Predicting Dreissena spp. presence and biomass as a function of Lake Huron environmental characteristics

Abstract

Invasive dreissenid mussels (Dreissena polymorpha and Dreissena rostriformis bugensis) have had multiple effects on Great Lakes ecosystems, changing benthic habitats and altering food web structure and nutrient cycling on lake-wide scales. The severity of these effects depends on dreissenid density and biomass distributions, therefore, dreissenid abundances must be well understood before their impacts can be accurately estimated and managed. This thesis explores whether environmental factors can quantify and predict dreissenid distributions and abundances on a lake-wide scale. Spatial trends in dreissenid distribution and abundance were examined using dreissenid mussel abundance data from the 2017 Lake Huron Coordinated Science and Monitoring Initiative (CSMI) survey. A suite of models was developed using ten environmental explanatory variables to interpolate and predict dreissenid presence and biomass. Specifically, I developed an empirical Bayesian kriging model (r2 = 0.61), boosted regression tree models for abundance of quagga mussels (r2 = 0.27) and both mussel species (r2 = 0.23), and a boosted regression tree model of dreissenid presence (ROC score 0.735). The most important explanatory variables for dreissenid biomass were February maximum bottom temperature, substrate, bathymetry, tributary influence, and April colored dissolved organic matter. Dreissenid presence was most well described by distance from the Saginaw River, bathymetry, and February maximum bottom temperature. The inclusion of bathymetry and temperature in all the best performing models indicates the high importance of these variables for dreissenid distribution and abundance. While the models developed performed adequately, future distribution models could be improved through increases in survey effort and improved information about mussel habitat characteristics and environmental constraints. Advanced survey technologies such as autonomous underwater vehicles may be particularly helpful to quickly produce better species abundance and habitat data. Future predictive modeling efforts will be enhanced by incorporating data for other important explanatory variables, such as ionic calcium concentration, currents, and benthic food availability.