Integrating structure-from-motion photogrammetry with geospatial software as a novel technique for quantifying 3D ecological characteristics of coral reefs

Introduction

The calcium carbonate skeletons deposited by reef forming corals provide the structural framework for highly diverse and productive coral reef ecosystems throughout tropical and subtropical oceans. These coastal ecosystems are among the most ecologically and economically productive on the planet, providing goods and services to humans through fisheries, coastline protection, building materials, biochemical compounds, and tourism (Moberg & Folke, 1999; Hoegh-Guldberg et al., 2007). Coral reefs are extremely vulnerable to stress and many have declined in integrity in the face of the increasing intensity and frequency of disturbances associated with climate change and human use (Harvell et al., 1999; Hoegh-Guldberg et al., 2007). Against this backdrop, it is critical to accurately evaluate and better understand how the structural integrity and ecological processes of these ecosystems respond to environmental changes.

Reef structural framework and reef-building capacity of biogenic components are primary drivers of the coral reef ecosystem functionality (Done, 1997; Fisher et al., 2007; Alvarez-Filip et al., 2009; Zawada, Piniak & Hearn, 2010). These factors have a reciprocal relationship with one another: the high level of architectural complexity provides a diverse range of microhabitats that support high biodiversity, productivity, and resilience; in turn, biogenic calcification is responsible for the reef-building processes that create the complex physical structure (Risk, 1972; Luckhurst & Luckhurst, 1978; Crowder & Cooper, 1982; Idjadi et al., 2006; Wilson, Graham & Polunin, 2007; Goatley & Bellwood, 2011; Graham & Nash, 2013). Furthermore, the extraordinary reef building capacity of scleractinian corals, enabled by their symbiotic partner Symbiodinium, is greatly influenced by physical characteristics of coral colonies. This is because physical characteristics of coral colonies like overall shape and topographic complexity influence key metabolic processes such as photosynthesis, respiration, calcium carbonate deposition, and reproduction (Smith, 1978; Crossland, Hatcher & Smith, 1991; Fisher et al., 2007). An accurate evaluation of the physical characteristics of coral colonies and coral reefs, involving 3-dimentional (3D) metrics, is critical in advancing our understanding of functionality of not only coral biology but also habitat availability, biogenic flux, and productivity of the entire coral reef ecosystem (Goatley & Bellwood, 2011; Pittman & Brown, 2011; Graham & Nash, 2013).

Recent advances in the fields of computer vision and photogrammetry, as well as improvements in data processing power, now make generation of 3D models and orthophotos from 2D imagery cost- and time-effective. Historically, collecting 3D environmental data has been a challenging undertaking in terms of logistics and cost. Oceanographic remote sensing tools, such as light detection and ranging (LIDAR), have been mounted to vessels for long-rage 3D underwater imaging, and in recent years, researchers have developed new techniques for high-resolution 3D reconstruction and visualization of underwater habitats using robotic vehicles (Kocak & Caimi, 2005; Pizarro, Eustice & Singh, 2009; Johnson-Roberson et al., 2010). Most of these technologies have not been widely used for small-scale coral reef studies due to the complexity of data collection, analytical time required, and prohibitive cost (Abdo et al., 2006; Bruno et al., 2011). The more recently developed Structure-from-Motion (SfM) photogrammetric technology offers a simpler, faster, and more affordable alternative for high-resolution topographic reconstruction (Westoby et al., 2012; McCarthy & Benjamin, 2014).

The SfM techniques have improved the quality of 3D data than can be derived from overlapping imagery by incorporating advancements in soft-copy triangulation and image-based terrain extraction algorithms (Westoby et al., 2012). Furthermore, SfM can accurately reconstruct scene geometry using high-resolution overlapping imagery obtained with single lens reflex (SLR) cameras and consumer point-and-shoot cameras, rather then relying on stereoscopic cameras, thus enhancing the accessibility and accuracy of 3D photogrammetric modeling for an array of uses. SfM techniques utilize the basic principles of stereoscopic photogrammetry, which is the science of obtaining reliable information about physical objects, the environment and terrain through processes of recording, measuring, and interpreting photographs or other images. However, the fundamental advantage of SfM is that the geometry of the photographed scene, camera positions, and orientation are evaluated without the need for a priori specification of targets with known 3D positions (Snavely, Seitz & Szeliski, 2008; Westoby et al., 2012). Rather, SfM photogrammetry determines these parameters simultaneously with a highly redundant and iterative bundle adjustment procedure, which is based on a dataset of invariant features extracted from multiple overlapping images (Snavely, Seitz & Szeliski, 2008; Westoby et al., 2012). These features are tracked from image to image, enabling initial estimates of camera position and object coordinates which are then refined iteratively using non-linear least squares minimization (Westoby et al., 2012; Fonstad et al., 2013). This process produces a point cloud of identifiable features present in the input photographs. Once georeferenced, this point cloud can be used to generate an array of digital elevation metrics to quantify 3D characteristics (Fonstad et al., 2013). Automating the process from identification of control points to 3D reconstruction of scene geometry makes SfM substantially more practical and cost-effective than traditional photogrammetric methodologies. Multiple studies have validated the accuracy of SfM techniques for high-resolution 3D topographic reconstruction and analysis, and in some cases found SfM to be highly comparable to substantially more expensive LIDAR techniques (Harwin & Lucieer, 2012; Delparte et al., 2014; Javernick, Brasington & Caruso, 2014).

