Atolls are found in tropical oceans, consisting of a carbonate reef platform (hereafter called reef flat) surrounding a central lagoon with subaerial islands on top of the reef (hereafter called motu) (Schlager and Purkis, 2013). The exact definition of what constitutes an atoll varies from the simple “annular reef whose rim develops at or near sea level and encircles a deeper lagoon” (Ladd, 2012) to more complex definitions such as proposed by Wiens (1962) in several pages or by Gischler (2016) as an “irregularly shaped, isolated oceanic reef structures at or near sea level that enclose a lagoon.” We take a broad definition for an atoll but focus on the common structures inherent in almost all of the atoll definitions: lagoon, shallow reef-structure, and islets atop the carbonate reef-structure.

Atolls are typically formed from coral reefs growing on subsiding dormant volcanic islands (Darwin, 1842) and are shaped by global changes in sea levels (Daly, 1925; Toomey et al., 2016) over hundreds of thousands of years. Atoll Islands, herein referred to as motu (aka cays or islets), are composed of carbonate sand, coral shingle, and rubble perched on top of the conglomerate reef flat encircling the inner lagoon (Woodroffe et al., 1999). For our satellite based analysis, the term motu encompass islands on atolls of any sediment size (Ortiz and Ashton, 2019), rather than limited to a particular class (Richmond, 1992; Brander et al., 2004; Kench et al., 2017). In addition, since we cannot easily differentiate land types, all sub-aerial landforms are classified as land and called motu in our analysis, regardless of potential elevation differences or sediment characteristics. These subaerial islets form and evolve on a shorter time scale, from decades to centuries (Perry et al., 2013; Hamylton et al., 2014; Kench et al., 2014a; Kench et al., 2014b; Kench et al., 2023), compared to the atoll itself, which operates on the timescale of millennia to hundreds of thousands of years (Toomey et al., 2013; Toomey et al., 2016). Motu initiation, formation, and evolution has occurred under current (Kench et al., 2014a; Sengupta et al., 2021), rising (Kench et al., 2005; Webb and Kench, 2010; McLean and Kench, 2015), and falling sea level conditions (Woodroffe, 2008; Kench et al., 2020). Evolution of motu can be episodic, with island change tied to specific events such as a tropical cyclone adding sediment (Duvat and Pillet, 2017) or land reclamation from human activities (Ford, 2011; Aslam and Kench, 2017; Duvat and Pillet, 2017; Duvat and Magnan, 2019).

Atolls are at risk from climate change due to many factors, including their low elevation, ocean acidification decreasing reef productivity, and changing storm intensities. Many atolls have a maximum elevation of less than 5 m with a mean elevation of 2 m, potentially at risk to estimated global sea level increases. Accelerated rates of sea-level rise (SLR) may outpace vertical reef flat accretion from the coral reefs. In addition, storm driven flooding will be increased by climate change (Storlazzi et al., 2015), driving increased flooding and inundation from swell waves generated by distant storms (Hoeke et al., 2013; Shope et al., 2017). Ocean acidification and other oceanographic stressors such as changing ocean temperature can reduce sediment production from the coral reefs, which puts the ability of motu and reef platforms to keep pace with SLR further at risk (Eyre et al., 2018). However, to understand the potential response of these landforms to changing climate risks, we must first understand the processes driving their evolution and their current status.

Although atolls are often described as circular or annular structures (Wiens, 1962; Ladd, 2012), their morphology varies widely (Stoddart, 1965; Gischler, 2016). The controls on atoll shape are still inconsistently quantified. Previous work by Stoddart (1965) used hand tracing of 99 atolls to measure their shape, demonstrating a “fundamental homogeneity of atoll shapes” including a tendency for elongated and more elliptical atolls, where the length of the atoll is typically 1.5–3 times the width, rather than the oft-described ring or circular shaped atoll. In this paper, we recreate Stoddart’s measurements both at the atoll scale and at the individual motu or reef-flat scale across our global atoll database (n = 154/623 atolls).

