Application of Regime Change Detection Methods to Productivity Analysis of Skeena Salmon Conservation Units (Draft, July 23, 2013)

, , , ,

Description

In this paper we apply a regime-shift detection methodology to uncover changes in the Ricker productivity parameter for salmon conservation units (CUs) in the Skeena watershed. In this paper, the term “regime shift” refers to the statistical concept of an abrupt change or discontinuity in the value of a model parameter, in the present case, the productivity parameter in the Ricker (1954) stock-recruitment model. The productivity parameter can help estimate sustainable harvest levels for individual CUs and inform appropriate management reference points. Therefore it is important to be able to detect, in a timely fashion, significant changes in productivity so that management decisions can be updated as appropriate.

Numerous studies have established that productivity varies with time for salmon stocks. For instance, Adkison et al. (1996) found that productivity for sockeye salmon stocks in Bristol Bay, Alaska increased rapidly in the 1970s. Peterman and Dorner (2012) demonstrated a sharp decrease in the productivity of Skeena sockeye stocks, beginning in the late 1980s and persisting through the 1990s. Using the methodology that we herein propose, Mueter et al. (2007) reported evidence for shifts in productivity for Pacific salmon stocks in 1974, and again in the late 1980s to mid 1990s.

We use the STARS algorithm developed by Rodionov (2004) to test for shifts in productivity. This method has several advantages over other regime-detection methodologies, notably that it can be used reliably toward the endpoints of time series, and has the ability to detect multiple shifts within the same series. The use of a discrete change framework, wherein productivity changes discretely and is constant between these changes, may also be of utility to stock management. For example, other techniques used to model changing productivity, such as the Kalman filter approach advocated by Peterman et al. (2000), require annual updating of productivity and hence of management targets. Alternatively, by modelling productivity as remaining constant between infrequent, discrete changes, management targets need only be updated when evidence of significant changes in productivity arrives.

–Excerpt from the Report’s Introduction