Predicting the occurrence of Salmonella , Escherichia coli , and pathogenic E. coli in agricultural ponds in

Abstract

Escherichia coli is one measure of fecal contamination in water, which is expected to correlate with the presence of fecal pathogens, such as pathogenic E. coli or Salmonella . Supplementing E. coli measurements with other water quality measurements, for example, temperature or pH, may better identify situations when foodborne pathogens are likely to be present in water. This work used generalized and multinomial regression and machine learning (K‐nearest neighbors, random forest (RF), and support vector machines) to create explanatory models for levels of E. coli , pathogenic E. coli , and Salmonella presence in irrigation ponds. Model performance was mixed, with RF models performing best for E. coli levels (root mean square error [RMSE] = 0.45) for all ponds ( n  = 1266), pond 1 (RMSE = 0.44, n  = 563), and pond 2 (RMSE = 0.71, n  = 505), while generalized regression performed best at pond 3 (RMSE = 0.41, n  = 198). Pathogenic E. coli presence for all ponds was best explained by RF (receiver operating characteristic area under the curve [ROC AUC] = 0.94, n  = 717). Salmonella presence for ponds 1 and 2 ( n  = 284) was best explained by multinomial regression (ROC AUC = 0.83). These results indicate that explanatory models struggled to identify situations conducive to pathogen occurrence in this system and were not useful in identifying factors associated with pathogen presence.