Many studies of atoll geomorphology use a combination of field and remote sensing technologies, including historic aerial photography and modern satellite imagery, focusing on one motu to a regional group of atolls. Our analysis provides a consistent method to expand studies of atoll geomorphology and track landscape change over time. Frequently, studies have relied on hand digitation of shorelines and atoll morphology (Webb and Kench, 2010; Kench et al., 2014a; Ford and Kench, 2014; Ford and Kench, 2015; Duvat and Pillet, 2017; Sengupta et al., 2021). Different features are used as a proxy for the shoreline location including edge of vegetation (Ford and Kench, 2015; Albert et al., 2016; Garcin et al., 2016), defining a stability line (Duvat and Pillet, 2017), using a GPS track at the time of the field work (Kench et al., 2014a), using supervised classification (Holdaway et al., 2021), and using an image analysis software with hand digitation to fix errors (Schlager and Purkis, 2013). Studies looking at changes in land area or shoreline position on atolls include (Kench et al., 2014a; 2018; Albert et al., 2016; Aslam and Kench, 2017; Duvat and Pillet, 2017; Duvat and Magnan, 2019; Nunn et al., 2020). For example, Ford and Kench (2014) examined Nadikdik Atoll in the Marshall Islands, and found a net increase of island area from 1945 to 2010. Rather than relying on time-intensive hand-digitization, our automated analysis of Landsat imagery provides a consistent and reproducible method for calculating land area at a given timepoint.

Duvat (2019) re-analyzed 20 different papers studying atoll land area change for 30 atolls (709 motu). They found that 88.6% of the islands were stable (within ± 3% area change) or increased in area over the time period of analysis. All the larger islands (>0.1 km²) were either stable or increased in area. This paper was one of the first studies to collate other published works and compare atoll island changes across decadal changes. However, Duvat (2019) did not create a common assessment protocol for atolls to strengthen data comparability. Our analysis presented herein provides a roadmap for a common protocol to calculate and assess atoll island evolution.

Holdaway et al. (2021) leveraged Landsat satellite imagery to quantify atoll land area change for 221 atolls with a mix of multiyear and annual temporal composites. Supervised classification, with a fluctuating number of landcover classes, was used to segment the images. For comparison between all the sites, a binary land/water class was created and total land area change was calculated on an atoll level. From 2000–2017, land area increased 6.1% for the 221 atolls, primarily in the Maldives and South China Sea mainly due to land reclamation (Holdaway et al., 2021). Their analysis provides one example of leveraging existing satellite imagery to track landcover change on atolls globally but still relied on relatively intensive oversight due to the use of supervised classification with varying landcover classes. In addition, they only investigated land water changes over time and ignored the reef flat. Our method provides a more automated process and measures morphometrics between landscape objects on the atoll (i.e., the width of the reef flat in front of a given motu). Moreover, our large database creates an important baseline for current atoll morphometrics.

Numerical modeling work by Ortiz and Ashton (2019) found that the width of the reef-flat in front of motu (the distance from the oceanside of the motu to the edge of the oceanside of the reef-flat) should reach an equilibrium distance dependent on the offshore wave climate. Using the open-source model XBeach in hydrodynamic mode, they investigated the potential response of sediment transport with the presence of a 1D motu to changing offshore wave climate and changing reef-flat widths in front of the motu. They found that once motu are present on the reef flat (i.e., once the sub-aerial land blocks sediment from reaching the lagoon), the motu would grow and accrete oceanward (thus narrowing the oceanside motu to reef-flat width) up to a certain point where the direction of sediment transport would reverse and direct sediment offshore. Their conceptual model of motu evolution predicts self-organization of motu prograding oceanward to a critical reef-flat width dependent on the offshore wave height. Similarly, 2D modeling by Shope and Storlazzi (2019) found that atoll islands orientated parallel to the deep-water wave direction would accrete towards the lagoon while eroding along shorelines exposed to direct wave action, bolstering the theory that offshore wave climate is key to motu evolution. Our analysis herein enables an initial quantification of global motu morphometrics, enabling us to further investigate the Ortiz and Ashton (2019) conceptual model.

While there are many studies of atolls island evolution, there is a critical need to establish a baseline of atoll morphometrics using a consistent methodology (Duvat, 2019). The aim of this research is to create a reproducible approach to evaluate atoll morphology on a broad spatial scale using satellite imagery and innovative data processing techniques. We use a constant time frame for the temporal composites, an automated classification technique to separate the atoll into parts (motu, reef flat, open water/lagoon), automatically segment the classified imagery, and calculate morphometrics of each object. We create a worldwide database of atolls and their morphometrics. This methodology removes spatial bias and enables a better understanding of current geographic patterns in atoll morphometrics and potentially identify first order patterns between atolls.

