Linking watershed modeling and bacterial source tracking to better assess E. coli sources

Science of the Total Environment 648 (2019) 164–175

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Linking watershed modeling and bacterial source tracking to better assess E. coli sources Jaehak Jeong a,⁎, Kevin Wagner b, Jaime J. Flores c, Tim Cawthon d, Younggu Her e, Javier Osorio a, Haw Yen a a

Blackland Research Center, Texas A&M AgriLife Research, Texas A&M University, 720 East Blackland Road, Temple, TX 76502, USA Oklahoma Water Resources Center, Oklahoma State University, 139 Ag Hall, Stillwater, OK 74078, USA Texas Water Resources Institute, 2260 TAMU, College Station, TX 77843, USA d Texas Commission on Environmental Quality, 2100 Park 35 Circle, Austin, TX 78753, USA e Agricultural and Biological Engineering Department/Tropical Research and Education Center, Institute of Food and Agricultural Sciences, University of Florida, Homestead, FL 33031, USA b c

H I G H L I G H T S

G R A P H I C A L

A B S T R A C T

• Diffuse sources of E. coli identified by a BST assessment were successfully incorporated into SWAT for a watershed scale assessment. • SWAT was implemented to identify critical source areas of E. coli in mixed land uses in south Texas. • Wildlife contribution is the major source of E. coli in streamflow and may remain to be significant after land use change with urbanization. • A combined effort that implements land managements and advanced treatment is needed to restore water quality for recreational and aquatic life uses.

a r t i c l e

i n f o

Article history: Received 9 March 2018 Received in revised form 27 June 2018 Accepted 6 August 2018 Available online 07 August 2018 Editor: Ouyang Wei Keywords: E. coli FIB SWAT Modeling Water quality Watershed

⁎ Corresponding author. E-mail address: [email protected] (J. Jeong). https://doi.org/10.1016/j.scitotenv.2018.08.097 0048-9697/© 2018 Elsevier B.V. All rights reserved.

a b s t r a c t Terrestrial fate and transport processes of E. coli can be complicated by human activities like urbanization or livestock grazing. There is a critical need to address contributing sources of bacterial contamination, properly assess the management of critical sources, and ultimately reduce E. coli concentrations in impaired water bodies. In particular, characterization of wildlife animal contributions and other “background” input sources of microbial pollution are highly uncertain and data are scarce. This study attempts to identify critical sources of E. coli and the efficacy of conservation practices for mitigating E. coli concentrations in the Arroyo Colorado watershed, Texas, using a process-based hydrologic and water quality model. We propose to incorporate a bacterial source tracking assessment into the modeling framework to fill the gap in data on wildlife and human contribution. In addition, other sources identified through a GIS survey, national census, and local expert knowledge were incorporated into the model as E. coli sources. Results suggest that simulated distribution of E. coli sources significantly improved after incorporating this enhanced data on E. coli sources into the model (R2 = 0.90) compared to the SWAT result without BST (R2 = 0.59). Scenario assessments indicate that wildlife contributions may remain significant despite land use change and urbanization, expected to mostly occur in agricultural and range lands. A combination of nonpoint source management measures, voluntary implementation of advanced treatment by wastewater plants where possible, and installation of aerators in the zone of impairment were demonstrated to be effective measures for restoring the recreation and aquatic life uses of the Arroyo Colorado. © 2018 Elsevier B.V. All rights reserved.

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

1. Introduction Urban and rural streams are increasingly contaminated with pathogenic bacteria. The Texas Commission on Environmental Quality (TCEQ, 2014) has found that over 67% of the streams on the 303(d) list are impaired by pathogenic bacteria, encompassing 27% of all stream segments in the state. Contamination of water by pathogenic organisms sourced to fecal waste is a major environmental concern (USFDA, 1995). Similar to water pollution by excess nutrients, water pollution by microbial pathogens can be caused by point and nonpoint sources (Dadswell, 1993; Garcia-Armisen and Servais, 2007; Jiang et al., 2007). Fecal Indicator Bacteria (FIB) are used as a proxy for pathogenic bacteria and are, therefore, used to evaluate the ability of a water body to support contact recreation uses. In freshwater, E. coli (Escherichia coli) is commonly used while enterococcus is typically used in tidal waters. Elevated concentrations of pathogenic bacteria signify an increased risk of contracting a gastrointestinal illness for those recreating in the water body (Neal et al., 1997). Point source fecal contamination of water normally results from direct entry of wastewater from a municipal treatment plant into a water body. It is difficult to identify nonpoint FIB sources, which can originate from animal production, land application of manure, humans through failing on-site sewage facilities (OSSF), or wildlife (Harmel et al., 2010; Garcia-Armisen and Servais, 2007; Parajuli et al., 2009; Pandey et al., 2012). Watersheds highly populated with humans, livestock, and wildlife are prone to water quality impairment due to high FIB concentrations. Terrestrial fate and transport processes and source loads of FIB can be complicated by human activities like urbanization or livestock grazing. Large quantities of FIB in stormwater runoff are perceived to be one of the most pressing issues in urban watersheds (Hardy and Koontz, 2010). Levels of FIB in stormwater can far exceed recreational water quality guidelines, often by several orders of magnitude, as stormwater picks up and transports a variety of chemicals and human and animal fecal wastes (Parker et al., 2010). OSSFs are designed to treat domestic wastewater using a septic tank for screening and pretreatment and a drain field where pretreated septic effluent is distributed for soil infiltration and final treatment by naturally existing microorganisms (Jeong et al., 2011). Poor installation or maintenance may cause OSSF failure, resulting in release of nutrients and pathogens into nearby water bodies (Ahmed et al., 2005; Siegrist et al., 2005). Grazing animals and wildlife can also negatively affect the quality of runoff and waterbodies with FIB contamination. The high concentration of FIB in waterbodies increase the risk of infection for people who use the water for drinking or various contact recreation purposes (Hubbard et al., 2004). Although it is well understood that high levels of livestock grazing can negatively impact stream water quality with elevated FIB concentrations (Gary et al., 1983), detailed data on the fate and transport of manure-borne FIB in soils, runoff, and streams remain insufficient (Jamieson et al., 2004; Harmel et al., 2010; Harmel et al., 2013). High populations of animals in preserved wildlife habitats may contribute to high FIB concentrations in streams (Stuart et al., 1971). Weiskel et al. (1996) found that waterfowl contributed 67% of the total annual loading of FIB along the east coast of the U.S. In Texas, non-avian wildlife, such as deer or feral hogs, are commonly found to be significant contributors of FIB to natural streams (Wagner and Moench, 2009). Process-based hydrologic/water quality models have advantages of simulating environmental outcomes in response to specific management practices (Jones et al., 2009; Guo et al., 2018a). Two commonly used models for FIB modeling are the Soil and Water Assessment Tool (SWAT) (Arnold et al., 1995) and Hydrological Simulation Program— Fortran (HSPF) (Bicknell et al., 1996). Paul et al. (2004) found in a HSPF study that simulated instream fecal coliform concentrations were most sensitive to the first-order decay coefficient in Salado Creek, implying that accurate simulation of FIB's fate and transport in the channels can be significant at the watershed scale. SWAT has an

