Effects of artificial lighting at night (ALAN) on predator density and salmonid predation

[A]Abstract Predation of juvenile salmonids within California’s Sacramento – San Joaquin Delta (the Delta) has been identified as a contributing factor to low survival during outmigration through the system. Artificial lighting at night (ALAN) may contribute to increased levels of salmonid predation by attracting predators and prey, increasing predator reaction distance, and foraging success. To assess ALAN effects on predator (piscivorous fishes) density and the relative predation risk of Chinook Salmon Oncorhynchus tshawytscha smolts in the Delta, we preformed field based experiments with introduced ALAN. We used ARIS (Adaptive Resolution Imaging Sonar) cameras to generate predator density estimates in light and dark treatments throughout nightly experiments at 30 minute intervals. We simultaneously deployed predation event recorders (PERs) to estimate the impact of ALAN intensity (lux) on relative predation risk of Chinook Salmon smolts. Early in the night (1-3 hours past sunset), predator density and relative predation risk of smolts was unrelated to ALAN. However, late in the night (3-5 hours past sunset) ALAN presence increased predator density and the relative predation risk of juvenile salmonids increased with increasing lux. Predation risk was also positively related to predator density, and increased late-night predator density under ALAN, coupled with late-night foraging benefits of ALAN, likely contributed to the lux risk relationship. The exact mechanism behind this discrepancy between early and late-night trends is unknown and could be a result of our experimental design or the predator community sampled here. However, if these temporal trends prove robust to future investigations, late-night lighting reduction campaigns during outmigration, could maximize the human benefits of ALAN, while minimizing the negative impacts on salmonids. Overall, our findings align

Management strategies that minimize out migrant mortality are needed to rebuild salmonid populations. Decreasing the interaction of juvenile salmonids and their predators may be a way to lower predation induced mortality. Removing anthropogenic contact and aggregation points between predators and prey is likely one way to decrease these interactions (Lehman et al. 2019). Major metropolitan areas located throughout the Delta (e.g., Sacramento, Stockton, and Antioch, CA), and elsewhere along rivers of the California Central Valley, likely produce an abundance of ALAN, and ALAN has been identified as a prevalent predator-prey contact point (Lehman et al. 2019). Therefore, we investigated whether ALAN affected piscivorous fish (predator) density and the relative predation risk of Chinook Salmon smolts in the Delta. This study was an important first step in determining if ALAN removal or intensity reduction may be a viable management strategy to help lower predation mortality in the Delta and other waterways. [A]Methods [C]Study System. -Prior to ocean entry, all out migrating Central Valley Chinook Salmon pass through the heavily modified Delta and San Francisco estuary (Nichols et al. 1986), typically in the spring (Williams 2006). During outmigration, nocturnal movements are preferred in all river reaches except the estuary ) and migration speed is slowest in the Delta . Salmon released in the Sacramento River take on average 12.4 days to reach the Delta and 13.2 days to migrate from the Delta to the ocean . Most Delta land is below water level, waterways are channelized and leveed, and riprap covers 73% of mainstream shorelines (Ingebritsen et al. 2000;Mount and Twiss 2005;Lehman et al. 2019). Invasive fishes and vegetation are prevalent in the Delta (Underwood et al. 2006;Brown

