Underwater macrophotogrammetry to monitor in situ benthic communities at submillimetre scale

Larval settlement and recruitment of sessile organisms are key ecological processes for population recovery and maintenance that occur at scales invisible to the human eye. Accordingly, proxies of recruitment have commonly been quantified using artificial substrata such as settlement tiles made of diverse materials and shapes, which are typically transported to the laboratory for examination. However, it is unknown how much bias is introduced with this sampling strategy and how recruitment quantified on tiles relates to recruitment on nearby natural substrata. Here, we applied techniques that combine macrophotography with photogrammetry (macrophotogrammetry) underwater to monitor benthic communities at submillimetre scale. This application allows the investigation of recruitment and community succession of the earliest life‐history stages in situ and on natural substrata. We tested the use of four different imaging systems, varying in costs from US$ 1400 to US$ 5440. While the most expensive SONY αRiv system provided the best visual output and ground resolution (up to 5 μm/pixel with a + 4 close‐up lens); regardless of systems, 3D models always had a ground resolution ≤23 μm/pixel and errors in planar measurements of submillimetre features were similar among systems. This level of resolution compares well with stereomicroscopy in the range of 5:1 to 10:1 magnification, while providing detailed 3D digital records through time. Using a coral reef example, we apply this approach to demonstrate how it can be used to monitor small reef areas (~300–600 cm2) through time, including the quantification of biophysical metrics such as cover of small facilitative and competitive organisms and microhabitat complexity. We further show that organisms as small 0.5 mm in size, such as 2‐month‐old coral settlers, can be located accurately within the 3D models and measured with a good level of confidence. This method can be readily applied to other benthic environments to elucidate drivers of early recruitment and recovery of benthic organisms following disturbance impacts at very fine scales, directly on natural substrata, to avoid biases inherent with laboratory‐based analyses of artificial surfaces.


| INTRODUC TI ON
Following large disturbances, the recruitment of new sessile organisms onto recently created spaces is a key ecological process allowing the recovery of both temperate (Roughgarden et al., 1988) and tropical marine communities (Connell et al., 1997). Most habitat building benthic organisms have a bipartite lifecycle in which recruitment involves three major phases: (1) larval supply, (2) larval settlement and (3) post-settlement survival until recruits become visible (Harrison & Wallace, 1990;Keough & Downes, 1982). Each of these phases, and especially phases 1 and 2, occur at scales that are typically invisible to the naked human eye (<1 mm), which has led to a proliferation of studies requiring examination of artificial surfaces in the laboratory with microscopy for quantification.
However, the use of artificial substrata to quantify recruitment introduces some level of bias related to: tile materials, tile orientation, duration of conditioning timeframe (and associated fouling communities), handling, change of environmental conditions during laboratory work and/or following bleaching and drying treatments (e.g. Hamizan et al., 2021;Harper et al., 2021). As microhabitats and fouling communities on tiles have been shown to interact with larval settlement and post-settlement survival on tiles (Doropoulos et al., 2016;Menge & Lubchenco, 1981;Wright & Steinberg, 2001), it remains largely unknown if this is consistent with recruitment patterns on nearby substrata. Therefore, elucidating patterns of recruitment of benthic organisms onto natural substrata requires investigation, particularly during the early life-history phases. This has been challenging due to the microscales and cryptic nature of the process, but new technologies exist that can be applied to monitor benthic communities in situ at micrometre ground resolution, similar to already existing methods at centimetre/millimetre ground resolution (Ferrari et al., 2017(Ferrari et al., , 2021Lange & Perry, 2020).
High-resolution imagery techniques have been developed and trialled to detect organisms on benthic substrata at millimetre scale, such as planar macrophotography (Edmunds, 2002) and fluorescence imagery (Baird et al., 2006;Zweifler et al., 2017), and at micrometre scale such as underwater microscopy (Mullen et al., 2016).
While these methods have proven to be successful, new technologies can be applied to further improve the resolution, depth of field (DOF; i.e. the distance between the nearest and farthest point in focus), and/or metrics in three dimensions (i.e. microhabitats, rugosity, growth). The proposed method aims to achieve similar visual output and resolution as common stereomicroscopes at a magnification up to 10:1, while maximizing DOF and providing a digital 3D record that can be reviewed through time with no handling impacts.
To achieve these aims, macrophotography techniques can be combined with photogrammetry to build a 3D model at high resolution, which is termed 'macrophotogrammetry'.
To take high-resolution and sharp photographs, it is necessary to use high-resolution sensor and a 'macro' lens (high focal length value >55 mm). The main challenge is typically keeping most of the target subject or area in focus and this is driven by the DOF that is dependent on three variables: the diaphragm aperture, the focal length and the distance to the target subject. The longer the focal length of a lens, the narrower the field of view, the less light reaches the sensor and the DOF will be smaller (London et al., 2016). Therefore, a balance between distance to the subject or area, aperture settings and light environment is required to improve the DOF of highresolution photographs.
Photogrammetry involves the two steps of (1) taking a series of photographs at different angles around a scene and (2) stitching the images by a structure-from-motion (SfM) software to reconstruct a model of the scene in 3D (Sturm & Triggs, 1996;Ullman, 1979). This technique provides a scaled record of a scene in time that is now widely integrated into ecological surveys for monitoring and restoration (Ferrari et al., 2021;Figueira et al., 2015;Kanki et al., 2021;Peterson, 2019;Storlazzi et al., 2016). Underwater photogrammetry is applied in both tropical and temperate reefs to monitor change in benthic communities at ~50m 2 scale, providing benthic orthomosaics (i.e. 2D flattened view of reconstruction) at centimetre to millimetre ground resolution (Edwards et al., 2017;Ferrari et al., 2021;Pedersen et al., 2019), and more recently 3D models to quantify growth of sessile organism such as corals at millimetre ground resolution (Ferrari et al., 2017;Figueira et al., 2015;Lange & Perry, 2020).
However, photogrammetry has not yet been tested using underwater macrophotography techniques to examine if the model reconstruction can reach micrometre ground resolution that is required for investigations related to benthic algal and invertebrate recruitment processes.
In these fields, resulting 3D models were shown to reach a ground resolution (i.e. the limit of detailed clarity in an image) of 5.1 μm/ pixel when studying lichens on tree trunks (Peterson, 2019), similar to stereomicroscopy at ~10:1 magnification. This study aims to impacts at very fine scales, directly on natural substrata, to avoid biases inherent with laboratory-based analyses of artificial surfaces.

