872North American Journal of Fisheries Management 18:872–885, 1998qCopyright by the American Fisheries Society 1998Inseason Forecasting of Southeastern Alaska Pink SalmonAbundance Based on Sex Ratios andCommercial Catch and Effort DataJIEZHENG*Alaska Department of Fish and GameCommercial Fisheries DivisionPost Office Box 25526, Juneau, Alaska 99802-5526, USAOLEA. MATHISENUniversity of Alaska Fairbanks, Juneau Center for Fisheries and Ocean Sciences11120 Glacier Highway, Juneau, Alaska 99801, USAAbstract.—We developed a sex ratio index that, together with cumulative catch of all gears orcumulative catch per unit effort of the seine fishery, estimated abundance and catch of pink salmonOncorhynchus gorbuscha in southern Southeast Alaska during a fishing season. We evaluated threeinseason forecast models—linear, nonlinear, and combined—using data from 1983 to 1997. Basedon a cross-validation evaluation of forecast accuracy, the nonlinear model generally outperformedthe linear and combined models. Cumulative catch per unit effort was a better predictor thancumulative catch in the first 3 weeks (statistical weeks 28–30) of a fishing season, but the relationwas reversed for the remaining 5 weeks. Inseason abundance estimations greatly outperformedthe preseason forecasts. Incorporating sex ratios into inseason forecast models correctly adjustedthe run timings during a large majority of years and thus improved overall forecasts starting inthe second week. In the second through fifth weeks (weeks 29–32), the best performing modelwith sex ratios improved forecasts more than 30% over the best model without sex ratios; im-provements included averages of absolute percentages of relative forecast errors, absolute devi-ations, and squared residuals. Average absolute percentages of relative forecast errors from thebest model were less than 20% before the midpoint of the run and less than 14% at and after themidpoint. Sex ratios improved inseason forecasts more in the recent years (1992–1997) than inthe earlier years (1983–1991).Pink salmon Oncorhynchus gorbuscha is themost abundant salmon species in southeasternAlaska and supports an important commercial fish-ery that takes place during a short period from lateJune to mid-September, although most catches oc-cur from the late July to late August, or statisticalweeks 31–34 (the first week starts with the firstSunday of the year). Like most salmon fisheriesin Alaska, pink salmon fisheries in southeasternAlaska are managed by a fixed escapement policythat aims to maximize harvests without jeopard-izing the population recruitment. To achieve a tar-geted escapement, managers must know the mag-nitude of the incoming spawning run. The accu-racy of abundance information acquired inseasonsubstantially affects the manager’s ability toachieve management objectives.Pink salmon abundance in southeastern Alaskafluctuates greatly from year to year and is noto-riously difficult to predict. The observed maximumabundance variation was greater than 10-fold dur-* Corresponding author: [email protected] the last three decades. Based on the informa-tion of the size of parent escapements and sub-sequent environmental conditions or on time seriesanalysis of catches (Quinn and Marshall 1989),preseason forecasts of pink salmon runs and catch-es lack sufficient accuracy for management pur-poses, with relative forecast errors in some yearsexceeding 200% (Zheng 1988). Consequently, in-season forecasts were initiated to update abun-dance estimates over the fishing season (Zheng1988).Pink salmon migrate from the ocean to thestreams of southeastern Alaska through four prin-cipal portals: Cross Sound, lower Chatham Strait,Sumner Strait, and Dixon Entrance (Figure 1). Theruns spread out from these entrances into branch-ing straits, channels, and fjords. Numerous taggingexperiments have demonstrated that northern(NSE) and southern Southeast (SSE) Alaska pinksalmon stocks are geographically separated bySumner Strait (e.g., Rich 1926; Davidson andChristey 1938; Elling and Macy 1955; Nakataniet al. 1975; Hoffman 1982). Thus, pink salmon873FORECASTING PINK SALMON ABUNDANCEFIGURE1.—Map of southeastern Alaska showing the four major entrance portals of pink salmon: Icy Strait/CrossSound, lower Chatham Strait, Sumner Strait, and Dixon Entrance. Note District 104.abundances are predicted separately for the NSEand SSE Alaska stocks. The pink runs to south-eastern Alaska can be characterized by locationand time of spawning as early, middle, and lateruns (Sheridan 1962). However, these three runsegments tend to overlap, so we treated them as asingle unit.Timing of SSE Alaska pink salmon runs variesgreatly from year to year (Mathisen and Zheng1994), and inseason information for run timingeach year is needed to accurately forecast runsizes. Typically, male pink salmon predominatethe early segment of the run, but this shifts tofemale predominance toward the end (Mathisenand Zheng 1994). The changes in sex ratios canbe indexed to the temporal progression of the run(McKinstry 1993; Mathisen and Zheng 1994). Us-ing a logistic regression of sex ratios against timeto estimate run timing, McKinstry (1993) showedthat adjusting run timing improved inseason fore-casts, although big improvement generally oc-curred only following the middle of the run. Be-cause of the small amount of data on sex ratiosduring a fishing season and nonlinear relationshipsbetween sex ratios and time, the logistic regressionused by McKinstry (1993) does not provide anaccurate estimate of run timing until after the mid-run date. Another drawback of McKinstry’s ap-874ZHENG AND MATHISENproach is that the same shape of run-timing curveswas assumed for all years, and only the timing ofthe mid-run changes each year. In reality, the shapeof a run-timing curve is affected by the relativestrengths of early, middle, and late-run segments;these relative strengths change each year. Becauseof this assumption and inaccurate run-timing es-timates before the mid-run date, inseason forecaststhat incorporated sex ratio information did notgreatly improve forecast accuracy.In our study, we incorporated sex ratio infor-mation into inseason forecast models to annuallyadjust timing and shape of the run-timing curvesfor SSE Alaska pink salmon to improve accuracyof inseason forecasts. First, we defined an indexfor sex ratio information from deviations of week-ly sex ratios and shapes of the sex ratio curves.Then, we