (ADCP) data. Initial locations for larvae that successfully reached Kenting (Fig. 1b), Orchid Island (Fig. 1c), and Green Island (Fig. 1d) were recorded along with their travel time.

Fig. 1.

Study sites: (a) Batanes, (b) Kenting, (c) Orchid Island and (d) Green Island. The yellow rectangle corresponds to the entire study region encompassing the four study sites.

(0.74MB).

Geo-referenced spatial habitat data for coral reefs and seagrass beds in 2019 were obtained from the Provincial Environment and Natural Resources Office (PENRO) in Batanes, Philippines. These datasets were rasterized in a 100 × 30 matrix (Fig. 2a and 2b). Since both coral reefs and seagrass beds are important for ecological connectivity (Dorenbosch, 2006; Nagelkerken et al., 2008), we considered the aggregate area of both habitats for future use in the model Fig. 2c).

Fig. 2.

Spatial distribution of (a) corals reefs, (b) seagrass beds and (c) total (coral + seagrass) habitat in Batanes, Phillipines.

(0.25MB).

Ocean current data from 1991 to 2018, obtained through an ADCP, was acquired from the Ocean Data Bank (ODB), Taiwan. The ADCP device is installed in Taiwan’s research vessels to measure the speed and direction of the ocean current, which is then stored in a database managed by ODB. Since coral reefs and seagrass beds are usually found in shallow waters (Guannel et al., 2016; Jagtap, 1998), the data records were filtered to exclude deep (>100 m) ocean currents. The filtered data records were then averaged and interpolated linearly for the two following seasons: summer (April to September) and winter (October to March), as shown in Fig. 3.

Fig. 3.

Averaged interpolated ocean surface currents (0–100 m depth) from 1991 to 2018 in. (a) summer and (b) winter.

(1.77MB).

Our results show that the southern part of Batanes seeds larvae to Orchid and Green Island only in summer (Fig. 5a) while there is no connection during winter (Fig. 5c). It was also found that Kenting does not receive virtual larvae from Batanes but may be connected to the Babuyan islands of northern Luzon, Philippines. The interpolated travel time from Batanes to Orchid and Green Island in summer ranges from 7 to 12 days, with larvae originating from the dark blue pixels in Fig. 5b taking 7 days to reach the sink zones, and larvae originating from the yellow pixels taking 12 days. One of the possible organisms that spawn during this time are corals from the family of Faviidae and Acroporidae (Jamodiong et al., 2018; Vicentuan et al., 2008).

Fig. 5.

Routes of virtual larvae for (a) summer and (c) winter. Colored dots represent the virtual larvae that reach Kenting (blue), Green Island (green) and Orchid Island (red). (b) Interpolated travel time for the summer connections to Orchid (northern area) and Green Islands (southern area).

(0.6MB).

Fig. 6 shows the MPAs selected by the optimization models we developed: Fig. 6a–d for the traditional approach (M1) and Fig. 6e–h for the integrated approach (M2). M1 selected ten cells that have the maximum total habitat area depending on the specified boundary length. For example, the most compact scenario (BL = 14, Fig. 6a) selected an MPA with a habitat area of 489.15 ha which is equivalent to 13.74% of the total habitat area in the whole Batanes region. Although some cells on the southwestern part of Batanes have the greatest habitat area, these are not part of the selection because of the compactness constraint that requires the ten selected cells to be arranged in a rectangle. When disaggregation is allowed (BL = 32, Fig. 6d), the selection favors individual cells with greater habitat area. The sum of the habitat area in this fragmented selection is equal to 725.90 ha (20.39%).

Fig. 6.

MPA selection (in red) and Z values (in blue) for models M1 (a–d) and M2 (e–h) under four compactness constraints (with boundary lengths BL = 14, 20, 26, 32).

(0.56MB).

When M2 is specified to have the same weights for habitat and travel time, the solution exhibits a totally different selection when high compactness is required (BL = 14, Fig. 6e). This is due to the shorter travel time on the southwestern side of Batanes (Fig. 5b). However, when greater disaggregation is allowed (BL = 32, Fig. 6h), the MPA selection for M2 tends to favor more of the same cells than M1, with a 60% similarity. This relative convergence of M1 and M2 under low compactness requirement is due to the structure of the dataset with six cells having both a high natural habitat area and short travel times.

To assess the difference between M1 and M2, we computed habitat (H) and habitat minus travel time (H-T) in both models. The objective function of M1 is H while for M2 it is H-T. To compare the behavior of both models with respect to its objective function, we manually computed the counterpart of the objective function in each model from the selection of the other model and plotted accordingly in Fig. 7. Indeed, despite excluding travel time in formulating M1, we still cannot ignore the fact that each selected cell in the M1 solution (Fig. 6a–d) has a corresponding travel time. With this, we can calculate H-T manually in order to compare its behavior with M2.

Fig. 7.

Z (objective) values (H (Z) and H-T (Z)) in comparison to their computed counterpart (H and H-T) from the selection of the other model across different boundary lengths (BL) in (a) model M1 and (b) model M2. Dotted grey lines highlight the Z values of the solutions presented in Fig. 6 for BL = 14, 20, 26, 32.

(0.22MB).

Fig. 7 shows that habitat values (H) are lower in M2 as compared to M1, but the integration of travel time in the objective function allows for much higher H-T values in M2, due to the fact that most of the selected cells in M1 do not connect and therefore suffer a penalty. The difference of H between M1 and M2 (the deviation from the compromised solution of M2 and the reference optimality of M1) can be interpreted as the opportunity cost of integrating connectivity into MPA planning.