Four-year temporal composites (2015–2018) are created from Tier One Landsat Images using the Google Earth Engine library (GEE) in Python similar to the methodology used by Ortiz et al. (2017). For a given atoll, all Tier one Landsat images available from 2015 to 2018 are collected, cloudy pixels are removed using in-built GEE cloud removal functions, and the remaining pixels from each image are composited using the 50th percentile on a per-pixel per-band basis. Tier one Landsat images are georectified, have atmospherically corrected surface reflection, and have a 30 m resolution (Google Earth Engine Team, 2015). Six bands (blue, green, red, NIR, SWIR1, and SWIR2) are retained in our final temporal composite. A count band, indicating the number of images used for the composite per pixel, and a mask band to show the geometry given to GEE are also included. For all 154 atolls, the average number of images used in the temporal composite is 68 with an average minimum number of images equal to 14. By using a 4-year temporal composite, we ensured that per-pixel, each composite is based on at least 14 non-cloudy Landsat images. The temporal composite method has been used by other researchers to remove cloudiness in a given Landsat image, a very common issue for atolls (Ortiz et al., 2017; Mateo-García et al., 2018).

We assume that the primary landcover classes on an atoll are: reef flat, motu (land), and water and use k-means clustering (assuming 3 clusters) for automated image classification. While previous authors have sometimes linked formation mechanisms or sediment type to different atoll island names, such as using sandy cay (Montaggioni et al., 2018) for low-lying sand dominated islets vs. motu for islets with a shingle ridge and gravel dominated (Woodroffe et al., 1999; Woodroffe and Biribo, 2011; Goldberg, 2016; Montaggioni et al., 2018), Landsat satellite imagery (at 30 m resolution) does not easily differentiate these landscapes, as such, we have chosen to combine all sub-aerial landforms be they vegetated, gravel, sand, conglomerate, or infrastructure, into one landcover class called land or motu. Given Landsat’s medium resolution (30 m), the classified image is cleaned by removing small groups of objects with less than eight isolated pixels (i.e., objects <7,200 m² are not analyzed). All motu smaller larger than 7,200 m² but smaller than 57,600 m² (64 isolated pixels) only have basic morphometrics calculated (area, perimeter, and location), while motu larger than 64 pixels have complex morphometrics calculated (such as width and length). The total area across our entire database accounted for by motu between 0.72 and 5.7 ha (i.e. 8-64 connected pixels, n = 2,036), is less than 3% of the total area (47 km²) measured by the remaining large motu (n = 1,752, for a total area of 1,789 km²).

A methodology for determining atoll morphometrics is created and implemented in Python scripts, publicly available on GitHub (https://github.com/ale37911/AtollGeoMorph). Our Python scripts are split into three distinct pieces: 1) create temporal composites (Section 3.1), 2) classify, clean, and segment the image then calculate morphometrics (Section 3.2), and 3) collate saved outputs from each atoll into larger dataframes for analysis and visualization (Section 3.3). For detailed description of the methods used in the morphometric analysis see Supporting Information Text S1.

Once the satellite image is classified, the number and approximate location of lagoons on the atoll is input by the user. Users manually close the lagoon for cases where our automated lagoon finder function is unable to match the lagoon number specified by the user. Morphometrics of the lagoons are calculated: area, perimeter, all the perimeter points (on a per-pixel basis), and the centroid. Atoll level morphometrics are also calculated: outside atoll perimeter (ocean perimeter), the atoll centroid, and shape factors (Horton Form Factor - F, Miller’s Circularity Ratio–R _c , Schumm’s Elongation Ratio–R _e , and the Ellipticity Index–I _e ) used by Stoddart (1965) (Supplementary Text S2). Area, perimeter, and centroid of each object (i.e., reef flat, motu, or atoll) are calculated and stored in pandas dataframe where each row is the perimeter point per object (motu or reef flat). All points are classified as an ocean-side or lagoon-side point based on relative distances to the lagoon or ocean (Supplemental S1).

For each point on an object, several morphometrics are calculated such as the cardinal position angle, defined relative to the position on the atoll compared to the centroid of the atoll. Multiple width morphometrics are also calculated within each object, i.e., width of a motu (Figure 2B– red lines) or total width of a reef-flat (Figure 2F– blue lines) per point from ocean-side to lagoon-side, or widths between objects, i.e., the width from a motu ocean-side point to a reef-flat ocean side points called the ocean reef width (Figure 2D– purples lines). The length of each motu is calculated using the cumulative distance along the center points of the motu width measurements (Figure 2C– yellow line). The cumulative distance between each of these points is used as the length measurement of a given object. We also calculate the shoreline length of each object by the cumulative distance of all ocean or lagoon side points, these are called the ocean-side and lagoon-side lengths respectively (Figure 2C– pink and purple line respectively).

