Quantifying impacts of stony coral tissue loss disease on corals in Southeast Florida through surveys and 3D photogrammetry

Since 2014, stony coral tissue loss disease (SCTLD) has contributed to substantial declines of reef-building corals in Florida. The emergence of this disease, which impacts over 20 scleractinian coral species, has generated a need for widespread reef monitoring and the implementation of novel survey and disease mitigation strategies. This study paired SCTLD prevalence assessments with colony-level monitoring to help improve understanding of disease dynamics on both individual coral colonies and at reef-wide scales. Benthic surveys were conducted throughout the northern Florida Reef Tract to monitor the presence/absence of disease, disease prevalence, and coral species affected by SCTLD. Observed SCTLD prevalence was lower in Jupiter and Palm Beach than in Lauderdale-by-the-Sea or St. Lucie Reef, but there were no significant changes in prevalence over time. To assess colony-level impacts of the disease, we optimized a low-cost, rapid 3D photogrammetry technique to fate-track infected Montastraea cavernosa coral colonies over four time points spanning nearly four months. Total colony area and healthy tissue area on fate-tracked colonies decreased significantly over time. However disease lesion area did not decrease over time and was not correlated with total colony area. Taken together these results suggest that targeted intervention efforts on larger colonies may maximize preservation of coral cover. Traditional coral surveys combined with 3D photogrammetry can provide greater insights into the spatiotemporal dynamics and impacts of coral diseases on individual colonies and coral communities than surveys or visual estimates of disease progression alone.


Introduction
Coral cover in the Tropical Western Atlantic (TWA) has declined over the last four decades [1,2], and coral diseases have been identified as one major driver of widespread coral decline throughout the region [3]. In the 1990s, white band disease dramatically reduced coral cover applications including habitat characterizations [32][33][34], surface and volume measurements [35,36], and growth measurements [37,38]. SfM photogrammetry was first implemented on coral colonies using 10-20 overlapping photographs to generate 3D models [39]. As technology has progressed and computational power has become increasingly affordable, SfM photogrammetry techniques for the purpose of assessing various coral metrics have been refined and improved upon [35,38]. Changes in image capture and model reconstruction have improved measurement accuracy and precision, especially in regards to more established surface area methods such as the foil-wrapping method [30,35,38,40,41]. As such, SfM photogrammetry has emerged as a powerful and commonly-used tool for coral reef researchers.
In this study, we adapted a SfM photogrammetry technique [41] to track disease progression in fate-tracked Montastraea cavernosa colonies in Southeast Florida to gain insight into the colony-level dynamics of SCTLD. The study was designed to provide insight into colony-and community-level dynamics of this poorly understood disease through a combination of roving diver surveys and colony fate-tracking using SfM photogrammetry. The ultimate goal of this work is to increase widespread application of this and similar techniques to improve the design, implementation, and success of coral disease intervention, mitigation, and management strategies.  Florida Fish and Wildlife Conservation Commission. The work conducted at JUP, PB, and LBTS was conducted under FWC SAL-17-1960-SCRP, issued by Florida Fish and Wildlife Conservation Commission. Following Hurricane Irma in September 2017, a rapid-response damage and disease survey effort was conducted throughout Southeast Florida [42]. Data from these initial surveys were used to select the sites for the present study based on occurrence of coral communities and prevalence of SCTLD (Fig 2).

Disease prevalence surveys
Roving diver disease surveys were conducted approximately monthly from November 2017 to June 2019. These surveys were designed to assess the greatest reef area possible, quantifying disease prevalence over an estimated range of 100-2000 m 2 per survey dependent on underwater visibility. SCUBA divers swam for 20 min and recorded the species and disease status of every living coral colony �10 cm in diameter, and SCTLD abundance and prevalence were calculated from raw counts data. To assess multivariate variation in disease prevalence among sites and survey times, permutational analysis of variance (PERMANOVA; 9,999 permutations) was conducted in the R package vegan [43,44], with Bonferroni-corrected pairwise comparisons using the package pairwiseAdonis [45]. Non-parametric tests were implemented for all analyses as datasets were non-normal and normal distributions could not be achieved through transformation.

