Gone with the wind: Seasonal distribution and habitat use by the reef manta ray ( Mobula alfredi ) in the Maldives, implications for conservation

1. Reef manta rays ( Mobula alfredi ) are one of the ocean's largest and most charismatic species. Pressure from targeted and bycatch fisheries coupled with their conservative life-history traits including slow growth, late maturity, and low fecundity has led to catastrophic declines of the global population. The species is now listed as Vulnerable to Extinction on IUCN Red List of Threatened Species. 2. The global M. alfredi population is widely distributed in highly fragmented subpopulations. The Maldives supports the world's largest known subpopulation that undergoes seasonal migrations which are thought to be linked to peaks in ocean productivity induced by the South Asian Monsoon. Although the species is protected from targeted fisheries in the region, increasing pressures from habitat degradation and unsustainable tourism activities mean their effective conservation relies upon knowledge of the species' habitat use, seasonal distribution, and the environmental influences on such movements. 3. Photo-ID sighting records and 2017 were used identify key aggregation sites throughout the archipelago, and multiple linear regression and prediction analysis identified the environmental variables affecting variations in the intra-annual sighting frequency of M. alfredi . Mobula were classified as key areas of habitat use. South-west monsoon winds and chlorophyll- a concentration predominantly affected the monthly percentage of M. alfredi sighted on the down-current side of the atolls. 5. In a country where climate change and touristic pressure are increasingly threat-ening this species and its habitat, the identification of key areas of habitat use and temporal changes in the use of these sites highlight the areas that should be prior-itized for protection enabling more effective conservation management.

Due to these declines, coupled with their conservative life history traits including slow growth, late maturity, and low fecundity (Lawson et al., 2017;Marshall & Bennett, 2010;Stevens, 2016;Stewart, Jaine, et al., 2018) they are now listed as Vulnerable to Extinction on the IUCN Red List of Threatened Species (Marshall et al., 2018).
Despite the economic value of M. alfredi to some local economies (Anderson, Adam, Kitchen-Wheeler, & Stevens, 2010;O'Malley, Lee-Brooks, & Medd, 2013), poorly managed tourism, development, and habitat degradation are increasingly impacting this species, especially at ecologically important aggregation sites (Murray et al., 2019;Rohner et al., 2013;Stevens & Froman, 2018;Venables, McGregor, Brain, & Van Keulen, 2016). Furthermore, the species is likely to be vulnerable to the impacts of climate change, such as rising sea surface temperatures, which have the potential to reduce the manta ray's food availability (Richardson, 2008). Therefore, to ensure the conservation of M. alfredi, there is a need to identify and effectively protect areas of important habitat for this species throughout its range (Stewart, Jaine, et al., 2018).
Site fidelity and migratory behaviour in M. alfredi have been linked to areas of high primary productivity and prey density (Armstrong et al., 2016;Jaine et al., 2014) and may also vary by sex and age-class (Couturier et al., 2011;Stewart, Nuttall, Hickerson, & Johnston, 2018. The 26 coral atolls that form the Maldives archipelago support the world's largest known subpopulation of M. alfredi (Kitchen-Wheeler, Ari, & Edwards, 2011;Stevens, 2016). The migratory behaviour of this subpopulation is strongly influenced by the South Asian Monsoon (SAM) (Anderson, Adam, & Goes, 2011), which drives currents that enhance productivity on the leeward side of the atolls through deep-water upwellings (Doty & Oguri, 1956;Sasamal, 2006), bringing nutrient-rich water into the euphotic zone (Deik, Reuning, & Pfeiffer, 2017;Sasamal, 2006). Mobula alfredi follow these productivity hotspots, migrating across the archipelago with the biannual reversal of winds and the concomitant ocean surface currents, exploiting the richest zooplankton feeding grounds (Anderson et al., 2010;Kitchen-Wheeler et al., 2011).
Although all ray species are protected from target fisheries in the Republic of Maldives (EPA, 2014), the total combined area protected by the government consists of 42 marine protected areas (MPAs) that cover just 116.3 km 2 , which is only 0.5% of the area (21,596 km 2 ) that falls within the boundaries of the 26 geographical atolls' outer rims (Stevens & Froman, 2018). Only one MPA, Hanifaru Bay, has a management plan (in place since July 2011), with on-site enforcement of the regulations (Stevens & Froman, 2018).
More effective protection of the Maldives' M. alfredi subpopulation is needed in the face of increasing pressures from habitat destruction, climate change, incidental bycatch, and tourism (Stevens & Froman, 2018). Effective protection relies heavily on a more detailed understanding of how the subpopulation utilise their environment and identification of the environmental factors that influence distribution. This study aims to assist conservation planning by using in-water manta ray photo-ID records, combined with environmental data, to: (1) identify locations used by this species throughout the archipelago; (2) identify the primary function of these sites for M. alfredi; (3) determine which environmental drivers influence site use; and (4) determine annual patterns in M. alfredi presence at these sites.

