Satellite‐ and field‐based facies mapping of isolated carbonate platforms from the Kepulauan Seribu Complex, Indonesia

Quantitative facies models from modern carbonate are essential for the interpretation of their fossil counterparts. The isolated carbonate platforms of the Kepulauan Seribu archipelago has many atoll‐like islands with reef belts exposed to bidirectional monsoon winds. Statistical analysis based on texture and composition reveal that there are four sedimentary facies; coral grainstone, coral packstone/grainstone, coral‐mollusc packstone and mollusc wackestone. The occurrence of mollusc wackestone in the lagoon is controlled by water depth, while the sand apron and reef front do not show significant facies separation with water depth. The co‐occurrence of these different facies in the same depth window is contrary to the common thought that changes in bathymetry should be reflected in facies changes. The studied reef systems therefore show aspects of random and ordered facies distribution with respect to water depth. A satellite derived environmental facies map generated by an image analysis algorithm indicates that environmental facies distribution is mainly controlled by water depth, density of seagrass cover and coral abundance. The sand apron can be subdivided into three environmental facies with no, sparse and dense seagrass cover. The deeper water zone can be separated into shallow and deep subtidal parts of lagoons and platform margins. In the lagoon, satellite derived environmental facies directly correlated with sedimentary facies. No direct correlation of environmental facies to sedimentary facies was possible in the sand apron due to the heterogeneity and complexity of the environment. However, the mean sediment grain size is significantly smaller in areas of the sand apron colonized by dense seagrass. This study aims to contribute towards a better understanding of modern equatorial Southeast Asian carbonate systems, delineate modern carbonate facies based on sediment texture and composition with the aid of multivariate statistical analysis combined with statistic based satellite mapping, and give insights regarding the correlation between depositional facies and water depth.

Conventional field-based mapping studies of carbonate platforms have long been used to gain dimensional data about sediment distribution patterns, while the satellitebased studies have the advantage of being able to cover larger areas in a more timely and cost-efficient manner. Satellite-based remote sensing has been established as a beneficial examination tool and has been extensively applied within the earth sciences, including the mapping of coral reef biota and bottom sediments (Gischler & Lomando, 1999;Kaczmarek, Hicks, Fullmer, Steffen, & Bachtel, 2010;Lyzenga, 1981;Madden, Wilson, & O'Shea, 2013;Purkis, Rowlands, & Kerr, 2015). Modern carbonate platforms are ideal for satellite-based studies as they generally are only covered by shallow and relatively clear water, which facilitates accurate characterization of satellite data in marine environments. The combination of statistical satellite image analysis and sediment data was recently proposed as a quantitative technique for producing satellite derived facies maps for carbonate platforms (Kaczmarek et al., 2010).
This study follows this approach intending to (i) contribute towards a greater understanding of modern equatorial Southeast Asia carbonate systems, (ii) compare modern carbonate facies with satellite derived environmental facies maps and (iii) test to what degree the carbonate facies distribution on small patch reef systems is related to water depth.

