A study of incorporating spatial autocorrelation into CPUE standardization with an application to Ommastrephes bartramii in the northwest Pacific Ocean

Catch per unit effort(CPUE)of a fishery is often used as an abundance index which is usually assumed to be proportional to the stock abundance. Observed fisheries CPUE data are, however, influenced by many factors, in addition to fish population abundance, including spatial-temporal factors such as area and season and environmental factors such as sea surface temperature(SST)and sea surface salinity. The impacts of these factors on CPUE may shift the assumed proportionality between observed CPUE and stock abundance. Thus, CPUE standardization is needed to remove the impacts of factors other than population abundance. Many statistical models have been developed for CPUE standardization such as General Linear Model(GLM)and General Additive Model(GAM). Generally, statistical methods always assume the independence of the observed CPUEs. However, this assumption is invalid for a fish school and distribution because of their spatial autocorrelation. Therefore, in this study, we take a CPUE standardization of red flying squid(Ommastrephes bartramii)as an example. Based on the fishing data in jigging fishery by Chinese fishing fleet and the corresponding SST data and the Chlorophylla data in the Northwest Pacific Ocean from June to November from 1999 to 2012, the spatial autocorrelation is incorporated into the standard general linear model(GLM). Four distance models(Gaussian, exponential, linear and spherical)are examined for spatial autocorrelation using the CPUE standardization of red flying squid. It is found that the four spatial-GLMs always produce the better goodness-of-fit to the data than that for the standard GLM. And the exponential model generates the best goodness-of-fit to the data in the four distance models. Therefore, it is suggested that spatial autocorrelation into CPUE standardization should be considered when the nominal CPUEs are strongly spatially autocorrelated.