Model-based Method to Estimate Bycatch Rates
Observer data are used to develop bycatch rate models for the US North Atlantic Bottom Trawl and Mid-Water Trawl fisheries. PSB staff members have developed bycatch rate models of white side dolphins and pilot whales using GLMs, GAMs, Classification Trees, and zero-inflated models. Many gear characteristics, fishing practices, spatial-temporal categories, and environmental factors were investigated to determine the best fitting bycatch rate model using observer data collected from multiple years. For example, a Poisson regression was used to model the bycatch rate (y/effort), where the number of takes (y) was a function of predictor variables:
where fi was a smooth function when the model was a GAM model or fi was a coefficient value when the model was a GLM; xi was a predictor variable describing an environmental, gear, or fishing characteristic: and the log(effort) was an offset variable with a coefficient set to 1.0.