Accepted Article
This article is protected by copyright. All rights reserved et al. 2007) and may contribute to poor survival of out migrating salmonids (Grossman 2016;Michel et al. 2018;Zeug et al. 2020).
[C]Experimental design.-We conducted our study during spring 2019 (April 22, 2019-June 7, 2019 at five sites in the Western Sacramento-San Joaquin River Delta (Figure 1). For each of the first two study weeks, predation experiments occurred on four consecutive nights within a single site (L3, L4, Figure 1). We randomly chose the light location within experimental reaches (upstream or downstream) without replacement to ensure that both locations were equally represented each week. After the first two weeks, we found an insignificant effect of light location on relative predation risk with a Cox proportional hazards model (p = 0.69, Cox 1972). Therefore, we only visited new experimental sites on two consecutive nights and we randomly selected the single light location before experiments began. Experiments occurred at three more sites (L6, L7, L8, Figure 1) using this design; however, predation at these sites was almost nonexistent (mean 2% predation). Therefore, we revisited the second site (L4) over two more weeks, and employed the two night experiment once each week.
We delineated our experimental reaches within each site with floating lines, measuring 200m in length (along-shore), and 25-50m in width (perpendicular to shore). All reaches had riprapped shorelines, and we defined within reach treatments as the 100m upstream and downstream sections of the reach. We mounted the artificial light source on a 4m pole and placed it at the waterline halfway within the upstream (50m into reach) or downstream (150m into reach) portion of the reach, depending on the experimental Accepted Article treatment. The light source was two LED floodlights, which emitted 20,000 lux, and was oriented parallel with the waterline (Figure 2).
To assess the relationship of ALAN intensity with the relative predation risk of juvenile salmon, we used predation event recorders (PERs, Demetras et al. 2016). Each PER was an independent drifting GPS enabled platform baited with a tethered hatchery-origin live Chinook Salmon smolt (87mm ± 0.01 SE mean Total Length [TL]) at 1 m depth. To tether each smolt, we looped fluorocarbon fishing line through their mouth and operculum. We attached this tether to a magnet that initiated a timer when it was pulled by a predation event (event time). To determine the exact location and time of predation events, we subtracted event time from deployment end time. On average, we had 82 (± 3.6 SE) PER deployments each night that typically began one hour after sunset and continued for four hours. To ensure that PERs traversed both light and dark treatments, we deployed PERs at the reach end where tidal flow carried them through the entire reach and spread them along the width of this end. When PERS reached the opposite end of the 200 m reach, they were collected and returned to the starting point. Before redeployment, we ensured that each PER had an active tethered smolt in good condition and used smolts 2-3 times, when no predation occurred.
Predation event recorders do not provide an estimation of absolute predation rates as tethered prey are less able to evade capture. However, PERs provide a cost-effective method to investigate drivers of predation and predator response that would be difficult to detect otherwise (Demetras et al. 2016;Michel et al. 2020a;Michel et al. 2020b). Our PERs were slightly modified versions of those described in detail in Demetras et al. (2016). Specifically, we constructed PERs with 5.08 cm diameter clear PVC pipe with the majority of components (GPS, timer, reed switch) contained within the PVC housing and sealed with a rubber end cap. Given the PERs' ability to capture precise predation locations and times, we were able to associate ALAN intensity (lux) with each observed predation event.
To compare predator density among light and dark treatments, we deployed ARIS (Adaptive Resolution Imaging Sonar, Sound Metrics Corp.) cameras in experimental reaches at 50 and 150m ( Figure 2). We positioned ARIS cameras at approximately 2 m depth, with one ARIS located directly offshore of the light source (light treatment) and the other 100m distant (dark treatment). Cameras continuously recorded from the start of PER deployments to the end, with a viewing window of 2 to 10 m, a -1° pitch, and a lens heading perpendicular to the reach length. In addition to among treatment comparisons, ARIS cameras provided predator density data that we incorporated into our predation risk models.

Accepted Article
This article is protected by copyright. All rights reserved [C]Light surveys.-After all PERs were retrieved, we surveyed surface light intensity (lux) within experimental reaches from a motorized vessel using an ILT2400 optometer. We performed survey transects at the inside floating line, the middle of the reach, and the outside line. We measured light attenuation with depth directly parallel to the light source at the nearshore and offshore limits of the study site. Starting at the surface, we lowered the optometer at 0.5 meter intervals until the bottom or 4 meters was reached and held it for 1 minute at each unique depth to record a mean lux value. To account for variation with distance from the light source and inherent variability among nights, we standardized lux at depth by dividing the value at each depth by the surface value from each cast. This value was light attenuation (At) and we fit the following exponential decay equation where kd (attenuation with depth) and kt (attenuation with turbidity [turb]) were fit with both coefficients or only kd. The model of best fit had a corrected Akaike Information Criterion (AICc, -467.772) within 2 units of the model with the lowest AICc (-469.411) and included the fewest parameters, kd (1.279) and depth.
We interpolated lux across experimental reaches using the R function autoKrige (Hiemstra et al. 2008) and assigned each PER GPS position a lux value at the water surface and at depth. This function generated an exponential variogram (we fixed this model type within the function) from each survey and used weighted least squares to select the best fitting values of nugget, range, and sill. Using this model, we interpolated water surface lux over a 500,000 cell grid, which resulted in smooth, fine-scale lux values across experimental reaches for each night ( Figure 3). We assigned interpolated surface lux values to each PER GPS position and used the above attenuation model to predict lux at 1 m depth; the approximate depth where tethered smolts drifted. If any PERs drifted outside of the survey bounds, we assigned the lux value from the nearest grid cell.
Our lux meter malfunctioned during the second night of sampling at L7, so this night was removed from PER predation risk analysis.
[C]ARIS data reduction and processing.-To post-process ARIS footage, we used Echoview version 10.2. This software removed background data and excess noise, and identified all fish or fish-like objects (targets) ≥ 200 mm in the ARIS footage (Boswell et al. 2008). We then manually reviewed each fish-like object and removed all non-fish before analysis. Although both ARIS cameras were deployed with similar settings each night, frame rate differed between and within cameras, given inherent processing speed differences between computers throughout sampling nights. To account for differing frame rate and instances of Echoview assigning multiple unique identifiers to the same fish (double counting), we used fish density