165

advantage of explicitly partitioning microorganisms into adsorbed and non-adsorbed forms within a spatially explicit GIS framework over other process-based models so that FIB transport processes are explicitly simulated between water-bound and sediment-bound forms (Jones et al., 2009). Bacterial source tracking (BST) techniques can be useful in identifying FIB sources (Petersen et al., 2005). BST determines the sources of FIB in environmental samples using DNA fingerprints or other phenotyping methods. FIB monitoring data, which are often scarce in quantity, have been complemented by BST assessment (Baffaut and Benson, 2003; Parajuli et al., 2009). BST can provide a groundtruth of the overall contribution of FIB sources to FIB loads at a stream outlet or an estuary; thus, it is expected to help improve watershed-scale FIB analysis and estimates. There is a critical need to address contributing sources of bacterial contamination of water, properly assess the management of critical sources, and ultimately reduce FIB concentrations in impaired water bodies. Major gaps in knowledge exist in identifying bacterial sources. In particular, characterization of wildlife contributions and other “background” input sources of microbial pollution are highly uncertain (Jamieson et al., 2004). Historically, bacteria models have performed poorly because wildlife is typically underrepresented due to lack of data (e.g., Parajuli et al., 2009; Coffey et al., 2013). In this paper, we attempt to overcome this by incorporating BST results (Casarez and Di Giovanni, 2015) and edge-of-field monitoring data collected in Texas from cropland (Harmel et al., 2013) and urban settings (Jones et al., 2016) into the model. In this study, we attempt to overcome the knowledge gap in identifying sources of FIB for watershed simulation using BST and evaluate if conservation practices can be effective to control FIB loads to streams. The main goal of this study is to develop a SWAT model to identify significant sources of bacterial contamination and evaluate possible benefits of conservation practice implementation on urban, agricultural, rangelands, and wildlife land uses. Specific objectives are to: 1) identify and quantify significant sources of E. coli using GIS, BST, and edge-offield monitoring results, 2) evaluate a watershed model for E. coli concentration and other water quality variables, 3) evaluate the effects of land use changes and conservation practices on the transport of E. coli, sediment, and nutrients in the Arroyo Colorado watershed, Texas. 2. Methods and materials 2.1. Description of the SWAT bacteria model The SWAT model is a watershed-scale, physically-based, continuous simulation model (Arnold et al., 1995). SWAT includes explicit simulation of various terrestrial and instream processes, including agricultural management practices (Christopher et al., 2017; Scavia et al., 2017), plant growth (Guo et al., 2015; Wang et al., 2017; Feng et al., 2017), urban processes (Jeong et al., 2012; Her et al., 2017c), water impoundments (Bosch, 2008), and evaluation of various conservation practices, on water quantity and quality (Cibin et al., 2012; Montgomery et al., 2014; Keitzer et al., 2016). Using a daily time step, SWAT can complete long-term simulations of rainfall-runoff (Wang et al., 2014), soil erosion and nutrient and chemical transport (Niraula et al., 2011; Guo et al., 2018b), algal growth (Millican et al., 2008), and bacteria loads (Sadeghi and Arnold, 2002). The bacteria submodel of SWAT has been extensively tested and validated. Baffaut and Sadeghi (2010) found SWAT reasonably simulated bacteria transport with (Nash-Sutcliffe efficiency) NSE values varying between −6.0 and 0.73 in eight watersheds in the U.S. and France. Kim et al. (2017) suggest that implementing E. coli resuspension from sediment improves model performance in their tropical mountain watershed, but the original SWAT model does not simulate E. coli during dry seasons. SWAT incorporates GIS to partition a watershed into multiple subbasins based on the formation of stream networks. Hydrologic and water

166

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

quality processes and FIB yields are estimated for each hydrologic response unit (HRU). An HRU is a lumped land area within the subbasin comprised of unique land cover, soil, and management combinations. Driven mainly by hydrologic cycles, terrestrial processes directly contribute to the flow, sediment yield, and nutrient and FIB concentration in the main channel. 2.2. Study area The Arroyo Colorado, located in the Lower Rio Grande Valley of South Texas (Fig. 1), originates just above Mission, Texas and flows northeast for 56 km through Cameron and Willacy counties before draining into the Lower Laguna Madre. The main channel has a freshwater segment (Segment 2202) and a tidally influenced marine segment (Segment 2201) downstream of the Port of Harlingen. Both segments of the river are impaired by high FIB levels. Perennial flow is sustained mainly by flows from municipal wastewater treatment facilities. Irrigation return flows and urban runoff supplement the flow on a seasonal basis. The watershed is characteristic of the Western Gulf Coast Plain– Lower Rio Grande Valley ecoregion. Predominant land use in the watershed is agriculture (~1000 km2) with major crops grown including grain sorghum, cotton, sugar cane, and citrus. However, the watershed is rapidly urbanizing with population growth approaching 40% in the major cities in the watershed. Population of the Lower Rio Grande Valley was estimated to be 1.2 million in 2010 and is projected to grow to over 2.5 million by 2050. Urban growth in the watershed is expected to primarily occur in areas that are currently cultivated and will likely influence the region's water quality. The upper two-thirds of the Arroyo Colorado is underlain by alluvium consisting mostly of muds and silts deposited by the Rio Grande; the lower third is underlain by barrier island deposits of mostly sand with some silt and clay. The soils in the Arroyo Colorado watershed are clays, clay loams, and sandy loams. Most soil depths range from about 1.5–2.0 m. The Harlingen, Mercedes and Raymondville soil series

consist predominantly of clay soils with low permeability and the Hidalgo, Rio Grande and Willacy soil series consist predominantly of sandy loam and sandy clay loam soils with moderate permeability. Hydrologic soil groups B and D dominate the watershed. The climate of the Lower Rio Grande Valley is characterized by diverging temperate and tropical climates and is semi-arid and subtropical. Average annual precipitation in the area is 624 mm. About 51% of the rainfall occurs in the summer months (June–September). Mean monthly temperature ranges from 15.9 °C in January to 29.1 °C in August with the maximum of 34.1 °C and the minimum temperature of 10.6 °C. In the Arroyo Colorado, E. coli is the FIB for the non-tidal segment, and enterococcus is used in the tidal segment. According to the 2014 303(d) List (TCEQ, 2014), contact recreation use is considered impaired in all assessment units in both segments if the FIB concentrations exceed the water quality standard. For E. coli, this standard is a geometric mean of 126 colony forming units (cfu) per 100 ml while the standard for enterococcus is 35 cfu/100 ml. Generally, FIB concentrations have been consistent over the last decade with many individual samples containing concentrations higher than the applicable water quality standard. 2.3. Monitoring data Daily precipitation and temperature data for 2000–2013 were collected at three local weather stations (Fig. 1). Streamflow, sediment concentration, nitrogen and phosphorus yield, water temperature, and dissolved oxygen data were available at two streamflow gages operated by International Boundary and Water Commission (IBWC) at El Fuste Siphon south of Mercedes, TX (IBWC 08-4703.00) and south of Harlingen, TX (IBWC 08-4704.00). Therefore, the Arroyo Colorado watershed (ACW) model was calibrated for these variables at the Mercedes and validated at the Harlingen gage. Measured data for ammonia-N were available only at the Mercedes and PO4-P data existed only at the Harlingen gage. Thus, these N and P variables were calibrated at the location where measured data were available but were not validated due to the lack of data.

Fig. 1. Study area is the Arroyo Colorado watershed located in the Lower Rio Grande Valley of South Texas.

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

The Arroyo Colorado was monitored for E. coli concentration from 2001 to 2012. Samples were collected and analyzed every few months on an irregular basis (Data available at: http://arroyocolorado.org/, last accessed on September 7, 2017). A total of 89 sample data points were downloaded from the Arroyo Colorado Watershed Partnership among which 51 data points collected at the TCEQ monitoring site 13081 (FM1015 in Weslaco) were used for model calibration and 38 data points at the TCEQ monitoring site 13074 (the Port of Harlingen) were used for validation.

2.4. Nonpoint sources of E. coli In the ACW, major sources of nonpoint FIB (E. coli) pollution include humans, livestock cattle, and wildlife (Casarez and Di Giovanni, 2015). Identification of critical sources of E. coli is of great interest for pollution control but remains a challenge because of the large spatial variability of human and animal populations in the watershed. A considerable number of OSSFs are distributed in suburban municipalities, domestic livestock (i.e., cattle) on RNGE (rangeland-grasses) and RNGB (range-brush) land uses, and non-avian wildlife throughout the watershed. In addition, large populations of migratory birds and waterfowl inhabit wetlands, mudflats, and beaches seasonally near the Laguna Madre. Such various sources provide challenges in estimating E. coli contributions. The population of wildlife was estimated to identify E. coli sources in the ACW. Other unaccounted or unidentified sources of E. coli were defined as “Unknown” sources. Table 1 summaries nonpoint sources of E. coli identified in the ACW.

2.4.1. Livestock The 2012 National Agricultural Statistics Service (NASS) census data reports cattle, goats and sheep counts at the county level. Cattle grazing is commonly practiced across the ACW watershed. The grazing practice command in SWAT is often used to simulate nonpoint manure contribution to lands. Livestock and wildlife are defined based on stocking rates in RNGE (rangeland-grasses) and RNGB (range-brush). In general, cattle stocking rates (animal units/ha) in the ACW range from 1:10 (or 0.25 AU/ha on rangeland-grasses) to 1:15 (or 0.17 AU/ha on rangeland-brush) depending on brush density (Paschal, J., personal communication). On irrigated grassland, the stocking rate can be high as 1:1, one animal unit per acre (Ronnie Ramirez of TSSWCB, personal communication). Animal populations reported in the 2012 NASS for Hidalgo, Cameron, and Willacy counties were used to estimate livestock numbers in the watershed based on the areal fraction of the ACW in each of these counties assuming that the countywide animal stocking rate is same as the watershed area belonging to the county. Then, livestock in the watershed were further assigned to RNGE and RNGB based on the areal fraction of these land use types. A manure production rate (Rmanure) of 59.1 lbs./AU/day (NRCS, 1995) was used for cattle, based on the average moisture contents of 85% in manure. Manure

167

production rate (Pi,manure) was then estimated as follows: P i;manure ¼ Si  Rmanure