| Data collection
The Maldives archipelago extends 870 km from 7 north to half a degree south of the equator in the Indian Ocean (Figure 1). During a 13-year study, from 2005 to the end of 2017, over 15,000 surveys were undertaken throughout the Maldives at known M. alfredi aggregation sites, and opportunistically at other locations, to photographically record the individuals present. Environmental data on the wind and primary productivity were obtained for the same period.

| Manta rays
Identification photographs (photo-ID) were taken of the ventral side of the manta rays at aggregation sites throughout the Maldives. The images captured the unique gill-plate spot pattern, which can be used to identify the individual throughout its lifetime (Kitchen-Wheeler,-2010). These images also allow the sex and physical condition of the individual to be determined (Kitchen-Wheeler, 2010). In the context of this study, a sighting is defined as a confirmed photo-ID of an individual M. alfredi on a given day at a defined location. When manta rays were encountered, where possible, photo-ID and behavioural activity of each individual was recorded. Behavioural activity was broken down into four major groups: (1) feeding; (2) cleaning; (3) cruising; and (4) courtship. During an encounter, an individual may undertake several different activities. In these situations, the activity that dominated the encounter was recorded as the primary behaviour. A typical survey during this study was performed via scuba or freediving from either a dedicated research vessel or commercial diving vessels. Scuba surveys lasted on average 60 minutes and ranged to a maximum depth of 30 m. Freediving surveys lasted on average 120 minutes.
Surveys were undertaken by one of the authors (Stevens, 2016), or by trained staff members or volunteers from the Manta Trust (www. mantatrust.org).
Surveys were performed at different times of day throughout the month in all months of the year. However, at the known M. alfredi aggregation sites, surveys were most likely to be undertaken during the period when sightings were most likely to occur, creating some sampling bias. Nonetheless, this dataset offers an opportunity to explore the distribution of this species throughout the Maldives in the most detailed way possible to date.

| Wind speed and direction
The Maldives south-west (SW) monsoon (season), or Hulhangu, occurs from May to October, while the north-east (NE) monsoon, or Iruvai, occurs from December to March each year, with the months of November and April considered as transitional periods of change in between (Anderson et al., 2010). However, the transition periods between the monsoons are highly variable, with reports that they also extend into October and March (Aslam & Kench, 2017). Daily mean wind direction and wind speed data were obtained from the Maldives' Meteorological Service (MMS) in Malé. These data were used to calculate the monthly wind direction as the percentage of days in a month that the wind direction represented the NE monsoon (0-90 ), or the SW monsoon (202.5-315 ) (Anderson et al., 2011). Mean monthly wind percentage was calculated to show the annual period (season) of each monsoon and identify the months in which transition between the monsoons occurs. Monthly mean wind speed was calculated separately for days classified as having NE, SW or 'other' (i.e. neither NE nor SW) wind direction.
The data were extracted from each location where reef manta rays were sighted using the Marine Geospatial Ecology Tools package (Roberts, Best, Dunn, Treml, & Halpin, 2010) via ArcGIS. Monthly mean Chl-a concentration (mg/m 3 ) was then calculated separately for the east and the west side of the atolls using values for the days and locations where reef manta rays had been sighted that month.