| STUDY AREA
Situated on the outer shelf margin of Sumatra in the Java Sea, Kepulauan Seribu (Thousand Islands) consists of numerous coral reef islands ranging in dimension from a few metres across to length greater than 7 km. The archipelago commencing about 25 km northwest of the Java coastline, extends for about 40 km into the Java Sea. Kepulauan Seribu sits on a NNE-SSW structural high, the Seribu platform (Figure 1), which is separated from the Sunda Basin towards the west by the still active N-S-oriented Seribu Fault (Park et al., 2010). The N-S orientation of the Seribu platform likely is also inherited from this structural grain (Park et al., 2010). The E-W-oriented inter island channels that can be in places more than 50 m deep are the remains of a late Pleistocene drainage pattern that was later modified and truncated by river capture. In this respect, the preceding topography has played an important role in establishing the location and nature of Holocene reefal carbonate build-ups in Kepulauan Seribu (Park et al., 1992). Little is known about the internal architecture of the Seribu platform. Based on three shallow cores from two islands, Park et al. (1992Park et al. ( , 2010 showed that modern reef growth initiated in the early Holocene. Vertical reef growth continued with a rate of 5-10 mm/year to around 4,500 years BP. This stage was followed by a slight late Holocene sea-level fall that led to lateral growth of the patch reef systems (Park et al., 1992(Park et al., , 2010. The controlling influence on modern reef growth and morphology in Kepulauan Seribu is believed to be the seasonal change in wind and current directions (Scrutton, 1976a). The Java Sea is a relatively shallow sea with water depth generally between 40 and 50 m and only limited connection to the ocean. The Java Sea is characterized by the monsoonal wind with a seasonal reversal between the west monsoon (December-February), and the  Fainstein, 1987) east monsoon (April-October). The east monsoon wind is two times more persistent than the west monsoon wind, but the west monsoon wind shows more extreme variations in wind strength (Poerbandono, 2016). The climate in Kepulauan Seribu is tropical with high air humidity of 80 mmHg, a mean temperature of 27°C and a seasonal cycle from 32.3°C to 21.6°C (Sachoemar, 2008). Salinity measurements in the west Java Sea indicate average values of 32‰ total dissolved salts, well below normal open ocean salinities. These relatively low salinities result from high seasonal rainfall and high runoff associated with the monsoonal climate. The tidal range in Kepulauan Seribu is micro tidal, averaging about 52 cm with a range from 15 to 82 cm and a typical diurnal tidal cycle with one high and one low stand per day (Wyrtki, 1961). Generally, surface currents in Kepulauan Seribu show a maximum speed of 0.5 m/s (Farhan & Lim, 2011). Wave height in Kepulauan Seribu varies between 0.5 and 1 m during the east monsoon and between 2 and 3 m during the west monsoon, while mean wave speed is only 1 knot due to the wave absorption by the island chain (Sachoemar, 2008).
This study focuses on the patch reef systems of Panggang Island, Pramuka Island and Semak Daun Island which are situated in the middle part of the Kepulauan Seribu chain (Figure 1). Panggang Island is the island in Kepulauan Seribu with the highest population density and is separated from the less inhabited Pramuka Island by a shallow inter-reef channel. Panggang is an atoll-like reef with an approximate area of 3.2 km 2 , and a central lagoon surrounded by a reefs and its sand apron. From the reef flat the water depth increases from about 1 m to about 20 m in the deepest part of the lagoon. Some of the reef on the reef crest will be exposed during low tide, showing various kinds of corals dominated by platy Acropora and foliated corals ( Figure 2). The sand apron on the eastern and western side of the lagoon is roughly two times wider compared to the sand apron on the northern and southern side ( Figure 3). Only one small inlet traverses the sand aprons allowing only very limited connection between the lagoon and the open sea. There are two small islands on the northeastern side of Panggang, preventing the reef system from direct wave exposure from the Java Sea. Pramuka is a fringing reef with an extensive reef flat on the eastern and F I G U R E 2 (a) Platy Acropora and foliaceous Montipora on the reef crest, south of Panggang lagoon, often exposed during low tide. (b) Sand apron with dense seagrass in Pramuka Island, individual corals often occur within the seagrass meadows. Underwater photograph of: (c) Sand apron with sparse short seagrass cover in Pramuka Island. (d) Sand apron with sparse seagrass, Halimeda, and small isolated coral heads in Panggang Island. (e) Sand apron without seagrass in Pramuka Island. (f) Coral framework on the reef front of Semak Daun Island northern part, while almost no reef flat exists on the western part (Figure 3). The island itself lies on the southwestern part of the platform, emerging about 2 m above the sea surface. Most of the reef flat is covered by less than 1 m of water with no distinct reef crest observed. The reef slope on the western side reaches depths of ca 20 m, while on the eastern side it falls off into much deeper water, reaching depths of about 40 m in the Java Sea. Semak Daun is a platform lagoon with an extensive reef flat surrounding a partially filled shallow lagoon with active coral growth in its centre (Figure 3). Similar to Panggang, the reef flat in Semak Daun is expanded broadly on the eastern and western side, leaving a narrow reef flat on the northern and southern side. Water exchange between the lagoon and the Java Sea is restricted to few inlets on its western side, while the eastern side of Semak Daun is directly exposed to the Java Sea.

