The implementation of numerous no-take marine reserves on the U.S. west coast has led to increased interest in further development of visual survey methods, particularly for demersal fishes inhabiting rocky reefs that are too deep for diver surveys. Deepwater rocky reefs are environments that are very challenging to sample because they often include many areas that cannot be effectively sampled with survey trawls (Zimmerman 2003; Williams et al. 2010). Visual surveys of deep-water rocky reefs on the U.S. west coast have typically been conducted with remotely operated or human-occupied vehicles (Stoner et al. 2008). Recently, video landers have been developed and tested as an alternative survey tool for these habitats (Hannah and Blume 2012; Easton 2013). A video lander is a simple rugged frame enclosing an underwater video system that can be dropped to the seafloor to record images at a fixed station for a fixed period. Landers are low-cost sampling devices designed to be used in very rugged rocky habitat and can be deployed from vessels equipped with only a hydraulic block (Hannah and Blume 2012). They have been shown to be useful for distinguishing differences in fish abundance and species assemblages and also for identifying species-habitat relationships (Hannah and Blume 2012; Easton 2013). The video landers used on the U.S west coast operate similarly to the remote underwater video stations (RUVs) that have been used successfully, either with or without bait as a fish attractant, to study reef fish populations in the southern hemisphere (Willis and Babcock 2000; Willis et al. 2000; Harvey et al. 2007). Previous video lander research that has been conducted in waters off of Oregon has not used bait as a fish attractant (Hannah and Blume 2012); however bait has been shown to increase visual sampling efficiency for some species in other waters (Harvey et al. 2007). Our primary objective was to determine how the use of bait would influence the species composition of fishes viewed with a video lander on a temperate rocky reef.

Video landers used on the U.S. west coast to date have typically utilized single-camera systems, requiring a pair of calibrated lasers to generate modest amounts of fish length data. Single-camera lander systems, even with paired lasers, cannot provide accurate estimates of the distance of counted fish from the camera, a potential measure of the area being sampled. Calibrated stereo video systems have been shown to be effective for estimating these parameters in other areas and applications (Harvey et al. 2001; Harvey et al. 2004; Williams et al. 2010; Langlois et al. 2012) but have not been evaluated on a video lander for sampling temperate deepwater rocky reef fishes. Therefore, our second objective was to evaluate the effectiveness of a stereo video lander for generating precise fish length and camera-to-fish distance estimates. In addition, stereo video capability facilitated objectives of evaluating the effect of bait on the size composition of demersal fishes viewed and the effect of bait on the distance at which fish were being counted and measured.

METHODS

Study area.—This study was conducted at Stonewall Bank, a large (approximately 5 km east-west and 42 km north-south) rocky reef complex located about 25 km off the central Oregon coast (Figure 1). Stonewall Bank was chosen as a study site based on available information from prior video lander surveys (Hannah and Blume 2012). Those surveys typically encountered acceptable water clarity and also provided data on habitat type at hundreds of locations, allowing some control for habitat type in selecting our sampling sites. The prior surveys also showed a wide variety of demersal fishes inhabiting this reef complex.

Study design.—We selected sampling sites for this study from the over 800 locations previously sampled with an unbaited video lander in 2009–2010 (Hannah and Blume 2012). We first systematically eliminated some of the possible sites to maintain a minimum distance of 400 m between all sites. This minimum distance was chosen to reduce the likelihood of viewing the same fish at two adjacent sites on the same day. Then we randomly selected 80 sampling sites each for baited and unbaited video lander deployments (total, 160 sites) focusing on sites with complex habitat types such as bedrock outcrops, large boulders, crevices or vertical walls that we considered more likely to hold abundances of demersal fish. The 160 sampling sites were then divided, based on proximity, into eight single-day surveys of 20 sites each, sites being evenly split between baited and unbaited deployments each day (Figure 1).

FIGURE 1.