Furthermore, this study also explored the opportunity cost of compactness which is determined by BL. This BL has always been considered crucial in MPA planning (Chen et al., 2015; Cheng et al., 2015; Fraschetti et al., 2009) which may usually cause some trade-offs with cost (Ball et al., 2000) or biodiversity (Chen et al., 2015). This trade-off between multiple objectives is common in optimization models (Eskelinen and Miettinen, 2012). A more compact MPA is usually preferred to a disaggregated one as it is thought to encompass more fish which can enhance the spillover effect (Laffoley et al., 2008). McLeod et al. (2009) summarized previous studies on the size of MPA networks and came to a conclusion that most studies suggest that the bigger the size of the MPA, the better. The reason for this is to protect higher marine habitats to shelter a larger portion of marine organisms that depend on these habitats for survival (Green et al., 2007; Mora et al., 2006; Palumbi, 2004). Our results show that given a specific size of 10 cells, a more compact MPA (BL = 14) protects a lesser percentage of habitat area (13.74%) as compared to a disaggregated one (BL = 32, 20.39%), highlighting the opportunity cost of compactness. This result contributes to the increased understanding of strategic approaches in MPA design (Cabral et al., 2020; Margules and Pressey, 2000; Sala et al., 2021).

The duration at which larvae survive in the water column differs greatly between species. For reef fishes, it was reported to range from 18 to 73 days (Victor and Wellington, 2000) while scleractinian coral larvae may survive up to 244 days before settlement (Graham et al., 2008). The computed travel time of the virtual larvae in the MPA selection shown in Fig. 6 (e–h) ranges from 7.08 to 7.75 days. This shows a higher chance that when a marine organism with a PLD greater than 7 days spawns in the southern part of Batanes, it is still viable upon reaching Orchid or Green Island. This result can hopefully spark a dialogue for both the Philippines and Taiwan to co-develop a transboundary MPA network.

Since this study is working on that idea, it is also important to consider the individual circumstances of the local players to ensure the effective management of each MPA site. This is where trade-off analysis comes into play. Despite the knowledge that a disaggregated MPA protects a higher habitat area, in the setting of a local scale (which can be the case in Batanes), it is likely more difficult to manage a disaggregated MPA due to the demands of management efforts (Chen et al., 2015), such as monitoring of MPAs (Abesamis et al., 2019), installation and replacement of buoys or markers, and patrolling which would be both financially costly and time-consuming when moving from one site to another.

Moreover, the trade-off analysis (Fig. 7) is essential when different sectors have contrasting preferences such as when environmentalists may require full protection or the protection of a greater habitat area, while fishers may argue that fishing regulations are already enough to be effective (Catalano, 2016). Alternatively, MPA managers may prefer a single compact MPA, while the national government prefers to protect as much habitat area as possible to achieve its conservation objectives. The tension that might arise from these situations can be eased with an effort to consider the views of both stakeholder groups (Harris et al., 2019), by showing the trade-off curve and providing the pros and cons of having different objectives to the decision-makers so they can weigh these factors according to their preference. The participation of these stakeholders is critical in the effective management of MPAs (Lucrezi et al., 2019) as well as in bridging the gap between science and implementation (Mills et al., 2020).

MPA networks are now being considered globally as a tool to enhance marine biodiversity and connectivity, and promote fishery management as they have been shown to be more effective than isolated MPAs (Palumbi, 2004; Metcalfe et al., 2015). The United Nations aims to protect at least 30% of the ocean by 2030. To aid in this objective, strategic MPA planning should be considered to efficiently maximize the benefit from marine waters. Transboundary MPA networks (Sala et al., 2021) and connectivity (Andrello et al., 2017) have been key considerations in this strategic global effort. However, the inclusion of connectivity in MPA design may require a wealth of biological information for both genetic approaches and biophysical modeling such as larval behavior and dispersal traits which may be unavailable, especially for large areas such as a transnational scale. This study proposes a simpler alternative for biophysical modeling to integrate connectivity in a possible future transboundary MPA network between the Philippines and Taiwan, with the aim of contributing to this global coordinated effort through M2, by which the potential source areas with higher chances of connectivity from Batanes to Orchid and Green Island were identified. Since we only mapped the well-represented habitats in the source (spawning) areas, our next step would be to determine which specific cells in the sink zones these source cells are connected to and how suitable they are for settlement. Settlement sites in Batanes should also be identified as larvae could be coming from the eastern side of the Philippines where a healthy coral reef is located (Nacorda et al., 2017). Similarly, the role of Taiwan as a stepping stone for coral connectivity along the Kuroshio current could be investigated by extending the study to Okinawa, Japan.

Furthermore, it is recommended, for future studies to: (1) include threats like coral bleaching and other anthropogenic activities to the model, (2) incorporate socio-economic features in the planning process such as fisheries and tourism, (3) expand the complexity of larval dispersal by adding stochasticity to the ocean current, and (4) validate this approach with empirical data.

This study could be a preliminary effort to establish a transboundary MPA network between the Philippines and Taiwan where negotiations among countries and other stakeholders should be considered, especially when deciding how the MPAs should be designed. Another related issue is the evaluation of the benefits of conservation in the Philippines for Taiwan, the willingness of local communities to support such a program, and the exploration of potential mechanisms to ensure equity and benefit sharing between these two countries.