ð1Þ

where Si is the stocking rate of the animal type i. This equation gives 0.67 kg/ha/day for RNGB and 1.0 kg/ha/day for RNGE. Reported values for fecal coliform levels in manure range from 1.38E+06 to 1.65E +07 cfu/g manure (ASAE, 2002; Doyle et al., 1975; Reddy et al., 1981). Manure contributions by sheep and goats were negligible, particularly when compared to cattle manure production. Thus, the contribution by sheep and goats were not exclusively considered in the SWAT model but rather considered as part of unknown sources. 2.4.2. Pets In Texas, approximately 44% of households own dogs with the average household owning 1.6 dogs (AVMA, 2012). There are 114,424 households within the watershed and approximately 80,554 dogs are estimated to exist in the watershed. These dogs are likely concentrated in areas of higher human population densities. It was assumed that many dog owners do not collect their dog's waste, especially in rural areas, and thus, this waste represents a potential contributor to E. coli in the Arroyo Colorado watershed. 2.4.3. Non-avian wildlife Deer population was incorporated into SWAT to simulate manure production and E. coli levels in the deer manure. Other non-avian wildlife contribution was then simulated in SWAT using “background” constant values based on BST and Harmel et al. (2013) data. Texas Parks and Wildlife Department (TPWD, 2012) estimates that the average deer density between 2005 and 2010 is 0.041 deer/ha in the South Texas Plains. Assuming manure production of a deer is 25.1 kg/day (Moffit, 2009), manure production rate is estimated to be 0.154 kg/ha/day. Based on the recommended FIB population in manure by Wagner and Moench (2009), we used 1.39E+06 cfu/g manure for deer. E. coli loads from unknown sources in rangelands, croplands, and urban lands are considered as direct input to surface runoff or to the main channel. 2.4.4. Avian wildlife With 615 bird species documented in Texas, it has the most species of any state; however, most are migratory. Mild winters, abundant food and protection of wildlife refuges make the Texas Gulf Coast prime winter habitat for nesting migratory birds. In the Texas gulf coast region, waterfowl and migratory birds may contribute to bacterial contamination of water primarily through direct deposition to surface waters, predominantly in the winter, thus resulting in seasonality being exhibited in FIB data. Fall migration starts in August and generally goes through October for late species with their return usually beginning in March. About 250 bird species frequent the regions along the Arroyo Colorado, with roughly 70% being migratory. Due to the favorable climate and habitat, many species permanently reside in the Lower Rio Grande

Table 1 Nonpoint sources of E. coli in the ACW incorporated in the SWAT model. Source

Discharge type

Notes

References

Cropland

Indirect

Harmel et al. (2013)

OSSFs

Indirect

Pets Urban stormwater Wildlife and feral animals Livestock

Indirect Indirect

Runoff of wildlife fecal matter, eroded soil, organic matter and fertilizer from cropland during rainfall and irrigation events. Failing or non-existent onsite septic systems. Colonias are implemented with greater failure rates. Deposition of fecal matter onto land in urban HRUs. Runoff of wildlife fecal matter, eroded soil, organic matter and fertilizer from urban lawns and impervious surfaces Deposition of fecal matter onto urban and agricultural lands and directly into the water. Deposition of fecal matter onto land by grazers' manure production in RNGE/RNGB HRUs.

Direct and indirect Indirect

Jeong et al. (2011), McCray et al. (2005), Olmstead (2004), RSY (2001) and EPA (2002) AVMA (2012) Jones et al. (2016), Leisenring et al. (2012) and Leisenring et al. (2014) Moffit (2009), Smith (2002), Smith et al. (2010), TPWD (2012) and Wagner and Moench (2009) ASAE (2002), Doyle et al. (1975), NASS (2012), NRCS (1995) and Reddy et al. (1981)

168

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

Valley even though they are considered migratory (Shackelford et al., 2005). Aerial surveys (Smith, 2002) of winter waterfowl populations in the Lower Texas Coast suggest that over 300,000 migratory waterfowl overwinter along the Lower Laguna Madre. Assuming the area along the Lower Laguna Madre is about 104,330 ha, the population of birds is estimated at 2.9 bird/ha. Presumed habitats for waterfowl are wetlands and open water throughout the watershed, which are primarily found near the coast in Cameron County. A recent BST study conducted at a bay area in the Texas gulf coast (Smith et al., 2010) found that migratory birds are a major source of E. coli over humans or cattle in the streams draining to the Gulf of Mexico. Fresh manure production rate for a duck is 0.15 kg/day (Barker et al., 2002). Wagner and Moench (2009) suggest that fecal coliform density of chicken ranges between 1.2E+06 and 1.83E+09 cfu/g of feces. Alderisio and DeLuca (1999) reported 4.82E+06 cfu/g of fecal coliform from Canada goose droppings from 236 samples collected in 1996–1997. Birds produce up to 680 g of feces a day (French and Parkhurst, 2009). Assuming 85% moisture content in the fresh manure, and 340 g manure production (50% of reported max value) in the water, the daily total E. coli rate per animal is estimated 1.55E+08 cfu/bird/day. A conversion factor of 0.63 was used between fecal coliform counts and E. coli. 2.4.5. OSSFs and colonias OSSFs are commonly used decentralized domestic wastewater treatment facilities in rural, suburban residential and commercial lands where central wastewater treatment service is not available. Effluent from an OSSF septic tank is dripped into soils for natural treatment by microorganisms. Aged or ill-maintained OSSFs often fail allowing wastewater to be discharged without proper treatment (Jeong et al., 2011). Per the EPA (2002), nationwide OSSF failure rates range from 10 to 20%. A Texas study found an average failure rate of 12% for the state (RSY, 2001). In the AWC, approximately 57% of OSSFs are associated with very limited soils and should be prioritized for evaluation. All OSSFs have the potential for adverse environmental impact if they are improperly functioning but failing systems in particular pose an elevated risk of exacerbating river water quality with nutrients and FIB from human waste. Those closer to streams pose even greater risk. Approximately 15–20% of Texas' border residents live in communities called colonias. The term means settlement or neighborhood and is commonly used to refer to unincorporated rural and peri-urban subdivisions along Texas' border with Mexico (Olmstead, 2004). These communities are generally characterized by a lack of physical infrastructure such as water and wastewater, storm drainage and paved streets. Additionally, colonias are often

constructed on the poorest land with poor natural drainage and low soil permeability; therefore, these colonias are subject to greater OSSFs failure rates and higher pollutant loads in stormwater. To estimate the number of OSSFs within the watershed, 911 address data for Cameron and Hidalgo counties and Google Map data for Willacy County were obtained. Due to missing addresses, especially in Hidalgo County, parcel GIS layers were obtained from the counties and used in conjunction with satellite imagery to fill any address gaps. A final address layer was then generated which consisted of 117,344 addresses in the watershed. The final OSSFs were determined to be those addresses outside the estimated wastewater treatment facilities (WWTF) service area boundaries (Fig. 2), assuming any addresses within the WWTF boundaries were connected to the central treatment facilities. Table S1 summarizes OSSF densities in subbasins incorporated into the SWAT model. In total, 19,969 addresses (around 17%) are outside the estimated WWTF boundaries and likely rely on OSSFs, among which 2875 addresses are within designated colonia areas. Based on Texas' OSSF failure rate (EPA, 2002), the mean OSSF failure rate was assumed 12%. Based on the priority classification by the Rural Community Assistance Partnership, OSSFs located in the colonias having a health hazard (red colonias) were assumed to have a greater failure rate (70%). Conversely, a 30% failure rate (determined based on local expert knowledge) was assigned to areas having the lower priority ratings (non-red colonias). Septic tank effluent rate was presumed to be 0.227 m3/capita/day assuming the majority of OSSFs serve single home residential houses (McCray et al., 2005). E. coli concentrations in the effluent were assumed to be 6.3E+06 cfu/100 ml based on EPA (2002).

2.4.6. Urban stormwater The Lower Rio Grande Valley is one of the fastest growing regions in the nation. With this growth comes increased impervious cover and increased stormwater discharges. Stormwater often contains pollutants that can adversely affect water quality. The mean E. coli concentration in urban stormwater recently reported in the International Best Management Practices (BMPs) Database is 1847 cfu/100 ml based on 10 sampling locations in the U.S. (Leisenring et al., 2012; Leisenring et al., 2014). A monitoring study on local stormwater quality (Jones et al., 2016) found an average E. coli concentration of 11,857 cfu/100 ml in stormwater during 2011–2013 in the city of McAllen, which is higher than other reported values. Such monitoring data reflect the unique local stormwater quality; thus, they were incorporated into urban HRUs in the ACW SWAT model.

Fig. 2. Location of OSSFs, colonias, and WWTF service boundaries.

