The Maldives is a 900 km-long archipelago (Naseer and Hatcher, 2004) that extends from 6°57 ´N to 0°34 ´S and is situated in the central Indian Ocean ( Figure 1 ) (Kench and Brander, 2006). It consists of 16 atolls, five oceanic faros and four oceanic platform reefs, incorporating 2,041 reef structures with a total area of 4,494 km² (Naseer and Hatcher, 2004). The Maldivian landmass is comprised of 1,190 coral reef islands, that are mid- to late Holocene in age, with a total area of 227.45 km² (Naseer and Hatcher, 2004; Kench et al., 2005; East et al., 2018). It is the lowest nation on Earth (< 1.5 m above mean sea level; MEE, 2017) and is, therefore, typically considered extremely vulnerable to the effects of climate change, particularly sea level rise (ref). This study is focused on Huvadhoo Atoll, which is located in the southern Maldives (0°32 ´N, 73°17 ´E) and is the nation’s largest atoll at approximately 60 km wide, 80 km in length and with an area of 3279 km² (Naseer and Hatcher, 2004). The atoll rim is discontinuous, with channels between the atoll rim platforms allowing ocean waves and currents to propagate across the lagoon, which has a depth of 80 m. The atoll has 241 reefs, with a combined area of 437.9 km² (Naseer and Hatcher, 2004).

The climate in the Maldives can be divided into two distinct monsoon periods, Hulhangu and Iruvai, which are marked by strong seasonal reversals in wind direction (Kench and Brander, 2006). The Hulhangu monsoon period takes place in April to November and is characterised by west to northwest wind directions with a mean wind speed of 5.1 m s^-1. The Iruvai monsoon takes place during December to March and is characterised by east to northeast wind directions with a mean wind speed of 4.9 m s^-1 (Kench and Brander, 2006). Oceanic swell waves advance from a south-easterly direction between November and March with a mean wave significant height (Hs) of 0.75 m and a south to south-westerly direction between April and November with a peak wave height of 1.8 m in July (Young, 1999). Thus, wave energies between November and March are greater than those between April and November. WaveWatch III model hindcasts for the time period 1979 – 2010 found that wave height and wave period (TO) were significantly higher and longer at windward sites (Hs = 1.6 ± 0.4 m, TO = 10.0 ± 1.6 s) compared to leeward sites (Hs = 1.4 ± 0.4 m, TO = 9.7 ± 1.5 East et al., 2018).

Sea level trends vary spatially within the Maldives as tide gauge records from 1987 to 2014 for Malé Atoll in the central Maldives show an increase of 3.2 mm/year, whereas Addu Atoll in the south shows an increase in sea level of 4.0 mm/year (Caldwell et al., 2015). However, NASA satellite data for Malé and Addu Atolls from 1993 to 2019 suggest sea level was rising 4.12 ± 1.22 mm/year and 3.07 ± 0.97 mm/year respectively (NASA, 2022). In Huvadhoo Atoll, sea level reconstructions from fossil coral microatolls indicate that sea level is rising by approximately 4.24 mm/year (Kench et al., 2020). Huvadhoo Atoll experiences a mean spring tide of 0.96 m and the tidal regime is semi-diurnal tidal with strong diurnal inequality (Wadey et al., 2017; Kench et al., 2022).

A total of 49 islands were analysed across Huvadhoo Atoll with 31 islands on the atoll rim (islands on the atoll periphery) and 18 located within the atoll lagoon. Islands were categorised into four different types ( Figure 1 ): (1) lagoon (n = 18) islands are those located on reef platforms within atoll lagoons; (2) leeward rim (n = 10) and (3) windward rim (n = 11) islands, based on their exposure to open ocean swell whereby windward islands were located on the south-western atoll rim and leeward islands were located on the northern-eastern atoll rim and; (4) anthropogenic (n = 10) islands were those with human modifications to their shorelines, such as land reclamation, harbours and coastal development. As such, these categories represented the variety of islands within the Maldives.