Stereo video lander study area at Stonewall Bank, showing the eight areas allocated 20 sampling sites each, at 10 deployments with bait as a fish attractant and 10 without bait.

Stereo video lander system.—Our stereo video lander used a frame that was very similar to that described by Hannah and Blume (2012). It consisted of a tubular aluminum frame (32-mm outer diameter tubing) enclosing the video equipment, lights, and battery housing (Figure 2). The frame system similarly used a weighted, sacrificial base made of 13-mm diameter mild steel rod designed to break away if the lander became snagged in rocky habitat. The system also incorporated a series of “break away” attachment points designed to fail in a sequence to tilt and then rotate the lander as pulling force increased; this improved the probability of successful retrieval from rocky habitat.

Although our lander frame and bases were very similar to those described by Hannah and Blume (2012), our camera system was quite different. We used a pair of stereo-calibrated high-definition Canon Vixia HFS21 video cameras instead of a single standard-definition Deep Sea Power and Light (DSPL) Multi-Seacam 2060. The use of a calibrated stereo video system allowed us to estimate the total length (cm) of many of the fish viewed, as well as the distance of these fish from the camera system (cm). Each camera was equipped with a wide-angle adapter (DVP-WA50-58 Pro Digital 0.5X) to increase the field of view. The video cameras were set to “progressive scan” and 24 frames/s to provide crisp single-frame still images for use in estimating fish length and camera-fish distance. Each camera was enclosed in an underwater housing, equipped with a dome port (Figure 2). The two camera housings were separated by 40.3 cm (measured between the centers of the camera housings) and were aimed inwards at a 4° angle to increase the overlap of the two video fields.

FIGURE 2.

Stereo video lander used to examine deepwater rocky reef fishes off the Oregon coast, showing the camera housings (A), the lights (B), the battery housing (C), the sacrificial base (D), the bait bag support pole (E) and the lander frame (F).

Our system illuminated the seafloor with two DSPL Sealite Spheres (3,000 lm, 6,000 Kelvin) as opposed to the two DSPL Mini-Sealites (850 lm, 6500 Kelvin) used by Hannah and Blume (2012). The more powerful lights were chosen because of the decreased low-light sensitivity of the high-definition video cameras compared with the DSPL Multi-Seacam 2060.

Calibration of the stereo camera system was completed using the camera calibration toolbox for Matlab (version R2011a) available at  http://www.vision.caltech.edu/bouguetj/calib_doc/ (Bouget 2008). Following initial calibration with a printed checkerboard grid, 49 underwater length measurements of known-size objects were taken in a seawater tank to determine the inherent accuracy of the calibrated stereo video system. Under these ideal conditions of lighting and water clarity, but within a distance limit inside the seawater tank of about 1.5 m, 42 of 49 length measurements were within 1 cm of the known length, which varied from 30 to 44 cm. The estimates were also slightly positively biased, with a mean measurement anomaly of 0.21 cm (SE = 0.10). Prior to final calibration, the cameras, camera mounting plates, housing ends and housing end-caps were all etched with fine, permanent alignment marks to ensure that the system could be disassembled to change camera batteries and then reassembled in the exact same physical orientation. To check for errors in length estimates from removing the cameras from the housings, the measurement of known-size objects was repeated before and after disassembly and reassembly of both housings, and the length estimates yielded very similar accuracy. We also filmed a digital stopwatch with both cameras simultaneously and compared frame counts at identical times at the start and after 10 min of filming to check for between-camera differences in clock “drift” and found none.

