No Nets, No Shocks, No Problem: Assessing Replicability and Disturbance Effects in Fish Monitoring Using Remote Video Cameras in Low Order Streams

Abstract

Conventional fish population sampling methods such as electrofishing and netting, pose risks to fish and are often restricted to small, shallow headwater streams—especially where endangered species may be present. Additionally, non-capture surveying (e.g., snorkelling) can disturb fish and make observation more challenging. This study evaluates the effectiveness and reproducibility of remote underwater video (RUV) surveys in a shallow (<0.5 m deep), freshwater stream. Additionally, fish response to disturbances (e.g., shadows, noise, surface disruptions) were characterized. Fish abundance was estimated by maxN (maximum number of individuals observed in a single frame) and used multiple cameras placed in the same habitat (pool). Findings indicated a high consistency in maxN when fish numbers were low (<5 individuals), with increasing variability at higher numbers (>15 individuals). This suggests that single camera setups can reliably detect minimum abundance. Fish responses to four disturbances (e.g., shadows, noise, surface disruptions, mink) were noted throughout. Typically, these responses were short-lived, with fish returning to pre-disturbance maxN values within minutes, with the most significant response to researcher-induced disturbance occurring immediately after RUV deployment. Overall, RUVs proved effective for passive, non-capture fish monitoring in shallow, sensitive habitats, producing replicable data with minimal impact caused by researcher disturbance. This technique can be added to our toolboxes for studying small-bodied fishes in challenging environments.

1. Introduction

Evaluating fish populations in shallow channels is essential to monitoring ecosystem health and making data-informed conservation and management decisions [1,2,3]. Common surveying techniques for assessing fish assemblages and abundance in streams include electrofishing, netting gear, and snorkelling surveys [1,2,4,5,6]. However, some of these techniques come with limitations including channel accessibility (i.e., where snorkelling is limited), and can impose harm to fish and inflict additional stress, and/or lead to increased mortalities [7]. Therefore, passive, non-capture sampling techniques are useful, specifically when working with evasive (shy) and/or endangered species.

Remote underwater video (RUV) cameras are a common, non-capture surveying method within marine environments and are becoming more prevalent in freshwater ecosystems [5,8,9,10,11,12]. RUVs allow researchers to determine fish abundance, diversity, and size as well as observe fish behaviour and habitat preference [5,8,11,13]. For example, Pereira et al. (2016) [5] compared reef fish abundance and activity from underwater visual surveys (snorkelling) and RUVs, and found no difference in species abundance or richness, but marked behavioural differences. Fish were observed to shelter during snorkel surveys, but were found to be passively swimming when observed through RUV [5]. In another study, when compared to conventional netting techniques within a 50 m reach, RUVs were able to detect a larger diversity of fish, observe more rare and flighty species that are typically missed during electrofishing surveys and were more applicable in a range of water depths and habitat types [9,14]. Another advantage is that RUV cameras can be deployed through the ice in winter to evaluate overwinter behaviour for fish when other surveying techniques were not feasible [8,10]. The use of capture surveying methods can cause harm to endangered species such as redside dace (Clinostomus elongatus) and therefore, it is difficult to acquire permitting to study reaches where these species are expected [14]. RUV surveys allow researchers to monitor at-risk fish populations and/or locations where at-risk species are part of the community, with lower costs (e.g., ~$100) and with more certainty (e.g., footage can be archived and reviewed by others). In addition, physical surveys, especially when footage is archived, can provide more information than other passive techniques (e.g., environmental-DNA) [1,4].

Small-bodied fish population abundance and habitat preference are difficult to determine due to site access and potential fright biases from monitoring, resulting in over or under estimating fish abundance [1,5]. Fright bias is identified by fish fleeing, hiding, or increasing vigilance within the channel due to researcher disturbances (e.g., moving equipment in and out of the water, noise along the bank), this can lead to lower fish population estimates with active monitoring techniques [5,15,16]. In addition to fright responses imposed by researcher disturbances, studies have found fish are attracted to or avoidant of monitoring equipment. For example, while using laser beams to estimate spatial scale during video surveys on fish behaviour within coastal rivers, short-lived fish species were more likely to chase the laser beams, potentially mistaking them for food, and thus artificially increasing observed abundance in these species [16].

