Can Juvenile Corals Be Surveyed Effectively Using Digital Photography?: Implications for Rapid Assessment Techniques

Environ Monit Assess DOI 10.1007/s10661-009-1282-1

Can juvenile corals be surveyed effectively using digital photography?: implications for rapid assessment techniques

Scott C. Burgess Kate Osborne · · Bianca Sﬁligoj Hugh Sweatman ·

Received: 1 April 2009 / Accepted: 3 December 2009 ©SpringerScience+BusinessMediaB.V.2009

Abstract The widespread decline of coral reefs Barrier Reef. On average, estimates of juvenile requires integrated management measures across densities from photographic images were lower, whole regions. Knowledge of demographic in both absolute and relative terms, than that processes of reef organisms is important for estimated from images. This was the case whether informed management, yet current techniques colonies <20 mm or <50 mm in diameter were for assessing such processes are time consuming, considered. Overall differences between methods making it impractical to gather relevant informa- were generally greater at reefs where recruitment tion over large scales. We tested the usefulness was higher, though proportional differences of digital still photography as a rapid assessment (density from images/density from direct visual technique to estimate coral recruitment—an im- census) still varied among reefs. Although the portant process in coral reef recovery. Estimates ranking of taxa, in terms of their densities, from of the density and diversity of juvenile hard corals the two methods were similar, the density of from digital images were compared with direct common genera was generally underestimated in visual estimates from the same plots made in the images, and the occurrence of ‘unknown’ taxa was ﬁeld. Multiple plots were sampled on four reefs higher. We conclude that photographic images do from a range of locations on Australia’s Great not constitute a reliable rapid assessment method for estimating the spatial patterns in the density or diversity of juvenile hard corals.

S. C. Burgess K. Osborne H. Sweatman Keywords Coral recruitment Digital · · · Australian Institute of Marine Science, photography Method comparison Monitoring PMB No.3, Townsville MC, Townsville, · · methods Rapid assessment methods Queensland 4810, Australia · B. Sfiligoj Life & Environmental Sciences, Introduction Deakin University, Princes Highway, Warrnambool, Victoria 3280, Australia The increasing pressure on coral reef health Present Address: worldwide calls for the management of coral reefs S. C. Burgess (B) across large spatial scales (Nyström et al. 2000; School of Biological Sciences, The University of Queensland, Brisbane 4072, Australia Wilkinson 2004;Sweatmanetal.2008). Monitor- e-mail: [email protected] ing data on the status and trends of hard corals Environ Monit Assess across large scales provides managers with a con- observers to provide estimates of observer error, text for assessing human impacts and a basis upon and (4) the images can potentially be analyzed by which to make decisions. Monitoring should not automated expert systems. only involve the assessment of the abundance and In principle, digital photography could be used distribution of the standing stock (mainly adult to estimate densities of juvenile corals, which colonies) but should also look into the underly- would increase understanding of spatial and tem- ing mechanisms of change. Recruitment (which poral variation in coral population dynamics represents the net result of larval supply, settle- across large reef systems. For digital photogra- ment, and early post-settlement mortality) is well phy to be useful, however, the accuracy and pre- recognised as a key demographic process influ- cision of estimates of juvenile density need to encing population decline and recovery in coral be compared with the best estimate of densities, reef ecosystems (Bak and Engel 1979;Harrison which currently come from detailed visual census and Wallace 1990;Caleyetal.1996;Connelletal. in the field, though both methods will be subject 1997;HughesandTanner2000;Milleretal.2000). to measurement error. Image quality and the two- Monitoring hard coral recruitment should there- dimensional planar view of digital images are two fore increase the ability to detect, understand, and factors that could potentially limit the utility of predict patterns of change on coral reefs. digital photography for reliably estimating density Estimating recruitment across large spatial and taxonomic composition. The aim of this study scales from the densities of juvenile hard corals was to investigate whether images taken with stan- should be a high priority in monitoring stud- dard, commercially available digital still cameras ies (Pearson 1981;Wells1995). However, tradi- provided estimates of density and taxonomic com- tional methods for measuring juvenile densities position of juvenile corals that were comparable in the field are time consuming, which makes to estimates obtained from visual census in the them impractical in monitoring programs de- field. signed to assess status and trends over large spatial and temporal scales (Edmunds 2000;Ruiz- Zárate and Arias-González 2004;Glassomand Methods Chadwick 2006). Coral reef researchers and monitoring programs require a standardised, rapid as- Juvenile densities of hard corals were esti- sessment method in the field, where any potential mated by counting the number of colonies in loss in precision is outweighed by the benefits of a34 34 cm (0.12 m2) quadrat with a scale × sampling a greater area. Many reef monitoring bar attached. Coral colonies less than 50 mm programs already use video and still photography diameter are generally categorized as juveniles to survey benthic adult populations and commu- (Chiappone and Sullivan 1996;Edmundsetal. nities on coral reefs (Done 1981; Abdo et al. 1998;Milleretal.2000)butsizedefinitionsvary 2003; Lirman et al. 2007)inordertomakerapid among studies, geographic locations, and coral surveys over large scales. The potential exists to taxa (e.g., Miller et al. 2000;Ruiz-Zárateand collect recruitment data in such surveys using Arias-González 2004). Two size classes of juve- similar photographic techniques, but this remains niles (<20 and 20–50 mm) were distinguished for untested. Modern digital cameras are cheap and this reason, which also meant that the effects of can produce high-quality images that can resolve colony size on detectability could also be assessed. objects the size of juvenile corals. Images can Small coral fragments <50 mm resulting from also be checked immediately to see that they are partial mortality of larger colonies were excluded. satisfactory (Edmunds et al. 1998). The poten- All juveniles within the quadrat were counted and tial advantages of digital photographic techniques identified to the highest possible taxonomic level, over visual census for surveying juvenile corals in which was usually genus. The genera Favia and the field include (1) reduced field time, (2) images Favites were grouped as “Faviids” for analyses can be collected by divers with no taxonomic ex- because juveniles of these genera could not be pertise, (3) images can be re-sampled by multiple distinguished reliably. Juveniles that could not be Environ Monit Assess