Atolls are analyzed individually, regionally, and globally with morphometrics summarized at a per-point, per-object, and per-atoll level. At the regional level, morphometrics are also binned by cardinal direction relative to the atoll centroid (cardinal position angle). While at the global scale, morphometrics are binned by absolute latitude. Summary tables are created for each atoll that include the area, mean widths, length, centroid location, cardinal directional bin (i.e., binned by the relative position of the point to the atoll centroid) for each motu, reef flat, and lagoon object. Primary morphometrics for all 154 atolls analyzed are available as a table (Supplementary Table S1). Finally, statistical analysis of our grouping and averaged morphometrics is analyzed using non-parametric variance analysis (Kruskal–Wallis implemented from the scipy. stats Python library) and the significance of the grouping scheme (i.e., cardinal binning of reef flat widths of French Polynesia) is analyzed with a post hoc Dunn Test (implemented from the scikit posthocs Python library) (Supplemental S3). All statistical analysis assumes an alpha of 0.05 and the calculated p-values are shown in the supplemental tables (Supplementary Tables S2–S8).

Faaite Atoll, French Polynesia is located at 16.758°S, 145.238°W in the Pacific Ocean south of the equator and east of the international date line (Figure 3). The atoll has one lagoon, two reef flats, and 27 motu. Faaite has a total lagoon area of 233.37 km², reef flat area of 52.27 km², and land area (motu area) of 11.48 km². The two reef flats are unequal in size with the primary reef flat accounting for more than 99% of the total reef flat area (Figure 3A). The large standard deviation on the effective reef flat width measurements for Faaite occurs because the distribution of widths is not unimodal. Seven of the motu (out of 27) are large enough to calculate complex morphometrics such as length and width measurements (Figure 3C). Similar to many atolls in our database, the lagoon-side motu length are mostly longer than the ocean-side motu length (6/7, Figure 3C). This is exemplified in motu M4, where the lagoon-side motu length is 75% longer than the ocean-side motu length (aka motu shoreline length), due to increased crenulation of the lagoon side shoreline compared to the smoother ocean-side shoreline. The larger motu (plotted in green, Figure 3) are distributed on the northern shoreline while the smaller motu (only simple morphometrics calculated, plotted in yellow in Figure 3) are found primarily on the southern shorelines. To quantify this observation, we also bin all morphometrics relative to relative cardinal position on the atoll (Supplementary Figure S1, Figure 2).

There are 60 atolls in French Polynesia with sufficient Landsat coverage to create a temporal composite and calculate morphometrics. These atolls are located between 14° 24′S-23° 21′S and 134° 29′W- 154° 41′W (Figure 1 – the large clump of pink dots south of the equator below the beige stripe on the eastern side of the pacific). These atolls have 1,930 motu (836 motu with full morphometrics calculated with areas >64 connected pixels or 5.76 ha) and 80 reef flats (Supplementary Table S1). In French Polynesia, there is on average 1.3 reef flats per atoll with the number of motu ranging from one to as many as 69 with an average of 14 larger motu per atoll.

On a per-point basis in French Polynesia, the ocean reef width has the most constrained distribution of measured widths (Figures 4A,E) while the total reef-flat width has the widest variation of measured widths (Figures 4A,C). When breaking out the data by cardinal position on the atolls, we see a clear trend that the northern shoreline of these atolls consistently has the smallest variance of widths while the southern and western shorelines tend to have the widest range of measurements. When looking at motu width, ocean reef width, and lagoon reef width, there is a strong unimodal distribution (Figures 4D,E). However, on the southern and western shorelines, the reef flat width measurements show a bimodal distribution (Figures 4C,F). Distinct variations in the distribution of measured widths for French Polynesia atolls highlights the spatial heterogeneity of landforms on these atolls.

Analyzing on a per-atoll basis (one data point per-cardinal bin per-atoll) weights small and large atolls equally (Figure 5; Figure 6; Table 1). Median motu widths range from 263 m in the south to 345 m in the west (Figure 5A). The width of the reef flat is narrower on the northern and eastern shores and wider on the southern and western with the largest variation of reef flat widths on the southern shores (Figure 5B). Moreover, in French Polynesia the total reef-flat width is statistically different on the Northern and Eastern sides compared to the southern and western sides (Supplementary Table S2, p-value <0.029). This trend is also observed in the effective reef flat width measurements, with the narrowest distribution of widths in the north and the widest distribution in the south (Figure 5E). The lagoon reef width (Figure 5C) is more variable than the ocean reef width (Figure 5D) with a wider distribution of measurements perhaps driven by increased crenulations of the motu shoreline on the lagoon side compared to their smoother ocean-side shores. The northern shores have the narrowest lagoon reef width of 100 m and the narrowest ocean reef width at 101 m. The south bin has the widest distribution of widths for both the mean lagoon reef width and the mean ocean reef width at 212 m and 185 m respectively. Consistently the northern and eastern shores of French Polynesia atolls have a narrower range of morphometrics than the southern and western shorelines.