Accepted Article
This article is protected by copyright. All rights reserved instead of fish counts for analysis. To calculate fish density, we exported total beam sampling volume and the total number of fish pings in 30 minute increments from each ARIS (light, dark ALAN treatments) on each experimental night. We then divided the number of fish pings by beam volume within the corresponding 30 minute time frame, to obtain fish per m 3 for each 30 minutes of sampling within each treatment on a given night. The number of fish pings and beam volume within a given time frame are inherently tied to frame rate.
For example, ARIS A has a frame rate of 4 frames/second and ARIS B has a frame rate of 8 frames/second and each ARIS samples 4 m 3 each frame. Assuming that one fish is continually present on both ARIS A and B for an entire 30 minute window, ARIS A would sample 28,800 m 3 and have 7,200 fish pings in 30 minutes, and ARIS B would sample 57,600 m 3 with 14,400 fish pings. The resulting 30 minute fish density for both hypothetical ARIS cameras would be 0.25 fish m -3 30min -1 . The ARIS in the light treatment malfunctioned during the sixth night of sampling at L4 and density data was not recorded for this night, therefore it was censored from all statistical analyses. Although we do not know large fish species, it is likely that these fishes are mainly piscivores (see Discussion), and fish density will be referred to hereafter as predator density.
[C]Statistical analysis.-To determine if ALAN altered predator density and if predator density was related to time of night, we analyzed ARIS data with a generalized linear mixed-effects model (GLMM) using the lme4 package in R (Bates et al. 2015; R Core Team 2019). We fit the GLMM using the Gamma family and a log link, given that the data distribution of 30 min predator density was non-normal and non-negative. To remove the few zeros in the dataset (n = 13, 5.75% of data) and allow the Gamma model to run, we added 1 x 10 -9 to all density measurements before GLM analysis. We included ALAN treatment, minutes past sunset (30 minute bin increments 1-8), and the interaction of ALAN and minutes past sunset as independent variables in the GLMM. To account for potential differences in baseline predator density among sampling nights and the fact that each ALAN treatment was resampled throughout each night, we included a random effect of sampling night in the GLMM. To investigate the interaction of minutes past sunset and ALAN on predator density, we split the data into early (1-3 hours past sunset) and late-night (3-5 hours past sunset) subsets and ran GLMMs without the interaction term.
To assess whether the relative predation risk of Chinook Salmon smolts was related to lux, time of night, and predator density, we evaluated Cox proportional hazards models with the R function coxph (Cox 1972;Therneau 2015). We included lux, minutes past sunset, mean 30 minute predator density among light and dark treatments throughout experimental nights, and the interaction of lux and minutes past sunset as independent variables. To investigate the interaction of minutes past sunset and lux, we split the data into Accepted Article early (1-3 hours past sunset) and late-night (3-5 hours past sunset) subsets and ran Cox models without the interaction term. We checked proportional hazards assumptions of Cox models using the cox.zph function in R (Therneau 2015). Although lux at depth was predicted from surface lux, we evaluated another set of Cox models using this variable instead of surface lux to demonstrate how predation risk responded to changes in lux at depth. To investigate the relative effect on predation risk of each variable, we also ran the Cox models using scaled (independent) variables. Neither of these approaches changed overall model fits or significance; however, they provided further insight that we might have missed in original model output. [A]Results We found that predator density was positively related to ALAN presence late in the night. When we analyzed all density data, the effects of minutes past sunset and ALAN presence on predator density were non-significant; however, a significant interaction between these two variables was present (Table 1). In the early-night model (1-3 hours past sunset), neither minutes past sunset or ALAN presence had a significant relationship with predator density. While in the late-night model (3-5 hours past sunset), predator density was predicted to increase by a factor of 3.96 in the presence of ALAN ( Figure 4A) and no relationship of minutes past sunset was detected (Table 1). This difference in the ALAN effect between early and late-night models demonstrates why the interaction of ALAN and minutes past sunset was significant in the overall model.
Relative predation risk of Chinook Salmon smolts increased with artificial illumination intensity (lux) late in the night and was positively related to predator density in all models (Figures 3, 4B, and 5). In the full night model, relative predation risk had a positive relationship with predator density, a negative relationship with minutes past sunset, and a positive relationship with the interaction of minutes past sunset and lux.
However, no overall relationship of relative predation risk with lux was detected (Table 2). Neither minutes past sunset or lux was significantly related to predation risk in the early-night model (1-3 hours past sunset), but the positive relationship with predator density remained. In the late-night model (3-5 hours past sunset), relative predation risk was predicted to increase by factors of 1.030 (scaled factor = 1.217) and 1.502 x 10 19 (scaled factor = 1.987) for each unit increase in lux and predator density, respectively, and no significant effect of minutes past sunset was detected (Table 2, Figure 5). For each unit increase of lux at 1m depth in the late-night model, the raw factor change was greater (1.111). However, this increase was because lux at depth only varied from 0 -20, while lux at the surface varied from 0 -72 (Table 2). Similar to the predator density Accepted Article models, the significant interaction of lux and minutes past sunset in the overall model was the result of a significant effect of lux late in the night with no effect early in the night on relative predation risk. [A]Discussion With continuing human development along rivers, estuaries and coastlines, it will be increasingly important to consider the impacts of artificial lighting at night (ALAN) on aquatic organisms and ecosystems (Davies et al. 2014;Jechow and Hölker 2019;Zapata et al. 2019). Elsewhere, ALAN has attracted young salmonids and their predators, resulting in elevated predation rates (Tabor et. al 2004;Tabor et. al 2017).
Similarly, introduced ALAN in the Delta increased predator density and relative predation risk of Chinook salmon smolts; however, these effects were only detected three hours past sunset and later. These results indicate that ALAN reduction may decrease predation rates and mortality of out migrating salmonids in the Delta and are an important first step in assessing ALAN impacts along outmigration routes in the California Central Valley and elsewhere.
As with any field experiment, there were notable environmental limitations and assumptions associated with our study. First and foremost, our metric of relative predation risk focused only on predators and does not take into account any prey effects, given that we used tethered smolts.

