Improving existing catch per unit effort (CPUE) models for construction of a fishery abundance index is important to the Alaska sablefish (Anoplopoma fimbria) stock assessment. Performance of statistical methods including Generalized Linear Models (GLM), Generalized Additive Models (GAM), and Boosted Regression Trees (BRT) were evaluated using CPUE data collected by observers from the sablefish longline fishery in the Gulf of Alaska, the Bering Sea, and the Aleutian Islands during 1995-2011. Due to the nonlinearity of several important covariates found during the diagnostics, GLM was dismissed as a potential method to standardize CPUE. Fitted GAM models for the Gulf of Alaska subregions: West Yakutat, Western Gulf, Central Gulf, and Southeast accounted for 42%, 29%, 30%, and 45% of total model deviance explained, respectively. BRT models accounted for 47%, 31%, 30%, and 46 %, respectively. For the Bering Sea and Aleutian Islands subregions, fitted GAM models accounted for 58% and 54% of total model deviance explained, respectively. BRT models accounted for 63% and 60% for the Bering Sea and the Aleutian Islands subregions, respectively. Predictive performance metrics (Root Mean Square Error) and 5-fold cross-validation results showed GAM and BRT models had similar predictive power. However, variance was significantly higher in GAM model predictions. In general, the BRT model performance was superior or equally robust to traditional methods such as GLM and GAM and should be considered as a potential statistical method for CPUE standardization.Improving existing catch per unit effort (CPUE) models for construction of a fishery abundance index is important to the Alaska sablefish (Anoplopoma fimbria) stock assessment.
|Title||:||A Comparison of Statistical Methods to Standardize Catch-per-unit-effort of the Alaska Longline Sablefish Fishery|
|Author||:||I. Mateo, Dana Henry Hanselman|