SWAT bacteria sub-model evaluation and application

Abstract

The overall goal of this study was to evaluate and apply the Soil and Water Assessment Tool (SWAT) model for fecal bacteria modeling. Methods were developed to characterize fecal coliform bacteria (FCB) from livestock, human, and wildlife sources to use as input in the model. Model sensitivity to predict FCB concentration was evaluated for the model parameters and input parameters using both SWAT 2000 and 2005 versions. Sensitivity of input parameters generally, ranked as Bacteria concentration ≥ TBACT > Wildlife source loads > Livestock stocking rate ≥ Livestock manure production rate > BACTKDQ for SWAT 2000 whereas it was ranked as BACTKDQ > TBACT > Bacteria concentration > WDLPQ > WDLPS for SWAT 2005. Sensitivity of model and input parameters were found changed from SWAT 2000. The SWAT (2005) model was calibrated and validated for daily flow, sediment, and fecal bacteria concentration using one year of measured data (January to December, 2004). The SWAT model predicted results with poor to very good agreement when compared with measured data with coefficient of determination (R2) and Nash-Sutcliffe Efficiency Index (E) range of 0.10 to 0.89 for daily flows, sediment, total phosphorus, total nitrogen and total FCB concentration. More extensive in-stream data are needed for more comprehensive model assessment. The SWAT model (2005) was evaluated for source-specific FCB modeling using three years (2004-2006) of observed modified deterministic probability of bacteria source tracking (BST) data. The FCB sources were modeled with three combinations (livestock and human, livestock and wildlife, wildlife and human) and each single source to evaluate the source-specific FCB concentrations. The SWAT model determined poor to good agreement for the combined source of FCB (R2, E range from -2.92 to 0.71) but determined generally decreased agreement for each single source of bacteria (R2, E range from -5.03 to 0.39) potentially due to BST uncertainty, spatial variability and source characterization. The SWAT model identified critical sub-watersheds in the watershed where implementing vegetative filter strips (VFS) could be most effective to abate fecal bacteria pollution. The targeting method of VFS application to the watershed sub-basins was found to be more effective in reducing both FCB (60% vs. 42%) and sediment yield (63% vs. 33%) as compared to a random approach. The FCB source characterization methods for modeling developed in this study are general and have the potential to be extended to other watersheds. The results of this study demonstrate that the SWAT model can be used to characterize the distribution of bacteria sources within a bacteria impaired watershed and assist with developing total maximum daily loads (TMDLs) and watershed restoration strategies.