K E Y W O R D S
3D, Agisoft Metashape, coral reefs, macrophotography, monitoring, photogrammetry, recruitment, structure from motion demonstrate the potential of underwater macrophotogrammetry (1) to provide 3D models at micrometre resolution, (2) for applications in benthic community succession, natural recovery and/or restoration and (3) for automated processing and future machine learning opportunities. First, we present a monitoring workflow to investigate georeferenced benthic communities at submillimetre scale through time by reconstructing 3D models, digital elevation models (DEMs) and orthomosaics of small benthic areas (100-631 cm 2 ) from in situ underwater macrophotographs, using noninvasive techniques and automated processing. Then, we provide a comparison of 3D models of the same scene taken by four different systems to highlight their advantages and disadvantages. Our workflow is illustrated with data collected on coral reefs but its application is appropriate for most benthic ecosystems.

| Field plot strategy and photography systems
Several reef areas were used to assess the feasibility of this method using permanently marked locations on the reef (reef areas) and settlement tiles (tile areas). Georeferenced reef areas varying in size from ~200 to 600 cm 2 were revisited through time at three time points, as camera systems were tested. Reef areas were permanently marked using stainless-steel nails and flexible plastic tags.
Small 'tile' areas (<200 cm 2 ) were only imaged once in June 2022 to test specific settings using one of the camera systems. Field work was conducted at Lizard Island, Australia (S 14° 40′ 4.9548″, E 145° 27′ 49.5972″) using SCUBA. Before photographing a scene, three to six 'mini' coded target stickers (6 mm diameter in size; Figure S1) fixed onto fishing weights were placed within the area of interest to improve photo alignment and scale the models (Agisoft LLC, 2021). Depending on the system used and area size, 50-300 photographs were taken around each scene at different oblique angles to improve the 3D dimensions, geometry and accuracy of SfM (Rossi et al., 2017).
The first system tested, hereinafter named 'TG6 handheld', was an Olympus TG6 (12MP) in the Olympus PT-059 housing with a 3000 lumens LED ring light (Kraken Weefine) mounted around the objective ( Figure 1a). The camera was used in microscope mode and handheld to take photos at oblique angles around the scene ( Figure S1b).
The second system, hereinafter named 'TG6 rail', was similar as the first one but tested the influence of focus stacking mode. Focus stacking mode composites 10 frames at varying depth field into one frame of maximized DOF. To use this mode effectively, the camera needs to be motionless during shooting, and was therefore mounted on a custom built 'quadropod' rail with a tripod head secured on a F I G U R E 1 Photos showing the TG6 systems with and without the sliding rail (a) and the SONY systems with and without the light quadrat and/or the close-up lens (b). (a) Olympus TG6 systems with LED ring light to be used handheld or mounted on rail with slider in microscope focus-stacking mode. (b) Sony systems with LED ring light, light-quadrat and optional close-up lens +4 for added magnification-handheld shooting. slider (Figure 1a; Figure S2). The quadropod rail was designed in SolidWorks (BIOVIA, Daussault Systemes, SolidWorks CAD v2019, San Diego), 3D printed and assembled ( Figure S2). Flexible legs (GorillaPod arm kit pro) were fastened to the quadropod rail and feet were weighted using lead to increase stability. During shooting, the LED ring light was switched on to 75% intensity and photographs were taken at 24 regular intervals around the scene with the camera inclined at 20° and 40° from the substrate plane and at 12 regular intervals with the camera inclined at 55° ( Figure S1a).
