Cook Inlet has the second highest tides in the world, and correspondingly strong tidal currents, requiring particularly careful engineering of the deployment systems for housing the tracking receivers (VEMCO VR2W). To address the key biological questions, we designed a marine telemetry array consisting of 70 acoustic receivers, deployed in early June to form a sparse grid starting at the western (offshore) edge of the eastside setnet (ESSN) fishery area and extending westward approximately 15 km or one-third of the way across Cook Inlet (Figure 1). The grid consisted of a series of six east-west lines spaced approximately 5 km apart. We deployed 10 receivers in each line spaced ca 1.7 km apart. We also sited two receivers midway between each line (for a total of 10 units) to provide greater resolution along the outer boundary of the ESSN fishing zone. We monitored the boundary to the ESSN rather than the fishing district itself because it was initially believed that Chinook entered the ESSN from the offshore more or less uniformly along the western boundary and were caught soon after entry. Second, the deployment and maintenance of acoustic equipment within the ESSN is difficult because the area is shallow, rocky, and exposed to strong currents and wave-driven surge.
We recovered and successfully uploaded 54 of the receivers in late summer (27 August to 5 September 2013). The remaining receivers either were displaced by fishing activities and then returned by members of the public prior to recovery of the marine array, or lost.
To monitor the freshwater phase of the migration, 11 receivers were deployed in the Kenai River between 9 June and 2 July. Single receivers were installed at RKms 2, 4, and 25.3, and paired receivers were installed at RKms 8.2, 13.8, 22.0, and 30.6. An additional two receivers were deployed in the Kasilof River at RKm 12.1. All freshwater deployments were successfully recovered and downloaded 8 to 24 August.
All acoustic tags used in this study (custom programmed VEMCO V16P-3H; 16 mm diameter × 64 mm long; 26 g in air) were equipped with a pressure sensor reporting depth at the time of transmission. Stated depth resolution (0.6 m) and accuracy (±6.8 m) for the sensors on the company’s website indicated rather large inaccuracies relative to the potential differences in depth of migration for Chinook and Sockeye salmon that were expected. However, in follow-up discussions with VEMCO staff it became clear that the resolution and accuracy data were provided by the sensor manufacture and there was no clear statistical definition accompanying the use of these terms.
In order to assess the accuracy of the pressure sensors in the acoustic tags, we deployed six tags previously returned from the fishery and eight receivers in Sproat Lake, B.C., between 23 October and 4 December 2013. Three tags were deployed at each of 5.2 or 3.0 m below the surface (close to the mean recorded depth of the Chinook and sockeye) on a fixed sub-surface mooring. Three major rainfall events occurred while this study was running which increased the water level in the lake and thus the depth of the tags. A water level gauge at the outflow of Sproat Lake (Environment Canada hydrometric station 08HB008) indicated that water levels there changed by a maximum of 0.9 m and thus should be closely correlated to water level changes above the tags. To limit the effect of water level changes, we used only sensor transmissions on days where the water level at the gauge was within 0.25 m of its level on the day the tags were deployed (23 October). Of the tags used, one depth sensor failed (although it continued transmitting), and one tag stopped transmitting before the end of the study. Apparently the tag that stopped transmitting had not been turned off when originally recovered from the fishery and the programmed kill time of 150 days post-activation was exceeded during the lake deployment. The results show that the average difference between the approximate deployed depth of the tags and the average depth reported by the remaining sensors was 0.3 m (range 0.1 to 0.58 m; see  for details). Because the tags were programmed to have a depth resolution of 0.6 m, this indicates that the tags were generally within 1 interval of true depth and that instrument errors were substantially smaller than the difference in species-specific mean depths measured in this study. There was also no indication that the accuracy of the tags’ depth sensors changed over time.
In addition to the acoustic tags, we used numbered red Peterson disc tags as an external marker. Both acoustic and disc tags were labeled ‘Return for Reward’ in case of capture by the fishery.