| METHODS
A total of 102 surface sediment samples were collected from three islands in Kepulauan Seribu; Panggang (May 2015, 45 samples), Pramuka (May 2016, 32 samples) and Semak Daun (May 2016, 25 samples). In shallow water, sediments were collected by hand and water depth was measured using a marked stick. Sampling in the inner lagoon and at the reef slope was conducted by scuba diving and water depth was determined using the diving gear. Latitude and longitude of sampling locations were measured with a global positioning system (GPS) and plotted on SPOT-6/7 satellite imagery. In the laboratory, samples were washed in fresh water and dried using an oven set to 70°C for 7 hr. Dried samples were first sieved through a 125 μm sieve and a split of the coarse fraction was stored in a glass bottle for later point counting analysis. The remaining portion of the >125 μm fraction was sieved through 250 μm, 500 μm, 1 mm and 2 mm sieves. All individual grain-size fractions were weighed. Mean grain size and sorting were calculated using the software Gradistat (Blott & Pye, 2001). Sorting of samples was quantified as graphic standard deviation using the equation of Folk and Ward (1957). Values below 2 indicate moderate sorting, while values <4 and >4 indicate poor and very poor sorting, respectively. Point counting analysis was done on the split of the >125 μm fraction by F I G U R E 3 Location and sorting of sediment samples counting 300 grains under a binocular microscope. Selected samples of the <125 μm fraction were further analysed using scanning electron microscopy in the SEM laboratory of FMIPA (Institut Teknologi Bandung) and the SEM/EDS laboratory of UPP Chevron-Prodi Geologi (Institut Teknologi Bandung). Correlation, cluster and principle component analysis were conducted using the software package PAST (Hammer, Harper, & Ryan, 2001). Sediment samples were classified using a modified Dunham (1962) classification following the method outlined in Gischler et al. (2017) which proposed a threshold value of 50% matrix to separate mud supported from grain-supported textures. Grainstone has 0%-10%, packstone 10%-50%, wackestone 50%-90% and mudstone 90%-100% matrix. The <125 μm fraction was defined as matrix.
The relationship between water depth and facies diversity was explored using the Shannon evenness index (Shannon, 1948). Rankey (2004) employed the maximum entropy concept whereby divergence from a state of disorder is statistically assessed using the Shannon evenness index to examine facies diversity. Evenness (E) for each water depth is calculated following Purkis et al. (2015) as: whereby n is the number of facies and p is the proportion of each facies within a depth interval. Evenness (E) can reach values from zero to one, with values near one indicating that only one facies is present, meaning that the facies present can be predicted with confidence for a given water depth (Purkis et al., 2015;Rankey, 2004). Because the sample distribution is skewed towards shallow water depth, the binning size of the depth intervals with water depth were increased following an exponential function (f(x) = 2 x ), resulting in depth bins from 0-1 m, 1-3 m, 3-7 m, 7-15 m, etc. A satellite-based facies map was produced by statistical analysis of the colour spectrum of high resolution SPOT-6/7 satellite images to quantitatively discriminate between combined benthic sediment and biota types across the study area. This study applied the simplified workflow of Kaczmarek et al. (2010) with slight modifications to generate SPOT-6/7 derived environmental facies maps. The SPOT-6/7 satellite image was used in this study because the spatial high resolution (1.5 m) is expected to cover benthic diversity in the relatively small (±24 km 2 ) study area. For this study, only the red, green and blue bands were used due to low level returns from subsurface features within the panchromatic band which was deemed unsuitable for work below the ocean surface (Riegl, Reichert, & Ullrich, 2006), and due to almost no penetration potential of the infrared bands because of increasing absorption of the longer wavelength by water (Kaczmarek et al., 2010). Thus deeper water areas contain less data on which to classify (Kaczmarek et al., 2010). Because of this limitation, no environmental facies could be derived for some sampling points at the deeper part of the reef front.