The next systems were all handheld but using Sony full frame mirrorless cameras. The third system tested, hereinafter referred to as 'SONY αiii', was the SONY αiii (24 MP) with a Sony 90 mm FE Macro G OSS lens in a SeaFrogs housing with 67 mm threaded flat port for a 90 mm macro lens (V.3 series 40 metres). The fourth system tested, hereinafter referred to as 'SONY αRiv', was the SONY αRiv

| Images processing and macrophotogrammetry workflow
Images were initially batch edited to improve contrast, brightness, sharpness, white balance and then converted into .tiff formats using Affinity photo software (Affinity Serif LTD, 2021). Sets of images were then processed in Agisoft Metashape software (version 1.8.1; Agisoft LLC, 2021), predominantly through a high-performance computing system (HPC) at CSIRO (Bracewell: Linux 64 bit, RAM: 251.81 GB, with two NVIDA P100 GPUs and two Intel Xeon 14-core CPUs). Processing was also done using a standard laptop computer (DELL Latitude 5400, RAM: 16.0 GB, CPU: Intel(R) Core(TM) i5-8365U CPU @ 1.60GHz) to provide a comparison in processing time between both computing systems. Macrophotogrammetry workflow using Agisoft Metashape (Agisoft LLC, 2021) followed work by Peterson (2019) and Lange and Perry (2020), which was further refined to utilize the best visual outputs using new algorithms from recent software versions, the Agisoft user manual, and the Agisoft online forum (https://www.agiso ft.com/forum/ index.php; Table 1).
The workflow to generate dense point cloud and 3D models was translated into python codes by adapting the existing Metashape workflow (Young et al., 2021) to our macrophotogrammetry workflow for batch processing on HPC (Gouezo & Slawinski, 2023).

| Scaling, alignment and measurements
Models were moved and rotated for a top-down view alignment (x, y, z) with the z axis set approximately perpendicular to the reef substrate ( Figure S1). Models were scaled by manually adding five to eight scale bars on the mini-target centre circles (radius = 1 mm; Figure S4). Most models were oriented manually using three small stainless-steel nails placed in the scene as permanent markers serving as a spatial reference. Small organisms and specific features present on the 3D model were marked using markers and their area measured using the polygon tool. Then, DEMs were built and rugosity measurements extracted using replicated polylines drawn across the DEM. Lastly, high-resolution orthomosaics of benthic areas were built, exported as .jpeg photos, and imported into ReefCloud (Gonzalez-Rivero et al., 2020) for percentage cover analysis of microscale benthic communities (Table S2). One hundred random points were allocated to orthomosaics and resulting percentage covers were compared visually using a stacked bar chart.
In photogrammetry, ground resolution and scale are two important concepts with different meanings. Ground resolution refers to the level of detail of the 3D models, defined as the distance between the centres of two consecutive pixels. In this study, it ranged from 5 to 23 μm depending on systems, meaning that a feature of 50 μm in size would be made up of ~2-10 pixels, which may be difficult to clearly detect depending on its shape and colours. The scale of the 3D models relates to the conversion of sizes between 'real life' and the model and is indirectly related to the resolution. High-resolution imagery provides a better representation of fine details and, therefore, more accurate measurements, increasing the scale's precision.
At the level of models' resolution in this study, the scale of the 3D models was found to be up to 10:1 (when zooming in into the model), allowing to clearly detect and accurately measure submillimetre (i.e. 500-800 μm) features with low levels of errors.