A total of 25 adult Chinook and 51 adult Sockeye salmon were caught and tagged in lower Cook Inlet. Initially, we fished in offshore regions using a commercial troller running six lures on each of the six troll wires; lures were roughly equally spaced from just off bottom to approximately 1 to 2 m below the surface. In this area, fishing effort was distributed across the Inlet and tags were applied approximately in proportion to the abundance of each species over the migration time period. Chinook capture rates were low, partly due to capture consistently occurring only on the deepest two hooks. These near-bottom lures were often taken by halibut, reducing their efficiency for Chinook. As a result, late in the tagging season we also chartered two inshore sport fishing boats to capture additional maturing Chinook using rod and line in very shallow (2 to 3 m) waters just off the beach on the eastern side of lower Cook Inlet, near Anchor Point.
Tagging tanks with recirculating pumps and aeration systems were set up. An artificial fish slime (Vidalife™) was introduced into the tank water and spread over tagging surfaces coming in contact with the fish. A light sedative dose of the anesthetic AQUI-S® 20E was put in the tank water. (The use of AQUI-S® 20E was approved under the Investigational New Animal Drug (INAD) program run by the U.S. Aquatic Animal Drug Approval Partnership (AADAP) Program). Because the adult Chinook and Sockeye salmon were found to be quite docile in the tagging sling once they were inverted and a hood covered their eyes, a decision was made in late July that sedation was unnecessary for the remainder of the gastric tagging (that is, for 8% of Sockeye and 24% of Chinook).
Captured animals were placed on their back in purpose-built tagging cradles and hooded, and a hose was placed in their mouths to supply a continuous flow of aerated pumped seawater (see  for full details_. Disc tags were attached through the musculature below the dorsal fin. Acoustic tags were implanted into the abdominal cavity of the first three Sockeye Salmon using surgical techniques (1 and 2 July 2013); however, we switched to using a gastric implantation technique for the remainder of the tagging (from 2 July) when we found that the surgical incision tended to gape, probably because of pressure from the developing gonads. After tagging, each fish was measured and then released to the ocean close to their capture location.
Receiver clock drift is linear and can be easily corrected; we corrected for drift using the automatic correction function in VEMCO’s VUE software.
Prior to analysis, we screen the accumulated detection data to identify the following: (1) single transmissions that were recorded on more than one receiver; (2) false detections; (3) the date of displacement for receivers pulled from position (likely by fishing activities) and returned by the public; (4) failed depth sensors; and (5) fish that were likely not of Kenai River origin.
The model of acoustic tags we used were powerful and a single transmission was occasionally detected by more than one receiver. We identified and removed these duplicates from the marine detections so that each successfully decoded transmission would only contribute one observation to the dataset. Duplicates were identified as detections of the same fish at the same depth on neighboring receivers that were recorded within the minimum transmission interval of tag (<15 s) after the receivers were corrected for clock drift. This screen identified 205 marine detections (0.94%) as probable duplicates.
False detection screening
All telemetry systems may record ‘false positives’, which are spurious records of the detection of tag ID codes not actually present. Although it can be difficult to unambiguously identify such detections, we screen the detections data prior to analysis to assess their possible presence. We identified and excluded any detection likely to be false using the First and Second Acceptance Criteria recommended by VEMCO (Pincock 2008; see http://www.vemco.com/pdf/false_detections.pdf) with a modification to the Second Criteria. Detections met the first criteria if there was at least one short interval (<0.5 h) between successive detections of an ID code on a receiver and if there were more short intervals (<0.5 h) between detections than long ones (>0.5 h). Detections not meeting the first criteria were then examined individually (second criteria) to determine if they were supported by detections on other sub-arrays in a temporally logical sequence (including release) along the migratory path and if they were recorded when the probability of collision between multiple tags was low (that is, at times when there was a silent interval of >5 min on at least one side of the detection in question). VEMCO acoustic tags generally have a very low false positive rate: we identified four false detections (0.016%) which were all of tag codes not released in this study.
