Do models parameterized with observations from the system predict larval yellow perch (Perca flavescens) growth performance better in Lake Erie?

Abstract

Growth performance can influence survival during early life. As such, a range of statistical to mechanistic modeling approaches has been used to predict growth performance, with few studies evaluating prediction accuracy. We tested the ability of three models to estimate observed larval yellow perch (Perca flavescens) growth and length in western Lake Erie (United States – Canada). We found that a general linear model developed using yellow perch data from western Lake Erie performed best followed closely by a semimechanistic individual-based model (IBM) specific to Lake Erie yellow perch and worse by a general multispecies IBM. We suspect that the statistical model performed better because, unlike IBMs, it does not require prey availability data, probably poorly represented by zooplankton samples, and because the IBMs are imperfectly parameterized. Our findings indicate that caution should be exercised when using general IBMs given that the models parameterized with observations from the system of interest outperformed the general IBM in providing accurate fish growth and length estimates, pointing to the need for research that can improve existing mechanism-based models of larval growth.