Shoreline change positions were derived from black and white historical imagery taken by the Royal Air Force from 1969, and high-resolution Google Earth satellite imagery from between 2006 and 2019 ( Table 1 ) (Testut et al., 2016). Imagery from 1969 was scanned from hard copies or negatives at minimum resolution of 900 dpi and had a scale of<1:25,000 (Aslam and Kench, 2017). Satellite images were georeferenced in ArcMap 10.5, using fixed natural (e.g., coral reefs and beach rock) and anthropogenic (e.g., corners of roads and corners of buildings) features as control points that were readily identifiable in both the source image and image undergoing georeferencing (Yates et al., 2013; Aslam and Kench, 2017).

The first stage in shoreline analysis was to establish a shoreline proxy. Shoreline proxies are indicators used to assess shoreline and island change (Duvat et al., 2017), such as the high-water line, base of the beach, debris line, edge of coastal structures, mean high water line and vegetation line (Biribo and Woodroffe, 2013). Following previous reef island shoreline change research, this study used the vegetation line as the shoreline proxy owing to its ability to be readily identified in all sources of imagery (Webb and Kench, 2010; Ford, 2013; Mann and Westphal, 2016; Aslam and Kench, 2017). The vegetation line was digitized for all images and saved as a shapefile in ArcMap 10.5. Shoreline change was analysed using the Digital Shoreline Analysis System (DSAS), which is available as an ArcGIS extension (Thieler et al., 2009). DSAS uses a baseline to calculate shoreline change metrics based on transects cast at 5 m intervals along the shoreline and a confidence interval of 2σ (95%) was applied when determining shoreline change rates (Thieler and Danforth, 2016). Two shoreline change metrics were selected for this study. Firstly, net shoreline movement (NSM, m) was calculated, which is a measure of the net change between the oldest and most recent shoreline year (Ford, 2012; Thieler and Danforth, 2016; Sengupta et al., 2021). Secondly, weighted linear regression (WLR, m/year) was calculated as a rate of change shoreline statistic where a greater emphasis is placed on data points where the position uncertainty is smaller (Thieler and Danforth, 2016). Consequently, shorelines with a higher degree of uncertainty have less influence on the WLR than those with lower uncertainty, and it thus provides a robust measure of shoreline change (Ford, 2012).

When analysing shoreline change from aerial photography and satellite imagery, errors and uncertainties were quantified. Three potential sources of error were identified: image resolution (pixel size), georeferencing and shoreline digitization (Ford, 2012; Duvat et al., 2017). Georeferencing error, the root mean square (RMS) value ArcGIS records during georeferencing, ranged from 0.0 m to 2.7 m. Image resolution error is the pixel size of each image, which ranged from 0.2 m to 5.0 m. Digitizing error was calculated as the average of digitizing the same strip of coastline by the operator 10 times and calculating the width at the widest points. Errors ranged from 0.2 m to 1.0 m for shoreline digitization, and an average was calculated as 0.54 m. An overall measure of uncertainty was then calculated as the square root of the sum of each individual uncertainty squared (RMSE). RMSE was calculated for 1969, 2006, 2011, 2013, 2014, 2015, 2016 and 2019 for each island shoreline and varied from ± 0.3 m to 4.2 m. Table 1 displays the average RMSE ± 1 S.D for each shoreline year. Islands that were considered eroding had values lower than average RMSE, and islands that were accreting had values higher than average RMSE. Islands that were classified as stable lied within the ± average RMSE boundary.