Last date of valid detection screening
When receivers are accidentally displaced from their deployment position (usually by fishing activity), they may be returned to Kintama by members of the public. We can download the data from these units; however, we do not always know the date and time they were displaced. Fishing crews are often able to provide dates when units were caught in their nets, providing us with accurate displacement dates, but receivers found floating or washed ashore may have been displaced much earlier.
When the date of displacement is not available, we estimate it by comparing the date and time of each tag ID logged with the date and time of the same tag ID on neighboring units that remained in position throughout the study period. The last date with a difference of less than 1 h between tag detections on neighboring receivers is accepted as the last date of valid detection, and otherwise valid detections recorded for later dates are excluded from any analysis that is sensitive to receiver position. This process can only be used for receivers that have data (empty units cannot be screened) and that have neighboring units that also recorded detections.
In 2013, there were four receivers returned to Kintama by members of the public. From these, we identified 35 detections (0.14%) as probably being recorded after the receiver was displaced.
Failed sensor screening
The pressure sensors malfunctioned in two tagged Chinook and one tagged Sockeye detected by the array (all recorded depths representing a constant depth above the ocean surface), so the pressure data for these animals was excluded from the depth analysis.
Stock of origin screening
This study focuses on Chinook and Sockeye salmon from the Kenai and Kasilof Rivers; however, fish were captured at sea and the stock of origin was not known at the time of tagging. Tissue samples were collected from each acoustic-tagged fish for genetic stock analysis, but results were problematic for Chinook. Prior to the start of our study, it was assumed that most fish captured in lower Cook Inlet in 2013 would return to the Kenai or Kasilof Rivers, similar to 2012. However, the recapture of a few tagged Chinook south of the tagging area provides evidence that some fish from other stocks were present in our sample. Although in most cases these fish would simply migrate elsewhere and not be detected by the array (that is, they would appear to be mortalities), it is possible that any that did encounter the marine array (for example, stocks from northern Cook Inlet) could exhibit different migration behaviors than Kenai stocks. However, the very consistent behavioral patterns observed for all tagged Chinook suggest that this is unlikely to have a large influence on the results.
To partially address this concern, we removed fish from the analyses (N = 2 Chinook; N = 3 Sockeye) that were recovered south of the release site; however, we could only remove individuals whose tags were returned. To further focus the results on the Kenai River, we also removed the one Sockeye salmon detected in the Kasilof River.
Distribution on marine array
To identify possible migratory pathways within Cook Inlet, we plotted the number of fish that were detected at each receiver on each of the six east-west lines deployed in the marine array. We also assessed the distribution of fish detected entering the ESSN by plotting fish counts at each receiver along the eastern boundary of the marine array. Because individual fish are usually heard at more than one receiver on a line, we allocated a proportion of each fish to each of the receivers on which it was detected (that is, if a fish was heard once at each of three positions, each unit was allocated 0.33 of a fish).
We calculated migration speed (km/day) as the ratio of the distance travelled over the travel time. Distance was measured for each fish along the shortest route in water. We calculated travel time for each fish from release until first detection on the marine array, from this first detection on the marine array until arrival at the Kenai River Mouth at RKm 2 (Snug Harbor), and from arrival at one detection site until arrival at the next for all sites in the Kenai River (RKms 2, 4.5, 8.2, 13.8, 22, 25.3, and 30.6). These estimates could only be made for fish detected at both detection sites bracketing the segment in question. Arrival was defined as the first detection at each detection site. For each species, we then used simple linear regression to assess if there was a relationship between migration rate and either tagging date or fork length at tagging.
In order to quantify how the density of Chinook and Sockeye salmon varied with water depth along the ESSN boundary, we calculated their cumulative depth distribution (CDD). We used each depth transmission as the unit of replication; however, because the number of depth measurements varied by individual fish, there was the possibility that unique behavior by one or a few individuals who were detected frequently could bias the results. Accordingly, we assessed the variability in the CDDs by calculating the jackknife distribution for the m fish of each species that were detected by resampling the data m times while successively leaving out all the detections from one individual fish. We then calculated the mean, minimum, and maximum values at each depth across these resampled cumulative distributions.