RUV use has grown in freshwater ecosystems for monitoring fish populations, studying at-risk species and determining fish behaviour and preferred habitat. Understanding reproducibility and potential bias will improve study veracity. Initially this study set out to evaluate fish populations during winter conditions, upon the conclusion of that work [17] we were curious as to whether the observations were reproducible, which led to this analysis and report. The study evaluates RUV surveys focusing on small-bodied fish to determine fish abundance by deploying multiple RUV cameras on various field dates and comparing abundance estimates (maxN). In addition, this study reported the impact of different researcher disturbances on fish abundances to understand their effect on fish presence and behaviour.

2. Methods

Digital action cameras were used to record underwater video footage within a small (<5 m wide), shallow (<0.5 m deep) pool in a headwater channel (watershed area is <3 km²) of Fourteen Mile Creek in Oakville, Ontario. Site visits occurred on a nearly bi-weekly basis between October 2020 and April 2021. Pools are common geomorphic units in headwater channels within this region [17]. The particular pool selected for this study has low-flow velocities (<0.1 m/s) and is easily accessible [17]. Low flow velocities were necessary to ensure visibility with minimal turbidity and to allow for equipment deployment using minimal infrastructure, limiting alternations to the water column. The RUV cameras used for this study included an Eken H8 (DS pool 1), a Dragon Touch 4K action camera (DS pool 2), and a GoPro Hero 3+ (DS pool 3) (Figure 1). The resolution for each RUV was set to 1080p at 60 frames per second to optimize video clarity and storage capacity [8,11]. DS pool 2 was positioned so that DS pool 3 was visible in the field of view (FOV) to observe fish interactions with the RUV cameras and verify fish abundance counts (Figure 1). RUV cameras were attached to a metal plate and gently lowered to the streambed to avoid disruption of the fine-grained sediments. The RUV cameras were placed flat along the bed and were left to record for 45 to 70 min. Data collected during field surveys occurring on days with clear skies and no rainfall in the 48 h prior were used in this study [17]. Allowing a 48 h buffer after rainfall ensured turbidity had moved through the system and water velocities were relatively close to base flow conditions given the watershed area (<3 km²). To confirm optimal visibility for RUV usage, turbidity was measured during each site visit using a YSI 9500 photometer (manufactured by YSI Inc. in Yellow Springs, OH, USA). This was performed to ensure good video clarity and quality (Figure 1d). Based on these criteria, eight field dates were utilized for analysis.

Recorded videos were analyzed using QuickTime Player 7 software on a MacIntosh computer. Footage was viewed at real-time speed, when no fish were present the footage was reviewed at 2× real-time speed [9]. The time of the first fish arriving, time until maximum fish abundance was observed, and the time of any disturbances were noted while reviewing video footage. Disturbances were classified after video review into one of four categories: researcher-induced shadows (e.g., caused by movement along the bank), research-induced noises (e.g., conversation), researcher-induced surface disruptions (e.g., deploying a camera) and naturally occurring disruptions caused by a mink moving through the camera FOV. Disturbance start-times were recorded immediately after the disturbance stopped (e.g., after a shadow left the FOV), and the timer was stopped when pre-disturbance fish activity resumed. Additionally, a record of each species in view and a complete list of species for each location and date was created. Fish species were determined using identification guides for the region: “The Royal Ontario Museum Field Guide to Freshwater Fishes of Ontario” [18] and “The Baitfish Primer” [19]. For each RUV camera deployment, maxN was determined and used to estimate the relative abundance of fish at each location. MaxN is the maximum number of individuals for each species present in a frame at one time [20,21]. This abundance estimate was used to prevent overestimating fish abundance within a video clip (e.g., double-counting fish).