| Biannual migration
To assess biannual migration, the east and west side of the atolls were established by creating a map of the Maldives in ArcGIS 10.5 including polygons for each of the 26 geographical atolls. Two atolls F I G U R E 1 The 48 sites identified as key M. alfredi habitats (>100 sightings) during NE (left map) and SW (right map) monsoon for the study period [2005][2006][2007][2008][2009][2010][2011][2012][2013][2014][2015][2016][2017]. Cleaning stations (blue circles) and feeding areas (red circles). Site numbers correspond to Table 1 (cleaning stations) and Table 2 (feeding areas) (Thiladhunmathi and Vaavu) have a 'lopsided' shape thus their polygons were subdivided giving a total of 30 polygons. Each atoll where sightings occurred was then divided into east and west by establishing the true centroid of each polygon using Calculate Geometry ( Figure S1), the polygon was then divided into angle segments using Data Management Tools. All the sighting locations were then projected, and those within 1-179 were classified as east, and those within 181-359 were classified as west. All sightings were then integrated and projected as monthly total sightings at each location. Any locations with 1-4 sightings in the month were excluded to reduce the bias a small number of sightings may have on subsequent analysis.
The effect on the total number of M. alfredi (log 10 (y+1)) sightings on each side of the atoll (east, west) and monsoon (SW, NE) was assessed by two-way analysis of variance (ANOVA) (R 3.5.2; R Core Team, 2013). The order of incorporation of the explanatory variables into the model was determined by Regsubsets of the 'leaps' R package (Lumley, 2017). An assessment of autocorrelation was made by visual inspection of the autocorrelation function (ACF) plot of residuals, followed by a Durbin-Watson test from the 'lmtest' library (Millo & Mitchell, 2017). Models that did not meet the requirements of the Durbin-Watson test (Field, Miles, & Field, 2012) were excluded from analysis. Models were then validated through an inspection of resid-

uals and the application of Global Validation of Linear Models
Assumptions of the 'gvlma' R package (Pena & Slate, 2006). Models that did not satisfy all assumptions were also excluded from the analysis.
An information theoretic approach was adopted to provide a quantitative measure of relative support via ranking and weighting of models thus allowing some inferences to be made about all models (Burnham & Anderson, 2002). Rank was established using corrected Akaike information criterion (AIC c ) test statistic, which is an asymptotically unbiased estimator of model quality (Burnham & Anderson, 2002). Models are not assessed by the absolute size of AIC c but by their relative values over candidate models, particularly the differences between AIC c values ( Δ AIC c ) (Burnham & Anderson, 2002). Δ AIC c is calculated using the following form where i is the model: The relative merits of the models were assessed based on the criteria specified by Burnham and Anderson (2002) where the model estimated to have the greatest support has while models with Δ AIC c < 2 are considered to have substantial support, models with 4-7 Δ AIC c have considerably less support, and those with Δ AIC c > 10 have essentially no support (Burnham & Anderson, 2002). Plausible models for the current study were identified as those with Δ AIC c < 2 and all other models were excluded from the analysis except the null model which was retained for comparison.
To effectively scale and interpret the Δ i values of the chosen models, Akaike weights ( w AIC c ) were calculated using the following form where R is the set of models: Therefore, w AIC c is relative to the set of chosen models and ranges from 0 (no support) to 1 (complete support). AIC c , Δ AIC c , and w AIC c were obtained using the 'MuMin' R package (Barto n, 2018).
The accuracy of an estimated parameter was inferred from 95% confidence intervals (CIs) for the slope (β). A narrower CI range implies a more precise estimation while a CI that does not span zero indicates that the null hypothesis may be rejected (Arnold, 2010). For the current study, CI was calculated using the 'MASS' R package (Brian, Venables, Bates, Firth, & Ripley, 2018).
To remediate uninformative parameters, model averaging using the full-model averaging approach was conducted using the 'MuMin' R package (Barto n, 2018) whereby the β is averaged across the set of competing models (Burnham & Anderson, 2002).
Model averaging calculates a weighted average of parameter estimates,β i across all models, those including and excludingβ i . The esti-

| Predictive models
The The prediction parameters were set using the environmental variables from each month, and the accuracy of predictions was assessed by comparing the results to the actual percentage of manta rays observed.
Error margins for the difference between the predicted and actual  Overall, 4,014 individual M. alfredi were sighted at the 48 key aggregation sites; 3,124 were sighted more than once, of which 2,588 were only sighted within the same atoll, and 755 were always seen at the same site. Of the 2,369 individuals sighted at more than one site, 1,352 were predominantly sighted at one location (>55% of sightings were at the same site).