SPOT standard products are delivered application ready; therefore, there is no need for pan sharpened and orthorectified imagery in natural colours. Discrimination of bottom benthic sediments into distinct thematic classes was conducted using the spatial analysis tool "isocluster unsupervised classification" in ArcGIS. This method assessed each pixel in the satellite image by conducting a calculation based on a compilation of the spectral values of an individual band. Applying a large number of thematic classes in the unsupervised classification is critical for capturing the wide range of variability in sediment reflectance (Kaczmarek et al., 2010). In this study, 30 spectral classes were utilized to produce an environmental facies map, resulting in six distinct environmental facies groups which were subsequently validated with field observation.

| Texture
Most sediment samples are grain-supported (Table 1) and can be categorized as either grainstone (0%-10% matrix) or packstone (10%-50% matrix) in nearly equal abundance. Wackestone (50%-90% matrix) is the only matrix-supported texture and forms only a small proportion of all sediment samples. In Panggang Island, the majority of sediment samples are grainstone, while packstones dominate in Pramuka and Semak Daun islands. Sorting is generally poor, but "very poorly sorted" samples also occur frequently (Figure 3). Only one sample classified as "moderately sorted" was found on the eastern reef flat of Pramuka Island.

| Sediment composition
Skeletal components, including fragments of corals and mollusc shells (Figure 4) constitute most of the sediment in Kepulauan Seribu. Less abundant skeletal components are echinoderm spines, Halimeda flakes, crustose coralline algae, the red algae Amphiroa sp., foraminiferal tests, sponge needles and tunicate spicule. The benthic foraminifera Homotrema sp. occurs pervasively, while Calcarina sp. is most abundant in sandy environments and Amphystegina sp. is most prolific at the reef slope. Nonskeletal grains generally are very rare, limited to only aggregates and peloids. Aggregates, usually ca 3 mm in size, consist of bioclasts bound together by marine fibrous or microcrystalline cement. Peloids were sporadically found in the inner lagoon as opaque and cemented faecal pellets. The statistical analysis shows several statistically significant (p < 0.05) and strong (r > 0.6) correlations (Table 2). The fine fraction (<125 μm) content increases strongly with water depth (r = 0.87), while the abundance of most bioclasts, and especially coral fragments (r = −0.77) decreases. Other components, which encompass mainly unidentified grains and sponge spicules, show a moderate positive correlation (r = 0.36) to the fine fraction. The reason for this is likely the small grain size of close to 125 μm for most of the unidentified grains. Foraminifera abundance shows a negative correlation (r = −0.51) to the fine fraction indicating that most foraminifera seem to be adopted to sandy substrates.

| Facies types
Principle component analysis shows three facies groupings that can be differentiated based on their grain size and the relative importance of coral and mollusc fragments (Figure 5). Relatively fine grained sediments containing abundant molluscs form a broad group on the right side of the diagram. The central part of the diagram shows a group of samples which are very poorly sorted and contain corals besides molluscs as an important component. A large group of relatively coarse grained samples on the right-hand side of the diagram is dominated by coral fragments. Using cluster analysis ( Figure 6) this group can be subdivided further into a grainstone (upper part) and a packstone/grainstone dominated facies. In summary, the combination of cluster and principal component analysis allows four facies to be distinguished: coral grainstone, coral packstone/grainstone, coral-mollusc packstone and mollusc wackestone (Figure 7, Table 1).