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

2.5. Point sources There are currently 25 active permitted wastewater discharges in the ACW of which 17 municipal and domestic facilities were identified as principal point source pollutant contributors (Table S2). For point source FIB loads, monthly self-monitored data were compiled from EPA's Integrated Compliance Information System (ICIS) databases (https://ssoprod.epa.gov/sso/jsp/ICIS_Login.jsp). Monthly discharge, sediment, nutrient, biological oxygen demand (BOD), and E. coli data for individual WWTFs were then aggregated and prepared as point source inputs at the subbasin scale. In the ACW SWAT model, these WWTFs contributed E. coli to 10 reaches in the main channel and one tributary channel with varying monthly average concentrations. Average daily flow from individual WWTFs ranged from as low as 0.083 megaliters per day (MLD) to as high as 26.5 MLD [equivalent to 7.0 million gallons per day (MGD)] or 20.7% of total WWTF flow to the Arroyo Colorado. Mean effluent total nitrogen (TN) concentrations ranged from 7.7 to 37 mg/l. Mean effluent ammonia-N (NH3-N) concentrations ranged from 0.07 to 13 mg/l and averaged 1.8 mg/l for all 17 facilities. The calculated average organic/phosphate phosphorus (PO4-P) concentrations ranged from 0.75 to 3.0 mg/l and averaged 1.2 mg/l. 2.6. Baseline model and scenarios A SWAT2009 model had been previously verified for flow, sediment and nutrient loads in the ACW (Kannan et al., 2010). The model was updated to SWAT2012 and recalibrated for E. coli simulation. This update involved: 1) transferring custom Fortran code snippets made for the ACW model from SWAT2009 to SWAT2012; 2) updating input files for SWAT2012; and 3) recalibrating flow, sediment, nutrients, and dissolved oxygen. A future land use change scenario was constructed to examine the effects of population growth and urbanization using the calibrated SWAT model and future land use map. The land use change scenario indicates that urban lands may grow nearly 10% while cropland (−7%) and rangeland (−2%) acres may decline (Table 2). Therefore, urban contribution of E. coli could become more dominant as urban stormwater discharge increases. Finally, three management scenarios were developed to assess the impacts of implementation measures recommended by local stakeholders. Future scenario 1 (FS1) focuses on actionable practices including land management on cropland, rangeland, and urban, and instream aeration. The second scenario (FS2) evaluates advanced wastewater treatment in conjunction with watershed management measures applied in FS1. The third scenario (FS3) evaluates restoration of spring flow and a detention reservoir in conjunction with FS1. Further details are provided in Section 2 of the Supplementary material. 2.7. Calibration strategies

169

combined with a modified stepwise single-pass method (Daggupati et al., 2015a, 2015b; Yen et al., 2016). The model was optimized using Integrated Parameter Estimation and Uncertainty Analysis (IPEAT, Yen et al., 2014). A multi-variate Dynamic Dimension Search algorithm (Tolson and Shoemaker, 2007) was used to screen parameter sets first, and then further parameter refinement was conducted manually by comparing observed and simulated monthly flow, sediment yield, ammonia-N, PO4-P, and water temperature during a 4-year period (2002–2005). Once the water quality (other than E. coli.) calibration was completed, the E. coli simulation of the model was calibrated using the optimization algorithm at the E. coli monitoring station in Mercedes (IBWC084703) and then validated at the monitoring station in south Harlingen (IBWC084704). The model parameters were then adjusted manually through investigating simulated E. coli statistics. In SWAT, the concentration of dissolved oxygen (DO) in streamflow is influenced by algal growth, nutrient dynamics, and BOD. Thus, no unique parameter set could be identified for DO, and the DO simulations were only evaluated based on measured values. Model parameters that were already optimized in the previous stage of calibration were not revised further at this stage. A total of 45 parameters were refined in the calibration (Table S3). Many FIB sources influence E. coli concentrations in the Arroyo Colorado. Point sources and nonpoint sources, like human wastewater discharges and domestic livestock, were identified based on reported discharge or local survey; however, E. coli loads contributed by wildlife were highly uncertain because their populations were unknown. Large populations of migratory birds, such as Canada Geese, are commonly found in the area during winter before flying north in early spring. Non-avian wildlife and waterfowl are highly populated in and near wetlands and waterbodies in the study area (Casarez and Di Giovanni, 2015). E. coli contributions by non-avian wildlife (other than deer) or background sources were configured in the SWAT model based on an edge-of-field monitoring study (Harmel et al., 2013), and results obtained from the Arroyo Colorado BST study (Casarez and Di Giovanni, 2015), as well as stream gage data. The process of incorporating the BST results and manually calibrating E. coli concentration involved iterative processes of SWAT parameterization, model simulation, and E. coli source allocation based on SWAT output. Distributed E. coli sources including OSSFs, wildlife, pets, and unknowns were directly written to the source code of the model. Therefore, the SWAT executable was recompiled during each iteration to provide updated distribution of E. coli sources in the watershed. Time-series plots comparing simulated and observed data were used to visually evaluate model performance. Model performance statistics such as NSE, the coefficient of determination (R2), and percent bias (PBIAS) were used as quantitative measures of model fit to supplement the visual evaluation. Geometric mean values were calculated to estimate mean E. coli concentrations and relative error was used to evaluate model performance on E. coli.

After updating to SWAT2012, the ACW SWAT model was recalibrated and validated using a multi-variable multi-site approach 2.8. Assumptions and limitation Table 2 Land uses dominant in the Arroyo Colorado watershed. Land covera

Agriculture Range-brush Range-grasses Urban Open water Wetland Total a

Present area

Projected area in 2025

Hectare

Percent

Hectare

Percent

98,919.9 27,150.4 4494.0 21,383.6 10,281.6 6906.7 169,136.3

58.5 16.1 2.7 12.6 6.1 4.1 100.0

87,278.9 23,488.2 3891.3 36,498.2 10,305.5 7674.1 169,136.2

51.6 13.9 2.3 21.6 6.1 4.5 100.0

Details of the land use survey can be found in Kannan (2012).

There are several assumptions made to efficiently link BST assessment to SWAT modeling in this study, including 1) SWAT performance can improve if the definition of E. coli sources is better defined through GIS, BST, and monitoring data, 2) E. coli sources defined in the model is reliable if simulated partitioning of sources match BST assessment, and 3) upland and in-stream bacteria processes available in SWAT2012 sufficiently describes the fate and transport processes of E. coli. Therefore, the proposed methodology or findings of the study do not apply in watersheds where no DNA fingerprinting data is available for source definition, or in areas where channel processes not available in SWAT2012, such as streambed resuspension of E. coli, are dominant in the fate and transport process of E. coli in channels.

170

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

3. Results and discussion 3.1. Model performance evaluation 3.1.1. Flow, sediment, and nutrient loads The calibrated ACW model simulated streamflow rates well over the entire flow regime. Overall model performance on streamflow simulation ranged between acceptable and excellent based on the evaluation criteria suggested by Moriasi et al. (2007) (Table 3). The model simulated intermediate and low flows very well; however, the performance was marginally satisfactory in predicting high flows (Fig. 3). Previous studies have found that SWAT under-predicts peak flows (e.g., Rostamian et al., 2008; Lin et al., 2015) after parameter optimization with commonly used statistic measures such as PBIAS or NSE as these performance indicators are not designed to fit extreme values. Therefore, it is likely that high peaks were attenuated to improve performance statistics during calibration. In addition, highly urbanized lands along the main channel between the cities of Mission and Harlingen provided significant amounts of flashy stormwater runoff, which may not be well represented by the model due to the simplified representation of urban catchments. For instance, storm sewers conveying runoff to tributary or main channels, which are not accounted for by SWAT, may greatly reduce flow travel time with minimal loss to infiltration. The underestimation of daily flow in the high-flow regime between zero and 20% exceedance may in part reflect this simplified urban representation by SWAT. Results indicate that annual runoff volume generated from urban lands is 157.5 mm, which is 27% and 23% greater than those of croplands and rangeland, respectively. Although many crops were irrigated, soils in croplands maintained relatively low soil moisture content due to high evapotranspiration rates in the ACW. SWAT uses a modified QUAL2E algorithm for simulating instream water quality processes (Neitsch et al., 2011). Water temperature is an elemental variable that influences instream nutrient dynamics and algal growth and usually provides seasonal variability in streamflow bio-chemical processes such as BOD, DO and FIB die-off and regrowth. Simulated water temperature showed good agreement with observed values (NSE = 0.86). Model results for sediment, ammonium-N and PO4-P yields compared well with observed average values, with low PBIAS for all variables in both calibration and validation periods with the exception of sediment yield (PBIAS N 15%). Such results imply that the central tendency of predicted water quality variables is reliable. However, model performance on overall curve-fitting varied between excellent and marginal. Visual inspection of the DO graph and PBIAS value indicates model performance was acceptable; however, the goodness-of-fit indicators suggest the temporal variability of simulated DO concentration was unsatisfactory. In general, the ACW model updated to SWAT2012 and recalibrated performed better for estimating

Fig. 3. Flow duration curves for simulated and observed daily mean flows at the Mercedes gage (RCH #6).

sediment yield and nutrient loads than the calibrated SWAT2009 used in a previous study by Kannan (2012).

