Investigating drivers of Cisco (Coregonus artedi) recruitment in Lake Superior

Abstract

Cisco (Coregonus artedi) are a pelagic freshwater fish native to the Laurentian Great Lakes Region. Successful year classes rarely occur, and recruitment is typically poor. Despite many studies investigating recruitment, researchers do not fully understand their erratic recruitment patterns. Cisco populations collapsed in the mid-1960s due to overharvest, climate change, and invasive species introductions. While Lake Superior’s cisco populations are recovering, recruitment remains unpredictable. In this study I analyzed a 43-year dataset of age-1 cisco catches from the United States Geological Survey (USGS) Lake Superior Biological Research Station and evaluated two statistical methods to predict poor recruitment using hypothesized drivers. Poor recruitment was defined as any year with fewer than 35 fish per ha. I used two statistical methods to identify drivers of poor recruitment and evaluated relationships with air temperatures, wind speeds, ice cover and an index of Rainbow Smelt (Osmerus mordax). Drivers were lagged according to the timing of hypothesized environmental drivers of early life survival of cisco prior to capture in bottom trawl surveys. The first model applied the Poor-Recruitment Paradigm to predict poor recruitment using extreme environmental thresholds one covariate at a time. The second model used a Random Forest approach to identify key covariates influencing recruitment from a multiple covariate approach. The first model’s findings indicate that a high ice cover during egg deposition, high May and June air temperatures, and low August and September air temperatures during the larval stage link to poor recruitment. These covariates predicted poor recruitment with 100% accuracy, with no instances of good recruitment occurring beyond the identified thresholds for these specific months. The second model identified December wind speeds, March, August and September temperatures and April wind speeds as some of the most influential factors based on the mean decrease in Gini Index over 100 instances of running the model with the Synthetic Minority Oversampling Technique (SMOTE). Analysis revealed important interactions between September air temperatures and December wind speeds, as well as March air temperatures and April wind speeds. This ability to predict poor Cisco recruitment using environmental covariates provides valuable insights for managers. By pinpointing specific environmental extremes that adversely affect recruitment, this approach improves forecasting of which year classes are likely to success or fail in reaching age-1.