Recruitment and formation of new juvenile coral colonies indicate good conditions for development and growth of coral reefs and is essential to their recovery (Glassom et al., 2004; Baird et al., 2006). Coral recruitment success is most commonly quantified by examination of artificial settlement plates. The plates are easy to use, as the recruits are inspected in the lab using a microscope under UV light. However, this method does not allow continuous monitoring because the plates are taken out for inspection (Mundy, 2000; Soong et al., 2003; Field et al., 2007). In addition, they were shown to be biased (Yerushalmi, 2016). Alternatively, in situ visual surveys are also conducted (Edmunds et al., 1998; Miller et al., 2000; Martinez and Abelson, 2013). Both of these methods are labor-intensive and time-consuming (Piniak et al., 2005; Shlesinger and Loya, 2016). Settlement plates require manual and microscopic examination of the surface, while many visual surveys rely on fluorescence and are therefore conducted at night using specialized equipment (Piniak et al., 2005; Roth and Knowlton, 2009; Salinas-de-Leon et al., 2011; Martinez and Abelson, 2013). Small-scale macro photography has also been used for imaging coral recruits (Edmunds et al., 1998). In the larger scale, using photographic surveys of the seabed is becoming a common method for mapping reef communities (Shihavuddin et al., 2013; Cardini et al., 2015). Advances in computer vision have enabled the processing and automated annotation of these large-scale coral reef surveys (Stokes and Deane, 2009; Beijbom et al., 2012, 2016), offering a powerful alternative to manually annotated visual surveys (Singh et al., 2004; Beijbom et al., 2015). However, such large-scale photographic surveys are impractical and are rarely used for coral recruits, which are very small and often cryptic under natural light, making them difficult to detect with traditional photographic techniques (Baird et al., 2006).

Most scleractinian coral and their symbiotic algae (Symbiodinium spp.) contain fluorescent pigments including GFP-like fluorescent proteins (FPs) and chlorophyll, respectively, that fluoresce under UV or blue light (Alieva et al., 2008). The FluorIS system was shown to acquire high resolution, wide field-of-view (FOV) in situ images of coral fluorescence during daytime (Treibitz et al., 2015). This system uses a modified consumer camera, which images both the green (520–630 nm), and red (630–800 nm) broadband components of the fluorescence spectra that correspond to the emission spectra of green, GFP-like fluorescence proteins and chlorophyll-a, respectively. The use of fluorescence has recently been shown to improve accuracy in automated image annotation of coral colonies (Beijbom et al., 2016). Fluorescence imaging is conducted through the use of filtered excitation strobes that emit short wavelengths, usually UV or blue light with emission maxima at 420–620 nm, and a barrier filter that is mounted on the camera lens to block reflected light (Mazel, 1997). At night, only fluorescence excited by the strobes is recorded. During the day, an additional additive signal from the ambient light is recorded (Treibitz et al., 2015). The reflected ambient component, which is not filtered by the barrier filter, appears in the image in green or yellow and is mixed with the fluorescence signal, as seen in Figure 1D. Consequently, most studies detecting fluorescence in situ take place at night to avoid ambient light illumination (Piniak et al., 2005; Baird et al., 2006). Treibitz et al. (2015) showed that fluorescence could be imaged during daytime by subtracting the ambient light image (captured with the strobes off).

In this study we optimized the FluorIS with specific components for surveying coral recruits under 2 cm in diameter (Figure 1), automated ambient light subtraction, and developed a custom software, “RecruitTracker” (Akkaynak, 2017), that improves the efficiency and accuracy of image analysis. We implemented the method in long-term field surveys for recording coral recruit settlement and survival. Our results show that this method reduces errors in identification caused by other fluorescent organisms and/or surveyor errors. In addition, during our surveys we were able to record the settlement, growth, and mortality of coral recruits from 1 mm in diameter and up. Our results indicate relative stability in coral recruit population in the Gulf of Aqaba, with high survival rates of 75% during the first month after settlement.

Compared to previous surveys of coral recruits, the fluorescence imaging method developed in this study enables in situ, high resolution, wide field-of-view daytime fluorescence imaging of coral recruits. We demonstrated that using the FlourIS imaging system with RecruitTracker improved surveys of coral recruits in a spatial and temporal scale. Our method can be used for various research applications and enables further ecological and physiological studies of coral recruits by simplifying large-scale daytime fluorescence surveys.