| Systems comparison
To deepen the comparison among systems, models of the same area were made and compared using four different systems: TG6 handheld, TG6 rail, SONY αRiv and SONY αRiv + Close-up lens. The surveyed area was a 10 × 10 cm crevice recruitment tile with metallic beads of 0.8 mm diameter glued at several locations on the tile ( Figure S4). A comprehensive comparison on system specifics, costs, model reconstructions outputs and planar measurement accuracy among these systems are presented in the results. To assess planar measurement accuracy, the diameter of n = 20 beads was measured on each model as well as five features of the tile (three cross-sections, the diameter of the centre hole and the width of TA B L E 1 Macrophotogrammetry workflow used to build models and extract data on benthic communities in situ at micrometre resolution.

Menu Functions Explanations and settings
All steps below can be batch processed using Python Use the radius of circle in the middle of target (=1 mm). Add another marker at the other end of radius (see Figure S4) the crevice crown, Figure S5). Real-life measurements of these tile features were done using a stainless 150 mm digital calliper with 0.01 mm resolution ( Figure S6). Errors in measurements (between model and real life) were quantified for each measurement replicate.
Absolute errors were compared among the four systems using a linear model on square root transformed response variable to conform to model assumptions of normality and homogeneity of variance.
The linear model analysis was conducted in R Statistical Software (R Development Core Team, 2023).

| RE SULTS
Overall, models created with the sharpest photos with a good DOF and high overlap (~70%) provided the best fine scale details and visual 3D model outputs. 3D models of underwater benthic areas (~200-600 cm 2 ) provided a ground resolution of ≤0.0218 mm/pixel (Table S1).

| High-resolution 3D models to monitor the survival and growth of small organisms
Meshes built from depth maps with the highest face count and high texture resolution (8152 × 8152 pixels, which could be higher depending on a computer's graphic card limit) resulted in the best detailed and textured models (Figures 2 and 3).
Models such as the one in Figure 2a that were repeatedly sam-

| DEM and orthomosaics
The DEM from the dense cloud model provided micro-rugosity details at a scale that could not be measured manually underwater ( Figure 4) and detailed data on the size and depth of individual crevices as small as a few millimetres can be extracted (Figure 4b). For

| System comparison
The tested systems produced models with a ground resolution ≤23 μm/pixel, with accurate reconstruction of tile crevices (1 cm deep) and planar measurements of the small beads with an average of 42 μm errors ( Figure 6; Table 2). Measurements of large features such as crevice tiles cross-section (7-9 cm in length) had measurement errors of up to, on average, 0.5 mm. There was no significant difference in measurement errors among the tested systems (Linear Model, p > 0.05, Figure S7).
The best model, showcasing highest resolution, highest point cloud density and overall best visual output, was imaged with the SONY αRiv system under well-lit conditions (Table 2). Adding the close-up lens enabled the photographer to get closer to the scene while increasing the light intensity provided by the LED ring light, which provided sharper images than without the close-up lens.
However, using close-up lenses decreases the DOF, and it is therefore recommended not exceeding +4 to +6 in strength ( Figure 6).
The colours and fine details better represented the crevice tiles 'real life' features when using the Sony set-up. The TG6 systems produced a more regular dense cloud, often less noisy than with the SONY systems, resulting in fewer holes in both the dense cloud and meshes ( Figure 6). Using the TG6 with the rail in focus stacking mode, led to sharp images with a high DOF and decreased the ground resolution by 50 μm/pixel (Table 2). Sampling effort was considerably lower when using the camera handheld than using the TG6 rail (Table 2). For example, three to four ~400 cm 2 plots were imaged in a 75 min dive using the TG6 rail, whereas six to eight ~400 cm 2 plots could be sampled in the same amount of time using the handheld camera. Macrophotogrammetry workflow processing time increased with image resolution. It ranged from ~1 h (TG6 rail) to ~6-10 h (SONY αRiv) using the HPC (Table 2) and between ~2 h (TG6) and >12 h (SONY αRiv) using a regular laptop. The advantage of using the HPC was that models could be set up as a batch process overnight. Once 3D models were generated, the manual steps to scale the model, extract data, generate DEM and orthomosaics generation took approximately ~30 min to 1 h per model.