| Coral grainstone
The coral grainstone facies is generally very poorly sorted and has the coarsest grain size (mean: 0.6 mm) of all facies in this area. The fine fraction content varies between 0%-15%, with a mean of 7% in packstones (20%) and grainstones (80%). Coral fragments are the most abundant components in this facies, on average reaching 60%. Molluscs are quite common and reach an average abundance of 17%, followed by foraminifera tests, which on average contribute 6%. Fragments of red algae, Halimeda, and echinoderms are also present in a limited amount of only 3%-4% each. Nonskeletal (0.3%) and other constituents (0.2%) are very rare. This facies occurs from the sand apron to the

| Coral-mollusc packstone
All samples in this group have 10%-50% matrix (mean: 23%) and are classified as packstone. The amount of corals (37%) and molluscs (21%) is moderate, followed by foraminifera (5%), echinoderms (4%), red algae (3%) and Halimeda (2.5%). Nonskeletal grains are very rare (0.4%). The mean grain size of this facies is 0.3 mm and it can be classified as very poorly sorted. This facies occurs in water depths between 0.4 and 18 m in the shallow part of lagoons, at parts of the reef front and sporadically on the sand apron of Panggang Island (Figure 7).

| Mollusc wackestone
This facies shows the highest fine fraction content with an average of almost 60%. The most common skeletal components in this facies are molluscs and corals. Foraminifera tests and echinoderm fragments contribute less than 2% and 4%, respectively. Halimeda, red algae and nonskeletal components are rare, contributing on average less than 1%. The mean grain size is ca 0.1 mm and samples in this facies are generally very poorly sorted. This facies is found in water depths between 3 and 15 m at the reef slope and in the lagoons of Panggang and Semak Daun Island (Figure 7).

| Satellite-derived environmental facies map
Six environmental facies ( Figure 8) have been derived by integrating SPOT-6/7 multispectral data, statistic-based isocluster unsupervised classification, and ground-truthing with field observation. The sand apron can be subdivided into three facies with no, sparse and dense seagrass cover. One facies is formed by areas with active coral growth. The deeper water zone can be separated into a shallow and a deep subtidal part of lagoons and platform margins. Variation in reflectance can be attributed to benthic biota and water depth. In general, the environmental facies map (Figure 8) indicates that the sand apron with sparse seagrass is the dominant environmental facies and the subtidal reef margin coral facies often occurs adjacent to the sand apron with dense seagrass cover. This might in part be due to the reflectance similarity of the chlorophyll-producing organisms which make coral reefs difficult to distinguish from loose sediments inhabited by seagrass (Kaczmarek et al., 2010;Luczkovich, Wagner, & Stoffle, 1993;Zainal, Dalby, & Robinson, 1993). Field observations show that the natural complexity of benthic organism associations cannot be completely resolved by the satellite data. This is most apparent on the sand apron where usually small solitary coral lumps occur among algae and seagrass (Figure 2). Cluster analysis has recognized four sedimentary facies, while six environmental facies have been derived from the satellite data ( Figure 8, Table 1). In general, sedimentary facies do not correlate directly with environmental facies except for the mollusc wackestone facies that only occurs in lagoonal environments. In contrast, there is no close correspondence between the density of seagrass cover and sedimentary facies in the sand apron. Generally, the texture associated with the subtidal lagoon is wackestone, while packstone and grainstone textures dominate all environmental facies on the sand apron (Table 1).

| Distribution of sediment, carbonate components and facies
Coral and mollusc are the most abundant sedimentary grains with minor contributions from Halimeda, red algae, foraminifera and echinoderms (Table 1). Nonskeletal components occur only in trace amounts in the form of aggregates and faecal pellets (Table 1). In Panggang Island, coral fragments occur mainly at the reef front, where corals are actively growing (Figure 8) and in the sand apron where the fragments are washed by waves and tides. In the lagoon, coral fragments are generally sparse (Figure 9a), except in areas where corals are actively growing (Figure 8). In Pramuka Island, coral fragments occur mostly in the sand apron near the island and reduce in numbers towards the outer sand apron and the reef front ( Figure 9b). A possible explanation for this trend indicated by satellite data (Figure 8) and field observations (Figure 2b,d) could be the local contribution from corals growing among the seagrass on the sand apron (Figure 2b,d).
