Predicting the abundance of marine fishes based on habitat models is often difficult due to the presence of large numbers of zero observations. The objective of this study was to analyze the ability of a 2-stage model to predict the presence and abundance of a rockfish species, shortspine thornyhead Sebastolobus alascanus. The data used for these analyses were collected during bottomtrawl surveys of the Gulf of Alaska ecosystem from 1993 to 2007 and in the Aleutian Islands ecosystem from 1994 to 2006. The presence of shortspine thornyhead was predicted from the 5th and 95th percentiles of the cumulative distribution function resampled over depth and temperature. The results predicted shortspine thornyhead would not occur at depths <176 m or > 671 m, and presence or absence was correctly predicted at 86.3% of the trawl survey stations. Environmental variables were then used to model shortspine thornyhead abundance at stations where presence was predicted. The best-fitting model of abundance included the variables depth, local slope, thermocline temperature, shrimp catch per unit effort (CPUE), and an index of predation refuge. The model explained 72.4% of the variation in 1993-2005 Gulf of Alaska survey data and 73.7% of the variation in the 2007 data from the Gulf of Alaska. The model explained only 23.9% of the variation in shortspine thornyhead CPUE from the Aleutian Islands bottom-trawl surveys from 1994 to 2006. The habitat model included important variables for survival and growth in order to provide more biologically meaningful results than with other modeling methods.
KEY WORDS: Rockfish · Fish habitat · Fish distributions · Environmental variables · Alaska
Resale or republication not permitted without written consent of the publisherMar Ecol Prog Ser 379: [253][254][255][256][257][258][259][260][261][262][263][264][265][266] 2009 gression or general linear model) or non-parametric (general additive model) model to predict abundance (Welsh et al. 1996, Barry & Welsh 2002. The general additive models or general linear models can quite effectively describe the data patterns and predict observations (Swartzman et al. 1992, Stoner et al. 2001, Walsh et al. 2004). However, the results of the analyses may be difficult to interpret in terms of the biology of the organism, and predictions of presence and abundance outside the region or years where the model was parameterized may be inaccurate. To provide more biologically meaningful results from these models, it would be preferable to utilize relationships based on the underlying processes affecting the organism's growth and survival; however, these are seldom measured directly.Niche theory predicts that a population's abundance should change along a resource continuum that defines its habitat (Hutchinson 1957). These relationships between the environment and population abundance have been described by linear relationships, densitydependent functions, and dome-shaped curves (May 1973, Murawski & Finn 1988, Friedlander & Parrish 1998, Iles & Beverto...