classiﬁed into a genus or family with conﬁdence nile were called “unknown.” Data were collected in back-reef habitats at four reefs representing outer-shelf (Cod Hole, ,7495%CI) Turner Reef), mid-shelf (Lizard Island), and in- shore habitats (Keppel Island). Between 7 and 33 quadrats were sampled at each reef (Table 1). Quadrats were placed haphazardly on the substrate and then searched thoroughly for juvenile corals. Photographs of the quadrat were then

taken using an 8.0 megapixel Canon Powershot (unknown genera) S80, except at the Cod Hole site, where a 5.0 megapixel Canon Powershot A95 had to be used. In both cases, the maximum number of pixels (“Large” image setting) and medium JPEG com- pression (“Fine” setting) was used with no ﬂash. These settings were chosen to maximize image quality and number of images that could be stored on the memory card. Programmed aper- ture, shutter speed, and ISO settings were used for model details) with the manufacturer’s “underwater” white bal- 2 20 mm) Juveniles (20–50 mm) Juveniles No. of genera

ance, which compensates for the attenuation of > red wavelengths with water depth. The estimate of juvenile density obtained from each quadrat by a single observer in the ﬁeld was compared with the average estimate obtained by careful scrutiny of the image of that quadrat by two observers in the laboratory. Two observers separately scrutinized each image to give independent estimates of the number of juveniles in order to assess observer bias. No observer surveyed the same quadrat with both methods. All observers were cross-calibrated to each other prior to data collection using images from a pilot study. To assess whether substrate 67)