Mollusc debris occurs in all sedimentary and environmental facies zones and its distribution shows no clear pattern (Figure 9c,d). This generally agrees with the notion of Scrutton (1976b) that molluscan debris is ubiquitous in the Seribu archipelago, a fact that distinguished the modern systems from their Neogene counterparts. Park et al. (2010) believed that the dominant easterly monsoon has formed a large bank consisting of argillaceous foram-mollusc packstone along the western margin of the platform. However, field observation, sedimentary facies analysis (Figure 7) and satellite-derived environmental facies study (Figure 8) show no evidence to support this argument. Mollusc fragments are distributed relatively equally along the western and eastern part of the platform (Figure 9c,d), suggesting the influence of bidirectional monsoon winds rather than an easterly domination.
Halimeda is generally a minor constituent (Figure 9e,f) that occurs mainly in the sand apron and to a lesser degree on the reef front. It is virtually absent from the lagoon. This study agrees with the findings of Jordan et al. (1993) who reported that Halimeda is usually found in the reef flats and only locally concentrated at the reef front and contrary to the findings of Scrutton (1976b) who observed that Halimeda appears to be prolific on the reef flank (approximate depth 10 to 40 m). Nevertheless, the observations reported here cover only the shallow part of the reef flank (down to 22 m) and therefore does not rule out more abundant Halimeda growth below this depth.
Generally, the sedimentary facies of the studied patch reef systems is comparable to other equatorial carbonate systems from Indonesia. Madden et al. (2013) investigated foreshore/backshore and intertidal deposits of the Kaledupa-Hoga in Tukang Besi Archipelago offshore Southeast Sulawesi, Indonesia. They found a similar association of sand-sized bioclasts dominated by coral and molluscs fragments and a general absence of fines and nonskeletal grains. They stress the role of waves and storms in the fragmentation and accumulation of these skeletal sands and coarse coral debris; a process also important in the monsoonal dominated system of Kepulauan Seribu. The shape of the mud-sized sediment particles in the Seribu Archipelago also suggests that maceration and other physical processes play an important role in the breakdown of the skeletal components such as coral, mollusc, echinoderm, red algae and foraminifera (Figure 4). Bioerosion, e.g. by the sponge Cliona seems to contribute to mud production ( Figure 4) but does not seem to play the dominant role suggested by Park et al. (2010). Clear evidence for other mud sources such as inorganic precipitation or bacterial induced precipitation as proposed by Scrutton (1976b) is lacking.
The dominance of sand-sized grains is also apparent in Figure 10 that shows distinct trends in the distribution of the fine fraction (<125 μm) and mean grain size. At Panggang Island, the fine fraction is concentrated in the centre of the lagoon (Figure 10a), which also corresponds to the finest mean grain size (Figure 10c). The sand apron is characterized by coarser grained sediments and a low fine fraction content, whereas the reef front shows a generally finer mean grain size and an increase in the fine fraction content. In Pramuka Island, the coarser mean grain sizes and low fine fraction contents occur in the west-southwest sand apron and reef front which is exposed to waves and currents from the Java Sea (Figure 10b). In contrast, the fine fraction and the finer mean grain sizes are concentrated in the east-northeast sand apron and reef front in more protected environments (Figure 10d). This asymmetry in grain-size distribution on Pramuka Island likely reflects the stronger wind force of the east monsoon compared to the west monsoon in the area. Poerbandono (2016) argued that this imbalance also results in a north-westward net drift of beach sediments. Another expression of the predominance of the eastern monsoon is likely the broader sand apron on the eastern side and the location of Pramuka Island on the more western side of the Seribu platform. In contrast, Panggang Island shows no pronounced morphological asymmetry in its east-west sand apron (Figure 7) and grain-size distribution, which likely results from its more sheltered position between the adjacent islands.