Fate-tracking of SCTLD-affected M. cavernosa
Benthic survey data indicated that disease incidence was too low at sites in Jupiter and Palm Beach, and that coral abundance was too low at sites in St. Lucie Reef, to conduct statistically robust fate-tracking studies in these locations. Consequently, three fate-tracking sites were established in Lauderdale-by-the-Sea (T328, BC1, and FTL4). These three sites are~12 km from the nearby Hillsboro Inlet, less than 500 m from shore, and have been previously used for benthic and coral monitoring [46-48] within the NFRT (Fig 2). Montastraea cavernosa was selected for colony fate-tracking due to the abundance of infected colonies within the study sites. This coral species is considered intermediately susceptible to SCTLD, with onset of tissue loss occurring weeks to months later than what has been reported for highly susceptible species (e.g. Dendrogyra cylindrus, Meandrina meandrites, Colpophyllia natans). Lesions on infected M. cavernosa generally progress slower as compared to highly susceptible coral species, with total mortality occurring within months to years [16]. Montastraea cavernosa comprised 11.6% of the reported cases of SCTLD in the NFRT in late 2017 [42].
Twenty-four colonies of M. cavernosa affected with SCTLD were tagged with uniquely numbered cattle tags across the three sites on 24-August-2018 (T 1 ). Sites were revisited on 11-Sep-2018 (T 2 ), 8-Nov-2018 (T 3 ), and 17-Dec-2018 (T 4 ). Continuous video was taken for 3D model generation to quantify total colony surface area and disease lesion area for each time point.
Fate-tracked colonies were filmed using methods outlined in Young et al. [41], with the following modifications: Canon G16 cameras in Fantasea underwater housings were set on "Underwater mode," 1080p and 60 frames per second (fps), and exposure was adjusted as needed based on ambient light conditions. One-meter, L-shaped PVC scale bars marked at 10 cm increments were placed at opposing right angles to frame the designated colony. A SCUBA diver swam approximately 1 m above the highest point of each coral colony and recorded continuous video while swimming repeated passes in a lawnmower pattern with the camera pointed directly downward. The number of adjacent passes varied depending on colony size, allowing for 60-70% overlap of filmed surface area. The camera was rotated 90˚at the end of the first set of adjacent passes and another set of passes was completed perpendicular to the first set. The two complete sets of passes for a single colony required between 1-3 min of dive time, depending on colony size. Each sampling event produced an average of 14 GB of.mp4 video files, or approximately 425 MB of video file size per coral colony.
Video processing and 3D model generation protocols are described in full in our GitHub repository [49]. In summary, the free software package, FFmpeg (www.ffmpeg.org), was used to extract still frame images from videos of the fate-tracked colonies at a rate of 3 fps. Still images were then imported into Agisoft Metashape Standard Edition (Version 1.5.2, Agisoft LLC) software, which uses a proprietary algorithm that incorporates SfM and Brown's lens distortion model to generate 3D models from 2D images [31,50]. Model generation in Metashape was conducted according to the manufacturer's protocol in four general steps: 1) camera alignment, 2) dense point cloud generation, 3) mesh generation, and 4) texture overlay. Models were rendered on a 2018 Apple MacBook Pro with a 2.9 GHz processor, 16GB of RAM and a Radeon Pro Vega 16 4GB graphics card. A single model took approximately 40 min to render depending on the number of still images generated. Generated models were then exported as a.obj file and imported into the software Rhinoceros 3D (Robert McNeel & Associates) for analysis; the mean model file size was 64 MB.
Models were scaled using the PVC scale bars, then total colony surface area and disease lesion surface area were measured by hand-tracing polygons around coral tissue. SCTLD disease lesion area was defined as the stark white newly-dead coral tissue or skeleton with sloughing tissue that is characteristic of the disease margin. Both total colony area and disease lesion area were generated directly within Rhinoceros 3D, while healthy tissue area was calculated by subtracting disease lesion area from total colony area. Proportion of loss and rate of tissue loss per week were calculated for each pair of consecutive time points. To identify significant effects of time on healthy area, diseased area, total area, the proportion of tissue loss and the rate of tissue loss, Friedman's rank sum tests were conducted using the package PMCMRplus [51]. Pairwise comparisons were made with Nemenyi tests in the package PMCMR [52]. Additionally, Spearman's rank correlation analyses were used to correlate diseased area and total colony area and rate of tissue loss and total colony area.