complexity accounted for any difference between =

the two methods, the substrate in each image was ! ( rated from 1 (flat) to 10 (complex with many holes and crevices). Estimates of juvenile corals from different observers or methods (photo vs field), were compared using generalized linear mixed models using a Poisson error distribution and a log link function. These models deal with the properties of count data well, namely that variances tend to increase with the mean, the response variable is bounded (>0), and errors are not normally dis- Characteristics and sampling of the four survey reefs and the ratio (95% confidence intervals in brackets) of the difference between estimates of juve tributed (Venables and Ripley 2002). “Quadrat” was modeled as a random effect to account for ReefCod Hole Shelf habitatKeppel Is. Outer DepthLizard (m) Is. InnerTurner No. Rf of Mid quadrats Outer Total 6–9 juvenilesParameters were 5–6 estimated from best Juveniles fits ( of generalised linear mixed effects 6–9 models 6–9 (see Table 7 33 10 17 0.49 (0.33, 0.74) 0.65 (0.49, 0.87) 0.31 (0.17, 0.57) 0.74 (0.49, 0.60 1.11) 1.00 (0.44, (0.77, 0.82) 1.32) 0.69 1.09 (0.45, (0.75, 1.05) 1.60) 1.28 (1.04, 1.58) 1.28 (1.04, 1.58) 1.28 1.28 (1.04, (1.04, 1.58) 1.58) 9.50 (2.15, 42.02) 1.11 (0.44, 2.79) 0.32 (1.17, 6.67 0.62) 2.44 (1.93, (1.11, 23) 5.39) 0.90 (0.63, 1.31) 0.85 (0.57, 1.27) 0.55 (0.34, 0.87) the grouping of observations within the same Table 1 density from images and visualof census that in estimated the from field visual at census) each reef (e.g., the average density of juveniles estimated from photographs at Cod Hole was 49% (33 Environ Monit Assess quadrat. “Method” and “Reef” were modeled as fixed effects. The minimal adequate model was 150 derived by stepwise deletion of terms and model selection was based on AIC (lower values indicate 100

a “better” model ﬁt to the data). To assess the photographic in )

images 2 signiﬁcance of terms in the ﬁnal model, the de- 50 pleted model was compared to the full model by calculating the deviance and comparing it to the

Density (m Density 0 value obtained from a chi-squared distribution. The p values presented here should be treated 0 50 100 150 Density (m2) in field surveys with caution because they can be unreliable with small sample sizes (Bolker et al. 2008). Point and Fig. 1 Relationship between the densities (per square interval estimates of model parameters are more meter) of juveniles estimated from photographic images informative and were extracted from the final and by visual census of the same quadrat in the field at Cod Hole (circle), Lizard Is (triangle), Keppel Is (plus symbol), model according to Pinheiro and Bates (2000). Be- and Turner Cay (ex symbol). Dashed line represents the havior of the variance and residuals was assessed theoretical concordance between methods (i.e., a slope of by examining plots of residuals vs fitted values and 1) normal quantile–quantile plots, respectively. Lin- ear regression was used to model the relationship between substrate complexity and the difference photographic images than that counted in field between methods. All analyses were performed in surveys of the same plot (Figs. 1 and 2). The mag- R2.4.1(RDevelopmentCoreTeam2006) nitude of the difference between methods tended to be greater at reefs where the average density of juvenile corals was high, though estimates from Results more reefs would be required to assess this for- mally (Fig. 2). Differences between observers Proportional differences (density in photographic images/density in field census) in juvenile There were no differences between observers in density varied across reefs for total juveniles and estimates of total juvenile density (deviance < <20 mm juveniles but not for 20–50 mm juve- 0.001, df 1, p 0.99), density of juveniles niles (Tables 1 and 2). At the Cod Hole, esti- = = <20 mm (deviance < 0.001, df 1, p 0.30), mates of the density of all juveniles and those = = density of juveniles 20–50 mm (deviance < 0.001, <20 mm were only 49% (33, 74 95%CI) and 31% df 1, p 0.32), or the number of genera = = identified (deviance < 0.001, df 1, p 0.38) = = in photographic images. Estimates from photos were averaged across observers for all subsequent analyses.