| Correlation between water depth, sedimentary and environmental facies
There is no close correspondence between the environmental and sedimentary facies in the sand apron. This lack of congruency partly arises from the low fine fraction content on the wave and tide swept sand apron. A similar discrepancy between sedimentary texture and environmental facies has been described for other mud lean settings such as in the Holocene tidal carbonates from offshore Al Ruwais in Qatar (Purkis et al., 2017). Despite the absence of a direct correspondence between sedimentary and environmental facies in the sand apron, the satellite-derived environmental facies map allow some predictions about the grain-size distribution. This is apparent from the predominance of the coral grainstones and the lack of coral-mollusc packstones in the sand apron without seagrass (Table 1). Three different environmental facies can be distinguished on the sand apron, based on the density of seagrass cover (Figure 8). Grainsize differences in areas without and with sparse seagrass are statistically not significant, but areas with a dense seagrass cover are characterized by a significantly (p < 0.05) finer sediment grain size compared to other areas on the sand apron ( Figure 11). The dense seagrass meadows effectively reduce current velocities and attenuate bed shear stress (Hansen & Reidenbach, 2012;Koch & Gust, 1999;Ward, Kemp, & Boynton, 1984) thus resulting in a baffling effect which decreases sediment erosion and enhances particle deposition (Koch, Sanford, Chen, Shafer, & Smith, 2006). Therefore, sediment in vegetated areas is usually finer and more organic rich compared to those in unvegetated areas (Fonseca & Koehl, 2006;Kenworthy, Zieman, & Thayer, 1982). This baffling effect results in a slight increase in the fine F I G U R E 8 Satellite-derived environmental facies map generated by an image analysis algorithm for unsupervised classification. Sedimentary facies is shown for comparison as coloured dots. Ocean and land area have been masked. Classification results were subsequently validated with field observation fraction with denser seagrass cover and the generally finer grain sizes in the sand apron environment with dense seagrass cover (Figure 11). An even clearer relationship between sedimentary facies and satellite derived environmental facies exists for the molluscs wackestone that occurs exclusively in the central lagoons of Panggang and Semak Daun (Table 1, Figure 8). This correlation results likely from the strong control of water depth on the satellite derived map (Figure 8) and the abundance of the fine fraction (Table 2). However, a distribution plot of the four sedimentary facies for the entire data set shows a large overlap of the four facies in water depth from 0 to 22 m. This results in a facies predictability of less than 40% and therefore an apparently low degree of ordering (Figure 12a,b). If the data from the reef front are not taken into account, a high facies diversity is only observed in shallow waters less than 7 m in depth (Figure 12c,d). On the other hand, a clear relationship between facies and water depth, and therefore a high predictability can be observed for water depths below 7 m (Figure 12c,d). The This in part random or mosaic-like facies distribution is very different from the observation of Scrutton (1976b) who stated that the boundaries between major facies units Recent studies on carbonate sediment, especially in shallow water, have focused on whether facies distribution is related to bathymetry, questioning the long-held concept in carbonate geology that changes in water depth can be recognized through analysis of the lithofacies (Bosence, 2008;Gischler et al., 2017;Purkis et al., 2015;Rankey, 2002Rankey, , 2004Wilkinson, Diedrich, & Drummond, 1996). These authors briefly outline that the understanding of where and how carbonate sediments are produced and accumulated has evolved from the rather simple concept of direct productivity-depth relationships to the recognition that carbonate depositional environments are influenced to some degree by depth, and also by a complex suite of autogenic factors that lead to facies mosaics especially in shallow waters where sediments migrate and superimpose on each other over short distances as a result of subtle environmental changes. Purkis et al. (2015) showed that observations on the relationship between facies and water depth can be subdivided into three groups: indistinguishable from random (Purkis & Riegl, 2005;Purkis et al., 2015;Rankey, 2004;Wilkinson et al., 1996), deterministic order with respect to depth (Bosence, 2008;Maloof & Grotzinger, 2012) and incorporating aspects of both random