Underwater SfM 3D reconstruction techniques have primarily been utilized for seafloor habitat characterization, bathymetry mapping, marine environment inspections and archeological surveys (Singh et al., 2002; Pizarro, Eustice & Singh, 2009; Beall et al., 2010; McCarthy & Benjamin, 2014). However, the SfM technology can also be applied to obtain a range of metrics pertaining to coral colony volume, biomass, and live surface area, all of which directly govern biological functions and ecology of corals. SfM photogrammetry also produces digital elevation models (DEMs) that can be analyzed using topographic software tools. This process enables quantification of intricate structural metrics such as surface complexity (3D/2D surface area), slope, and curvature. These 3D structural metrics are important predictors of organismal abundance, biomass and diversity, and also affect benthic current velocities associated with the food particle supply for suspension feeding corals (Kostylev et al., 2005; Guinan et al., 2009; Pittman, Costa & Battista, 2009; Pittman & Brown, 2011). The ability to quantify these topographic features will greatly enhance both biological and ecological investigations of coral reef ecosystems.

This study details a suite of methods for quantifying 3D characteristics of a coral reef using SfM photogrammetry in conjunction with geospatial software tools. This novel approach enables the extraction and quantification of biologically and ecologically meaningful data from 3D reconstructions made with SfM techniques. We tested the described methods at a study site in French Frigate Shoals, the largest atoll in the Papahānaumokuākea Marine National Monument (Northwestern Hawaiian Islands). The objectives of this study were to: (1) create a 3D reconstruction of the benthic habitat; (2) quantify several metrics pertaining to ecologically meaningful 3D characteristics; and (3) determine if each benthic component in the surveyed habitat will exhibit significantly distinct 3D structural characteristics.

Methods

3D model generation

3D models were constructed with the acquired imagery using Agisoft PhotoScan modeling software. PhotoScan was chosen because of the affordable cost, practical user interface, capability to determine camera parameters intrinsically and perform calibration, local processing, extremely detailed high-resolution 3D models, and the multiple exportable file formats compatible with geospatial software. There are a number of SfM programs that range from free software utilizing cloud processing to expensive professional-grade software that require powerful computing capabilities. Useful software reviews are available in order to determine the best program for the objectives and budget associated with a given research project (Remondino et al., 2012). The Agisoft modeling software is capable of generating the 3D digital surface model in three primary stages, (1) photo alignment (bundle adjustment), (2) geometry building, and (3) texture building. Photoscan aligns the series of photos from an individual survey using algorithms to automatically detect invariant features (“keypoints”), which overlap in the images, to create a system of geometrical projective matrices and determine the position and orientation of each camera position (Verhoeven, 2012; Westoby et al., 2012). The software constructs the 3D geometry on the 2D image plane using the extrinsic parameters that are calculated during the photo alignment (i.e., camera position and feature points) in conjunction with the intrinsic parameters and focal length of the camera that are extracted from the metadata of each image (Stal et al., 2012). The photo alignment (bundle adjustment) creates a sparse 3D point cloud that results from the projection and intersection of pixel rays from the different positions and oriented images in a 3D space. The x, y, z values from the GCPs are used to create a local coordinate system for accurate scaling and orientation of the point cloud, and at least 3 GCPs are needed for accurate 3D reconstruction (Verhoeven, 2012). Markers were annotated onto each of the GCPs using the Photoscan software. The location of each marker was manually reviewed and corrected in all 82 photos containing the GCPs (out of 441 total photos). This step required approximately 1-hour, and is dependent on both the amount of GCPs used and the number of photos containing the GCPs. Manually calibrating the location of GCPs using the Photoscan software ensures accurate interior and exterior orientation of the resulting model. After manually inputting the local coordinates of the GCPs, the model geometry is corrected by using the Photoscan optimize alignment tool. After correcting the interior and exterior orientation, the matching features are used to complete the final phase of the geometry building and generate a dense point cloud, which in turn can be used to build a continuous surface or mesh. The resulting mesh can be triangulated and rendered with the original imagery in order to build textured 3D mesh and create the final digital surface model (Fig. 1).