Accepted Article
This article is protected by copyright. All rights reserved 2020a). If ALAN sufficiently slows outmigration, increased water temperatures, low river flows, and prolonged exposure to predators along the outmigration corridor could also reduce salmonid survival (Henderson et al. 2019;Michel et al. 2020a). Although our study found a response of predators to ALAN, future investigations focused on prey responses will help elucidate if ALAN reduction is a viable restoration strategy.
While we were not able to identify the large fishes used for predator density calculations to species, it is likely that many were piscivores, given the positive relationship between density and relative predation risk. Furthermore, we conducted opportunistic hook and line sampling on a few occasions prior to nightly experiments, and both Largemouth Bass Micropterus salmoides and Striped Bass Morone saxatilis were caught and released at site L4. These species are common predators of salmonid smolts in the Delta (Sabal et al. 2016;Michel et al. 2018) and were likely a major component of the predator community in this study.
Because this study was conducted at night, we could not use traditional video recordings to identify PER predators. Coupling underwater infrared light sources with traditional video (Mazur and Beauchamp 2003) may aid in nighttime predator community identification without introducing additional light in the visible spectrum (Jordan and Howe 2007;Horodysky et al. 2010;Mitchem et al. 2018). However, sampling area of traditional video would likely be less than our ARIS cameras and would be problematic in turbid waters.
Combining ARIS with this infrared assisted video analysis could be advantageous for low light predator density and community analysis. Our predator density metric also did not differentiate between individual fish remaining under the light source or new individuals being attracted to the light, distinctions that are potentially important, but are pooled in our analysis. Future work should investigate differences in fish behavior in dark and illuminated treatments to determine how ALAN may change the behavioral response of fishes.
Predator density was greater in ALAN treatments and relative predation risk of Chinook Salmon smolts increased with increasing ALAN intensity (lux) three hours past sunset and later. However, no ALAN impacts on density or predation risk were detected early in the night. These discrepancies may be driven by ambient light levels and the relationship of predation success with light. Foraging efficiency and predation success of piscivorous fishes increases with increasing light levels, but fish can also successfully forage in low light (McMahon and Holanov 1995;Mazur and Beauchamp 2003;Mazur and Beauchamp 2006). Mazur and Beauchamp (2003) suggested that salmonids have a foraging threshold around 1 lux, with success continually increasing until it asymptotes around 20 lux. However, for Largemouth Bass this threshold could be as low as ambient moon light (0.003 lux, McMahon and Holanov 1995). Early in experimental nights, ambient light Accepted Article levels may have been sufficient for successful foraging and depredation throughout experimental reaches, resulting in similar predator densities and predation risk throughout the reach. After three hours past sunset, an ambient light foraging threshold may have been reached where predation success was diminished in unilluminated reach sections. The additional light provided by ALAN likely allowed for continued unimpeded predation success in illuminated portions of the reach. This increased success likely led to greater predator density in ALAN treatments and increased predation risk with increasing lux late in the night. Therefore, the overall decrease in predation risk with elapsed minutes past sunset may be explained by decreasing successful foraging area with time.
The late-night relationship of elevated predation risk with increasing lux was also likely driven by increased predator density in ALAN treatments, given that predation risk was positively correlated with predator density in all models. The increased foraging success of predators in ALAN would have also contributed to the elevated predation risk with lux. However, decoupling these two drivers was not possible in our experiment. To elucidate foraging thresholds and quantify the effects of ambient illumination on predation risk, future studies should deploy a light meter throughout the night to monitor background illumination.
The lack of ALAN effects on predation early in the night could also be a result of the predator community we sampled and our experimental lights. As mentioned above, predation risk in our study was higher earlier in the night, but ALAN did not affect density or predation risk during this time. Assuming that the majority of our predators were roving channel associated fishes (e.g. Striped Bass), it may have taken time for them to cue into the light, which we established each night. If our light was a permanent fixture, perhaps fishes and other piscivores would have established nocturnal feeding habits under this fixture early in the night. For example, harbor seals that fed on out migrating salmonids under an illuminated bridge, were most abundant 1 -1.5 h after nightfall and decreased as the night progressed (Yurk and Trites 2000). It is also possible that our experimental ALAN attracted prey fishes, which in turn attracted predators (Becker et al.