Field methods.—Sampling was conducted between April and September, 2013. Sites for baited and unbaited lander deployments were sampled in an AABB pattern each day. To balance out any effects from time of day, four of the eight surveys were started with baited deployments and four were started with unbaited deployments. A nominal duration of 12 min on the seafloor was used for both baited and unbaited deployments. This duration was chosen to allow as much time as possible for fishes to respond to the bait while still completing 20 sites/d and thus complete a total of 160 sites within 8 d of vessel time. For baited deployments, an orange mesh bait bag enclosing 0.9 kg of chopped Pacific Sardine Sardinops sagax, chopped Pacific Herring Clupea harengus, or a mixture of these species was suspended in front of the video cameras at the end of a 152-cm aluminum pole (Figures 2, 3). For unbaited deployments, an identical empty bait bag, that had never been in contact with bait, was suspended in the same position. Although the mix of bait species varied some between survey days due to availability, the same mix of bait was used during each sampling day. Prior to each deployment, a running digital stopwatch showing hundredths of a second was held in full view of both cameras to allow later synchronization of the video feeds to the individual frame.

Video analysis.—The video from each deployment was imported into Adobe Premiere Pro CS5.5 so that it could be easily reviewed on a full size computer screen. Each deployment was scored for water clarity and the quality of the view of the surrounding habitat, using the criteria from Hannah and Blume (2012). Fish that could be positively identified to species or a species group were counted, based on a maximum count of that species or group in any single video frame (MaxN; Harvey et al. 2007). This approach resulted in conservative counts, but was necessary because it was often impossible to tell if a fish newly entering the frame was a different individual than one seen seconds or minutes before. The primary habitat (most common habitat in the view) and secondary habitat (second most common habitat type in the view) types were classified for each deployment, using the criteria from Hannah and Blume (2012) to verify that differences in habitat type were adequately controlled for between baited and unbaited deployments.

Each counted fish was next evaluated to determine if estimates of fish length (cm) and distance from the camera (cm) could be made. Estimates were made whenever a pair of synchronized frames could be identified in which the fish in view was known to be a unique individual, the nose and tip of the tail could be clearly seen on both video feeds and the fish's body was approximately straight, not bent. Length and distance were calculated from paired, synchronized video frames using a custom Matlab program (entitled “Stereomeasure” developed by the National Marine Fisheries Service; Williams et al. 2010). To evaluate measurement precision with the stereo video system, three separate measurements of each fish were taken by sequentially selecting the nose and tail of each fish in each paired image with a computer mouse, generating replicated estimates of length, as well as distance from the camera system to both the nose and tail of the fish. The distance measurements for the nose and tail were then averaged to generate a mean fish distance from the camera system. As an indicator of the precision of the length and distance measurements for each fish we used the range of the three replicate measurements.

FIGURE 3.

The view from one of paired video cameras in good water clarity and moderate view quality (partially obstructed) in a baited deployment at Stonewall Bank; see the orange bait bag hanging near the center of the view. Three Rosethorn Rockfish are on the right and a Yelloweye Rockfish is emerging from behind a rock on the left.

Statistical analysis.—Except for reporting simple totals, we restricted our analysis to the 13 most commonly encountered species in our study (sum of MaxN > 10). However, some analysis was focused solely on demersal fish species. Semipelagic rockfishes, such as Widow Rockfish Sebastes entomelas, Blue Rockfish S. mystinus, or Yellowtail Rockfish S. flavidus, can be abundant at Stonewall Bank; however, they are often distributed ≥10 m above the seafloor and therefore may not be effectively sampled for relative abundance or size composition with a video lander.

We evaluated how bait influenced the assemblage of species viewed using a one-way ANOSIM, as implemented in PAST 2.14 (Hammer et al. 2001). We used the nonparametric onetailed Wilcoxon test to examine whether the presence of bait increased the mean MaxN per station. To determine how bait influenced the time at which MaxN was measured, we compared the elapsed time (min) of baited versus unbaited deployments with the frame where MaxN was established, in this case using a two-tailed Wilcoxon test. To determine the effect of bait on fish length and distance, we first screened the mean length and distance estimates to eliminate ones that were not precisely estimated. We included fish mean length and distance estimates in which all three replicates differed by no more than 3 cm for length and 20 cm for distance. For all of the demersal fish that met the screening criteria and for the three most frequent species in this group, we used the nonparametric Wilcoxon test to compare their mean distance from the stereo video lander for baited versus unbaited deployments. For demersal fish species in which a total of at least 25 acceptably precise individual mean length and distance estimates were made, we compared mean length between baited and unbaited deployments using a single-factor ANOVA.