The following steps were used to obtain estimates of groundwater depth and volume for 2030, 2050 and 2100. Annual precipitation for GDh. Kaadehdhoo island in 2019 was obtained from the 2020 Maldives Statistical Year Handbook (National Bureau of Statistics, 2021). Daily potential evapotranspiration was calculated using the Thornthwaite method (1948), as applied in (Bailey and Jenson, 2014; Bailey et al., 2014), and aquifer recharge was calculated using the Griggs and Peterson method (1993). Estimates of future island width were calculated using the width of each island from 2019 and the corresponding island’s average rate of shoreline change (weighted linear regression, m/year), resulting in widths for the years 2030, 2050 and 2100. By using each island’s average rate of shoreline change, we assume future sea level rise rates were similar to historical rates. These island widths were used to calculate the volume of the freshwater lens (Oude Essink, 2001). Firstly, the deepest position of the fresh-salt water interface was calculated.

where B is the width of the island (m), α = 0.025, k is hydraulic conductivity and f is recharge (m/day). Once H m a x was calculated, the freshwater lens volume (V) was estimated.

where n e   is effective porosity (m³/m).

For each island where horizontal hydraulic conductivity was 75 m/day (Bailey et al., 2014), effective porosity was 0.2 m³ m^-3 (Anthony, 1987), and alpha was 0.025 (Oude Essink, 2001). Freshwater lens volumes were estimated based on changes in precipitation for 2030, 2050 and 2100, and under different emission scenarios for each time point (Relative concentrating pathway (RCP) 2.6, 4.5, 7.0 and 8.5). A summary table for changes in total precipitation relative to 1995-2014 for each time point, emission scenario and model output can be found in Supplementary Table 2 . These estimates in total precipitation were obtained from the IPCC WG1 Interactive Atlas using CMIP6 dataset with advanced uncertainty (Gutiérrez et al., 2021). Supplementary Table 3 lists all 33 global climate models (GLMs) used in this analysis. The models used in the Atlas generate a range of statistical values therefore to reflect model output uncertainty. Volumes were generated based on the 5th percentile (P5), median and 95^th percentile (P95) values (Gutiérrez et al., 2021).

Shoreline change was analysed by comparing changes in shoreline positions in historical aerial imagery from 1969 and satellite images from between 2006 and 2019 at Huvadhoo Atoll and results show varying patterns of shoreline change ( Figure 2 ). Out of the 49 islands studied, 26 islands accreted (53%), 12 eroded (25%) and 11 remained stable (22%) ( Table 2 ). Out of a total of 14,043 transects, 7,688 accreted (55%), 5,223 eroded (37%) and 1,133 exhibited no change (8%). Accretion dominated anthropogenic and lagoon islands whereby 90% and 61% accreted, respectively. Windward islands primarily accreted (37%) or remained stable (37%). A higher proportion of leeward islands underwent erosion (40%) or remained stable (40%). Overall, NSM was 4.13 ± 23.99 m and WLR was 0.13 ± 0.60 m/year, with anthropogenic islands displaying substantial rates and magnitudes of change ( Table 3 and Figure 3 ). Upon exclusion of anthropogenic islands, which highly favour accretion due to land reclamation, 44% of islands accreted, 28% eroded and 28% remained stable. Shoreline change data for all 49 islands are provided in Supplementary Table 4 .

The rate and magnitude of shoreline change varied substantially amongst the different island types. Anthropogenic islands displayed the highest magnitude and rate of island change with NSM of 16.44 ± 45.21 m and WLR of 0.52 ± 1.34 m/year, largely as a result of sediment input arising from land reclamation projects. Land reclamation was observed on nine anthropogenic islands, which included: the increase in island size, construction of passages between islands, and, in one case, the joining of islands. Moreover, eight islands experienced the construction of harbours, boat channels and sea walls. Substantial island development meant that large rates of shoreline change (up to 2.65 m/year) were observed over the study period. Land reclamation was observed on the lagoon side of all anthropogenic islands resulting in advancement of lagoon shorelines due to the constructions of harbours and artificial shorelines. Notably, one anthropogenic island eroded over the study period with NSM -8.30 ± 30.28 m and WLR -0.07 ± 0.60 m/year, although variation across monitoring transects was high. In terms of human modifications, it had the least developed shorelines with the construction of an airport and a harbour dredged into the reef flat. Morphological change is proceeding at greater magnitude and rate on anthropogenic islands compared to natural islands as lagoon islands displayed an NSM of 1.84 ± 18.02 m and WLR of 0.06 ± 0.48 m/year and windward islands had a NSM 1.79 ± 20.60 m and WLR 0.02 ± 0.40 m/year. However, leeward islands eroded throughout the 50-year time period by -1.50 m ± 14.30 m, at an erosional rate of -0.03 ± 0.27 m/year.