| Wind direction
Mean monthly wind direction percentage indicated that that the SW monsoon occurred from April until November, and the NE monsoon runs from December to March. The transition months appeared to be November/December between the SW and NE monsoon, and March/April between the NE and SW monsoon (Figure 2).

| Biannual migration
More sightings of M. alfredi were on the east side of atolls during the SW monsoon and more on the west side during the NE monsoon ( Figure S2). The significant interaction between the side of atolls and the monsoon period (F 1, 44 = 55.59, P < 0.001) supports this conclusion. These results support the biannual east-west and west-east migration pattern reported by Anderson et al. (2011).

| Environmental influences
The influence of the environmental factors measured (Table 3)    The 95% CI for the explanatory variables of each of the models and the averaged model ( Figure 5) showed that the CI for SWWS was consistent throughout the models and had the narrowest range, which did not span zero, indicating that the null hypothesis for this variable may be rejected (Arnold, 2010 25.8 0 n/a n/a n/a plausible models within the Δ AIC c < 2 thresholds were produced (Table 5). Model E15 is the highest-ranking model ( Δ AIC c = 0, w AIC c = 0.367) which indicated increasing SWWS and monthly mean Chl-a concentration on the east side of the atolls (ECHLA) increased EMAN ( Figure 6).
The model explained 20% of the variation in EMAN (F 2,106 = 14.80, R 2 = 0.20, P << 0.001). All four models within the Δ AIC c < 2 thresholds contain SWWS and ECHLA and explained a similar amount of variation, but with the mean NE monsoon wind speed, monthly NE monsoon wind frequency, and WDSW as additional variables in models E25, E17, and E18, respectively, which indicated that these variables did not improve model E15. Multimodel inference via CI and model averaging (Figure 7), provided evidence that the variables of model E15 were useful parameters as neither spanned zero (Arnold, 2010), although the CI of ECHLA was relatively wide. In all other models, the additional variables spanned zero; thus, E15 may be considered the most plausible model.

| Prediction models
The models W8, W15, and E15 were used to predict the monthly percentage of manta rays on one side of atolls. These predictions were then compared with the actual monthly percentages observed; when assessing the accuracy of predictions, an absolute difference of <15% between predicted and observed percentage was deemed accurate and 15-20% acceptable. Prediction accuracy was assessed across three different time periods (months, years, monsoons). For model W8 (WMAN~SWWS + WDSW), 80 months of the 153 for which data were available were accurately predicted (|prediction-actual dif-ference| < 15%), 16 months were acceptable (|prediction-actual differ-ence|15-20%) and the remaining 57 months had differences >20% (Table S1). There was no significant difference between W8 predicted F I G U R E 3 Relationship between the monthly percentage of manta rays on the west side of the atolls (WMAN) and the variables identified by model W8 (monthly mean SW monsoon wind speed, SWWS+ the percentage of days each month that the wind direction represented the SW monsoon (202.5-315 , WDSW) with regression plane of best fit to data points F I G U R E 4 Relationship between the monthly percentage of manta rays on the west side of the atolls (WMAN) and the variables identified by model W15 (monthly mean SW monsoon wind speed, SWWS+ mean chlorophyll-a concentration on the west side of the atolls, WCHLA) with regression plane of best fit to data points Monsoon months were predicted accurately for nine of the 13 years; in the other years, predictions were acceptable. However, for the transition only months (March, April, November, and December), the differences were > 20% in all 13 years.
Predictions differences for the monsoon months were accurate for six years (2006, 2010, 2013, 2014, 2015, and 2016) and acceptable for five more years (2005, 2007, 2009, 2011 and 2017 3 0 n/a n/a n/a F I G U R E 6 Relationship between the monthly percentage of manta rays on the east side of the atolls (EMAN) and the variables identified by model E15 (monthly mean SW monsoon wind speed, SWWS + mean chlorophyll-a concentration on the east side of the atolls, ECHLA) with regression plane of best fit to data points F I G U R E 5 Point estimate of the monthly percentage of manta rays on the west side of the atolls (WMAN)j (y − mean (y))jwith respective +/− 95% confidence interval of each variable in the west models within the Δ AIC c < 2 threshold. SWWS = monthly mean SW monsoon wind speed, WDSW = the percentage of days each month that the wind direction represented the SW monsoon (202.5-315 ), WCHLA = monthly mean chlorophyll-a concentration on the west side of the atolls transition months (March, April, November, and December), predictions were acceptable for three years (2009, 2011, and 2014).