Until recently, fluorescence imaging systems were limited either by resolution (Winters et al., 2009), spatial coverage of the measurements and/or ease of operation (Mazel, 1997; Mazel et al., 2003). Using a high-resolution camera with a wide-angle lens enabled us to identify small recruits (from ~1 mm in diameter) in a relatively large field-of-view (20 × 20 cm). The system is built with off-the-shelf components and is easy to use and replicate, allowing standardization and widespread use.

The ability to conduct daytime surveys has several advantages for the diver, including safety, ease of navigation, coverage of larger areas, spotting subjects from a distance and reducing fatigue (Zawada and Mazel, 2014). Thus, one of our main goals was to achieve daytime fluorescence imaging. This required overcoming two challenges. First, we implemented automatic registration of the fluorescent and ambient light images, which has not been previously demonstrated. This was not an easy task as the images are from different sources. Nevertheless, our algorithm (RecruitTracker) achieved high success rate of 92%. As opposed to nighttime fluorescence images, the fact that ambient light signal exists in the original fluorescence image works in our benefit and aids in the automatic registration (Figure 2). Second, when imaging during daytime, if the fluorescence signal is significantly lower than the ambient sunlight, the subtraction results in a noisy, less useful, image. To overcome this, we used a wide-angle lens (rather than a macro lens), which allowed us to place the camera very close to the seabed (30 cm). This resulted in high irradiance from the strobes on the coral and hence a strong fluorescence signal in comparison to the subtracted ambient signal (Figure 3). In this case, it was important to use a camera that had high sensitivity and bit-depth. These two system characteristics yield a high signal-to-noise-ratio in the fluorescence image (Treibitz et al., 2015).

Another challenge we faced while developing the system was the fact that fluorescence is not species-specific and there are other organisms besides corals that contain fluorescent pigments, e.g., algae, sponges, and worms. This, combined with user error, could lead to high error rates (Piniak et al., 2005). Many taxa have diverse fluorescent pigments, which helps to distinguish them from scleractinian recruits. In addition, there are corals that express a very weak GFP signal if any, (e.g., Porites sp.) (Roth et al., 2010) whose identification was only possible by their texture and shape in the fluorescence images. Light intensity is known to modulate pigmentation in corals of the species Acroporidae, Merulinidae, and Pocilloporidae (D'Angelo et al., 2008). However, here we do not aim to follow quantitative changes, but use the fluorescence to find the recruits. If high taxonomic resolution is required, utilizing the FluorIS along with a standard reflectance camera is essential (Baird et al., 2006). We followed this recommendation and compared each recruit identified in the fluorescence image to a high-resolution reflectance image. One inherent limitation of in situ imaging surveys is the fact that we can only photograph the surface of the reef, and cannot see recruits that settle on the side or beneath rocks.

We demonstrated the higher precision of this system in comparison to regular images and visual surveys. The advantages of using an imaging method over a visual survey are that field-time is reduced, images can be taken by a diver with no taxonomic expertise (or an autonomouse vehicle) and re-examined by multiple observers for accurate analysis, and potentially be analyzed by automated annotation programs (Edmunds et al., 1998; Piniak et al., 2005; Burgess et al., 2010). We have shown here that by using our fluorescence images we found more coral recruits compared to the amount we would have found using only reflectance images and diver observations. In addition, due to the camera's high resolution and modification carried out on the FluorIS, expanding spectral range in the long wavelengths, errors in identification caused by other fluorescent organisms and/or surveyor errors were significantly reduced (Figure 4). Moreover, the number of non-fluorescent recruits was ~11%. This suggests that to speed identification, it can be done only on the fluorescence images, with a cross-check to the reflectance images to verify that the fluorescent organism is indeed a coral recruit.

In this study, we covered most of the known reproduction periods of the dominant corals in the northern part of the Gulf of Aqaba. The spawning is known to be a year round process, in which the different coral species have different spawning dates (Shlesinger et al., 1985, 1998). We found relative stability in recruitment/mortality rates with high survival rates at the two surveyed sites (Figure 6). The coral reefs in the Gulf of Aqaba have been reported in decline and at risk of collapse, due to both local and global stressors (Glassom et al., 2004, 2006; Loya, 2004; Abelson et al., 2005). Nevertheless, our results are in line with a recent study (Shlesinger and Loya, 2016) which suggested that the reefs in Eilat (Gulf of Aqaba) are recovering. It is likely that our findings are an underestimation of the actual rates since we only surveyed a one-year reproduction period. Therefore, consistent in situ monitoring studies should be performed to reveal the full picture of the settlement and survival of coral recruits in order to estimate the true status of the reef.