Table 4 describes the eight types of geomorphic change that were observed at Huvadhoo Atoll across the study period ( Figures 4 , 5 ). Notable types of adjustment include island migration upon the lagoonal reef platforms which was observed in 11 study islands. Further, advance of the lagoon shoreline was detected in nine study islands. Additionally, seven islands underwent parallel migration, five of which were leeward islands. Overall, lagoon shorelines underwent greater magnitudes of geomorphic change with 10 islands exhibiting morphological change on their lagoon shorelines whereas ocean shoreline change was only observed on five islands.

Geomorphic change varied between island type as on lagoonal islands migration was widespread because this behaviour tends to favour islands located on lagoonal platforms which results in an increase in island size (Aslam and Kench, 2017). Migration was also found in smaller (< 10 ha), circular (< 1.5) islands which is expected as these types of islands are typically found on lagoonal platforms. Larger (> 10 ha), more complex islands favoured parallel migration whereby different erosion and accretion of the island resulted in alongshore migration and these tend to be found on the atoll rim. As such, the majority of leeward islands underwent parallel migration, which could explain their overall erosional response as when sediment is eroded from one side of the island and transported to the other, it is possible that some sediment is lost along the way, resulting in net island erosion. However, ocean migration dominated windward islands which could be because of the larger wave energies, due to being exposed to oceanic swell, providing the energy to transport sediment to oceanward island shorelines. Or, because windward islands have a larger sediment budget that has enabled accretion of oceanward shorelines. Finally, our study found that advancement of lagoon shorelines was prevalent in anthropogenic islands due to being heavily modified with land reclamation and harbours resulting in expansion of lagoon shorelines. Geomorphic changes of all 49 islands are provided in Supplementary Table 5 .

Island area varied from 0.3 ha to 33.2 ha with island size averaging 5.3 ± 6.6 ha. Larger islands were situated on the atoll rim with leeward and windward island area averaging 8.9 ± 6.8 ha and 6.7 ± 9.4 ha respectively, whereas island area in the atoll lagoon averaged 2.5 ± 1.5 ha. Windward islands exhibited the largest variability in island size with a range of 32.5 ha followed by leeward at 18.9 ha and lagoon at 5.7 ha. Overall, smaller islands (0 to 10 ha) accreted by 0.56 ± 16.77 m with at a rate of 0.02 ± 0.39 m/year and larger islands (> 10 ha) accreted by 3.65 ± 23.38 m at a rate of 0.06 ± 0.46 m/year. No significant correlation was found between net shoreline movement and island size (r = 0.093, P = 0.573, n = 39; Pearson’s correlation). Similarly, no significant correlation was found between weighted linear regression and island size (r = 0.014, P = 0.931, n = 39; Pearson’s correlation).

Shape data ranged from 0.4 to 8.1, with average island shape 2.4 ± 1.8. No significant correlation was found between net shoreline movement and island shape (r = 0.026, P = 0.811; Pearson’s correlation). Equally, no significant correlation was found between weighted linear regression and island shape variables (r = -0.021, P = 0.899; Pearson’s correlation).