3. Results and Discussion

3.1. Reproducibility of Estimating Fish Abundance Using RUV Surveys

Field observations occurred bi-weekly from October 2020 to April 2021, totaling 18 field visits and producing over 125 h of video footage. To evaluate RUV efficacy, video footage from days with high quality visibility and fish observations were selected. This resulted in video footage (19 h) on eight field days at three camera locations in a pool habitat in the small, headwater channel being selected for this study. Over the field dates selected, eight fish species were observed and identified: blacknose dace (Rhinichthys atratulus), bluntnose minnow (Pimephales notatus), brook stickleback (Culaea inconstans), common shiner (Luxilus cornutus), creek chub (Semotilus atromaculatus), fathead minnow (Pimephales promelas), longnose dace (Rhinichthys cataractae), and redside dace (Clinostomus elongatus). Blacknose dace and creek chub were the most abundant species observed, but all species listed above were observed on at least one of the eight days used in this study. There were limited discernable differences in behaviour between species, so to evaluate RUV survey reproducibility, maxN for all fish was used.

Fish abundance values for each camera deployed within the pool were compared to determine single camera reproducibility using maxN. Results showed when fish abundance within the channel was low, between 0 and 10 individuals, maxN varied little between RUVs (+/−3 individuals) (Figure 2, e.g., field visits on 22 December 2020, 15 and 17 February 2021). The data points from all three camera locations are very close together and overlapping in some instances (Figure 2). As fish abundance increased, a larger variation in maxN estimates between the cameras occurred. With high fish abundance, >15 individuals, maxN determinations between the cameras varied largely (+/−10) (Figure 2, e.g., field visits on 19 November 2020, 23 February, and 25 March 2021). When looking at Figure 2, the data points from the three camera locations on these field dates are dispersed from each other, indicating variation in fish abundance. Using multiple RUV cameras allowed for the verification of fish presence and provided a range of maxN values when fish abundance was high. A single RUV camera within a habitat is reasonable to identify fish presence and an estimate of the minimum fish abundance. Replicating the process over multiple field dates/field conditions will produce the best estimates of fish abundance, but a single site visit can produce a reasonable baseline abundance measure, even when the channel is covered with ice (Figure 2, e.g., 26 January 2021).

3.2. Researcher and Naturally Occurring Disturbances

Disturbances that fish may interpret as a threat (e.g., movement along the bank, within the water, noise, shadows) could lead to surveys over- or under-counting fish abundance at a particular RUV station. During field observations at the site, we observed fish responding to human-made shadows leading to the decision to evaluate researcher disturbance. Four disturbance types were classified including overhead shadows, sounds such as talking and auguring into the ice (January observations only), surface disruptions including deploying RUV cameras and monitoring equipment, and finally, animal presence. The disturbance impacts were examined by reviewing video footage when a researcher-induced disruption was observed in the FOV for a given camera. The time to fish returning after a disturbance as well as the time to fish returning to pre-disturbance abundances was noted. The time to fish returning was determined as the first fish observed after a disturbance occurred. Whereas the time to fish returning to pre-disturbance abundances was defined as when the number of fish present was relatively (+/−5%) similar to pre-disturbance abundances and previous fish activity had resumed (

Table 1. Researcher-induced and naturally occurring disturbances imposed on fish showing the average (and range) of time to first fish arrival after a disturbance and time to pre-disturbance abundances (mean (range) ± SD), the average and range of maxN pre- and post- disturbance (mean (range) ± SD), the Mann–Whitney U statistic, effect size (r), p-value, and total sample size. Note: a p-value for animal disruptions was not calculated as there were only four observations.

	Shadows	Sounds	Surface Disruptions	Animal Disruptions
Sample size	21	14	9	4
Time to arrival post-disturbance (seconds)	1.3 (0–8) ± 2.6	30 (17–52) ± 14.2	45 (7–79) ± 25.0	160 (120–200) ± 40.4
Time to pre-disturbance abundance (seconds)	5 (0–21) ± 3.2	60 (28–70) ± 13.8	180 (21–200) ± 37.4	215 (140–300) ± 80.4
MaxN pre-disturbance	13.9 (4–26) ± 8.4	19 (2–33) ± 10.1	22.4 (5–44) ± 10.7	7.3 (1–15) ± 7.1
MaxN post-disturbance	13.6 (4–25) ± 8.0	16 (5–24) ± 5.5	17 (7–28) ± 8.1	5.3 (1–13) ± 6.7
Mann–Whitney U	247.31	100.73	42.25	n/a
Effect Size (r)	0.104 (small)	0.127 (small)	0.159 (small)	n/a
p-value	0.50	0.298	0.189	n/a