RESULTS

All 160 planned sites were sampled with the stereo video lander. To achieve an acceptable view of the surrounding habitat, six sites had to be resampled, primarily due to the lander coming to rest with the cameras pointing straight down or directly at a large rock at close range. Water clarity was typically rated as “good” (157 of 160 sites; Figure 3) or “moderate” (3 sites). The quality of the view of surrounding habitat was rated as “good” at 144 sites and “moderate” at 16 sites. Only two of the sacrificial bases were lost in the process of sampling all 160 sites. Due to a variety of technical problems with one or more of the video cameras, synchronized, stereo video from which fish length and distance could be estimated was obtained for only 140 of the 160 stations. Of the 20 stations for which synchronized stereo video was not obtained, 12 were unbaited deployments and 8 were baited deployments. The primary and secondary habitats encountered at the 160 stations were generally similar for baited and unbaited lander deployments (Figure 4). Baited deployments encountered slightly less flat bedrock, bedrock outcrop and large boulder, and slightly more small boulder and cobble, as primary habitat types. Baited deployments encountered slightly less flat bedrock, small boulder and large boulder, and slightly more bedrock outcrop and cobble, as secondary habitat types (Figure 4).

FIGURE 4.

Comparison of percent composition of primary and secondary habitat types at Stonewall Bank between baited and unbaited deployments, April through September 2013. Habitat characteristics were identified via analysis of stereo video lander recordings.

TABLE 1.

Summary of Stonewall Bank, Oregon, fish-count metrics observed with a stereo video lander with and without bait as a fish attractant, April through September 2013, and one-tailed P-values comparing mean abundance with and without bait (significant differences in bold italics).

In total (sum of maxN), 625 fish of 24 different species or groups were counted at the 160 sites, 415 in baited deployments and 210 in unbaited deployments. The most commonly encountered demersal fish was Canary Rockfish Sebastes pinniger (118 observed) followed by Rosethorn Rockfish S. helvomaculatus (77), and Northern Ronquil Ronquilus jordani (75). Other demersal fishes that were encountered included Yelloweye Rockfish S. ruberrimus (47), Kelp Greenling Hexagrammos decagrammus (30), Lingcod Ophiodon elongatus (20), Spotted Ratfish Hydrolagus colliei (18), Quillback Rockfish S. maliger (18), unidentified sculpins (14, Cottidae) and Silvergray Rockfish S. brevispinis (11). A variety of semipelagic rockfishes, including Blue Rockfish (66), Widow Rockfish (65), and Yellowtail Rockfish (49) were also frequently encountered and were very abundant at a few sites.

Effectiveness of Stereo Video

As derived from the 140 successful stereo video lander deployments, fish length and camera-fish distance estimates were generated for 206 fishes of the 13 most commonly encountered species (Table 1). The precision in fish length and distance estimates was influenced by the distance of fish from the camera system (Figure 5). Generally, whenever the mean distance of a fish was less than 200 cm, acceptably precise estimates of both length and distance were obtained. Between 200 and 500 cm from the camera system, the percentage of acceptably precise estimates of length and distance declined, and beyond 500 cm they were infrequently obtained. We eliminated estimates in which three replicate measurements varied by more than 3 cm for length or 20 cm for distance; this left a total of 151 fishes for which length and distance estimates were considered acceptable, or about 29% of those that we were able to identify and count (sum of MaxN = 521). Of the 55 pairs of length and distance estimates that were rejected, 51 (93%) were rejected based solely on the <3-cm length precision criteria. Baited lander deployments accounted for 71.5% of the acceptable estimate pairs. For demersal fishes, the stereo video system performed somewhat better. Acceptable length and distance estimates were obtained for 122 demersal fishes from 140 lander deployments, or 34.3% of the 356 that we were able to count. Baited lander deployments accounted for 71.3% of these measurements.

