Predicting submerged aquatic vegetation cover and occurrence in a Lake Superior estuary

Abstract

Submerged aquatic vegetation (SAV) provides the biophysical basis for multiple ecosystem services in Great Lakes estuaries. Understanding sources of variation in SAV is necessary for sustainable management of SAV habitat. From data collected using hydroacoustic survey methods, we created predictive models for SAV in the St. Louis River Estuary (SLRE) of western Lake Superior. The dominant SAV species in most areas of the estuary was American wild celery (Vallisneria americana Michx.). Maximum depth of SAV in 2011 was approximately 2.1m. In regression tree models, most of the variation in SAV cover was explained by an autoregression (lag) term, depth, and a measure of exposure based on fetch. Logistic SAV occurrence models including water depth, exposure, bed slope, substrate fractal dimension, lag term, and interactions predicted the occurrence of SAV in three areas of the St. Louis River with 78–86% accuracy based on cross validation of a holdout dataset. Reduced models, excluding fractal dimension and the lag term, predicted SAV occurrence with 75–82% accuracy based on cross validation and with 68–85% accuracy for an independent SAV dataset collected using a different sampling method. In one area of the estuary, the probability of SAV occurrence was related to the interaction of depth and exposure. At more exposed sites, SAV was more likely to occur in shallow areas than at less exposed sites. Our predictive models show the range of depth, exposure, and bed slope favorable for SAV in the SLRE; information useful for planning shallow-water habitat restoration projects.