Digital annotation of benthic features

The 2D orthophoto image file was exported from the Agisoft modeling software and uploaded into geospatial software for annotation and subsequent topographic analyses. The orthophoto provides a high-resolution photo-mosaic of the surveyed substrate, and is geometrically corrected such that each cell contains accurate 3D information and can be used for measurement of topographic parameters (Fig. 2A). The orthophoto was manually digitized in order to create unique polygon shapefiles for all individual living coral colonies (Acropora cytherea, Montipora capitata, M. patula, Pocillopora meandrina, Porites compressa, and P. lobata) and abiotic features (rock/rubble and sand) present in the surveyed benthic habitat. Benthic features were digitized using the ArcGIS (ArcMap 10.1; Environmental Systems Resource Institute, Redlands, California, USA) editor tools. A ‘feature class layer’ was created within ArcGIS, and each digitized polygon was assigned to a specific sub-type that represented each benthic feature in order to catalogue the benthic substrate. This post-processing procedure enabled spatial data to be quantified for each individual coral colony (2,290 total colonies) and abiotic feature (Fig. 3 and Table 1). Manually digitizing all features in the orthophoto required approximately 20-hours to complete. Annotating the orthophoto provided the means to compare topographic complexity data among all the living coral colonies and abiotic features that comprise the surveyed benthic habitat at French Frigate Shoals.

Table 1:

Values of structural metrics compared among the coral species and two substrate types present at the French Frigate Shoals study site.

	Structural metrics
Benthic component	Mean surface complexity (±S.E.)	Percent cover	Mean percent slope (±S.E.)	Combined curvature (±S.E.)	Profile curvature (±S.E.)	Planform curvature (±S.E.)
A. cytherea	4.59 ± 0.68	0.32%	3,614 ± 536	0.32 ± −0.09	−0.41 ± 0.06	−0.09 ± 0.05
M. capitata	1.71 ± .0.06	0.19%	323.3 ± 42	0.07 ± 0.02	−0.09 ± 0.01	−0.02 ± 0.01
M. patula	1.81 ± 0.13	0.11%	546 ± 159	0.16 ± 0.05	−0.16 ± 0.03	0.00 ± 0.02
P. compressa	3.02 ± 0.29	1.07%	1,299 ± 182	0.12 ± 0.05	−0.20 ± 0.03	−0.08 ± 0.02
P. lobata	2.12 ± 0.04	10.96%	733.7 ± 30.6	0.17 ± 0.01	−0.19 ± 0.01	−0.02 ± 0.00
P. meandrina	2.30 ± 0.41	0.03%	1,438 ± 744	0.42 ± 0.28	−0.49 ± 0.17	−0.08 ± 0.14
Rock/rubble	1.35 ± 0.05	74.75%	568 ± 131	−0.02 ± 0.02	−0.03 ± 0.01	−0.05 ± 0.01
Sand	1.29 ± 0.05	12.56%	275.5 ± 68	−0.02 ± 0.01	−0.05 ± 0.01	−0.06 ± 0.01

DOI: 10.7717/peerj.1077/table-1

Results

The ground sample distance (resolution/pixel) of the resulting 3D model was 0.00109-m/pix with an error of 0.501 pix, thus our designated DEM cell size of 1.5-cm was well within the range of accuracy of the model. 100% of the images were matched with a median of 11715.7 keypoint matches per image and a total of 965,640 tie-points.

The SfM process renders textured 3D models in multiple file formats that serve as powerful visualization tools for assessing the dynamics of structural complexity in coral reef environments. Symbology gradients were applied to the DEMs in order to visualize variability in structural topography of the surveyed habitat for both percent slope and depth (Figs. 2C and 2D). These 3D models serve as a useful tool for visually representing physical characteristics, such as depth and slope, of the benthic habitat.

