Application of the SWAT Model to Bacterial Loading rates in Kranji Catchment, Singapore

Abstract

Despite its tropical climate and abundant rainfall, Singapore is classified as a water scarce country. To protect its limited freshwater resources for both consumption and recreation, Singapore's Public Utilities Board (PUB) has created the Active, Beautiful, and Clean (ABC) campaign. In light of this program, the Massachusetts Institute of Technology (MIT) and Nanyang Technological University (NTU) in Singapore have partnered for various water quality research projects, including sampling of Choa Chu Kang, Bras Basah, Verde, and agricultural areas throughout Kranji Catchment in January 2011. Currently, bacterial levels in Kranji Reservoir are measured by sampling, which is labor intensive and delayed. As an alternative, a model of the surrounding watershed was constructed to estimate bacterial loading to the reservoir as driven by changing weather conditions. The watershed stream network was recreated using ArcSWAT, a version of the Soil and Water Assessment Tool used with geographic information system software. This model is based on a model previously created by Granger (2010). A major improvement is the specification of bacterial loading rates by land use and agriculture type. In order to estimate land-use-specific loading rates, numerous field samples were collected and analyzed for bacterial concentration in January 2011. Nonpoint source bacteria concentrations were estimated from field sample concentrations and applied to the land continuously in the model. Using weather data from January 2005 to February 2007, the model was run twice on a daily time step. The first run included only nonpoint sources, while the second included 23 sewage treatment plant point sources throughout the catchment. Simulated results were compared to independent samples taken in 2009 by Nshimyimana (2010) and indicate a general agreement of order of magnitude, with most measured values within the predicted range. The magnitudes of the nonpoint source run achieved a better fit with field data, although the point source run produced concentration frequency distributions that are approximately lognormal, a characteristic typical of environmental bacteria concentration distributions.