TABLE 2.

Mean elapsed time to MaxN for all demersal fish species counted at Stonewall Bank, Oregon, with a stereo video lander at sites with and without bait, April through September 2013, and two-tailed P-values comparing means with and without bait (significant differences in bold, italics).

FIGURE 5.

Relationship between the range of three replicate measurements of a fish's length (upper panel) and fish-to-camera distance (lower panel) as a function of the fish-to-camera distance as measured with a stereo video lander.

Bait Effects on Length and Distance Estimates

The mean fish-camera distance at which demersal fishes were measured was reduced by the presence of bait (P < 0.01). When bait was not present, demersal fishes were measured at a mean distance of 264 cm, but when bait was present, this distance was reduced to 200 cm, a difference of 64 cm. The combined effect of bait on demersal fish counts and the distance at which they were measured resulted in baited lander deployments being much more efficient for generating precise length and distance data (Figure 6). From the 68 unbaited deployments that resulted in synchronized stereo video, 35 acceptable measurements of demersal fish length and distance were obtained, or 0.51 combined measurements per deployment. With bait present, this rate increased to 1.21 measurements of length and distance per deployment, or 237% of the unbaited level. This effect was most pronounced for Yelloweye Rockfish, Silvergray Rockfish, Canary Rockfish and Lingcod. Oddly, more acceptably precise estimates of length and distance were obtained for Spotted Ratfish when bait was absent.

At least 25 acceptably precise measurements of length and distance were obtained for just three demersal species, Canary Rockfish (N = 32), Yelloweye Rockfish (27) and Rosethorn Rockfish (26). For these fishes, the presence of bait had no significant effect on mean fish length (P > 0.05, Figure 7), but did decrease mean distance (Table 3) for Canary Rockfish (P < 0.01) and Rosethorn Rockfish (P < 0.05), but not for Yelloweye Rockfish (P > 0.05). However, it should be noted that statistical power is very limited with these small sample sizes.

FIGURE 6.

The percentage of fish, by species or group, for which acceptable length and distance estimates were generated (number of acceptable estimate pairs per sum of maxN) with stereo video lander deployments at sites with and without bait.

DISCUSSION

This study shows that the use of bait as a fish attractant improves the efficiency of a stereo video lander system as a survey tool for demersal reef fishes in Oregon waters in several ways. Counts of many demersal fish species were increased, an important improvement for surveying populations of less abundant species such as Quillback Rockfish or overfished species like Yelloweye Rockfish. This finding is consistent with previous studies of the effect of bait to attract fishes to RUV stations (Harvey et al. 2007), which showed that some species were encountered in higher numbers when bait was present, while sampling of other species was still effective. In our study, bait was also shown to reduce the average distance at which fish length and distance measurements were taken, suggesting a general effect of bringing some species closer to the camera system, which should improve both the precision of length and distance measurements from paired images and the ability to identify species from video. Although these effects from bait are not surprising, it is interesting that they are not reflected in all of the demersal species. We saw no effect of bait on either Kelp Greenling or Canary Rockfish counts, even though they were two of the more common species encountered. Interestingly, bait did improve the ability to generate length and distance estimates for Canary Rockfish with the stereo video system, while not increasing mean counts.

FIGURE 7.

Estimated total length (mean of three measurements) of Canary Rockfish, Rosethorn Rockfish, and Yelloweye Rockfish measured with a stereo video lander in deployments with and without bait.

TABLE 3.

Mean fish-to-camera distance for Canary Rockfish, Rosethorn Rockfish, and Yelloweye Rockfish for which acceptably precise estimates of fish length and distance were obtained with a stereo video lander at sites with and without bait, April through September 2013, and two-tailed P-value comparing mean distance with and without bait (significant differences in bold italics).