3.1.2. E. coli loads A BST study conducted in the main channel of the Arroyo Colorado during 2014–2015 revealed that wildlife was the most dominant source of E. coli representing 68% of the E. coli isolates identified (Casarez and Di Giovanni, 2015). With the BST results incorporated, the ACW SWAT model performed very well in predicting the order-of-magnitude of E. coli concentration as indicated by geometric mean values (i.e. 5.8% lower than the geomean of the observed E. coli concentration values at RCH#6 and 9.5% greater than observed at RCH#10). In addition, the partitioning of the nonpoint/point sources simulated by SWAT was highly correlated to the BST result (R2 = 0.90). Before incorporating the BST, simulated distribution of E. coli among sources was weakly correlated to observed distribution based on the BST result (R2 = 0.59). However, the calibrated model did not produce reasonable performance indicator value for fitting temporal variability (R2 = 0.14 for combined data for both sampling locations). The short duration of the monitoring data, uncertainty in FIB sources definition in the model, and anthropogenic activities such as urban and agricultural management are likely contributing factors to the marginal performance of the model. Literature indicates that calibration of temporal variability in streamflow E. coli concentration remains a challenge. For example, Cho et al. (2010) reported NSE values for E. coli between −0.99 and 0.01 and Coffey et al. (2013) reported unsatisfactory to fair performance of SWAT with R2 = 0.03–0.35. Alternatively, order-of-magnitude approach is favored in E. coli modeling assessments (Pandey et al., 2016) and predictions of E. coli within 1 order of magnitude is suggested to

Table 3 Model performances on monthly flow and water quality. Type

Stream flowa (m3/s) Sediment yielda (ton/day) Water temp.a (°C) Ammonium-Na (ton/day) Phosphate-Pb (ton/day) DOb (mg/l) a b c

Calibration

Validation

Obsc

Simc

NSE

R2

PBIAS

Obsc

Simc

NSE

R2

PBIAS

4.71

4.34

0.74

0.81

7.79

7.89

6.71

0.7

0.69

15

58.74 25.27

69.52 25.5

0.44 0.86

0.58 0.86

15.51 −0.74

199.34 25.94

187.82 22.96

−1.78 0.43

0.44 0.84

5.78 11.48

0.83

0.84

0.62

0.73

−4.82

–

–

–

–

–

0.14

0.13

−1.3

0.37

4.55

–

–

–

–

–

6.83

7.48

−0.3

0.14

8.65

–

–

–

–

–

Calibrated at Mercedes and validated at Harlingen. No validation data were available for Ammonia-N. Calibrated at Harlingen. No validation data were available for Phosphate-P. Average monthly amounts.

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

provide credible information in instream bacterial assessments (Dorner et al., 2006). Seasonality in stream FIB concentrations are often found to be more or less correlated with water temperature (e.g., Cho et al., 2012). However, this seasonal trend was not found in either predicted or observed E. coli values in the ACW, which might be partially attributed to migratory birds that are abundant in the study area in the winter. Daily E. coli concentration plots suggest that point source E. coli loads exhibit little variability, often forming a background concentration (Fig. 4). In contrast, E. coli loads from nonpoint sources are highly variable, varying roughly between one to two orders of magnitude, providing a large quantity of E. coli to the stream, exceeding the compliance level of 126 cfu/100 ml for contact recreation required by Texas Administrative Code 309.3(h). Overall, significant nonpoint source contributions are attributed to flashy stormwater and sheet flows during storm events. 3.2. Identification of water quality sources SWAT modeling results indicate the predominant sources of nitrogen and phosphorus are cropland and point sources (i.e. WWTFs) in the middle of the watershed, which together contribute over 90% of the nutrient loads (Table S4). OSSFs in the highly populated suburban residential areas contributed to significant nitrogen loads. Total phosphorus (TP) loads were highest in the uppermost reaches of the watershed, while high TN loads were distributed

171

throughout the watershed. The predominant sources of sediment were cropland and rangeland erosion, predominately in the lower reaches of the watershed. Implementation of conservation measures in farming areas to reduce nutrient runoff from croplands, along with improved wastewater treatment and/or increased wastewater reuse are needed to reduce instream nutrient concentrations and improve DO conditions. E. coli loading was highest in the lower half of the watershed, where SWAT output suggests the primary E. coli sources are wildlife with smaller contributions from domestic animals (i.e., cattle and dogs), septic systems and point sources (Fig. 5). It should be noted that wildlife includes small mammals, such as rodents, raccoons, opossums, and skunks, as well as waterfowl and other wild birds whose densities can be very high in riparian zones and are likely to have direct deposition of fecal material into waterways. These small animals may also contribute to fecal loading in urban runoff. Although rain events can greatly increase E. coli levels in water, BST consistently identified wildlife as a major contributor for each month and at each monitoring station. Further, conservation measures implemented in both rural and urban settings to treat runoff from those land uses by enhancing infiltration, filtration, detention, and retention are effective means for reducing E. coli from wildlife as well. Despite significant E. coli contributions from wildlife, human fecal pollution still poses the greatest human health risk (Casarez and Di Giovanni, 2015). SWAT modeling results, which showed that 12% of the FIB originated from OSSFs and WWTFs, suggest that solutions to

(a) E. coli concentration at Mercedes (RCH#6)

(b) E. coli concentration at Harlingen (RCH#10)

(c) Log-log scale plot of E. coli concentration between measured and predicted (left) and geomean concentration values for Mercedes (# samples=53) and Harlingen (# samples38) (right) Fig. 4. E. coli concentration calibrated at the Mercedes gage and validated at the Harlingen gage.

172

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

loads in the main channel of the Arroyo Colorado to varying degrees. However, only future scenario 3 (FS3) showed sufficient E. coli reduction to meet state water quality standards (Table 4). Fig. 6 presents percent differences in the mass load of instream water quality variables with future baseline (FB) and future scenarios (FS1, FS2, FS3) relative to the current baseline (CB) scenario at the outlet of the non-tidal segment near the Port of Harlingen.

Fig. 5. Predominant sources of water pollutants in the Arroyo Colorado watershed.

failing OSSFs and sewage releases should be a priority to protect human health in the ACW. Also, a strong correlation was found between water yield and E. coli load and between TN and TP with R2 of 0.46 and 0.62, respectively; however, other combinations between flow, sediment, TN, TP, and E. coli did not show significant correlation. This suggests that conservation practices which reduce discharge to streams would be efficient to control E. coli load to channels, whereas practices that are specific to sediment or nutrient control may not be effective in reducing E. coli loads. 3.3. Land use change effects SWAT results for the future baseline (FB) scenario, developed from the calibrated current baseline (CB) scenario by modifying land use fractions according to the future land use map, suggest that stormwater runoff will increase by 13.8% and groundwater recharge will decline by 9.5% due to increased impervious cover. As a result, the total water yield from drainage areas is projected to increase by 4.3%. Other nonpoint source contaminant loads including sediment, TP, and E. coli increase by 16.8%, 2.1%, and 32.8%, respectively. In contrast, the nonpoint source TN load decreases by 11.1% because cropland and rangeland are converted to urban and fertilizer applications and livestock contributions are reduced. When these changes in nonpoint source loads are combined with increased discharges from point sources, the average daily flow in the main channel, along with TN, TP, and E. coli concentrations, would increase by 13.6%, 2.9%, 18.5%, and 11.6%, respectively at the Port of Harlingen (RCH#10), while sediment concentration is reduced by 30.3%. The large increases in TN and TP concentrations are largely attributed to the significant increase in point sources discharges. The significant increase in E. coli loads from upland nonpoint sources is contributed by urban land developments. However, model output indicates that increased volume of well-treated point source discharges would dilute the concentration of E. coli in the main channel. After landuse changes, the E. coli concentration at the outlet of RCH#10 was estimated to increase by only 11.6%, which is relatively less than the 32.8% increase in E. coli load from nonpoint sources. The probability of E. coli concentration exceeding the single sample maximum (N394 cfu/100 ml) increased to 11.0% (i.e. 0.9% greater than CB). However, the probability of exceeding the 7-day geomean standard (N126 cfu/100 ml) was reduced to 9.5% (i.e. 1.8% lower than CB), implying that the contribution of nonpoint sources carried by stormwater runoff will become more significant. 3.4. Management scenarios SWAT scenarios indicate that the implementation of management measures combined with increased wastewater discharge resulting from population growth will help address E. coli concentrations and

