Northwest Fishery Resource Bulletin: Assessment of Two Methods for Estimating Total Chinook Salmon Encounters in Puget Sound/Strait of Juan de Fuca Mark-Selective Chinook Fisheries. Manuscript Series Report No. 2.

Study Background and Intent

Washington Department of Fish and Wildlife’s Puget Sound Sampling Unit currently uses two methods to estimate total Chinook encounters in marine recreational mark-selective fisheries. Though both approaches are designed to estimate the same quantity, they often yield differing results, which poses a challenge to the interpretation of post-season estimates of fishery impacts. The two approaches are:

• Method 1 (M1) â€" M1 estimates of total Chinook encounters are derived from the combination of dockside observations of landed catch and angler interview responses about salmon releases; thus, the accuracy of Method 1 estimates depends heavily on the ability of anglers to correctly recall and report the number of Chinook they actually encountered and released¹.
• Method 2 (M2) â€" M2 estimates of Chinook encounters are obtained using a combination of creel estimates of legal-size, marked (LSM) Chinook harvest based on dockside sampling and test-fishery data on the relative abundance of LSM Chinook in the targeted (i.e., vulnerable to angling gear) Chinook population. The M2 estimator was derived assuming that anglers retain all LSM Chinook encountered, therefore, its accuracy depends on the extent to which angler behavior deviates from this idealized case.

Over several mark-selective Chinook fisheries prosecuted in the marine waters of the Puget Sound/Strait of Juan de Fuca between 2003 and 2007, M1 and M2 estimates of total Chinook encounters have deviated substantially, with the former typically being larger than the latter. Observed differences in M1 and M2 estimates are a product of differing biases in estimates (i.e., relative to the true but unknown number of Chinook encountered in a given fishery) and random sampling variation. With respect to the first factor, past studies and the literature indicate that the assumptions affecting the accuracy of both M1 and M2 estimates are routinely violated to varying degrees. For Method 1, for instance, recall and/or reporting errors are a common feature of angler responses to interview questions requiring precise, quantity-based responses. Of relevance to the accuracy of Method 2 estimates, earlier work demonstrates that anglers occasionally (and purposefully) release LSM Chinook during fishing trips.

The fact that Methods 1 and 2 yield comparable estimates of Chinook encounters in only a subset of cases poses challenges to post-season fishery evaluations and limits the utility of such results to other applications. For example, the existence of diverging estimates of encounters and mortalities renders post-season comparisons of "observed" to projected fishery impacts on unmarked Chinook somewhat subjective. Further, reliable estimates of stock-specific total mortality are needed for catch accounting, cohort run reconstruction, and abundance forecasting. With two divergent estimates, it is difficult to determine which supplies the best estimate to use for these applications. Motivated by these issues, our primary goal was to determine if an adjustment to either Method 1 or Method 2 total encounter estimates can be produced that provides an acceptable single estimate of total Chinook encounters, and subsequent total release mortality, for selective fisheries conducted in the marine areas of Puget Sound. To do this, we i) characterized seasonal differences in selective fisheries with respect to parameters that may influence the potential for bias in estimates; ii) described patterns in the differences between M1 and M2 estimates and quantified their association with fishery characteristics; iii) identified and attempted to quantify the potential sources of bias in M1 and M2 estimates; and iv) evaluated bias "correction" possibilities for the two estimate types. Based on this combination of objectives and their associated analyses, we identified a potentially suitable bias correction that could be applied to Method 2 estimates of Chinook encounters to obtain a single "best" estimate of total Chinook encounters.

Supporting Analyses and Results

Seasonal Selective Fishery Patterns

We compared Method 1 and Method 2 estimates of total Chinook encounters for 44 discrete time/area/year strata in which Chinook selective fisheries were conducted in the marine waters of Puget Sound and the Strait of June de Fuca between 2003 and 2007. The estimates included both summer (May through September; catch areas 5, 6, 9, 10, and 11) and winter (October through April; catch areas 8-1 and 8-2) fisheries.

Differences (M1-M2) between M1 and M2 estimates of total Chinook encounters ranged from -107 to 7,963 encounters for summer selective fisheries (mean = 3,073) and ranged from -1,903 to 4,049 encounters for winter selective fisheries (mean = 336). The M1 estimate was greater than the M2 estimate 73% of the time across the set of fisheries analyzed. The difference between the estimates exceeded 1,000 fish in nearly half of the cases (45%).

From this evaluation of general M1 and M2 patterns, we concluded that differences in estimates are quite often large enough to limit their utility to management processes. A single, best estimate of total encounters is therefore needed.

Comparison of Seasonal Patterns in Differences and Associations with Fishery Metrics

We compared metrics describing fishery conditions (angling effort and per-capita encounter rates) between summer and winter selective seasons and examined associations between these metrics and differences in M1 and M2 estimates. Results from these analyses demonstrate that there are:

1. Seasonal differences in fishery characteristics, with summer selective fisheries exhibiting consistently higher angler effort and lower salmon release rates than winter selective fisheries.
2. Seasonal differences in M1 and M2 differences. Absolute differences in M1 and M2 estimates (DIFF = M1 â€" M2) were greater and more variable for summer than winter fisheries; in contrast, relative differences in M1 and M2 estimates (RATIO = M1/M2) were similar for the two season types.
3. Significant relationships between M1 and M2 differences and fishery conditions. Differences in estimates were positively correlated with both fishing effort and salmon release rates but were more strongly related to effort for summer fisheries and more strongly related to release rates for winter fisheries. General patterns of association of fishery predictors with M1 and M2 responses (absolute and relative differences in estimates) were similar; however, the effort and encounter-rate predictors accounted for a greater proportion of the variation in DIFF than for RATIO.

