A Spatially Explicit Model to Predict Walleye Spawning in an Eastern Lake Ontario Tributary

Abstract

Field egg collections coupled with physical habitat measurements were used to develop, apply, and validate a spatially explicit model to predict the spawning sites of walleyes Sander vitreus in an eastern Lake Ontario tributary. We collected 22,246 walleye eggs using passive traps and measured the attendant habitat variables, including substrate composition and heterogeneity, water depth and velocity, and cumulative spring water temperature. The squared term that we used for water depth in our regressions and selected over linear and polynomial fits better represented the relationship between spawning probability and depth. To predict walleye spawning likelihood, we developed 10 candidate models based on the available spawning data; following an information-theoretic approach, we compared the models by means of Akaike's information criterion. The best model had an area under the receiver operating curve of 0.89 and a weight of evidence of 0.32, where the probability of spawning (egg presence) was associated with predominance of course substrate and shallow depth and timing was associated with the accumulated spring water temperature. The model was applied to the study stream and validated by additional egg collections. The model correctly classified egg presence/absence in 66.7% (10 of 15) of egg traps in the initial validation sample. After a high-flow event, classification success decreased to only 20% (3 of 15 traps), probably because of the redistribution of eggs to less suitable habitats in depositional areas. Prediction of walleye spawning distribution allows for reach-scale assessment of critical habitat to help guide spawning habitat restoration efforts.