3.4.1. FS1: implementation of key management measures The FS1 scenario, as described in Section 2.2 of the Supplementary materials, attempts to control terrestrial nonpoint sources with conservation practices. The main goal of this scenario is to control E. coli loads by reducing runoff through implementing feasible conservation practices. With the implementation of conservation measures, the average daily E. coli concentration of 182 cfu/100 ml in FB is reduced to 151 cfu/100 ml, 12 cfu/100 ml lower than CB. Therefore, targeted implementation of key management measures was found to be effective at reducing E. coli concentrations. However, noting that E. coli concentrations in the CB scenario already exceed state standards for contact recreation, more aggressive FIB management was required. Statistical analysis indicates that there still exists a 10% probability of E. coli concentrations exceeding the single sample maximum concentration (Pmax N 394 cfu/100 ml) and 9.5% probability of exceeding 7-day geomean standard (Pgeo N 126 cfu/100 ml) under FS1. If Pmax is an indicator of controlling peak E. coli loads, Pgeo represents normal flow conditions when recreational activities are expected to happen. Therefore, from a practical point of view, implementation of conservation practices in FS1 is effective by reducing the probability of E. coli concentration exceeding the water quality standard from 9.5% to 7.8% (see Table 4). 3.4.2. FS2: implementation of advanced wastewater treatment The FS2 scenario evaluated voluntary utilization of enhanced wastewater treatment (using tertiary treatment mechanisms) and reuse by local WWTFs, in addition to conservation measures considered in FS1, to reduce the loading of pollutants to the Arroyo Colorado. Results suggest that enhanced treatment of point sources is highly effective at reducing nitrogen and phosphorus loads in the main channel. With additional treatment of nutrients, TN and TP concentration in the channel are reduced by 46% and 94%, respectively. As a result, average daily DO concentration slightly increases to 8.17 mg/l. However, this additional advanced wastewater treatment is predicted to have little effect on reducing E. coli concentrations. 3.4.3. FS3: restoration of Llano Grande Lake and spring The FS3 scenario introduces restoration of Llano Grande Lake in addition to the key conservation measures outlined in FS1. This restoration provided an added 3 million m3 of detention capacity to the main channel of the Arroyo Colorado along with restoration of a natural spring providing fresh water at the anticipated rate of 5500 m3/day to the stream at the location of Llano Grande Lake. Reduction in E. coli concentration was greater in FS3 (34%) due largely to the added detention storage, but also to the restored spring flow. The probability of exceeding the single sample maximum E. coli concentration (Pmax N 394 cfu/100 ml) was greatest in FB (11%) and lowest in FS3 (8%). The significant reduction in Pmax in FS3 is mainly attributable to the mixing and attenuation of E. coli with extended detention in the restored Llano Grande Lake. 4. Conclusions The impaired main channel of the Arroyo Colorado watershed was evaluated using the SWAT watershed model to assess E. coli load reductions resulting from implementation of various scenarios. Initially, SWAT was limited for analyzing the variety of FIB sources in the watershed; therefore, several FIB sources such as avian and non-avian

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

173

Table 4 Effect of land use change and conservation managements on water quality (Site 13074). Conditions

Flow (m3/s)

Sediment (mg/l)

TN (mg/l)

TP (mg/l)

DO (mg/l)

E. coli (cfu/100 ml)

Pmax⁎

Pgeo⁎⁎

Current baseline Future baseline Scenario 1 (conservation practices) Scenario 2 (low N/P load from PS) Scenario 3 (Llano Grande Lake)

6.70 7.61 6.60

137.19 95.65 72.66

4.92 5.07 4.13

0.70 0.83 0.63

8.13 7.93 8.12

162.91 181.76 151.01

9.9% 11.0% 10.2%

11.3% 9.5% 7.8%

6.60

72.66

2.73

0.05

8.17

151.01

10.2%

7.8%

6.66

36.92

1.69

0.14

8.26

120.00

8.1%

8.6%

⁎ Probability of exceeding single sample max (N394 cfu/100 ml). ⁎⁎ Probability of exceeding 7-day geomean (N126 cfu/100 ml).

wildlife, which contribute E. coli loads directly to the main channels, tributaries, or wetland waterbodies, stormwater E. coli loads, and OSSFs, were inputted to the FORTRAN source code of the model. Relative significance of E. coli sources was successfully incorporated from a BST analysis into SWAT using an iterative parameterization process. The incorporation of BST into SWAT modeling reduced uncertainty in simulated E. coli sources and helped refine critical source areas. As demonstrated in the case study of the Arroyo Colorado, data gaps in watersheds where wildlife contributes a significant amount of E. coli load to streamflow can significantly overcome by informing a simulation model with DNA foot-printing. The informed watershed model then can be used to support watershed management decisions. SWAT was able to predict E. coli concentration reasonably well in order-of-magnitude. Sources of model error included approximated agricultural lands representation, simplified irrigation canals, irrigation timing and amounts, lack of urban sewer lines representation, and simple representation of wildlife contributions. Measurement uncertainties of E. coli concentration are comparatively larger than other water pollutants, varying between 10% and 55% (McCarthy et al., 2008). Transport of E. coli from the landscape into river network relies on watershed's hydrological processes. Due to the dependency of E. coli transport to local hydrology, uncertainties that were transferred from hydrologic simulation to water quality variables and then to FIB output may have accumulated during the multi-year simulation, negatively influencing the accuracy of the output (Her et al., 2017a; Her et al., 2017b) in terms of predicting E. coli concentration, and moreover, making management decisions based on model output.

Fig. 6. Effects of land use change and conservation practices on instream water quality and E. coli concentration at the outlet of the non-tidal segment (Port of Harlingen; RCH#10). FB is the scenario for future landuse change (future baseline); FS1 is the upland sources management scenario (future scenario 1); FS2 is the implementation of advanced wastewater treatment in addition to FS1; and FS3 is the scenario that evaluates the effect of the restoration of spring flow from Llano Grande Lake in addition to FS1.

Predominant sources of sediment yield were cropland and rangeland. Croplands and point sources contributed the most to nitrogen and phosphorus yield, while non-avian wildlife was the biggest source of E. coli. The combined avian and non-avian wildlife contribution was estimated to be over 67% of the total E. coli load. The prevailing contribution of E. coli by wildlife makes it difficult to reduce instream E. coli concentrations to achieve compliance. Increase in wastewater discharges and urban lands due to population growth resulted in increased concentration of E. coli in the main channel. Implementation of key agricultural, WWTF, OSSF, urban and instream management measures, and increased inflow from wastewater treatment plants were found to be effective in reducing 7-day geometric mean E. coli concentration. Acknowledgements This study was supported by the Texas Commission on Environmental Quality [grant number 582-14-40161]. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.scitotenv.2018.08.097. References Ahmed, W., Neller, R., Katouli, M., 2005. Evidence of septic system failure determined by a bacterial biochemical fingerprinting method. J. Appl. Microbiol. 98 (4), 910–920. Alderisio, K., DeLuca, N., 1999. Seasonal enumeration of fecal coliform bacteria from the feces of ring-billed gulls (Larus delawarensis) and Canada geese (Branta canadensis). Appl. Environ. Microbiol. 65 (12), 5628–5630. Arnold, J.G., Williams, J.R., Maidment, D.R., 1995. Continuous-time water and sedimentrouting model for large basins. J. Hydraul. Eng. 121 (2), 171–183. ASAE, 2002. ASAE Standard D384: Manure Production and Characteristics. ASAE, St. Joseph, Mich. AVMA, 2012. U.S. Pet Ownership & Demographics Sourcebook 2012. The American Veterinary Medical Association (ISBN: 978-1-882691-29-6). Baffaut, C., Benson, V.W., 2003. A bacteria TMDL for Shoal Creek using SWAT modeling and DNA source tracking. Total Maximum Daily Load (TMDL) Environmental Regulations II. American Society of Agricultural and Biological Engineers. Baffaut, C., Sadeghi, A., 2010. Bacteria modeling with SWAT for assessment and remediation studies: a review. Trans. ASABE 53, 1585–1594. Barker, J., Hodges, S., Walls, F., 2002. Livestock manure production rates and nutrient content. North Carolina Agricultural Chemicals Manual. Bicknell, B.R., Imhoff, J.C., Kittle, J.L., Donigian, A.S., Johanson, R.C., 1996. Hydrological Simulation Program-FORTRAN User's Manual. vol. 11. USEPA, Athens, GA. Bosch, N.S, 2008. The influence of impoundments on riverine nutrient transport: an evaluation using the soil and water assessment tool. J. Hydrol. 355, 131–147. Casarez, E., Di Giovanni, G., 2015. Arroyo Colorado River Watershed Bacterial Source Tracking (Presented at the Arroyo Colorado Steering Committee, October 29, 2015). Cho, K.H., Pachepsky, Y., Kim, J.H., Guber, A., Shelton, D., Rowland, R., 2010. Release of Escherichia coli from the bottom sediment in a first-order creek: experiment and reach-specific modeling. J. Hydrol. 391, 322–332. Cho, K.H., Pachepsky, Y.A., Kim, J.H., Kim, J.W., Park, M.H., 2012. The modified SWAT model for predicting fecal coliforms in the Wachusett Reservoir Watershed, USA. Water Res. 46 (15), 4750–4760. Christopher, S.F., Tank, J.L., Mahl, U.H., et al., 2017. Modeling nutrient removal using watershed-scale implementation of the two-stage ditch. Ecol. Eng. https://doi.org/ 10.1016/j.ecoleng.2017.03.015.