There were statistically significantly differences in the mean values of percent slope and surface complexity among the benthic features (biotic—corals, abiotic—rock/rubble and sand) of the reef at French Frigate Shoals (MANOVA, F = 22.27, p < 0.001). Further univariate analysis showed that percent slope, calculated as the tangent of θ multiplied by 100, was significantly greater in the coral, A. cytherea, than all other corals and abiotic features (ANOVA, F = 12.39, p < 0.001, Fig. 3B). Surface complexity, calculated as the ratio of 3D surface area to 2D surface area, was significantly greater in the corals exhibiting plating and branching morphologies (A. cytherea, P. compressa, and P. meandrina) compared to the encrusting and mounding corals (M. capitata, M. patula, and P. lobata), and significantly lower in the abiotic substrate (ANOVA, F = 30.63, p < 0.001, Fig. 3C). The values of combined, profile, and planform curvature varied among the biotic and abiotic benthic features. A. cytherea and P. meandrina, coral species with branching and plating morphologies, exhibited distinctly high values of both combined and profile curvature (Fig. 3A and Table 1). All coral colonies showed higher values of combined curvature compared to the abiotic features. Values of percent cover showed that the benthic habitat was predominantly comprised of abiotic features, and P. lobata was the dominant coral species (Table 1). The average Rugosity value of the surveyed habitat was 1.52 ± 0.03 S.E. (Fig. 2E).

Discussion

This study utilized Structure for Motion (SfM) photogrammetry techniques in order to characterize the 3D topographic structure of a coral reef at French Frigate Shoals, an atoll located in the Papahānaumokuākea Marine National Monument. Development of cost- and time-effective methods for quantifying 3D characteristics of coral reefs is critically needed considering that for the past six decades ecologists have found multiple measures of complexity to be primary drivers of biodiversity in these ecosystems (Goreau, 1959; McCormick, 1994; Knudby & LeDrew, 2007; Walker, Jordan & Spieler, 2009; Dustan, Doherty & Pardede, 2013). Integration of the 3D products derived from SfM photogrammetry with geospatial software tools, as developed in this study, provides a novel method for accurately characterizing structural complexity of coral reef habitats for ecological applications. Because this process requires integration of a number of steps, multiple variables should be carefully considered when determining the feasibility and accuracy of this approach to suit the needs of each study. For example, camera and lens, calibration, software options, DEM scale, annotation, and environmental conditions are some of the many factors that can affect the resulting models and quantification of 3D metrics (Bianco et al., 2011; McCarthy & Benjamin, 2014). Our approach was optimized for working in a remote field location with the goal of conducting a rapid survey with minimal equipment that would facilitate 3D reconstruction to enhance on-going studies of coral health and community structure.

We quantified 3D metrics that are informative of topographic structural complexity of a coral reef. The 3D structure of an environment directly affects important ecological parameters such as habitat provisioning and light availability and can therefore strongly influence ecosystem function—the biological, geochemical, and physical processes and components occurring within an ecosystem that drive biomass production, energy flux and other ecological services (Crowder & Cooper, 1982; Alvarez-Filip et al., 2009; Graham & Nash, 2013). Despite the known importance of structural complexity for ecosystem function, the difficulty in collecting 3D data has meant that the majority of ecological studies have been limited to utilizing two-dimensional (2D) planar survey techniques. While reducing dimensionality in sampling is beneficial for the cost and time constraints associated with ecological surveys, it limits the quality of data and is incapable of detecting specific 3D structural characteristics that are drivers of habitat facilitation and key ecological services (Goatley & Bellwood, 2011). The 3D metrics such as slope, curvature, and surface complexity of corals not only affects the diversity and abundance of species hosted by the coral community, but also may influence hydrodynamics and the amount of light irradiance received by photosynthetic organisms (Kostylev et al., 2005; Guinan et al., 2009; Pittman, Costa & Battista, 2009; Pittman & Brown, 2011). Previous studies have found that the specific 3D metrics used in our study were important predictors of vegetation distribution on terrestrial landscapes and influenced the distribution, biomass and diversity of vertebrate and invertebrate species in marine systems (Friedlander & Parrish, 1998; Kostylev et al., 2005; Pittman & Brown, 2011). Our study obtained these 3D metrics from 2,290 individual living coral colonies and surrounding abiotic features. We found that plating and branching corals (Acropora cytherea, Porites compressa, and Pocillopra meandrina) exhibited the greatest levels of surface complexity, slope, and curvature (Fig. 3 and Table 1). Furthermore, all three of these more complex species yielded large values for profile and planform curvature, indicating the coral colony surfaces are primarily concave in the vertical plane and convex in the horizontal plane (Fig. 3A). Such combination of concave and convex structure likely plays a substantial role in creating more microhabitat space per unit area for organisms using corals as shelter, and therefore may support more diversity, species richness and abundance of organisms (Alvarez-Filip et al., 2009; Graham & Nash, 2013). Among the coral species found to be structurally complex, A. cytherea had the greatest values of slope (Fig. 3B), which is indicative of substantial transitions in terrain surface. The large slope values are derived from the distinct plating morphology of this species, and are associated with the overhanging of the plates. The data presented here suggests that other reefs located at French Frigate Shoals with coral communities dominated by A. cytherea, P. meandrina, and P. compressa provide the most complex habitat and potentially support higher levels of diversity and productivity. In summary, utilizing the 3D approach described here can help to identify key coral species that contribute to high levels of productivity and diversity in coral reef ecosystems. Furthermore, the 3D characterization can track the effects of environmental perturbations on this ecosystem. Corals exhibiting complex branching, plating and corymbose morphologies, such as Acropora, Pocillopora, and branching Porites are generally more susceptible to disturbance than slow-growing corals that exhibit massive morphologies (Gates & Ainsworth, 2011; Woesik et al., 2011). Quantifying the relationship between the 3D structural characteristics of corals and assemblages of epibenthic organisms can help to elucidate potential changes to ecological functionality associated with local and global stressors that cause mortality and loss of coral cover.

