Flexing machine learning muscle: predicting lake calcium to prioritize invasive mussel monitoring in Alaska
摘要
Invasive zebra (Dreissena polymorpha) and quagga (Dreissena rostiformis bugensis) mussels (ZQM) are spreading across western North America, but are not yet known to be established in Alaska. If they invaded Alaska, ZQM could disrupt freshwater ecosystems supporting some of the world’s most productive wild salmon fisheries. This study examined a key factor limiting ZQM invasion risk, the concentration of dissolved ionic calcium (Ca2+), by synthesizing existing water quality data from published and unpublished sources and collecting new samples to fill data gaps for important salmon-bearing lakes across Alaska. We assembled a geospatial dataset including observed dissolved Ca2+ concentrations for 1406 lakes and used a randomForest model to predict Ca2+ concentrations for 1182 additional lakes using other water quality parameters as predictors. The most important predictors of Ca2+ were Mg, conductivity, pH, Si, Na, and K. Both predicted and observed Ca2+ values were highly variable, ranging from 0.1 to 35.8 mg/l. A total of 888 lakes had moderate or high risk for ZQM (Ca2+ values > 12.0 mg/l), including many important salmon-bearing lakes, and 1700 lakes had minimal (Ca2+ values of 0.0–8.0 mg/l) or low (Ca2+ values of 8.1–12.0 mg/l) suitability. No region of the state had consistently low suitability for ZQM based on calcium levels, meaning prevention efforts will need to focus at the scale of individual lakes rather than assuming that entire regions are safe from invasion. This study assists in prioritizing prevention, education, and early detection and rapid response to focus on lakes with the greatest potential for ZQM invasion of highly valued ecosystems and fisheries.