How to identify the drivers of population connectivity remains a fundamental question in ecology and evolution. Answering this question can be challenging in aquatic environments where dynamic lake and ocean currents coupled with high levels of dispersal and gene flow can decrease the utility of modern population genetic tools. To address this challenge, we used RAD-Seq to genotype 959 yellow perch (Perca flavescens), a species with an ~40-day pelagic larval duration (PLD), collected from 20 sites circumscribing Lake Michigan. We also developed a novel, integrative approach that couples detailed biophysical models with eco-genetic agent-based models to generate 'predictive' values of genetic differentiation. By comparing predictive and empirical values of genetic differentiation, we estimated the relative contributions for known drivers of population connectivity (e.g., currents, behavior, PLD). For the main basin populations (i.e., the largest contiguous portion of the lake), we found that high gene flow led to low overall levels of genetic differentiation among populations (F_ST = 0.003). By far the best predictors of genetic differentiation were connectivity matrices that were derived from periods of time when there were strong and highly dispersive currents. Thus, these highly dispersive currents are driving the patterns of population connectivity in the main basin. We also found that populations from the northern and southern main basin are slightly divergent from one another, while those from Green Bay and the main basin are highly divergent (F_ST = 0.11). By integrating biophysical and eco-genetic models with genome-wide data, we illustrate that the drivers of population connectivity can be identified in high gene flow systems.

For the biophysical model, we used a Lagrangian particle tracking model previously developed to study the transport of larval cod (Churchill et al., 2011; Huret et al., 2007), where three-dimensional current velocities and turbulent diffusivity were output from the application of the Finite Volume Community Ocean Model (FVCOM). A random-walk scheme for spatially varying vertical diffusivity was used, including a vertical floating/sinking/swimming velocity (Gräwe, 2011; Rowe et al., 2016). Particles were designated to be either 1.) neutrally buoyant or 2.) have an upward vertical swimming velocity of 0.0003 m/s. We chose to use an upward vertical swimming velocity because yellow perch larvae are more likely to be collected in the upper layers of Lake Michigan (Martin et al. 2011). The Lagrangian particle tracking simulations were forced by output from FVCOM simulation of Lake Michigan-Huron (Anderson & Schwab, 2013) incorporating exchange currents in the Straits of Mackinac. Horizontal grid resolution varied with finer resolution nearshore and in regions with complex coastlines (e.g., 100 m in the Straits of Mackinac to 2.5 km in the center of the lakes), and each horizontal grid was discretized into 20 terrain-following (sigma) layers. 

The files consist of text files (*.txt) and inage files (*.png). A simple text editor and photo imager should be suffifient to open all files.