174

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175

Cibin, R., Chaubey, I., Engel, B., 2012. Simulated watershed scale impacts of corn Stover removal for biofuel on hydrology and water quality. Hydrol. Process. 26 (11), 1629–1641. Coffey, R., Dorai-Raj, S., O'Flaherty, V., Cormican, M., Cummins, E., 2013. Modeling of pathogen indicator organisms in a small-scale agricultural catchment using SWAT. Hum. Ecol. Risk Assess. Int. J. 19 (1), 232–253. Dadswell, J.V., 1993. Microbiological quality of coastal waters and its health effects. Int. J. Environ. Health Res. 3, 32–46. Daggupati, P., Pai, N., Ale, S., Douglas-Mankin, K.R., Zeckoski, R.W., Jeong, J., Parajuli, P.B., Saraswat, D., Youssef, M.A., 2015a. A recommended calibration and validation strategy for hydrologic and water quality models. Trans. ASABE 58 (6), 1705. Daggupati, P., Yen, H., White, M.J., Srinivasan, R., Arnold, J.G., Keitzer, C.S., Sowa, S.P., 2015b. Impact of model development, calibration and validation decisions on hydrological simulations in West Lake Erie Basin. Hydrol. Process. 29, 5307–5320. https:// doi.org/10.1002/hyp.10536. Dorner, S.M., Anderson, W.B., Slawson, R.M., Kouwen, N., Huck, P.M., 2006. Hydrologic Modeling of Pathogen Fate and Transport. Environ. Sci. Technol. 40 (15), 4746–4753. Doyle, R.C., Wolfe, D.C., Bezdicek, D.V., 1975. Effectiveness of forest buffer strips in improving the water quality of 1246 TRANSACTIONS OF THE ASAEmanure−polluted runoff. Managing Livestock Wastes: Proc. 3rd Inter. Symp. on Livestock Wastes. ASAE, St. Joseph, Mich., pp. 299–302 (ASAE Publ. Proc−275). EPA, 2002. Onsite Wastewater Treatment Systems Manual. U. S. E. P. A. Office of Research and Development. Feng, Q., Chaubey, I., Cibin, R., Engel, B., Sudheer, K.P.P., Volenec, J., 2017. Simulating establishment periods of switchgrass and Miscanthus in the soil and water assessment tool (SWAT). Trans. Am. Soc. Agric. Biol. Eng. 60 (2012), 1621–1632. https://doi.org/ 10.13031/trans.12227. French, L., Parkhurst, J.A., 2009. Managing Wildlife Damage: Canada Goose (Branta canadensis). Garcia-Armisen, T., Servais, P., 2007. Respective contributions of point and non-point sources of E. coli and enterococci in a large urbanized watershed (the Seine river, France). J. Environ. Manag. 82 (4), 512–518. Gary, H.L., Johnson, S.R., Ponce, S.L., 1983. Cattle grazing impact on surface water quality in a Colorado front range stream. J. Soil Water Conserv. 38 (2), 124–128. Guo, T., Engel, B.A., Shao, G., Arnold, J.G., Srinivasan, R., Kiniry, J.R., 2015. Functional approach to simulating short-rotation woody crops in process-based models. BioEnergy Res. 8 (4), 1598–1613. Guo, T., Cibin, R., Chaubey, I., Gitau, M., Arnold, J.G., Srinivasan, R., Kiniry, J.R., Engel, B.A., 2018a. Evaluation of bioenergy crop growth and the impacts of bioenergy crops on streamflow, tile drain flow and nutrient losses in an extensively tile-drained watershed using SWAT. Sci. Total Environ. 613, 724–735. Guo, T., Gitau, M., Merwade, V., Arnold, J., Srinivasan, R., Hirschi, M., Engel, B., 2018b. Comparison of performance of tile drainage routines in SWAT 2009 and 2012 in an extensively tile-drained watershed in the Midwest. Hydrol. Earth Syst. Sci. 22 (1), 89–110. Hardy, S.D., Koontz, T.M., 2010. Collaborative watershed partnerships in urban and rural areas: different pathways to success? Landsc. Urban Plan. 95 (3), 79–90. Harmel, R., Karthikeyan, R., Gentry, T., Srinivasan, R., 2010. Effects of agricultural management, land use, and watershed scale on E. coli concentrations in runoff and streamflow. Trans. ASABE 53 (6), 1833–1841. Harmel, R.D., Wagner, K.L., Martin, E., Gentry, T.J., Karthikeyan, R., Dozier, M., Coufal, C., 2013. Impact of poultry litter application and land use on E. coli runoff from small agricultural watersheds. Biol. Eng. Trans. 6 (1), 3–16. Her, Y., Chaubey, I., Frankenberger, J., Jeong, J., 2017a. Implications of spatial and temporal variations in effects of conservation practices on water management strategies. Agric. Water Manag. 180, 252–266. Her, Y., Jeong, J., Bieger, K., Rathjens, H., Arnold, J., Srinivasan, R., 2017b. Implications of Conceptual Channel Representation on SWAT Streamflow and Sediment Modeling. J. Am. Water Resour. Assoc. 53 (4), 725–747. Her, Y., Jeong, J., Arnold, J., Gosselink, L., Glick, R., Jaber, F., 2017c. A new framework for modeling decentralized low impact developments using Soil and Water Assessment Tool. Environ. Model. Softw. 96 (Supplement C), 305–322. Hubbard, R.K., Newton, G.L., Hill, G.M., 2004. Water quality and the grazing animal. J. Anim. Sci. 82 (13_suppl), E255–E263. Jamieson, R., Gordon, R., Joy, D., Lee, H., 2004. Assessing microbial pollution of rural surface waters: a review of current watershed scale modeling approaches. Agric. Water Manag. 70 (1), 1–17. Jeong, J., Santhi, C., Arnold, J.G., Srinivasan, R., Pradhan, S., Flynn, K., 2011. Development of algorithms for modeling onsite wastewater systems within SWAT. Trans. ASABE 54 (5), 1693–1704. Jeong, J., Kannan, N., Arnold, J.G., Glick, R., Gosselink, L., Srinivasan, R., et al., 2012. Modeling sedimentation-filtration basins for urban watersheds using Soil and Water Assessment Tool. J. Environ. Eng. 139, 838–848. Jiang, S.C., Chu, W., Olson, B.H., He, J.W., Choi, S., Zhang, J., Le, J.Y., Gedalanga, P.B., 2007. Microbial source tracking in a small southern California urban watershed indicates wild animals and growth as the source of fecal bacteria. Appl. Microbiol. Biotechnol. 76 (4), 927–934. Jones, C.A., Wagner, K., Di Giovanni, G., Hauck, L., Mott, J., Rifai, H., Srinivasan, R., Ward, G., Wythe, K., 2009. Bacteria Total Maximum Daily Load Task Force Final Report. Texas Water Resources Institute. Jones, K.D., Garza, A., Balakrshnan, V., Zhang, J., Falade, A., Adeniyi, A., 2016. Texas Nonpoint Source Management Program McAllen Innovative Storm Water Detention Facilities. (Final Report v11-11-13, Institute for Sustainable Energy and the Environment). Texas A&M University, Kingsville. Kannan, N., 2012. SWAT Modeling of the Arroyo Colorado Watershed. Texas Water Resources Institute (TR-426, June 2012).