The east and the north shores of French Polynesian atolls have the longest motu while the south and west shores have shorter ones (Figure 6A) as can be seen qualitatively in the k-means cleaned image of Katiu, French Polynesia (Figures 2 – Tropical Atoll). The north and east sides also have a larger percent of their reef flats covered by motu (Figure 6B), and are statistically distinct from the southern shorelines (p-value <0.001, and 0.003, Supplementary Table S2). In addition, the northern and eastern sides have a narrower overall reef-flat width, which are again statistically different from the southern and western shores (Supplementary Table S2, p-value <0.005). Lastly the southern shores are dominated by statistically distinct morphometrics from any of the other cardinal bins (Supplementary Table S2), with shorter motu and wider reef flats (Table 1). Overall, in French Polynesia the northern and eastern shores have longer motu and narrower reef flats for all reef measurements (reef flat width, lagoon reef width, ocean reef width, and effective reef flat width) compared to the southern and western shores.

On a per-object basis (aka averaging all the points for each motu then analyzing across all motu in French Polynesia categorized into cardinal bins), the ocean reef width (reef width in front of the motu) reaches a near constant width once the motu reaches a certain length (Figure 7). For all French Polynesia motu longer than 1 km, the mean ocean reef width is 150 ± 60 m (n = 324/795), which decreases to 108 ± 50 m for all motu longer than 10 km (n = 54/795, Figure 7B). When grouping purely by cardinal bins (so not per-motu, but per atoll and per cardinal bin) as shown in Figure 7A, for the sum of motu longer than 1 km within each cardinal bin, the mean ocean reef width is 145 ± 85 m (n = 222/231), which decreases to a mean of 131 ± 86 m for motu longer than 10 km (n = 83/231), similar to our analysis as per-motu per atoll per cardinal bin (Figure 7B). In both analyses (Figures 7A,B) analyzing either per motu per cardinal bin or per cardinal bin, we see that the larger motu with the near constant ocean reef flat width are clustered on the northern and eastern shores with fewer on the southern and western shorelines (blue and green dots compared to olive and yellow) and these are statistically distinct populations (Supplementary Table S3).

There are clear trends of increased numbers of the largest motu (>10 km) on the northern and eastern shores of the French Polynesia Atolls (37% and 52% respectively compared to only 4% and 7% on the Southern and Western sides, Figure 7). For smaller motu, i.e., with lengths <1 km (n = 471/795), the mean ocean reef width is larger, 238 ± 73 m. The relative or normalized ocean reef width in front of the motu was calculated to ensure that smaller overall atoll footprints were considered (Figure 7C). The same trend of a near constant ocean reef width is exhibited again, dominated by northern and eastern motu (blue and green dots, Figure 7C). For motu that occupy more than 10% of the total reef-flat length, there is an average of 18% of the total reef-flat width open in front of the motu (n = 97/795), with the majority (70/97) of these motu found on the N and E shores (72%). The consistent pattern of a near constant width in front of large motu implies that self-organization may be driving a critical reef-flat width for French Polynesia Atolls.

Morphometrics are calculated for 154 atolls (Figure 1), and overall, there are 3,786 motu (1,752 with all morphometrics calculated) and 593 reef flats (Table 2) in our analyses. The total land (motu) area is 1,836.47 km² and the total reef flat area is 7,387.43 km².

Atolls are binned by their absolute latitudes, with the least number of atolls in the equatorial latitudes (Table 2). The equatorial atolls (n = 31) are located from 4.7°S to 4.7°N, the mid tropical atolls (n = 57) are between 4.7°S and 14°S and 4.7°N - 14°N, and the high tropical atolls (n = 66) are located greater than 14°S and 14°N below the Tropics of Capricorn and Cancer respectively (see Figure 2 for an example of an atoll in each region). In our analyses, the high tropical atolls are all located in the Pacific Ocean in the southern hemisphere and are dominated by the atolls of French Polynesia (n=60/66). The equatorial atolls have the least number of motu per atoll with an average of 12.7 motu (5.9 large motu with all morphometrics calculated) and 3.8 reef flats. Mid tropical have an average of 24.5 motu (12.2 large motu) and 6.1 reef flats, while high tropical atolls have an average of 30.2 motu (with 13.2 large motu) and 1.9 reef flats with statistically distinct populations between the latitudinal atoll groups (Supplementary Table S4).