Although bait improved the efficiency of our stereo video lander for many demersal species, adding bait as a fish attractant had the disadvantage of requiring longer seafloor deployment times. We used nominal 12-min deployments of the lander system to allow the bait plume to spread and attract fish. Because we restricted sampling to just daylight hours, this limited the number of sites that could be sampled to about 20/d. Much of the prior work with an unbaited lander system utilized 4–5 min deployments sampled only during daylight hours, allowing about 30–40 sites/d to be sampled (Hannah and Blume 2012; Easton 2013). This tradeoff between maximizing demersal fish counts and the number of sites sampled per day can perhaps be optimized with a set of two video lander systems, deployed in an alternating manner. If the second lander system can be deployed at a nearby site while the first system is recording video on the seafloor, the number of sites sampled with a nominal 12-min bottom time can probably be increased to about 30–35/d (daylight hours only).

Despite bait improving the ability of our stereo video system to generate fish length and fish-camera distance estimates, our system was relatively inefficient at generating these data, especially in comparison with capture-based survey methods such as hook-and-line sampling. Only about 1.2 acceptably precise combined measurements of length and distance were generated per deployment of the baited lander system, equivalent to about 24 combined estimates per day of sampling at 20 sites/d. Although the ability to catch fish with hook-and-line gear can be quite variable, it still seems likely that capture-based sampling would generate considerably more length data per day than our stereo video lander system. Although less efficient than capture-based methods, the stereo video lander system can generate some accurate length frequency data for areas in which capture-based survey methods cannot be used, such as some marine protected areas. Hook-and-line sampling in particular can also have size-related and species-related biases. These biases can make it difficult to accurately characterize species assemblages or generate unbiased length-frequency data. The data from our study suggest that these biases may be reduced some with a baited stereo video lander. For example, hook-and-line sampling at Stonewall Bank typically does not yield catches of Northern Ronquil or Spotted Ratfish, both of which were shown with the video lander to be fairly common. With the stereo video lander system we were able to generate 15 combined length and distance estimates for these two species.

It may be possible to improve the efficiency of the stereo video lander at generating precise fish length and distance estimates. The seawater tank we used for calibration limited the distance at which calibration images could be collected to about 1.5 m from the cameras, possibly limiting the precision of fish length or distance estimates made at greater distances. Calibrating the system in a larger seawater tank could possibly help address this. Also, the Stereomeasure estimation program we used could be modified to increase the size of the images viewed on a computer screen or a larger computer screen could be used, potentially improving the repeatability of length estimates through more precise identification of the location of the edges of each fish.

However, there are also some inherent limitations for length estimation with a stereo video lander. To obtain length and distance estimates, the nose and tail of a fish must be fully visible on both video feeds, and readily distinguishable from the background. For some smaller species or those that blend in well with the substrate, this is simply not possible at times. This problem limited our ability to generate length and distance estimates for Northern Ronquil in particular. In other work with stereo video RUVs, a larger separation between the two cameras (1.4–1.5 m) yielded better performance at generating precise length estimates (Harvey and Shortis 1996; Harvey et al. 2001). For our lander system however, the baseline separation of the cameras was limited by the size of the lander frame (Figure 2), which must be small enough to be readily handled on deck and deployed repeatedly over the side of a vessel (Hannah and Blume 2012). Lander systems are also purposely built for deployment into deep and very rugose rocky habitat. Retrieval of the lander system frequently resulted in dragging of the frame through rocky habitat, requiring that the camera housings be well inside the frame edges to be sufficiently protected, limiting the camera separation that can be used. These limitations, the modest efficiency of the stereo video system at generating precise length and distance estimates, and the increased cost and complexity of a stereo video lander, suggest that adding stereo capabilities to a video lander will probably be worthwhile only for surveys in which length and distance estimates are considered critical.