Differences between methods

A total of 421 juveniles were recorded in the field compared with 296 recorded in the digital photographic images of the same plots covering a total of 7.75 m2.Theaveragejuveniledensityestimated in the field (all sites and taxa pooled) was 54.36/m2 Fig. 2 Relationship between the mean ( SE) density of ( 4.84 SE), whereas the equivalent estimate from ± ± juvenile corals and the mean ( SE) difference between the images was 36.67/m2 ( 2.92 SE). Estimates ± ± methods at each reef. Positive differences indicated higher of total juvenile density were generally lower in estimates from visual census in the field Environ Monit Assess

Table 2 Results from model simpliﬁcation showing the main and interactive effects of method (ﬁeld or photo) on estimates of juvenile corals Response Term !AIC Test Total juveniles (<50 mm) Method * reef 5.6 Deviance 11.6, df 3, p 0.009 = = = Juveniles (<20 mm) Method * reef 7.9 Deviance 13.9, df 3, p 0.003 = = = Juveniles 20–50 mm Method 3.7 Deviance 5.7, df 1, p 0.017 = = = Juveniles of unknown genera Method * reef 3.9 Deviance 9.9, d.f 3, P 0.02 = = = No. of genera Method * reef 4.5 Deviance 10.5, df 3, p 0.015 = = = !AIC is the decline in AIC values from a full model compared to a model with the term removed. p values are obtained from a chi-squared distribution

(17, 57 95%CI), respectively of those estimated Cay, an outer-shelf site in clearer water, Table 1, from direct visual census in the field; while at Fig. 3). Turner Reef, these same estimates from images Substrate complexity did not explain any dif- were comparable to visual census in the field ferences between methods for estimates of the (Table 1). Proportional differences between meth- density of total juveniles (r2 < 0.001, p 0.77), = ods in estimates of 20–50 mm juveniles were <20 mm juveniles (r2 < 0.001, p 0.87), 20– = similar among reefs, but slightly higher in images, 50 mm juveniles (r2 < 0.001, p 0.82), or the = suggesting false identification (Tables 1 and 2). number of genera identified (r2 < 0.001, p 0.56). = Differences between methods in their estimates Although the rank order of the densities of of juveniles categorized as “unknown” also varied taxa was similar for the two methods, the pho- among reefs and was up to 9.5 (2.15, 42.02 95%CI) tographic method generally underestimated the times higher in estimates from images com- relative density of common genera, especially for pared to estimates from visual census in the field Acropora spp. and Porites spp. (Fig. 3). This was (Tables 1 and 2;Fig.3). Overall, photographic probably because more juveniles were allocated images only provided similar estimates to visual to the “unknown” group when using the photo- census in the field at one of four reefs (Turner graphic method, reflecting a reduced ability to identify juveniles to the level of genus in images compared with visual census in the field.

Discussion

We found that rapid assessment estimates from photographic images consistently underestimated the density and diversity of juvenile hard corals compared to that estimated from detailed searches of the same plots in the field. Propor- tional differences between methods also varied among reefs, meaning that spatial patterns of relative densities of juvenile coral would not be preserved in estimates from photographic images at the sampled reefs. This occurred across a range of reef types with different coral communities and Fig. 3 Comparison of the proportional density of the most environmental settings. At present, we do not rec- common genera of juvenile corals <50 mm in field samples ommend the application of digital photographic and in photographic images of the same plots. Taxa are techniques to the rapid assessment of spatial pat- plotted in rank order; black bars represent data from field surveys (421 juveniles in total); white bars represent data terns in the density (absolute or relative) or diver- from photographic images (296 juveniles in total) sity of juvenile corals. Environ Monit Assess