In Huvadhoo Atoll, wave exposure is highest (< 124.35 J m^-3) off the western margin of the atoll rim while the most sheltered locations are on the leeward side of the westerly and south-westly atoll rim platforms (> 0 J m^-3). The atoll lagoon is subject to a cross-atoll gradient with wave exposure increasing from west (< 0 J m^-3) to east (<~95 J m^-3) due to the dominant westerly wind direction, which means fetch length increases as wind blows from west to east across the atoll lagoon ( Figure 6 ). Lagoon islands experienced on average wave exposure of 60.8 J m^-3, leeward 96.0 J m^-3 and windward 120.0 J m^-3. No significant relationship was found between net shoreline movement and wave exposure (R² = 0.006, P = 0.766; linear regression). Additionally, no significant relationship was found between weighted linear regression data and wave exposure (R² = 0.043, P = 0.410; linear regression). Finally, no clear correlation was found between wave exposure on natural islands (leeward, windward and lagoon), and the magnitude and rate of shoreline change despite the varying degrees of wave exposure experienced at each island type.

Visual plotting suggested that island type was the strongest predictor of shoreline change, and subsequent testing of all factors simultaneously showed this to be the case ( Figure 7 ). Linear Mixed Models were fitted to both NSM and WLR within the nlme package [R ‘nlme’ package, function lme(), Pinheiro, 2023)], separately, in which type was controlled for as a random factor (random intercept), and wave exposure, island size, and island shape were included as fixed factors. No significant effects of wave exposure, size or shape were found, with either NSM or WLR as the response variable (p< 0.05, Supplementary Table 6 ). For further confirmation we removed the island type variable and fitted analogous Generalised Linear Models (GLMs, R ‘stats’ package, function glm(), R Core Team, 2021) to the same data, with wave exposure, size of island, and shape of island as predictors, i.e. omitting island type as an explanatory variable altogether. Here again no continuous variables were found to be significant predictors (p< 0.05, Supplementary Table 7 ) of either NSM or WLR (also reflecting the single variable regressions reported in 3.2.1 to 3.2.3 above), so we proceeded by further analyzing the effect of island type on shoreline change alone, using ANOVA/MANOVA testing.

In these final analyses of shoreline change predictors, island type was identified as a significant predictor of NSM and WLR ( Figure 7 ). The effect of type was confirmed both via (one way) ANOVA tests on the two response variables separately- NSM (F (3,45) = 5.891, p = 0.002), WLR (F (3,45) = 4.411, p = 0.008)- and when combined into one model using MANOVA (Pillai’s trace = 0.30, F (3,45) = 2.62, p = 0.022).

Contrary to previous studies, this study found no relationship between island size and shoreline change. Multiple studies in the Pacific have documented a relationship between island size and shoreline change, suggesting that smaller islands (< 10 ha) undergo erosion whereas larger islands tend to be more stable (Duvat, 2019; Sengupta et al., 2021). In the Maldives, Aslam and Kench (2017) found that islands in Huvadhoo Atoll behaved very similar to islands in the Pacific as a significant relationship was discovered between island size and shoreline change whereby smaller islands (< 10 ha) eroded and larger islands were dominated by accretion.

Despite the variety of island shapes at Huvadhoo Atoll, there was no significant relationship between island shape and shoreline change with results contradictory to the findings by Kench and Brander (2006) in the Maldives, who found island shape impacted morphological change more than wave energy, whereby the beaches on circular islands were more susceptible to shoreline oscillations than elliptical islands. However, the observed geomorphological responses were over seasonal time frames, whereas this study investigates shoreline change over multi-decadal time scales. Yet, some studies have found that circular and elliptical reefs are more prone to retain sediment, compared to linear reefs (Mandlier and Kench, 2012; Mandlier, 2013).

As Huvadhoo Atoll is situated close to the equator, it rarely experiences tropical storms, however, it is exposed to oceanic swell driven by high latitude storms (Aslam and Kench, 2017). Previous studies suggested that wave exposure is a key control on island morphology (Fletcher et al., 2003; Webb and Kench, 2010; Yates et al., 2013; Duvat and Pillet, 2017; Duvat et al., 2017), yet results from the present study found no relationship between wave exposure and shoreline change.

