Indicator bacteria at five swimming beaches—analysis using random forests

“Random forests,” an extension of tree regression, provide a relatively new technique for exploring relationships of a response variable like the density of indicator bacteria in water to numerous potential explanatory variables. We used this tool to study relationships of indicator density at five beaches to numerous other variables and found that day of the week, indicator density 24 h earlier, water depth at the sampling point, cloud cover, and others were related to density at one or more of the beaches. Using data from the first 52 days of measurement allowed predicting indicator densities in the following 10 days to order of magnitude at some of the beaches. Our analyses served to demonstrate the potential usefulness of this analytic tool for large data sets with many variables.