The results of such comparisons between pho- ancies between the two methods. However, the tographic and visual census methods depend on use of a flash was deliberately avoided in this image quality and resolution, which can be af- study, since the aim was to assess the suitability fected by camera performance and ambient con- of digital photographic images for rapid assess- ditions. Other things being equal, the extent to ment. This militated against flash photography which one can zoom in on a digital image and still for two main reasons: (1) use of flash would in- discern small corals is a function of the number of crease the minimum time between consecutive pixels per unit area. The quality of digital images photographs due to the flash’s recharge time and may also depend on the age of the camera as color (2) increased demand on battery power would re- processing algorithms, especially in relation to un- duce the number of images that could be recorded derwater environments, have been continuously on a dive. Furthermore, the addition of a flash improving. Reduced image quality and resolution would still not have avoided problems associ- may help explain differences at the Cod Hole site ated with a planar view of a three-dimensional where the most significant discrepancies between habitat. Indeed, a flash may create more prob- the two methods occurred. Images collected at the lems by casting shadows when substrate is topo- Cod Hole had a resolution of 2,592 1,944 pixels graphically complex and use of the built-in flash × and were taken with an older camera, while im- would amplify backscatter from particles in the ages taken at all other reefs that had a resolution water. of 3,264 2,448 pixels. Images taken with higher This study suggests that photographic images × resolution may have been better able to provide do not constitute a reliable rapid assessment sufficient detail. method for estimating the density and taxonomic Another important determinant of image qual- composition of juvenile corals in the range of ity is the size of the quadrat, which affects the habitat types sampled, regardless of the size de- minimum distance that the camera must be held finition of a juvenile coral. This concurs with from the substratum. The further the camera is Edmunds et al. (1998)whousedafilmcamera held from the substratum, the more water (and as- in the Caribbean where different taxa dominate sociated particles and turbidity) separates the lens juvenile communities (Miller et al. 2000). Dis- from the subject and the poorer the image qual- crepancies between the two methods in this study ity. External factors like water clarity and light most likely resulted from of a reduced ability are particularly relevant when comparing samples (1) of observers to detect juvenile corals in two- from sites at different distances from terrestrial dimensional images of complex habitats that can influences. The largest quadrat size possible was be more easily searched by a diver and (2) to iden- used (in an attempt to reduce variability caused tify juveniles beyond a limited taxonomic level by small scale variation in juvenile distribution) in images. Higher resolution images and a flash that would still allow the ability to zoom in and may provide better quality images that would al- discriminate juveniles. Whether that quadrat size low more juveniles to be detected and identified, (0.12 m2) was suitable to estimate densities with but the problems associated with the planar view the desired level of precision depends on how and the requirement for the method to be rapid widely dispersed juveniles were and how many in the field remain. Based on this study, we would quadrats needed to be sampled at each site. How- recommend that densities and composition of ju- ever, determining the correct quadrat size and venile corals be estimated from visual census in number was not the objective of this study. As the field. This has implications for environmental Edmunds et al. (1998) noted, the conflicting re- assessment studies and the design of monitoring quirements of resolution and sampling scale may programs such that the benefits of greater accu- preclude the use of images to estimate densities of racy of visual census in the field for surveying juvenile corals. juvenile corals should outweigh the additional Aphotographicflashmayhaveprovidedbetter costs of longer dive times and the requirement for quality images, especially in poor light conditions taxonomic expertise in the field that are associated and clear water, and might have reduced discrep- with this method. Environ Monit Assess