In addition to the values representing 3D structural complexity, traditional benthic parameters including relative cover of coral species and abiotic substrate types and rugosity were analyzed using the orthophoto and DEM with the ArcGIS topographic software. Analyzing relative cover provides a useful construct perspective from which to assess reef structure and complexity. While the corals, A. cytherea, P. compressa, and P. meandrina were the most structurally complex, they only occupied a total of 1.42% of the surveyed space. On the other hand, Porites lobata was the dominant coral, occupying 10.96% of the benthos, thus overall biotic reef structure is predominantly driven by P. lobata at this site. Abiotic substrate (rock, rubble, sand) occupied 87.31% of the benthos, therefore indicating the habitat at this particular site is primarily composed of abiotic features. Assessing the 3D structural dynamics of a benthic habitat in conjunction with percent cover is critical for characterizing the site in an ecological context. Determining how each feature is distinctly different in terms of habitat structure is equally as important as identifying dominant features, as the dominant features will likely have the largest impact on associated species assemblages. It is critical to determine the morpho-functional groups of corals in order to understand how shifts in coral dominance will induce changes in biodiversity and ecosystem services (Alvarez-Filip et al., 2011). Rugosity, measured as the ratio of a straight line transect to a flexible chain draped over the substrate, has been the universally accepted measure of habitat complexity for several decades (Risk, 1972; Friedlander & Parrish, 1998; Dustan, Doherty & Pardede, 2013). While this technique is simple and inexpensive, the chain can easily tangle as well as disrupt the benthic habitat, and this method only captures a crude measure of complexity at a small scale. The SfM technique described in this paper allows for a rugosity profile to be virtually measured along any segment of the surveyed area (Fig. 2E). Quantifying multiple permutations of rugosity provides a much more accurate value of this metric, and is far more efficient than a single-chain extrapolation.

The utility of the SfM photogrammetry and associated geospatial analysis of reef complexity described in this paper has wide applications in coral reef research and management. Future studies can utilize these metrics to decipher the specific elements of habitat complexity that drive biodiversity, thus enhancing the ability to model future changes in ecosystem functionality associated with changes or loss in coral structure. Long-term monitoring of coral reef habitats can also be enhanced with SfM photogrammetry techniques. This method not only provides visual tools for assessing changes over time, such as the orthophoto and photo-mosaic, but the point clouds can also be compared across temporal scales in order to precisely quantify changes occurring in the volumetric and structural properties of corals. This will be critical for accurately investigating parameters such as coral biomass and habitat availability for reef fishes and invertebrates, all of which plays a vital role in coral reef productivity (Luckhurst & Luckhurst, 1978; Abdo et al., 2006; Wilson, Graham & Polunin, 2007). Examining changes in coral biomass and structure is also essential for accurately determining how coral reefs will respond to exponentially increasing global stressors, such as climate change and ocean acidification (Goatley & Bellwood, 2011; Graham & Nash, 2013). It is our hope that this study will provide a methodological platform for future studies to accurately quantify the 3D properties of coral reefs to improve our collective understanding of these important ecosystems.

Supplemental Information