| DISCUSSION
This study identified 171 M. alfredi aggregation sites throughout the Maldives archipelago; 48 of which were feeding areas or cleaning stations that, based on the high number of individuals sighted, were considered areas of key habitat use. Cleaning stations provide essential benefits for M. alfredi, such as parasite removal, as well as social and reproductive interactions (Stevens, 2016;, while feeding hot-spots provide the concentrated food source required for their energetically efficient foraging strategies (Armstrong et al., 2016;Stevens, 2016). It is likely that more key aggregation sites exist, especially in regions of the country (the F I G U R E 8 Mean prediction difference between the actual monthly percentage of manta rays on the west side of the atolls (WMAN) observed each year and the WMAN calculated by R predict() function using the variables identified by model W8 (monthly mean SW monsoon wind speed, SWWS + the percentage of days each month that the wind direction represented the SW monsoon (202.5-315 , WDSW). Shown as mean annual prediction difference, prediction difference for transition months only (March, April, November, and December) and prediction difference for monsoon months only (January, February, May-October), all with +SE. The red line shows the acceptable mean prediction difference threshold (20%) F I G U R E 7 Point estimate of the monthly percentage of manta rays on the east side of the atolls (EMAN)j (y − mean (y))jwith respective +/− 95% confidence interval of each variable in the east models within the Δ AIC c < 2 threshold. SWWS = monthly mean SW monsoon wind speed, WDSW = the percentage of days each month that the wind direction represented the SW monsoon (202.5-315 ), ECHLA = monthly mean chlorophyll-a concentration on the east side of the atolls, NEWS = monthly mean northeast monsoon wind speed, WDNE = the percentage of days each month that the wind direction represented the NE monsoon (0-90 ) northernmost atolls) where surveys were less frequently undertaken.
However, the extensive nature of this study, both spatially and temporally, means that many of the key aggregation sites, within the shallow (<30 m) reef systems of the Maldives, will have been recorded.
The current study also provides quantitative evidence that M. alfredi migrates east-west and west-east biannually, supporting previous observations (Anderson et al., 2011;Kitchen-Wheeler et al., 2011). The results of MLR modelling and prediction analysis F I G U R E 1 0 Mean prediction difference between the actual monthly percentage of manta rays on the east side of the atolls (EMAN) observed each year and the EMAN calculated by R predict() function using the variables identified by model E15 (monthly mean SW monsoon wind speed, SWWS + mean chlorophyll-a concentration on the east side of the atolls, ECHLA). Shown as mean annual prediction difference, prediction difference for transition months only (March, April, November, and December) and prediction difference for monsoon months only (January, February, May-October), all with +SE. The red line shows the acceptable mean prediction difference threshold (20%) F I G U R E 9 Mean prediction difference between the actual monthly percentage of manta rays on the west side of the atolls (WMAN) observed each year and the WMAN calculated by R predict() function using the variables identified by model W15 (monthly mean SW monsoon wind speed, SWWS + mean chlorophyll-a concentration on the west side of the atolls, WCHLA). Shown as mean annual prediction difference, prediction difference for transition months only (March, April, November, and December) and prediction difference for monsoon months only (January, February, May-October), all with +SE. The red line shows the acceptable mean prediction difference threshold (20%) suggests that this distribution pattern was predominantly influenced by the SW monsoon winds and Chl-a concentration. The model for the east side of the atolls (E15; mean SW monsoon wind speed, SWWS + mean Chl-a on the east side of the atolls, ECHLA), linked the increase in the percentage of manta rays on the east side of the atolls (EMAN) to productivity enhanced by the strong ocean surface currents induced by the SW monsoon winds (Deik et al., 2017;Sasamal, 2006). On the west side of the atolls, productivity is increased by the NE monsoon winds (Sasamal, 2006); however, these winds are dominated by the onset and retreat of prevalent SW monsoon (Schott & McCreary, 2001). Both of the plausible models for the west side (W8; SWWS + SW monsoon wind frequency, WDSW. W15; SWWS + monthly mean Chl-a concentration on the west side of the atolls, WCHLA) identified the effect of SW monsoon windsdecreased wind speed increased primary production on the west side, increasing the percentage of manta rays observed (WMAN).