The total materials costs, including an educational licence to
Agisoft Metashape software, ranged from US$1400 (TG6 handheld) to US$5440 for the SONY αRiv system (further detailed in Table S3).
The light quadrat developed is recommended to use with the SONY system, as the aperture setting is narrow (f/22) and focusing distance is higher than with TG6 systems (Table 2), leading to a loss of light. This would add a cost of ~US$660 (Table S3).

| DISCUSS ION
In this study, we show that macrophotogrammetry can be applied underwater to monitor benthic communities at micrometre ground F I G U R E 3 Time series of benthic models (a) and source photographs (b, c) at the same location 2 days before (i) and after coral larval enhancement interventions on the reef (ii, iii), displaying coral recruits ≤1 mm in diameter visible on both models (a.iii) and source photographs (b,c.iii).
resolution. We tested the use of four different camera systems varying in cost from US$1400 to US$5440. While the most expensive

| Application
This method was initially designed for application on coral reefs during the early stages of recovery, to investigate recruitment processes occurring in situ directly on the natural reef substrata, as opposed to using artificial substrata. Recruitment tiles may not replicate well natural substrata and introduce an unquantified level of bias that may have major implications for our understandings of ecosystem recovery. This method allows the sampling of the same plots through time to quantify the succession of facilitative and competitive organisms following a large disturbance, when space becomes available and ecological succession occurs. On coral reefs, applying this method in areas that have shown extremely rapid recovery (<6 years) would help to elucidate the key facilitative drivers that enable successful recruitment and rapid growth of juvenile corals, topics which are still debated (Doropoulos et al., 2022;Edmunds, 2018;Gilmour et al., 2013). This method could also be applied during coral larval restoration (dela Cruz & Harrison, 2017;Harrison et al., 2021) to investigate larval supply density-dependence effects on early stages of survival and growth of settled corals, previously done using tiles (e.g. Cameron & Harrison, 2020;Doropoulos et al., 2017;Sampayo et al., 2020) and thus avoiding the associated biases.
With this level of resolution, the application of the method is not limited to scleractinian coral communities, as it could also be readily applied to investigate benthic ecological interactions that limit or facilitate early stages of recruitment and regeneration of kelp reefs, rocky shores or mussel beds (Dayton et al., 1992;Keough & Downes, 1982;Lubchenco, 1983). For example, the approach could be applied to better describe the ecology of micro algal turfs (Connell et al., 2014) and their global proliferation in both coral and temperate reefs (Airoldi et al., 2008). Our method also allows researchers to quantify for the first-time the micro-rugosity features of the natural substrata using DEM models, which cannot be measured at this level of accuracy manually (<1 mm). Quantifying this metric at the same scale as benthic organisms at their early stages is key to testing the influence of cryptic microhabitats in providing refugia immediately after settlement (Doropoulos et al., 2016;Lubchenco, 1983).

| Benthic plot photography strategy
The strategy adopted to photograph benthic plots differed between the TG6 and SONY systems The advantage of using the TG6 rail is that the camera was motionless during shooting and therefore focus F I G U R E 6 Models of crevice tiles used to accurately compare the tested systems: TG6 handheld (a-c.i), TG6 rail (b, c.ii), Sony αriv (b, c.iii), Sony αRiv with close-up lens +4 (a.ii, b, c.iv).
stacking could be used to increase the DOF of photographs (Santella & Milner, 2017), leading to a resulting model with higher resolution than using the TG6 handheld without the focus stacking mode. In addition, by moving the camera in small increments around the rail, the overlap between photographs was regular (≥70%), which is key requirement for good model reconstruction ( Figure S1). The main disadvantage of using the TG6 rail was that it took time to set up and shoot in focus stacking mode, therefore only three to four plots could be sampled in a 75 min dive or snorkel in a shallow reef habitat.  diver with good buoyancy control, which contributed to decreasing motion blurring on photographs. Therefore, if the approach is conducted by inexperienced divers, in situ imaging practice or the use of the TG6 rail is necessary for accurate data collection.
The light environment when taking photographs is extremely important in underwater photography. As depth increases, water absorbs different wavelengths starting with the red-light spectrum.
In this study, we found that using high levels of constant artificial light (i.e. >10,000 lumens around an area <0.25 m 2 ) with external video lights (as opposed to using strobes) gave the best imagery output. The lighting environment facilitated highly detailed imagery and definition to fine-scale features of the organisms being documented for precision outlining and identification, as well as effectively lighting shaded areas of microhabitat scale complexity where algae and invertebrates typically prefer to recruit into. We found it was best to avoid very shallow depths (i.e. <2 m) on sunny days, where sunlight glare is intense between 11 AM and 2 PM and can overwhelm the steady light environment created by artificial lights.
We recommend that the size of benthic plots should not be much larger than 600 cm 2 . Recruitment is highly variable through space (Adjeroud et al., 2007;Gouezo et al., 2020), even at a scale of a site (Bauman et al., 2015); therefore, it is best to have high replication of many small areas than few replicates of a larger area. Additionally, using this sampling approach, smaller-size models are faster to process and easier to navigate and analyse at such high resolution using a standard computer.

