UW Aquatic & Fishery Sciences Quantitative Seminar

Mark Scheuerell

NOAA Fisheries, NWFSC and Skip McKinnell (PICES), Jim Irvine (DFO), Eli Holmes (NOAA), Eric Ward (NOAA)

Using dynamic factor analysis to identify large-scale patterns in adult abundance and marine survival of Pacific salmon

Hare et al. (1999) described an "Inverse Production Regimes" hypothesis for Pacific salmon whereby adult returns of populations from northern latitudes (i.e., Alaska) appeared to vary inversely with those from southern locations (e.g., WA, OR, CA). Much of the variance was attributed to shifts in large-scale oceanographic patterns (i.e., the Pacific Decadal Oscillation), and this seminal research shifted much of our focus to the importance of the ocean environment. Capitalizing on hindsight and recent developments in statistical models, we see two potential shortcomings in their assessment. First, their data were indices of adult abundance, which integrate variance over the entire life cycle rather than only during the period of ocean residency. Second, they used principal components analysis (PCA) to identify similar groups of salmon, but PCA restricts the trends to be linear and ignores information implicit in the time ordering of the data. We present an approach that (1) includes estimates of actual marine (i.e., smolt-to-adult) survival, and (2) uses a multivariate time series technique called dynamic factor analysis (DFA) to identify common, time-varying trends in the data. Our survival data consisted of time series of smolt-adult survival estimates for 95 populations with freshwater spawning habitats from around the North Pacific, including Japan, Russia, the United States, and Canada. Time series were 15-35 years long, and some included missing data. The data include ocean-entry years 1970-2007. Chinook and coho dominated the species composition, but chum, pink, and sockeye were also represented. In cases without any effects of additional covariates (e.g., PDO) 3-4 common trends were strongly supported by the data. The role of species was negligible (i.e., trends did not fall out simply by species), but the effect of geographic region was obvious. As expected, we observed large-scale patterns in that Asian populations were different from Alaskan stocks, which were different from more southerly ones. Perhaps more striking, however, was that at smaller scales salmon from the Salish Sea (i.e., Puget Sound and Strait of Georgia) were considerably different than fish from Vancouver Island, the Washington coast, and the Columbia River basin. These results point to a dynamic ocean environment whereby juvenile salmon that enter the ocean in areas relatively close to one another can exhibit very different trends in marine survival. Additional analyses focus on including the effects of environmental covariates believed to be important (e.g., SST, PDO), and using DFA to examine whether adult abundance data show similar spatial groupings to those posited by Hare et al. (1999).