The lack of significant relationships between island size, island shape, wave exposure and shoreline change contrasts findings and suggestions from previous research (Webb and Kench, 2010; Aslam and Kench, 2017; Duvat and Pillet, 2017; Duvat, 2019; Sengupta et al., 2021). As such, findings from this study highlight that reef island shoreline dynamics are very site-specific and controlled by a range of different variables. For example, in addition to the variables analysed in this study, variability in island sediment grain size, vegetation type, elevation, reef platform morphologies, bathymetry are all additional variables that may be interact to control the magnitudes of reef island shoreline change. From a management perspective, the site-specific nature of reef island shoreline dynamics makes island planning decisions problematic as it is complex to predict which islands may be the most and least mobile over decadal timescales.

A significant difference in shoreline change was found between island types whereby shoreline change was significantly higher on anthropogenic islands than all other island types. A key reason for this was due to the extreme modifications of anthropogenic island shorelines, which meant that they were undergoing morphological change at a greater magnitude and rate than natural islands. While nine out of the 10 anthropogenic islands accreted, the least modified island eroded. Therefore, small-scale modifications may enable erosion to be more clearly identified, whereas erosion may be occurring on other islands, but large-scale modifications, such as land reclamation, may be masking any natural morphological changes within the data.

Large island communities (>1,000 people) tend to inhabit islands that have undergone significant modifications (i.e., coastal development or hard engineering) and the future stability of these islands is still unclear. While these anthropogenic islands may have experienced a substantial increase in size, modifications have been relatively recent and there is a lack of long-term data available which would enable us to understand how such severe modifications may impact island shoreline dynamics. Further, modifications have been found to not only impact shorelines on the modified island, but those of other islands within the atoll. For instance, when investigating the impact of coastline modification and sea level rise on island shorelines in North Malé Atoll, Maldives, Rasheed et al. (2020) found that increased flow rate in the channel between Hulhumalé and the neighbouring island Kanifinolhu influenced erosion of the island Kudabandos. This increased erosion was due to land reclamation blocking sediment transport across the lagoon to Kudabandos’s shorelines and sediment was instead transported to the channel.

Modification of reef island shorelines is widespread in the Maldives and a recent investigation by Duvat and Magnan (2019) across 608 islands found that the capacity of inhabited and modified islands to respond to climate pressures had severely decreased over the last decade. An increase in human activities has meant that in one decade 56 natural islands were exploited. Further, significant population growth has occurred on 47 inhabited islands meaning that the number of islands experiencing a substantially high human footprint increased by 9.2%. Similarly, this study found that island modification occurred on rapid timescales with major developments appearing over the last decade (from 2006 to 2014). The Maldives have experienced a surge in population over recent years as in 2021 the population was 543,620 whereas in 2011 the population was 380,493 (World Bank, 2022). A rapidly increasing population will place extra pressure on islands as land becomes increasingly sparse, thus, islands may undergo further modification to combat the growing population. In addition to a surge in population, a rise in tourism has led to an increase in visitors each year with 1,321,855 people visiting the country in 2021. As such, it has become the one of the dominant industries in the region accounting for 26% of the country’s GDP (Maldives Bureau of Statistics, 2020). This has led to a rise in the number of resorts, hotels and guests houses and, as of December 2020, 159, 13 and 638 existed respectively (Statistical Yearbook, 2021). As the tourism industry increases in the Maldives, more islands may undergo substantial transformations to accommodate the rise in popularity. Expansions in population and tourism could lead to the construction of additional island infrastructure both globally and in the Maldives. Consequently, a challenge for the future adaptive capacity of atoll nations is to develop infrastructure that can accommodate for naturally shifting reef island shorelines which is implemented quickly enough to ensure that reef island physical response to sea level rise is not compromised.