and deterministic (Gischler et al., 2017;Purkis & Vlaswinkel, 2012). Study area scale and sampling density emerged to be a significant factor as well because large-scale studies usually show a distinct relationship between facies and water depth, whereas smaller areas and depth range do not seem to represent a clear cut relationship (Harris, Purkis, & Ellis, 2011;Purkis et al., 2015;Rankey, 2004) and high-resolution sediment maps are commonly more heterogeneous with facies mosaics instead of broad, homogenous facies belts (Kaczmarek et al., 2010;Purkis, Harris, & Ellis, 2012). The results obtained here for the relatively small patch reefs of the Seribu archipelago indicate elements of randomness but also deterministic ordering (Figure 12). The lagoonal sediments in water depths of more than 7 m are distinct from the sand apron or reef front. The nearly complete lack of tidal inlets in the relatively wide sand aprons prevents flushing of fine sediments from the interior lagoon. This is in contrast to other studies where currents prevent deposition of fines in the relatively deep lagoons of atoll-like carbonate systems (Betzler et al., 2015;Gischler, 2006). Three factors were likely important for the formation of the continuous sand aprons. Firstly, the late Holocene sea-level fall is thought to have led to widespread progradation of the sand aprons and partial infilling of the lagoons (Park et al., 2010). Secondly, bimodal monsoonal winds lead to a relatively symmetrical distribution of the sand aprons without pronounced lee effects. Thirdly, the small size of the patch reefs of the Seribu archipelago leads to a high ratio between the length of the productive reef front and the area of the platform interior (Purdy & Gischler, 2005). The sediments that are produced by the reef and transported by monsoonal winds and currents to the sand apron therefore can fill in the available accommodation space rapidly. Sea-level fall, monsoonal climate and size therefore all contribute to the effective isolation of the lagoon and the development of a distinct facies.

| CONCLUSION
Four sedimentary facies occur in isolated carbonate platforms of Kepulauan Seribu: coral grainstone, coral packstone/grainstone, coral-mollusc packstone and mollusc wackestone. Six environmental facies could be distinguished using statistical satellite image analysis. These include the deeper lagoon/platform margin, shallower lagoon/platform margin and the subtidal reef margin/coral. The sand apron can be subdivided into three environmental facies based on the density of seagrass cover: sand apron with dense seagrass, with sparse seagrass and without seagrass. The distribution of the environmental facies is mainly controlled by water depth, density of seagrass cover and coral abundance. Only one sedimentary facies (mollusc wackestone) can be directly correlated with satellite-mapped environmental facies (lagoon). In contrast, there is no close correspondence between the density of seagrass cover and sedimentary facies due to the heterogeneity and complexity of the environment. Although no close correspondence was found between the seagrass cover and sedimentary facies, there is a clear influence of seagrass density on the grain-size distribution. In areas of the sand apron with dense seagrass meadows, the grain size is significantly smaller compared to areas with sparse or no seagrass cover. The mean grain size and the abundance of coral fragments exerts a strong control on sedimentary facies. Both parameters are negatively correlated with water depth. However, sedimentary facies cannot be predicted based on water depth alone if the complete data set is taken into account. This can be explained by the stronger wave exposure at the reef front compared to similar water depth in the lagoon. The analysis shows that the shallow carbonate depositional environment (<7 m) is influenced by a complex suite of autogenic processes, resulting in an apparently random facies distribution. For water depths below 7 m, the sedimentary facies can be predicted reliably if the reef front samples are excluded from the analysis. This facies dependence on water depth is in contrast to several other facies studies from atoll structures that showed random facies distributions at similar water depth. This difference can be explained by a lack of tidal inlets at this study site, leading to homogeneously low energy conditions in the lagoon. The relatively closed nature of the lagoons is interpreted to result from a late Holocene sea-level fall, leading to progradation of the sand apron under high energy monsoonal conditions and a partial infilling of the accommodation space on these kilometre-scale carbonate platforms.