Accepted Article
This article is protected by copyright. All rights reserved Another pertinent direction for future research that would benefit ALAN management is to determine a minimum lux value that does not impact fishes. Our experiment did not empirically test variable lux intensity; however, we did sample a large range of lux values that may provide insight into predator response.
If we fit a penalized spline to the lux predictor in our late-night Cox model, predation risk does not increase until after 8 lux is reached. A similar trend can be seen in the top right panel of Figure 2, with greater increases in predation percentage after 10 lux. However, we did not use this spline in our final model, given that the linear fit was much better (∆AIC c = 12.33), so this value should be interpreted with caution.
Additionally, this minimum value may only be relevant to the system and predator community studied here.
Previous work has suggested ALAN intensities should remain as low as possible (< 0.1 lux) to mitigate impacts on salmonids during outmigration (Tabor et al. 2004;Tabor et al. 2017). To determine a minimum lux management goal, future experimental tests of ALAN intensity with variable treatments among experimental nights are needed.
The discrepancies between early and late-night ALAN effects on predator density and predation risk could prove useful for ALAN management. For humans, artificial illumination is one of the building blocks of modern society and many social, recreational, and economic benefits result from ALAN (Doll et al. 2006;Hölker et al. 2010a;Gaston et al. 2014b). However, human activity is still highest during the day and early hours of the night, decreasing as the night progresses (Monsivais et al. 2017;Martín-Olalla 2018; Bhattacharya and Kaski 2019). Our results suggest that ALAN effects on predator density and predation may be minimal early in the night, but this trend could be driven by our experimental light design and predator community. If these temporal trends prove robust to future investigations, late-night lighting reduction campaigns during outmigration could maximize the human benefits of artificial illumination, while minimizing the negative impacts on out migrating salmonids. Overall, our findings align with others and suggest that ALAN increases juvenile salmonid predation (Tabor et al. 2004;Tabor et al. 2017). Although many questions remain unanswered, it appears that reducing artificial illumination is a practical management strategy to reduce predation on out migrating salmonids. [A]Acknowledgements

Accepted Article
This article is protected by copyright. All rights reserved

Accepted Article
This article is protected by copyright. All rights reserved 1 SE) within both light and dark treatments (B) and the estimated nightly mean PER predation risk (filled circles) from the overall Cox model. Continuous lines under experimental nights represent consecutive experimental nights at a given site.