Estimating the age-structure of fish populations is required to evaluate year-class strength and make informed management decisions. Obtaining accurate age estimates often requires substantial resources and the sacrifice of a large number of fish. Age-length keys are widely used to reduce the number of fish that must be physically aged; however, they rely on limited information to extrapolate age to remaining samples. Random forest analysis is a nonparametric approach that has been shown to perform well using a large number of predictor variables and a small number of samples. It allows for the incorporation of multiple types of variables to predict the response as a result of majority vote among “trees” and assess the importance of individual predictors. We utilized variables collected during sampling to calibrate random forest models for fish age prediction. Prediction accuracy and time investment from the random forest analysis was then compared to those from traditional age-length keys. Results of this examination allow us to compare a novel technique to a widely used application in an effort to increase rapidity and accuracy of fish age estimates.
|Original language||English (US)|
|State||Published - 2014|