Functional Evaluation of Three Manure-Borne Indicator Bacteria Release Models with Multiyear Field Experiment Data

Modeling the fate and transport of Escherichia coli is of substantial interest because of how this organism serves as an indicator of fecal contamination in microbial water quality assessment. The efficacy of models used to assess the export of E. coli from agricultural fields is dependent, in part, on submodels they utilize to simulate E. coli release from land-applied manure and animal waste. Although several release submodels have been proposed, they have only been evaluated and compared with data from laboratory or small plot E. coli release experiments. Our objective was to evaluate and compare performances of three manure-borne bacteria release submodels at the field-scale. Models evaluated included the exponential release model (EM), the two-parametric Bradford and Schijven model (B-S), and the two-parametric Vadas-Kleinman-Sharpley model (VKS). Each model was independently incorporated and tested as a submodel within the export model KINEROS2/STWIR, using data of E. coli in runoff. Dairy manure was uniformly applied via surface broadcasting once a year for six consecutive years on a 0.28-ha experimental field site. Two irrigation events followed each application: the first immediately followed the initial application and the second occurred 1 week later. Manure and soil samples were collected before and after irrigation, respectively, and manure, soil, and edge-of-field runoff samples were analyzed for E. coli. Model performance was evaluated with the Akaike criterion, coefficients of determination, and standard error values. The percentage of exported manure-borne E. coli varied from 0.1 to 10% in most cases, generally reflecting the lag time between initiation of irrigation and initiation of edge-of-field runoff. The export model performed better when using the VKS submodel which was preferred in 52% of cases. The B-S and EM submodels were preferred in 24 and 6% of cases, respectively. Submodels were equally efficient in 18% of cases. Two-parametric submodels were ultimately preferred over the single parameter submodel.