Similar to Stoddart (1965), we see variation across all 154 atolls in the overall morphology of the atoll, i.e., the outer shape of the atoll itself, but with overall homogeneity of morphology (Figure 8; Figure 2–example atolls shown at bottom). We can predict the perimeter of an atoll with high certainty given it is overall surface area (r ² = .96, Supplementary Figure S2). Atolls cluster around a mean elongated shape as shown by the mean of the shape factors (F = 0.39 ± 0.15, R _c = 0.55 ± 0.17, R _e = 0.69 ± 0.14, I _e = 2.3 ± 1.3, Figure 8). The distributions of atoll shape do vary across our three latitudinal groupings (equatorial, mid-tropical, and high-tropical) with statistically significant differences for all calculated shape factors except the Ellipticity Index (Supplementary Table S5, p-value <0.029). The mid tropical and equatorial atolls are more elongated than the high tropical atolls as shown by the low Horton Form Factors and Miller Circularity Ratios (Figures 8A,B). The high tropical atolls are dominated by more circular, less complex atoll shapes compared to either of the equatorial or mid tropical atolls.

Across all our atolls analyzed, the ocean reef width (distance between the ocean-side of the motu and the ocean-side of the reef-flat) has the narrowest distribution of widths indicating that it varies the least across all atolls (Figure 9A) in comparison to our other width measurements. The total reef flat width has the widest distribution with the largest range of measurements especially among the equatorial atolls (Figure 9C). Equatorial atolls consistently have lower peak densities, and thus have a wider distribution of widths compared to our other two groups of atolls (Figures 9B–F) and are statistically distinct from mid-tropical and high-tropical atolls (Supplementary Table S6, p-value <0.001). High tropical atolls have the narrowest distributions of widths compared to our other two latitude categories. The mid tropical atoll morphologies lie between the equatorial and high tropical atolls with motu width and ocean reef width more closely aligned with the high tropical atolls, while the effective reef flat width and lagoon reef width more closely match the equatorial atolls. Ocean reef width for the equatorial atolls has a bimodal distribution with a skew toward wider widths. Conversely, both mid and high tropical atolls show unimodal distributions (Figure 9E). The equatorial atolls also exhibit distinctly more skew towards wider motu (Figure 9B). Overall, we see a strong tendency for distinct groupings of atoll morphometrics at these different latitudes.

High tropical atolls have a narrower distribution of motu area than either equatorial or mid tropical atolls with the larger motu dominated by longer (up to 20 km) and relatively narrow motu (Figure 10C). However, equatorial motu have a very large range in motu area (Figure 2 - purple line) dominated by very long motu lengths (more than double the lengths of large motu in other regions) (Figure 10A). The mid tropical motu are dominated by relatively short and stubby motu that are less wide and long than either the equatorial or high tropical motu (pink 2D histogram of motu width and length, Figure 10B). The equatorial motu are statistically different for motu area and width compared to both the mid-tropical and high-tropical motu (Dunn Test, p-value <0.016, Supplementary Table S7).

In order to validate our methodology, we compared the calculated morphometrics presented in this paper with values reported in previous studies on atolls (Figure 11). Similar to Holdaway et al. (2021), we performed this validation at the atoll scale for the land area, reef-flat area, and lagoon area using the bias (mean error between previously reported morphometrics and our calculated morphometrics) and Brier Skill Score (estimate of error in our morphometrics to variance of previously reported morphometrics) (Gharagozlou et al., 2020). In general, our estimates of area are close to previously reported values especially for total land area and lagoon area (Figure 11). While, our estimates of both reef-flat area and lagoon area (Figures 11B,C) are consistently under-predicting the total area with a bias of 26.5 km² and 4.4 km² respectively. Our morphometrics over-predict total land area per atoll on average by 3.1 km² (Figure 11A). This bias is perhaps driven by the inability of our Landsat composites (at 30 m resolution) and subsequent k-means classification to easily differentiate narrow shallow channels (15–50 m) between motu and instead merge these into one larger motu, as happened with Faaite, M5 (Figure 3). However, all our calculations of total area per atoll have an excellent Brier Skill Score and follow reported landcover areas well.