Fig 1.
Map of study site at the Pillar Point Harbor.
Basemap reprinted under a CC BY license, with permission from CARTO (https://carto.com/), original copyright 2018.
Fig 2.
Time series of FIB data collected during the sampling event.
Log10 transformed TC, EC, and ENT concentrations are presented in the top three subplots. Gray area surrounding the points represents the 95% confidence interval. The dashed lines represent the regulatory threshold, and samples below the LOD are plotted with a value of 0. Tide level during each sampling point is plotted in the final subplot where gray shading represents periods during the night.
Table 1.
Summary of FIB data collected every 30 minutes during the ‘main’ sampling event.
Table 2.
Summary of FIB data collected every 1 minute during the ‘sprint’ sampling event.
Fig 3.
Mean δi was calculated using the sprint data (i = 1m) as well as samples downsampled from a 30 minute sampling interval (30m) to one hour (1H), two hour (2H), three hour (3H), six hour (6H), 12 hour (12H), and 24 hour (24H) intervals. As downsampling interval increases, there are multiple choices of the first sampling time point. As such, the mean value across all versions of the downsampled data for a given interval (dots) as well as the standard deviation of those values (shaded regions) are plotted.
Table 3.
The first two columns list each environmental variable and the associated lag with the strongest Spearman correlation to log10-transformed EC concentration. daytime and hours_from_noon were excluded from the temporal lag process. chl and turb values were log10-transformed and all variables were normalized (i.e. made unitless) prior to model fitting. The Concentration Model columns show the variable coefficients and associated p-values of the generalized least squares regression model; values where p-value is less than 0.05 are bolded and variables with no value were removed from the model due to high variance inflation factor (VIF). Note than because TC was never measured below the LOD, no Binary Model (i.e. random forest) was fit.
Table 4.
EC descriptive (hurdle) model.
The first two columns list each environmental variable and the associated lag with the strongest Spearman correlation to log10-transformed EC concentration. daytime and hours_from_noon were excluded from the temporal lag process. chl and turb values were log10-transformed and all variables were normalized (i.e. made unitless) prior to model fitting. The Binary Model column shows the non-zero permutation feature importances (i.e. the change in model accuracy upon fitting a model after randomly shuffling the variable’s data five times) of the random forest model. The Concentration Model columns show the variable coefficients and associated p-values of the generalized least squares regression model; values where p-value is less than 0.05 are bolded and variables with no value were removed from the model due to high variance inflation factor (VIF).
Table 5.
ENT descriptive (hurdle) model.
The first two columns list each environmental variable and the associated lag with the strongest Spearman correlation to log10-transformed EC concentration. daytime and hours_from_noon were excluded from the temporal lag process. chl and turb values were log10-transformed and all variables were normalized (i.e. made unitless) prior to model fitting. The Binary Model column shows the non-zero permutation feature importances (i.e. the change in model accuracy upon fitting a model after randomly shuffling the variable’s data five times) of the random forest model. The Concentration Model columns show the variable coefficients and associated p-values of the generalized least squares regression model; values where p-value is less than 0.05 are bolded and variables with no value were removed from the model due to high variance inflation factor (VIF).
Table 6.
The Binary Model columns display the number of samples (out of 96) that were measured below the LOD and the accuracy of the random forest in classifying samples above or below the LOD. Note than because TC was never measured below the LOD, no Binary Model was fit. The Concentration Model columns display the R2, root mean square error (RMSE) and Durbin-Watson statistic of the GLS models which predict log10-transformed FIB concentration of samples measured above the LOD. The Overall Model columns display the R2 and root mean square error (RMSE) of the hurdle models which predict log10-transformed FIB concentration of all samples.