). This was performed to determine the time length of different disturbances and evaluate the potential impact of disturbance. MaxN values before and after disturbances occurred were compared using a Mann–Whitney U test and effect size (r) to examine how these disruptions impacted fish abundance. The Mann–Whitney U test requires a minimum of five observations to run the test and therefore, was not conducted for the animal disruption as the sample size was too small. The fish were observed returning to the RUV field of view within seconds to minutes, depending on the disturbance type (Table 1). And values close to pre-disturbance maxN abundances were usually achieved (Table 1). Animal disruptions were observed on four occasions, as a mink was observed on video swimming through the channel and in front of the RUV cameras. Although this was only observed four times, this disturbance was included to evaluate how fish respond to a natural disruption versus a researcher-induced one.

Overhead disturbance such as researcher shadows moving across the water surface, was the most common disruption noted, had the quickest recovery time, and had the lowest impact on fish abundance. Shadows were created across the channel by standing within direct sunlight and slowly moving an arm back and forth above the water surface. Initially, this happened by accident, and then purposefully to note responses. Researcher-made shadows produced instantaneous movement by fish (Table 1). Once the shadow was removed, pre-disturbance maxN occurred on average within 5 s (Table 1). In addition, fish displacement from the observed location was minimal, as shown by the small effect size score and with fish abundance values remaining statistically similar to pre-disturbance (p = 0.50, sample size = 21). When there was a prolonged unmoving shadow, fish returned to pre-disturbance activities while the shadow was still present. Shadows over the channel caused by researchers could be perceived by fish to be similar to shadows caused by trees moving in the wind or clouds moving across the sky. Given these similarities, this might explain why the fish response was minimal, and future RUV survey studies should work to minimize shadows. However, if this type of disturbance did occur, the impacts appear to be minor.

Researcher generated sounds noted throughout the study included talking, walking along the banks (vibration), and surface ice auguring activities. Sound-generated disturbance resulted in an increased fright response time compared to overhead shadows (Table 1). These interferences resulted in a delayed fright response, with fish leaving the camera field of view 5 to 10 s after the noise initiated. These disruptions resulted in fish movement away from the sound, with fish returning on average within 30 s of the noise subsiding and returning to pre-disturbance abundance on average within 60 s (Table 1). Field dates with higher fish activity resulted in a reduced fright response, with fish either returning while the noise was still occurring or returning sooner, this resulted in the large standard deviation noted (Table 1). In addition, maxN were not statistically significantly impacted by sound disruptions with (p = 0.289, sample size = 14), and the effect size remained small. In future work, minimal researcher activity along the stream bank is recommended, but if it does occur, impacts on maxN are negligible.

Surface disruptions, such as deploying cameras and inserting monitoring equipment (e.g., reference rod or metre-stick for scale) into the water column resulted in the longest researcher-induced responses (Table 1). Fish were observed leaving the location when the water surface was broken by the equipment deployment. The time of first arrival of fish following surface disruptions was on average 45 s (Table 1). However, pre-disturbance abundances were not recorded until on average ~3 min after the cameras were deployed (Table 1). Between these time periods fish presence was sparse and fish exhibited fright responses by moving rapidly across the channel. Once fish acclimated to the presence of the RUV cameras within the channel, they did not show any signs of fright towards the cameras. Fish abundance after the fright response had subsided were similar to pre-disturbance abundances (p = 0.189, sample size = 9), and the effect size was small (Table 1). Future RUV survey studies should look to minimize surface disruptions, however if they do occur and are stopped, their impact is minimal. During future studies, to ensure surface disturbances do not impact fish abundance estimates, maxN estimations should occur after the anticipated 3 min response time. Furthermore, these results align with other studies that have reported the need for an acclimation period, and/or sufficient recording time when using RUV surveys [8,11].