Acknowledgments This work forms part of the Long- Glassom, D., & Chadwick, N. E. (2006). Recruitment, Term Monitoring Program at the Australian Institute of growth, and mortality of juvenile corals at Eilat, Marine Science. Logistic support from the crew of the RV Northern Red Sea. Marine Ecology Progress Series, Lady Basten is acknowledged, particularly Indigo for help 318,111–122. making quadrats. We thank Ray Berkelmans, Julian Caley, Harrison, P. L., & Wallace, C. C. (1990). Repro- Tim Cooper, and Michelle Jonker for helpful comments duction, dispersal, and recruitment of scleractinian that improved earlier versions of the manuscript. corals. In Z. Dubinsky (Ed.), Ecosystems of the world, coral reefs (Vol. 25, pp. 130–207). New York: Elsevier. Hughes, T. P., & Tanner, J. E. (2000). Recruitment fail- References ure, life histories, and long-term decline of Caribbean corals. Ecology, 81,2250–2263. Abdo, D., Burgess, S., Coleman, G., & Osborne, K. Lirman, D., Gracias, N. R., Gintert, B. E., Gleason, (2003). Surveys of benthic communities using un- A. C. R., Reid, R. P., Negahdaripour, S., et al. (2007). derwater video. Long-term Monitoring of the Great Development and application of a video-mosaic sur- Barrier Reef Standard Operating Procedure No. 2 vey technology to document the status of coral com- (3rd revised ed.). Townsville: Australian Institute of munities. Environmental, Monitoring, and Assessment, Marine Science. 125,59–73 Bak, R. P. M., & Engel, M. S. (1979). Distribution, abun- Miller, M. W., Weil, E., & Szmant, A. M. (2000). Coral dance, and survival of juvenile hermatypic corals recruitment and juvenile mortality as structuring fac- (Scleractinia) and the importance of life history strate- tors for reef benthic communities in Biscayne National gies in the parent community. Marine Biology, 54, Park, USA. Coral Reefs, 19,115–123. 341–352. Nyström, M., Folke, C., & Moberg, F. (2000). Coral reef Bolker, B. M., Brooks, M. E., Clark, C. J., Geange, S. W., disturbance and resilience in a human-dominated en- Poulsen, J. R., Stevens, M. H. H., et al. (2008). Gener- vironment. Trends in Ecology and Evolution, 15(10), alized linear mixed models: A practical guide for ecol- 413–417. ogy and evolution. Trends in Ecology and Evolution, Pearson, R. G. (1981). Recovery and recolonisation of 24,127–135. coral reefs. Marine Ecology Progress Series, 4,105– Caley, M. J., Carr, M. H., Hixon, M. A., Hughes, T. P., 122. Jones, G. P., & Menge, B. A. (1996). Recruitment Pinheiro, J. C., & Bates, D. M. (2000). Mixed-effects models and the local dynamics of open marine populations. in S and S-plus.Berlin:Springer. Annual Review of Ecology and Systematics, 27,477– RDevelopmentCoreTeam(2006).R: A language 500. and environment for statistical computing.RFoun- Chiappone, M., & Sullivan, K. M. (1996). Distribution, dation for Statistical Computing, Vienna, Austria. abundance, and species composition of juvenile scle- ISBN 3-900051-07-0. Retrieved from http://www.R- ractinian corals in the Florida reef tract. Bulletin of project.org. Marine Science, 58,555–569. Ruiz-Zárate, M. A., & Arias-González, J. E. (2004). Spatial Connell, J. H., Hughes, T. P., & Wallace, C. C. (1997). A 30- study of juvenile corals in the Northern region of the year study of coral abundance, recruitment, and dis- Mesoamerican Barrier Reef System (MBRS). Coral turbance at several scales in space and time. Ecological Reefs, 23,584–594. Monographs, 67,461–488. Sweatman, H. P. A., Cheal, A. J., Coleman, G. J., Emslie, Done, T. (1981). Photogrammetry in coral ecology: A tech- M. J., Johns, K., Jonker, M., et al. (2008). Long-term nique for the study of change in coral communities. monitoring of the Great Barrier Reef. Status Report Proceedings of the 4th International Coral Reef Sym- Number 8.Townsville:AustralianInstituteofMarine posium, 2,315–320. Science. Edmunds, P. J. (2000). Patterns in the distribution of ju- Venables, W. N., & Ripley, B. D. (2002). Modern applied venile corals and coral reef community structure in statistics with S (4th ed., p. 495). Berlin: Springer. St John, US Virgin Islands. Marine Ecology Progress Wells, S. M. (1995). Science and management of coral Series, 202,113–124. reefs: Problems and prospects. Corals Reefs, 14,177– Edmunds, P. J., Aronson, P. J., Swanson, D. W., Levitan, 181. D. R., & Precht, W. F. (1998). Photographic versus Wilkinson, C. R. (Ed.) (2004). Status of coral reefs of the visual census techniques for the quantiﬁcation of ju- world: 2004. Townsville: Australian Institute of Ma- venile corals. Bulletin of Marine Science, 62,937–946. rine Science.