Determining threat status for data-limited fisheries based on catch-only stock assessment models

Catch-only stock assessment methods have been developed to manage data-limited fisheries where only catch data is available. This research evaluated the ability of four catch-only stock assessment methods to correctly classify a stock of concern based on population trends. To accomplish this, true trends from simulated stocks and the trends produced by the models were used to classify stocks into threat categories based on percent change. ROC curves and PR curves were then used to test the effectiveness of the four models as classifiers. ROC curves indicated that the models performed well under most scenarios. However, the confusion matrices and PR curves revealed low precision values for all models. The high number of stocks falsely classified as threatened were masked in the ROC analysis by the imbalance of few threatened stocks compared to numerous non-threatened stocks. This is an important caveat, as it could lead to inappropriate threshold selection.