We recorded 16% less survival during the first month at the IUI site in comparison to the MPA. The two sites are only a few 100 m away from each other. Nevertheless, they differ in the following parameters: the MPA was declared by the Israeli Nature and Parks Authority 20 years ago and has been closed to the public since, whereas the IUI area is subject to frequent diving activity. In addition, the topography of the reef at the MPA is more complex and three-dimetional whereas at the IUI the reef is mostly flat. These facts can explain the lower survival rates during the first month after settlement at the IUI in comparision to the MPA. Extensive human activity, structural complexity and availability of substrate for settlement was reported to have an important impact on the recruitment process at different locations (Hughes and Connell, 1999; Alvarez-Filip et al., 2011). Based on our data, we estimate ~72 recruits settled per m² year⁻¹. This result significantly differs from other studies conducted on natural substrate such as Shlesinger and Loya (2016), who conducted their surveys at the same sites and found ~15 recruits per m² year⁻¹. Other studies found ~7 recruits per m² year⁻¹ at the Great Barrier Reef (Connell et al., 1997; Hughes et al., 2007) and ~4 recruits per m² year⁻¹ in the western Atlantic (Smith, 1992; Miller et al., 2000).

Significant differences between similar studies using different methods were demonstrated by Abelson et al. (2005) and Yerushalmi (2016), as well as inconsistent findings in recruitment studies using different experimental protocols. Yerushalmi (2016) showed significant differences in number and in species distribution between artificial and natural substrate of coral recruit settlement, the author reported more recruits on artificial substrate as opposed to more species diversity on natural substrate. Moreover, several studies that examined coral recruitment processes in the Gulf of Aqaba using in situ settlement plates followed by microscopic analysis (Glassom et al., 2004, 2006; Abelson et al., 2005; Martinez and Abelson, 2013), presented significantly different recruit numbers at the same study site. In this study, we demonstrated these biases by showing different recruit counts when using different methods. We counted more recruits with our method due to its higher precision. Hence, there is a crucial need for a standardized method of recruitment assessment in coral reefs worldwide (Abelson and Gaines, 2005), which could be fulfilled by our method.

In this study, we also showed that various phenomena and morphological changes in recruits could be tracked using our imaging system (Figure 7). For example, we were able to record fusion, also known as a chimerism, of two recruits. This could open the way for a larger genomic study on chimerism in coral reefs. Chimerism is known to exist in several marine colonial animals from different phyla including sponges, cnidarians, bryozoans and tunicates, as well as in protists and plants. To date, the amount of chimerism in reef corals in situ is unknown (Heyward and Stoddart, 1985; Rinkevich and Weissman, 1987; Puill-Stephan et al., 2009). In addition, the ability to calculate growth rate with a non-distractive method in situ is crucial for large scale monitoring. The growth rate of Porites sp. recruit presented here (~0.036 cm² year⁻¹) corresponds with the growth rates of Porites sp. known from the Great Barrier Reef and the Gulf of Aqaba (Lough and Barnes, 1997; Al-Rousan and Felis, 2013). We have to note, that since corals are three dimensional whereas the images taken by this method are two dimensional, natural changes in the environment such as seasonal algal blooms, which cover large areas of the substrate, changes in the quadrat positioning and camera angle can affect the calculation.

Coral recruitment, particularly settlement patterns and survival rates, has been long identified as a vital process in the ability of reefs to recover from anthropogenic and natural disturbances (Hughes et al., 2007, 2010; Urvoix et al., 2012). Therefore, understanding the recruitment process is essential for developing suitable reef conservation and management strategies. We believe that our low-cost, non-destructive, and easy-to-use method will help standardize surveys and long-term monitoring of coral recruits. In addition, it can be applied via underwater vehicles for rapid and automated surveys. In the future, we plan to develop automatic annotation methods to speed up image analysis. The recorded data can be easily uploaded to create universal databases contributing to improved understanding of this vital and delicate life stage in the coral's life cycle. This will help predict the destiny of coral reefs that are at risk of decline due to numerous environmental stressors.

TT conceived the project, built the system, and advised algorithm development. AZ led lab and in situ experiments, performed data analysis, and provided consent for the publication of the identifiable images. TM helped and advised throughout the experiments and contributed to experiment design and data collection and analysis. DA wrote the image annotation code and helped in field experiments. AZ. TT, TM, and DA wrote the manuscript.