A mink was filmed swimming through the channel and past the RUV field of view four times and this activity was classified as a naturally occurring disturbance. The mink in the channel produced the longest response time with an average of 160 s to fish returning and an average of 215 s to pre-disturbance abundances (Table 1). Although there were only four occurrences, maxN values were comparable before and after the animal disruption. It is also possible that while the mink no longer appeared on camera, it could still be within the channel, causing fish to find shelter. Even with these limited observations of naturally occurring disturbances, it seems this disturbance type had the largest impact on fish, although fish abundance returned within 5 min. While conducting RUV surveys, naturally occurring disturbances are difficult to avoid; by having the equipment in the water for a sufficient time-period, if a disturbance were to occur, it is possible for these disturbances to have minimal impact on the observations.

During fish observations, fright responses are inherent regardless of the monitoring technique utilized. However, active sampling methods, even non-capture ones (e.g., snorkelling) result in prolonged fright response as researchers continuously disrupt the environment throughout the survey [1,5,15,16,21]. This potentially leads to underestimating fish abundance and reduced species diversity as fish hide [15,16]. Determining fish response to the RUV survey technique is important for ensuring fish abundance estimates are reasonable and that the technique could be successful in small, shallow channels, inhabited by evasive fish species.

Previous studies have advised that a quiet period along the bank was needed (~20 min) prior to surveying to ensure minimum disturbance [8,11,21,22]. However, disturbances can occur at any time and potentially impact the survey. The results from this study, focused on small-bodied, freshwater fish suggest that if a disturbance were to occur, its impact is minimal, and at most may impact results for several minutes. Similarly, other studies have reported that smaller fish species with shorter life spans were less risk adverse [16]. With the short response time in fish returning to previous abundances after a disturbance, decreasing future camera deployment times to ~30 min is reasonable and requires little if any quiet period when studying small-bodied (often shorter life span) fish.

3.3. Camera-Induced Bias

There are biases associated with all sampling fish abundance methods, fish behave differently in the presence of electrofishing and netting equipment, as well as manual snorkelling surveys [1,15]. Human-made objects create a visual stimulus and may result in trade-offs that resemble those associated with predation risk [16]. The cameras used in the present study were encased in a waterproof housing that reflects light at certain angles. In addition, a red-light flashes on the camara indicating the camera is recording, which may induce an attractive or agitative response in fish [16]. Understanding fish attraction or avoidance to RUV cameras helps quantify sampling bias with this surveying method. Fish response to RUV presence was compared using a silver-coloured GoPro Hero 3+ (manufactured by GoPro Inc., San Mateo, CA, USA) and a black-coloured Dragon Touch 4K action camera (manufactured by Yoidesu, in China). The RUV cameras were angled so the other camera was visible in the field of view. While video analyses occurred, a record was made of fish showing either attraction to the RUV cameras or avoidance by remaining further from the camera lens or quickly darting by it. Fish were not observed to avoid the RUV cameras after the initial fright response discussed prior had subsided. Anecdotally, one to two creek chub (Semotilus atromaculatus) were observed on two field visits, including 17 February and 25 March, repeatedly bumping into the camera and opening and closing their mouths, towards the GoPro Hero 3+ lens (silver). Although these observations indicated that creek chub were attracted to or agitated by the GoPro Hero 3+, the maxN and abundance values compared to other RUVs on these field dates remained within the same range (Figure 2). The limited maxN variability across days when creek chub interacted with the RUV compared to days when they did not indicate that this behaviour, while interesting and worth noting, did not impact overall maxN consistency. This indicates that the RUV cameras themselves, created minimal disruption and thus maxN is not over- or under-counted and a sampling bias was not detected.

4. Conclusions

This study aimed to ensure reproducibility when RUVs are used to estimate fish abundance and to evaluate the impact of researcher disturbances on fish abundance and behaviour. Although increasing fish presence results in larger uncertainties regarding fish abundance, a single camera is sufficient for estimating a minimum maxN value within a habitat. Based on the results from this study, fright response in fish as a reaction to researcher-induced disturbances was minimal. In addition, fish attraction or agitation with the RUV cameras did not impact abundance estimates. This indicates that RUV surveys are an effective means at estimating fish abundance minimums. A single RUV over multiple field visits will give a better estimate to fish abundances but is still likely represent a minimum. Using a single camera to estimate fish abundance in future work will reduce resource needs (e.g., costs, time (video recording review)).