Based on these findings, we concluded that effort and salmon encounter rates (average number of salmon released per angler trip) have a moderating influence on the difference between the M1 and M2 estimates.

Review of Bias in Estimates

For Method 1, we reviewed evidence that suggested a combination of digit bias and prestige bias contributes to M1 over-estimating the true number of Chinook encounters. The first form of recall error involves anglers reporting the number of salmon they encounter as a rounded approximation of what they actually release (e.g., reporting that 10 salmon were released when the true number was actually 9 or 11). Indicative of digit bias, we found that anglers preferentially report salmon releases in numbers ending in 5 and 10 when encounter rates are high. In these instances, we also observed large differences in M1 and M2 estimates (i.e., M1 >> M2). With respect to literature-based estimates of bias due to the over-reporting of releases, the only study available for Puget Sound salmon fisheries suggested that this form of bias can lead to a 40% overestimate of salmon releases, on average.

For Method 2, we evaluated evidence indicating that LSM Chinook release occurs on both an intentional and unintentional basis. Voluntary trip report (VTR) and dockside-collected data indicate that anglers intentionally release between four and eight percent of the LSM Chinook that they encounter, with this rate being nearly double for winter compared to summer selective fisheries. Using a novel framework, we estimated the magnitude of unintentional LSM Chinook releases (e.g., due to errors in measurement made by anglers) and determined that it occurs at a similar level (i.e., 4-8%). In contrast to the intentional rate, however, it was estimated to be higher in summer than winter selective fisheries. In combination, intentional and unintentional releases likely contribute to a 12-13% underestimate of actual (true but unknown) encounters by M2.

Consideration of Bias-Correction Possibilities

To identify a single, reliable estimate of total Chinook encounters for selective fisheries, we considered possibilities for correcting bias in estimates generated by both approaches. First, we considered two Method-1 bias corrections, one which attempts to eliminate digit bias (the Beamanâ€"Vaske method) from the raw interview data and another which corrects fishery-total M1 estimates for assumed levels of positive bias levels (combined effect of digit and prestige bias expressed in final estimates). Given that estimates were generally insensitive to the Beamanâ€"Vaske bias correction and that recent, relevant field data on reporting bias (i.e., for use in M1 fishery-total bias correction) are unavailable, M1 bias-correction does not appear practicable at the present time. In contrast, M2 bias correction based on recently collected data on LSM Chinook release rates has promise for application. Finally, we conducted a parameter grid search to identify hypothetical combinations of M1 and M2 bias corrections that minimized summed differences in bias-corrected Chinook encounters estimates. Results from this exercise confirmed that the independent field estimates of M1 bias (+40%, on average, from a prior WDFW study) and M2 bias (-12 to -13%, from VTR and dockside studies) were consistent with M1 and M2 divergence patterns in the observed data.

Recommendations

Based on our analyses and practical considerations, we recommend Method 2 with a correction for the release of legal-size marked Chinook as the preferred method for estimating total Chinook encounters in mark-selective Chinook fisheries. In particular, an "unbiased" estimate of total Chinook encounters could be obtained under Method 2 using:

Bias-Corrected M2 = Original M2 Estimate / (0.87)

We recommend that this bias correction be applied to all Method 2 estimates produced through intensive mark-selective fishery monitoring. Also, it may be possible to use this bias correction in cases where an estimate of the total number of LSM Chinook harvested is obtained through less intensive survey approaches (e.g., estimates of total Chinook harvest resulting from the WDFW Catch Record Card [CRC] system, coupled with field estimates of LSM Chinook relative abundance). In addition, to maintain and/or increase the reliability of modified selective fishery estimates in the future, we recommend:

1. If the proposed 13% correction is deemed acceptable, past estimates of total Chinook encounters should be updated and included in a historical data appendix in future post-season reports.
2. Means for reducing the variance of the Method 2 estimates without expanding sampling efforts should be explored (e.g., evaluate variance contributions from dockside and test fishing components).
3. Sampling should continue in mark-selective fisheries so that the data needed to estimate LSM Chinook release rates are periodically obtained. This will enable routine calibration of the proposed bias correction, which may be necessary if/when major changes occur in either fishery regulations or fish populations that might affect the value of this parameter. Also, where and when it is feasible, fishery-total estimates of intentional LSM Chinook releases should be produced and evaluated relative to the proposed M2 bias correction.
4. As additional CRC estimates of total Chinook harvest become available for the selective fisheries reviewed in this report, estimates of total Chinook encounters generated using the intensive Murthy approach should be compared with those derived from the CRC system. In particular, such an analysis should emphasize understanding the utility of the bias-corrected M2 estimator for generating unbiased estimates of total Chinook encounters with CRC harvest data as the starting point.
5. In fisheries characterized by a large catch-and-release component (mark-selective, salmon, or otherwise), we recommend that estimates of total encounters generated from angler interviews be interpreted cautiously. This recommendation is particularly relevant to situations where apparent encounter rates are high.

¹ Estimates of Chinook salmon harvest are the same for both M1 and M2.