3D model validation
Accuracy of model-generated surface area metrics was assessed by a validation experiment using a standardized square cupola prism with equal dimensions and surface features to represent a coral. The prism was constructed from ½" PVC using a top square of 25.4 x 25.4 cm, a bottom square of 47.6 x 47.6 cm, and height of 30.5 cm (S1 Fig). Angled sides were achieved using 45˚PVC tees to better represent a mounding coral colony. Additionally, polygons of known surface areas (square: 40.32 cm 2 ; rectangle: 12.90 cm 2 ; circle: 45.60 cm 2 ) were printed to scale on waterproof paper, and affixed to the top and sloped sides of the weighted prism.
The prism was filmed in a pool and at the three coral fate-tracking sites in Lauderdale-bythe-Sea (Fig 2) to quantify model error in pristine (pool) versus reef conditions (e.g. variable turbidity, currents, surge). Eight videos were recorded for the prism and boundary frames across the three reef sites (2-3 replicates per site), while six prism videos were captured in the pool. The prism was filmed at a depth of 2.8 m in the pool, 8.8 m at BC1, 5.2 m at T328, and 6.7 m at FTL4; filming distance remained the same for all videos. To assess accuracy of surface area measurements, 3D models were generated and analyzed using the same protocols described earlier, with polygons traced for each of the prism shapes based on the four corners for squares and rectangles, and the entire circumference for circles. Replicate surface area measurements of shapes were made for each model according to face of the prism (top and four sides), and number of shapes on each face (four for square and rectangle, one for circle), for a total n = 229 measurements for square, n = 232 for rectangle, and n = 59 for circle. Shape error was calculated as the absolute difference between measured shape surface areas and the corresponding template area. Shape error measurements were found to be highly right-skewed, therefore a square-root transformation was applied prior to statistical analyses. Multivariate homogeneity of variance was first tested using the betadisper function of the package vegan, and following identification of heterogeneous variance among sites due primarily to lower variance at site FTL4. A single-factor PERMANOVA was conducted across sites using the adonis function of the package vegan.
Additionally, a subset of four models from the colony fate-tracking dataset were extracted at three different frame rates (3-5 fps) across their respective time points to determine if frame rate affected model accuracy. A Kruskal-Wallis test was used to determine if model size differed across the different extraction frequencies.

Fate-tracking of SCTLD-affected M. cavernosa
Friedman's rank sum tests indicated a significant decrease in total colony area (cm 2 ) and healthy tissue area over time ( Table 2, Fig 4). Disease lesion area varied significantly through time, however pairwise comparisons revealed that one time point comparison (T 2 -T 4 ) was driving the variation ( Table 2). A Friedman's rank sum test revealed significant differences between rates of tissue loss through time (df 2,71 , Friedman χ 2 = 6.25, p = 0.044). Average tissue loss (mean ± SEM) between T 1 -T 2 was -98.2 ± 40.7 cm 2 wk -1 , -25.0 ± 7.7 cm 2 wk -1 between T 2 -T 3 , and -45.1 ± 19.6 cm 2 wk -1 between T 3 -T 4 (Fig 5A). Rate of tissue loss was not significantly different among sites between any time point (S1 Table). On average, the proportion of tissue lost was 37.1 ± 7.2% over the course of this study, with three colonies experiencing complete mortality. There was no correlation between total colony area and disease lesion area (Spearman's rank correlation r s = 0.15, p = 0.154, Fig 6), or between rates of tissue loss and total colony area (Spearman's rank correlation r s = -0.046, p = 0.704, Fig 5B).
To assess potential effects of different frame rates used during model generation on colonyscale surface area measurements, a single-factor Kruskal-Wallis test indicated that frame rate had no significant impact on measured colony surface area (Kruskal-Wallis, H = 0.12, p = 0.94). This suggests that higher frame rates can be used from the initial video to improve poorly constructed models without affecting downstream spatial analyses. To keep processing time as low as possible, however, stills were extracted at 3 fps unless otherwise necessary.

Disease prevalence surveys
SCTLD is a unique coral disease due to its broad geographic extent, the number of coral species affected, and rapid rates of disease progression and spread [16,17,21,53]. SCTLD has been reported across~400 km of the FRT, and now has been observed in at least 12 other territories throughout the TWA [20]. The geographic distribution of SCTLD continues to expand with  new observations throughout the Caribbean [19,20]. SCTLD has been reported as far west as Belize and as far southeast as Martinique which are 1300 and 2500 km respectively from the northern terminus of the NFRT at St. Lucie Reef, FL [20]. Other coral diseases, such as white syndrome, have been observed over broad spatial extents (~1500 km) in the Great Barrier Reef [54]. Likewise, black band disease has been observed on coral reefs across ocean basins [55]. The overall mean disease prevalence for the NFRT observed in this study was 6%, which is lower than recorded in post Hurricane Irma surveys in 2017 [42] and previous SCTLD surveys in SE Florida [53]. Sites in Jupiter and Palm Beach had relatively low disease prevalence (1.7% and 1.0%) consistent with background disease levels (1-3%) within the TWA [56]. The highest values observed in this study were between 21-43% at SLR, but no site reached the highest reported disease prevalence values of 60% observed near Miami in 2014 [17]. The lower prevalence values reported in this study may be due in part to differences in species composition between the NFRT and southern regions of the FRT. The most abundant and susceptible species such as M. meandrites, D, stokesii, and Pseudodiploria strigosa [17,53] were comparatively sparse within our NFRT surveys. However, previous SCTLD impacts before our monitoring sites were established may also have affected the relative abundance of susceptible species [53,57]. We observed an increase in disease prevalence during the spring of 2018 which was unexpected, as prevalence for other described coral diseases such as white syndrome, white band, black band, and white pox often increases during the summer months as water temperatures increase [58][59][60][61][62]. SCTLD prevalence does not appear to have a positive correlation with temperature [53,58] as has been observed for other coral diseases [63][64][65], but potential environmental cofactors that may drive SCTLD prevalence need to be examined further.