To further investigate the influence of individual animals, we recalculated the CDDs and assessed their variability as described above using individual fish as the unit of replication. For this approach, we allocated a proportion of each fish to each of its depth transmissions (that is, if a fish was detected 100 times, each detection was weighted as 0.01). Thus, the total number of detections for each individual summed to one. Using the fish as the unit of replication in this manner reduced the influence of individuals with high detection counts, but in exchange, individuals for whom we have little information were weighted the same as those whose depth distributions are well known. Results were very similar to those based on individual detections and are detailed in .
We then used density histograms in a trellis plot to show the relative distribution of depth detections for both species in relation to daylight and stage of tide. We defined ‘high’ tides as the top 20th percentile of tide heights predicted for the Kenai River mouth during the interval the tagged fish were migrating over the marine array, and ‘low’ tides as the bottom 20th percentile of tide heights during this same interval. Times of sunrise and sunset were calculated as the time when the upper edge of the sun’s disc coincided with the ideal horizon (that is, ignoring surface topography and variations in weather conditions on actual light levels) at Kenai Airport.
Changes in catch rates with changes in net dimensions
Because we observed a significant difference in migration depth between the species (see Migration depth under Results), we were able to explore how catch rates might vary with modification to fish gear. We started with a general model where the number of fish caught in a net of length L and depth Z is given by
Here, e(x,z) is the efficiency of the net as a function of distance offshore, x, and depth, z, while ρ(x,z) is the relative abundance of fish as a function of distance and depth. (We neglect fishing time for simplicity).
Lacking receivers in the interior of the ESSN, we assumed that the horizontal distributions of Chinook and Sockeye salmon were uniform within the ESSN zone, and that fish density varied with depth everywhere inside the ESSN as ρ(z), which we have measured along the ESSN boundary (we examine this assumption in the Discussion). If we also assume that the fishing efficiency of the net was constant with distance offshore and depth, ē, this leads to:
Here, ē ⋅ k is the catch per unit length of net at a given fish abundance, k. (We assume here that k is time-invariant to focus on the relative catch rates with different net depths, but this assumption is easily relaxed.)
In (3) we have specified a maximum allowable net length, Lmax, and maximum net depth, Zmax. The relative abundance of salmon in the water column, ρ(z), can be approximated from the frequency distribution of all depth measurements for a given species using telemetry tags.
For simplicity, let the proportion of the cumulative depth distribution above some reference depth z′ ≤ Zmax be:
Then the predicted ratio of catches between a net of standard depth and a net of maximum depth z′ will be:
Regulations stipulate a maximum depth of 45 meshes for setnets with individual meshes of ≤6” (15.9 cm), and a maximum length of 35 fathoms (64 m); see http://www.adfg.alaska.gov/static/fishing/PDFs/commercial/12uciregs.pdf. For this paper, we have assumed that the standard fishing depth is 5.5 m (18’) for a 45 mesh net. Assuming that net length and fishing time are not allowed to change, this simplifies to a predicted catch relative to the standard net of:
The predicted relative catch thus depends purely upon the depth distribution of each species as the other terms cancel in this development. We used equation (5) and the CDDs of Sockeye and Chinook salmon along the ESSN to show the expected baseline harvest for nets of varying depths relative to the 5.5 m (45 mesh) standard; and equation (4) to show the projected harvest if net length was increased to exactly compensate for the change in net depth.
Availability of supporting data
Data collected in this study are included as Additional files 5 and 6. Data were also submitted to the Alaska Department of Fish and Game, and are freely available from the authors, with no restrictions placed on use.
Use of eugenol (AQUI-S) was approved under an INAD Study Numbers 11-741-13-143 F, 11-741-13-144 F, and 11-741-13-144FA. Fish collection and tagging were conducted under permits issued by the Alaska Department of Fish and Game.