The model results highlight the prominent role of the SW monsoon in driving productivity, which supports the Maldives M. alfredi subpopulation. In particular, the longer duration of the SW monsoon means a comparatively longer period of enhanced primary production (Strutton et al., 2015). Moreover, as primary productivity can be suppressed during the NE monsoon due to the inflow of low-salinity surface waters from the eastern Indian Ocean and Bay of Bengal (Bruce, Johnson, & Kindle, 1994;Schulte, Rostek, Bard, Rullkötter, & Marchal, 1999), there might be greater food availability during the SW monsoon. Reef manta fecundity is linked to food availability (Ramirez-Llodra, 2002;Stevens, 2016), and productivity peaks that occur towards the end of the SW monsoon (Schulte et al., 1999) coincide with reproduction (Stevens, 2016).
Climate change has historically influenced primary production in the Indian Ocean through the intensification of the SAM winds (Gupta, Singh, Joseph, & Thomas, 2004). The modern SAM is influenced by anthropogenic climate change (Roxy et al., 2015). For example, emissions over India have subdued warming over land masses, reducing the land-sea thermal gradient (Roxy et al., 2015;Turner & Annamalai, 2012). The reduced thermal contrast affects the seasonal migration of the inter-tropical convergence zone, the shift of which is an essential component in the onset and retreat of the SW monsoon (Yadav, 2013). The strength of the SW monsoon is influenced by meteorological teleconnections, including two synopticscale jets, the low-level jet and tropical easterly jet (Kalapureddy, Rao, Jain, & Ohno, 2007). The low-level jet has been following a weakening trend since the 1950s; 'weak' spells in the SW monsoon with reduced wind speeds have increased by 30% (Joseph & Simon, 2005). A similar trend has been observed for the tropical easterly jet, which has been attributed to the cooling of land masses and warming of the Indian Ocean (Abish, Joseph, & Johannessen, 2013). The weakening of the SW monsoon winds intensifies ocean stratification and inhibits upwelling, thus lowering primary production in the Indian Ocean (Singh, Jung, Anand, Kroon, & Ganeshram, 2018). Enhanced stratification in the region is also driven by rising sea surface temperature, which is suggested to have decreased marine phytoplankton by up to 20% in the last 60 years (Roxy et al., 2016).
The predictable utilization of key aggregation sites and the large number of M. alfredi present suggest increased vulnerability to anthropogenic activities at these locations. Having identified the primary function of these key sites for M. alfredi, future conservation measures can now be focused more effectively. For example, the establishment of no-take MPAs at feeding locations would reduce manta ray vulnerability to fishing gear entanglement (Stevens & Froman, 2018), which can result in disfigurement and disablement (Deakos, Baker, & Bejder, 2011). It would also protect them from boat strikes and propeller injuries, which are common (Stevens & Froman, 2018) and have been highlighted as a major concern for M. alfredi subpopulations (Germanov & Marshall, 2014;Graham et al., 2012;Stewart, Jaine, et al., 2018). These threats also extend to other charismatic species in the Maldives (Stevens & Froman, 2018), including whale sharks (Rhincodon typus), with reports that as many as 40% of whale sharks encountered in the South Ari Atoll MPA bear injuries and scars caused by ocean vessels or other anthropogenic activities (Collins, 2013). At cleaning stations, MPAs can reduce damage from anthropogenic activities, which may degrade the habitat such as the intentional destruction of coral reefs to allow boat access and contact damage caused by divers and snorkellers (Stevens & Froman, 2018). Habitat degradation reduces live coral cover and in turn reef fish abundance (Jones, McCormick, Srinivasan, & Eagle, 2004) and cleaner wrasse activities (Arnal, Kulbicki, Harmelin-Vivien, Galzin, & Morand, 2002;Triki, Wismer, Levorato, & Bshary, 2018), potentially influencing reef manta ray visitation patterns (Barr & Abelson, 2019). If disruption of the mutualistically symbiotic relationship between cleaner fish and M. alfredi occurs, it could compromise the manta ray's fitness (Côté, 2000). Furthermore, at Although a short-term response to human interaction, disturbance is cumulative, and thus can incrementally develop into significant impacts (Venables et al., 2016). Semeniuk, Bourgeon, Smith, and Rothley (2009) (Stevens, 2016). Similarly, displacement from cleaning stations may impact breeding success, as these sites are important aggregation sites for reproductive activity .