Fate-tracking of SCTLD-affected M. cavernosa
At the colony level, disease progression was highly variable across fate-tracked M. cavernosa colonies near Lauderdale-by-the-Sea. As expected, total colony area and healthy tissue area decreased significantly over time, demonstrating the impact SCTLD can have on living coral cover over just a few months. Over the course of this study (115 days), colonies lost on average 37.1 ± 7.24% of their tissue surface area. Using a similar 3D photogrammetry method, Meiling et al. [58] saw proportional losses of 2.0 ± 0.11% day -1 of six SCTLD-infected M. cavernosa colonies in the US Virgin Islands. Additionally, Aeby et al. [25] noted subacute tissue loss in M. cavernosa was 34 ± 8.7% over 1 year using a semi-quantitative method to estimate tissue.
Disease lesion area did not correlate with total colony area, suggesting that larger colonies do not exhibit a higher proportion of diseased tissue compared to smaller colonies. As SCTLD progresses on coral colonies, loss of healthy tissue appears to be a more important indicator of disease virulence, rather than increase in disease lesion area. Importantly, rates of tissue loss did not correlate with total colony area. Therefore, larger corals may have a more favorable time horizon for potential intervention actions that could prevent total colony mortality. Conversely, smaller infected colonies may succumb more quickly to SCTLD. Considering the level of effort required for SCTLD interventions [66], intervention efforts focused on larger colonies with sufficient area of uninfected tissue remaining may result in more successful disease mitigation, and therefore reduction of coral cover loss due to colony mortality.

3D model generation as a fate-tracking method
This study tested a relatively inexpensive and rapid 3D model generation methodology described in Young et al.
[41] as a means to track disease progression on a colony-level without substantial increases in expense, time, or computational requirements compared to observational and photographic techniques. Healthy and diseased surface area of 24 coral colonies were quantified and compared over time. Additionally, this 3D modeling technique was validated through the quantification of model error using a mock coral colony with standardized surface shapes. Shape error (i.e. variation between measured surface areas and template shapes) did not vary across different depths and turbidity conditions (S2 Fig). This The relative error of 6.13 ± 0.27% is likely an inflated estimate of model error, as the prism shapes used for accuracy assessments were printed on waterproof paper that could move with currents, rather than representing solid surfaces. There were some instances, however, where model generation was poor due to anomalies in the Metashape algorithm, inconsistent underwater filming, or most common, high turbidity, all of which may impact surface area measurements and therefore affect model error. Typically, these issues could be rectified by extracting still images from the video at an increased frame rate to ensure higher overlap among images, or by collecting replicate videos for each target. Notably, extracting at a higher fps improved poor models, but did not disproportionately affect surface area measurements from models that were constructed well using images extracted at 3 fps.
Due to Metashape's proprietary algorithm, further adjustments within the model generation process to rectify poor models are limited. Stable, high-resolution image collection is therefore integral to successful downstream model generation. Alternatives to the lawn-mower video path used in this study may be more effective but also more time-intensive, particularly for tall (>1 m), highly rugose coral colonies. Discrete, overlapping photographs could be taken instead of continuous video [37], or video could be taken while swimming a circular pattern around the coral at varying angles [67]. Critically, while disease monitoring via linear extension measurements may be able to determine potential differences among colonies or treatments over time [16,28,68], linear extension does not accurately quantify tissue loss and may underestimate the progression of disease lesions on coral colonies. Quantitative 3D approaches such as the method presented here will improve our understanding of the ecology and impacts of SCTLD and other diseases on coral reef ecosystems, and may guide colony selection in future disease intervention strategies.

Conclusions
SCTLD affects a broad range of host taxa over broad geographic ranges relative to other described coral diseases. Based on the results from this study, larger colonies should be prioritized for SCTLD mitigation measures (i.e. disease intervention approaches) to maximize effort and potential treatment success when logistics and resources may be limited. Since SCTLD is a progressive and necrotic infection, the area of tissue loss, or proportion of tissue loss, may represent more impactful metrics for quantifying the severity of infection as opposed to disease lesion area or percent affected tissue. Additional studies incorporating longer timescales and multiple species are desirable to confirm these patterns of disease progression across all susceptible species. Rapid, cost-effective, and accurate methods using 3D models for quantifying coral surface area are a valuable approach for colony fate-tracking if high-resolution imagery can be obtained. It is recommended that managers and intervention specialists-particularly those focusing on SCTLD-adopt photogrammetric methods to enhance colony tracking methods and facilitate comparability across future studies and intervention trials.