Can data on major phytoplankton functional group concentrations improve the estimation of E. coli concentrations in agricultural pond waters?

Irrigation ponds provide a substantial amount of water for crop production.  An increasingly large body of evidence has linked microbial impairment of these resources to foodborne outbreaks. Therefore, monitoring the microbial quality of irrigation ponds is especially prudent for food safety and reducing the incidences of illnesses and deaths resulting from contamination events. Escherichia coli (E. coli) is used worldwide as an indicator for microbial contamination of water resources as concentrations are usually indicative of pathogen presence and/or cases of illnesses.

Algae and cyanobacteria (collectively henceforth referred to as phytoplankton can comprise large fractions of the overall biomass in waterbodies. Phytoplankton are important in water quality monitoring because they directly affect water quality metrics such as dissolved oxygen and pH as well as potentially producing toxic compounds. The interaction between E. coli concentrations and phytoplankton in environmental waters has received relatively little attention and has not been studied in ponds providing water for irrigation. The objective of this work was to see if phytoplankton can be used as predictors of E. coli concentrations in irrigation ponds.

Two irrigation ponds in Maryland, USA were sampled and sensed eleven times on the permanent spatial grid during the 2017 and 2018 growing seasons. A YSI sonde was used to measure water quality variable (WQV) concentrations of pH, dissolved oxygen (DO), specific conductance (SPC), temperature(C) , turbidity (NTU), phycocyanin, Chlorophyll a (CHL),and dissolved organic matter (FDOM). Total carbon (TC), and total nitrogen (TN) were measured in the laboratory. Phytoplankton functional groups (PFG) were green algae, diatoms, cyanobacteria, and dinoflagellates. Identification and enumeration of PFG was performed with laboratory microscopy.  The random forest (RF) algorithm was used to predict E. coli concentrations and rank variables by importance using three predictor sets including water quality variables (WQV)+PFG, PFG only, and WQV only on the 2017, 2018, and 2017+2018 datasets.

For both ponds, the WQV predictor set alone provided the best model performance metric results (R²= 0.671 and 0.812, and RMSE= 0.321 and 0.374 log concentrations for Ponds 1 and 2, respectively). The combined phytoplankton and WQV predictor sets provided very close results to the WQV results alone and in all the phytoplankton variables alone as predictors showed the worst performance. The top predictors in the PFG+WQV for Pond 1 were CHL, TN, pH, NTU, and FDOM which was similar to the WQV only set. Flagellates ranked among the most important predictors in the PFG+WQV (6^th) and PFG predictor sets (1^st). In Pond 2, the top predictors in the PFG+WQV were TC, C, pH, DO, and TN. Diatoms were found to be the leading predictor in the PFG-only dataset in Pond 2.

Results of this work indicate that in studies of water bodies the effect of phytoplankton on E. coli concentrations is well represented by the water quality variables, and concentrations of the phytoplankton groups per se do not add information for improvement of the prediction of microbial water quality evaluated by E. coli concentrations using the usually very efficient machine learning predictive random forest algorithm.