Analyzing fish spatial and temporal distribution patterns using uncalibrated acoustic data from multiple vessels

No abstracts are to be cited without prior reference to the author.

Acoustic density data were collected from three commercial fishing vessels participating in the 2003 eastern Bering Sea walleye pollock (Theragra chalcogramma) fishery. Unique challenges encountered with these data included: 1) uncalibrated data from each vessel, 2) no formal sampling design, 3) data volumes that exceeded computing capacity, 4) unbalanced temporal and special coverage, 5) a high degree of spatial autocorrelation, and 6) heteroskedastic density distribution between light and dark hours. We assess spatial and temporal structure of pollock and present a stratified random sampling technique for subsampling the data and addressing unbalanced data collection, while retaining the spatial and temporal attributes of the pollock distribution within the data. A general additive mixed model (GAMM) with an internal correlation structure and heterogeneous variance was used to inter-calibrate the data and to evaluate spatial and temporal distribution of pollock density on the eastern Bering Sea shelf. Covariates investigated included vessel, daylight phase, week of the year, bottom depth, and sea surface temperature. Initial optimum GAMM configuration was evaluated using small sample Akaike information criterion (AICc). Resampling cross-validation is presented as a means to counter possible overparameterization of smoothing splines where a GAMM was not initially optimized over the full range of data. This study introduces a new data source and presents analytic methods to quantify factors influencing fish distributions and dynamics at spatial and temporal scales that were not previously possible