| 3D reconstruction and accuracy
The ability to model small reef plots at such high ground resolution using a handheld camera while on SCUBA exceeded initial expectations. The 3D reconstructions of these small reef areas take time to process but provide more data than just taking planar photographs such as (1) a larger ground sampling area up to 600 cm 2 (as opposed to maximum of 63 cm 2 ,  Figure S3).
However, meshes should always be cross-referenced with the dense cloud as the latter is the truest representation of the surface. These small error levels allow the mapping of organisms as small as ~0.5 mm and monitoring their growth from a size of ~0.5-0.8 mm with a high level of confidence (Figures 2 and 3; Table 2).
This outcome builds on and improves previous 3D modelling work on adult coral growth (Ferrari et al., 2017;Figueira et al., 2015; Lange & Perry, 2020) with measurement errors ranging from 1 to 5.8 mm. In this study, the 2D size of small organisms (≤1-1.5 mm) was measured as the points making up the cloud were either not dense enough or had little holes reducing confidence to truly represent their volume at such small size ( Figure S8). Using this study workflow, we recommend measuring volume of organisms only when >3-5 mm in size ( Figure S8), which is larger than juvenile corals quantified in Quigley (2022) using a dental scanners in ex situ laboratory settings.

| Limits
While levels of accuracy of reconstructions and precision of models were overall very high in this study, macrophotogrammetry still has some limits that need to be considered for future use and development. First, there can be instances where some parts of the models are not well represented because of areas with missing data in the scene. This can be due to limited overlap between some images (i.e. <50%), poor DOF on images (i.e. blurry parts) and/or when camera view angles are not possible due to obstructions. The quality of the photographs is key; therefore, this aspect needs to be carefully planned and should not be rushed while diving. Second, photogrammetry in general does not perform well when objects within the scene are moving. In the proposed context of investigating benthic recruitment, if parts of the benthos are covered with elongated fleshy algae that is prone to moving during imaging within the small ~500 cm 2 scene for example, the model reconstruction will be very poor due to moving algae and the algae will obstruct parts of the substrata.
Lastly, the high resolution of these models come at a cost of processing time, which can be over 12 h on a standard computer for the high-resolution SONY αRiv system. If models cannot be processed using a HPC as in this study, we suggest that a computer with a sufficient amount of RAM and GPU card supported by photogrammetry software (e.g. gaming computer) is used for model processing. We acknowledge that this can be a limiting factor and therefore the lower resolution and less expensive TG6 imaging system may be more cost and time efficient for use by some researchers and monitoring teams.

| CON CLUS ION
The method developed in this study has shown that macrophotogrammetry can achieve excellent outputs underwater when investigating ecological succession in situ. This technique does not require installation/retrieval of equipment nor access to laboratory microscopes, becomes inexpensive following initial set-up costs, and provides a permanent 3D record that characterizes small benthic organisms and bio-physical interactions occurring directly on the substrata. This study tested five different camera systems, which resulted in model's resolution ≤23 μm/pixel. The choice of camera system will depend on funding and computing power availability.
The TG6 system is more cost and processing efficient, whereas the SONY system provides better image quality and finer resolution but is costlier and requires more computational power.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors have no conflict of interest.

PEER R E V I E W
The peer review history for this article is available at https:// w w w.web of scien ce.com/api/g atew ay/wos/p e er-revie w/ 10.1111/2041-210X.14175.

DATA AVA I L A B I L I T Y S TAT E M E N T
Animations of 3D models can be accessed via the first author pro-  (Gouezo & Slawinski, 2023).         Figure S8. Dense cloud of a 2.90 mm bead. Table S1. Summary of macrophotogrammetry results of small benthic plots and tile models using the four different systems. Table S2. Micro-community benthic codes used to analyse the percentage cover of small benthic organisms on orthomosaics in