For an MPA to be effective, the designation of protection must also be accompanied by a comprehensive management plan which includes a code of conduct (CoC) and active enforcement (Venables et al., 2016). In the Maldives, a CoC and 10-step guide to sustainable tourism was published in 2017 (Murray et al., 2019) to help mitigate the impacts of touristic pressure (https://swimwithmantas.org/).
However, due to the lack of government enforcement, compliance with these regulations is mainly voluntary (Murray et al., 2019). Without active enforcement, CoC compliance has been shown to be limited (Allen, Smith, Waples, & Harcourt, 2004;Murray et al., 2019) and the rate of compliance diminishes the longer the regulations remained unenforced (Schleimer et al., 2015). The protection of M. alfredi and its aggregation sites will also ensure protection for many other marine wildlife (Roff & Evans, 2002). Moreover, MPAs may increase resilience to the impacts of climate change (Roberts et al., 2017) by reducing anthropogenic stressors that can increase susceptibility (Cabral, Fonseca, Sousa, & Leal, 2019). For example, increased sea surface temperatures can cause temperature-driven increases in metabolism, thus increasing food requirements (Pistevos, Nagelkerken, Rossi, Olmos, & Connell, 2015). Acclimation is possible via physical or behavioural adaptation (Pistevos et al., 2015); however, human stressors such as organic pollution operate synergistically with increased temperature (Cabral et al., 2019), increasing sensitization (Sokolova & Lannig, 2008) and hindering the animals' ability to adapt (Cabral et al., 2019).  (Stevens & Froman, 2018).
These actions would significantly improve protection for M. alfredi in the Maldives, as well as assist the government achieving its commitments to the Convention on Biological Diversity Aichi Biodiversity Targets. Furthermore, the establishment and maintenance of larger MPAs are more cost effective than smaller MPAs (McCrea-Strub et al., 2011). Currently, due to the absence of management plans and active enforcement at all but one of the Maldives MPAs, they are little more than 'paper parks' that do not offer adequate protection (Mohamed, 2007;Rife, Erisman, Sanchez, & Aburto-Oropeza, 2013).
Therefore, the introduction of species-and area-specific management plans for current and future MPAs, and the active enforcement thereof is also required which will greatly assist the protection of M. alfredi and the coral reefs in the Maldives.
To further the current study, future research would benefit from the use of tagging to track M. alfredi movements and identify other key aggregation sites (Stewart, Nuttall, et al., 2018). Site use should also be demographically defined to identify essential requirements for the species reproductive success and overall fitness. Modelling techniques, such as ecological niche factor analysis, could then be used to highlight sites that may be crucial for M. alfredi survivorship. Furthermore, the potential impacts of climate change, such as the weakening of the SW monsoon, should be investigated as a matter of priority.

| CONCLUSION
The current study used multiple linear regression and prediction analysis to develop the current understanding of the relationships between M. alfredi seasonal movement in the Maldives and environmental variables. The quantitative evidence presented confirms that M. alfredi movements are predictable both spatially and temporally.
While this is advantageous to the tourist economy, the current lack of specific protections for the species throughout the majority of the Maldives archipelago has been shown to lead to direct injury, displacement, and habitat degradation due to anthropogenic stressors.
This study identifies how and when sites are most likely to be used by M. alfredi, thus highlighting areas of conservation concern. This knowl-