Kannan, N., Jeong, J., Srinivasan, R., 2010. Hydrologic modeling of a canal-irrigated agricultural watershed with irrigation best management practices: case study. J. Hydrol. Eng. 16 (9), 746–757. Keitzer, S.C., Ludsin, S.A., Sowa, S.P., Annis, G., Daggupati, P., Froelich, A., Herbert, M., Johnson, M.V., Yen, H., White, M., Arnold, J.G., Sasson, A., Rewa, C., 2016. Thinking outside the lake: how might Lake Erie nutrient management benefit stream conservation in the watershed? J. Great Lakes Res. 42, 1322–1331. Kim, M., Boithias, L., Cho, K.H., Silvera, N., Thammahacksa, C., Latsachack, K., et al., 2017. Hydrological modeling of Fecal Indicator Bacteria in a tropical mountain catchment. Water Res. 119, 102–113. Leisenring, M., Clary, J., Hobson, P., 2012. International Stormwater Best Management Practices (BMP) Database Pollutant Category Summary Statistical Addendum: TSS, Bacteria, Nutrients, and Metals. International Stormwater BMP Database. Leisenring, M., Clary, J., Hobson, P., 2014. International Stormwater Best Management Practices (BMP) Database Pollutant Category Statistical Summary Report TSS, Bacteria, Nutrients, and Metals. International Stormwater BMP Database. Lin, B., Chen, X., Yao, H., Chen, Y., Liu, M., Gao, L., James, A., 2015. Analyses of landuse change impacts on catchment runoff using different time indicators based on SWAT model. Ecol. Indic. 58, 55–63. McCarthy, D.T., Deletic, A., Mitchell, V.G., Fletcher, T.D., Diaper, C., 2008. Uncertainties in stormwater E. coli levels. Water Res. 42 (6), 1812–1824. McCray, J.E., Kirkland, S.L., Siegrist, R.L., Thyne, G.D., 2005. Model Parameters for Simulating Fate and Transport of On-Site Wastewater Nutrients. Ground Water 43 (4), 628–639. Millican, J.S., Back, J.A., McFarland, A., 2008. Nutrient bioassays of growth parameters for algae in the north Bosque river of central Texas. J. Am. Water Resour. Assoc. 44 (5), 1219–1230. Moffit, D., 2009. Documentation of nitrogen and phosphorus loadings from wildlife populations. Available at. USDA Natural Resources Conservation Servicehttp://www.nrcs. usda.gov/Internet/FSE_DOCUMENTS/nrcs143_013181.pdf. Montgomery, A.K., Wang, R., Brouder, S.M., Chaubey, I., Volenec, J.J., 2014. Water Quality Effects of Cellulosic Biofuel Crops Grown on Marginal Land. (In 2014 Montreal, Quebec Canada July 13–July 16, 2014). American Society of Agricultural and Biological Engineers, p. 1. Moriasi, D., Arnold, J., Van Liew, M., Bingner, R., Harmel, R., Veith, T., 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans. ASAE 50 (3), 885–900. NASS, 2012. Agricultural Statistics 2012. National Agricultural Statistics Service, United States Department of Agriculture. United States Government Printing Office, Washington, D.C. ISBN: 978-0-16-091518-5. Neal, K.R., Hebden, J., Spiller, R., 1997. Prevalence of gastrointestinal symptoms six months after bacterial gastroenteritis and risk factors for development of the irritable bowel syndrome: postal survey of patients. BMJ 314 (7083), 779. Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2011. Soil and Water Assessment Tool Theoretical Documentation Version 2009. Texas Water Resources Institute. Niraula, R., Kalin, L., Wang, R., Srivastava, P., 2011. Determining nutrient and sediment critical source areas with SWAT: effect of lumped calibration. Trans. ASABE 55 (1), 137–147. NRCS, 1995. Animal manure management. RCA Issue Brief #7 December 1995. Available at. https://www.nrcs.usda.gov/wps/portal/nrcs/detail/nj/technical/cp/cta/?cid= nrcs143_014211. Olmstead, S.M., 2004. Thirsty colonias: rate regulation and the provision of water service. Land Econ. 80 (1), 136–150. Pandey, P.K., Soupir, M.L., Haddad, M., Rothwell, J.J., 2012. Assessing the impacts of watershed indexes and precipitation on spatial in-stream E. coli concentrations. Ecol. Indic. 23, 641–652. Pandey, P.K., Soupir, M.L., Ikenberry, C.D., Rehmann, C.R., 2016. Predicting Streambed Sediment and Water Column Escherichia coli Levels at Watershed Scale. J. Am. Water Resour. Assoc. 52, 184–197. Parajuli, P.B., Mankin, K.R., Barnes, P.L., 2009. Source specific fecal bacteria modeling using soil and water assessment tool model. Bioresour. Technol. 100 (2), 953–963. Parker, J.K., McIntyre, D., Noble, R.T., 2010. Characterizing fecal contamination in stormwater runoff in coastal North Carolina, USA. Water Res. 44 (14), 4186–4194. Paul, S., Haan, P., Matlock, M., Mukhtar, S., Pillai, S., 2004. Analysis of the HSPF water quality parameter uncertainty in predicting peak in-stream fecal coliform concentrations. Trans. ASAE 47 (1), 69. Petersen, T.M., Rifai, H., Suarez, S., Stein, M.P., A. R., 2005. Bacteria loads from point and nonpoint sources in an urban watershed. J. Environ. Eng. 131 (10), 1414–1425. Reddy, K.R., Khaleel, R., Overcash, M.R., 1981. Behavior and transport of microbial pathogens and indicator organisms in soils treated with organic wastes. J. Environ. Qual. 10, 255–266. Ronnie Ramirez of TSSWCB, personal communication Rostamian, R., Jaleh, A., Afyuni, M., Mousavi, S.F., Heidarpour, M., Jalalian, A., Abbaspour, K.C., 2008. Application of a SWAT model for estimating runoff and sediment in two mountainous basins in Central Iran. Hydrol. Sci. J. 53 (5), 977–988. RSY, 2001. Study to Determine the Magnitude of, and Reasons for, Chronically Malfunctioning On-site Sewage Facility Systems in Texas. Reed, Stowe & Yanke, LLC, Austin, TX. Sadeghi, A.M., Arnold, J.G., 2002. A SWAT/microbial sub-model for predicting pathogen loadings in surface and groundwater at watershed and basin scales. (Paper presented at the Total Maximum Daily Load (TMDL)). Environmental Regulations, Proceedings of 2002 Conference. Scavia, D., Kalcic, M., Muenich, R.L., Read, J., Aloysius, N., Bertani, I., Boles, C., Confessor, R., DePinto, J., Gildow, M., Martin, J., Redder, T., Sowa, S., Wang, Y., Yen, H., 2017. Multiple models guide strategies for agricultural nutrientreductions. Front. Ecol. Environ. 15, 126–132. https://doi.org/10.1002/fee.1472.

J. Jeong et al. / Science of the Total Environment 648 (2019) 164–175 Shackelford, C.E., Rozenburg, E.R., Hunter, W.C., Lockwood, M.W., 2005. Migration and the Migratory Birds of Texas: Who They Are and Where They Are Going (Texas Parks and Wildlife PWD BK W7000-511 (11/05). Booklet, 34 pp.). Siegrist, R.L., McCray, J., Weintraub, L., Chen, C., Bagdol, J., Lemonds, P., Van Cuyk, S., Lowe, K., Goldstein, R., Rada, J., 2005. Quantifying Site-scale Processes and Watershed-scale Cumulative Effects of Decentralized Wastewater Systems, Project No. WU-HT-00-27 (Prepared for the National Decentralized Water Resources Capacity Development Project, Washington University, St. Louis, MO, by the Colorado School of Mines). Smith, E.H., 2002. Redheads and other wintering waterfowl. In: Tunnell Jr., John W., Judd, Frank W. (Eds.), The Laguna Madre of Texas and Tamaulipas. 177. Texas A&M University Press, College Station. Smith, A., Sterba-Boatwright, B., Mott, J., 2010. Novel application of a statistical technique, Random Forests, in a bacterial source tracking study. Water Res. 44 (14), 4067–4076. Stuart, D.G., Bissonnette, G.K., Goodrich, T.D., Walter, W.G., 1971. Effects of multiple use on water quality of high-mountain watersheds: bacteriological investigations of mountain streams. Appl. Microbiol. 22 (6), 1048–1054. TCEQ (Texas Commission on Environmental Quality), 2014. 2014 Texas Integrated Report for Clean Water Act Sections 305(b) and 303(d). Tolson, B.A., Shoemaker, C.A., 2007. Dynamically dimensioned search algorithm for computationally efficient watershed model calibration. Water Resour. Res. 43 (1), 1–16. TPWD (Texas Parks and Wildlife Department), 2012. White-tailed Deer (WTD) Federal Aid Report Charts and Tables (Report WL-127-R. Austin, TX).

175

USFDA, 1995. Sanitation of shellfish growing areas. National Shellfish Sanitation Program Manual of Operations, Part I. United States Department of Health and Human Services, Food and Drug Administration, Office of Seafood, Washington, D.C. Wagner, K., Moench, E., 2009. Education Program for Improved Water Quality in Copano Bay Task Two Report. Wang, R., Kalin, L., Kuang, W., Tian, H., 2014. Individual and combined effects of land use/ cover and climate change on Wolf Bay watershed streamflow in southern Alabama. Hydrol. Process. 28 (22), 5530–5546. Wang, R., Bowling, L.C., Cherkauer, K.A., Cibin, R., Her, Y., Chaubey, I., 2017. Biophysical and hydrological effects of future climate change including trends in CO 2, in the St. Joseph River watershed, Eastern Corn Belt. Agric. Water Manag. 180, 280–296. Weiskel, P.K., Howes, B.L., Heufelder, G.R., 1996. Coliform contamination of a coastal embayment: sources and transport pathways. Environ. Sci. Technol. 30 (6), 1872–1881. Yen, H., Wang, X., Fontane, D.G., Harmel, R.D., Arabi, M., 2014. A framework for propagation of uncertainty contributed by parameterization, input data, model structure, and calibration/validation data in watershed modeling. Environ. Model. Softw. 54, 211–221. https://doi.org/10.1016/j.envsoft.2014.01.004. Yen, H., White, M.J., Arnold, J.G., Keitzer, S.C., Johnson, M.V., Atwood, J.D., Daggupati, P., Herbert, M.E., Sowa, S.P., Ludsin, S.A., et al., 2016. Western Lake Erie Basin: softdata-constrained, NHDPlus resolution watershed modeling and exploration of applicable conservation scenarios. Sci. Total Environ. 2016 (569–570), 1265–1281.