Journal of Environmental Quality
TECHNICAL REPORTS Urban Pollutants
Modeling the Effects of Onsite Wastewater Treatment Systems on Nitrate Loads Using SWAT in an Urban Watershed of Metropolitan Atlanta Nahal Hoghooghi,* David E. Radcliffe, Mussie Y. Habteselassie, and Jaehak Jeong
O
nsite wastewater treatment systems (OWTSs) collect, treat, and release wastewater effluent from about 26 million homes, businesses, and recreational facilities in the United States (USEPA, 2002). However, failing or high density of OWTSs can be a source of nitrogen (N) inputs to both surface and ground waters. Conventional OWTSs are not designed to remove nitrate (NO3). Oakley et al. (2010) found that N loads were reduced by only 10 to 20% before discharge to the soil. From 2000 to 2010, the population in metropolitan Atlanta, GA, increased by ~24%, the second highest increase among the largest metropolitan areas in the United States. In this region, ~26% of homes are on OWTSs, and the number is expected to increase with population growth (MNGWPD, 2006). Gold et al. (1990) and Rich (2005) found that OWTSs were the main source of NO3 ground and surface water contamination. Onsite wastewater treatment systems are listed as potential sources of NO3–N in total maximum daily load (TMDL) reports of nutrient impairments by the Georgia Environmental Protection Division (Georgia EPD, 2013) and the USEPA (2007). In metropolitan Atlanta, we examined the effect of OWTS density on NO3–N concentrations under baseflow conditions in 24 small watersheds from 2011 to 2012 (Hoghooghi et al., 2016). We found that NO3–N concentrations increased linearly with increasing OWTS density above a threshold of ~75 OWTSs km−2. Watershed scale models such as the Soil and Water Assessment Tool (SWAT) can be used in TMDLs to predict water quality and develop a watershed management plan (White and King, 2003), but most models use a simplified approach for the effect of OWTSs (Geza et al., 2010). Recently, an algorithm created by Siegrist et al. (2005) was incorporated into SWAT version 2009 to simulate the impact of OWTSs and Jeong et al. (2011) described it in detail. In a case study in the Hoods Creek Watershed in North Carolina, the performance of the SWAT version 2009 biozone algorithm was tested by Jeong et al. (2011).
Abstract Onsite wastewater treatment systems (OWTSs) can be a source of nitrogen (N) pollution in both surface and ground waters. In metropolitan Atlanta, GA, >26% of homes are on OWTSs. In a previous article, we used the Soil Water Assessment Tool to model the effect of OWTSs on stream flow in the Big Haynes Creek Watershed in metropolitan Atlanta. The objective of this study was to estimate the effect of OWTSs, including failing systems, on nitrate as N (NO3–N) load in the same watershed. Big Haynes Creek has a drainage area of 44 km2 with mainly urban land use (67%), and most of the homes use OWTSs. A USGS gauge station where stream flow was measured daily and NO3–N concentrations were measured monthly was used as the outlet. The model was simulated for 12 yr. Overall, the model showed satisfactory daily stream flow and NO3–N loads with Nash– Sutcliffe coefficients of 0.62 and 0.58 for the calibration period and 0.67 and 0.33 for the validation period at the outlet of the Big Haynes Watershed. Onsite wastewater treatment systems caused an average increase in NO3–N load of 23% at the watershed scale and 29% at the outlet of a subbasin with the highest density of OWTSs. Failing OWTSs were estimated to be 1% of the total systems and did not have a large impact on stream flow or NO3–N load. The NO3–N load was 74% of the total N load in the watershed, indicating the important effect of OWTSs on stream loads in this urban watershed.
Core Ideas • We examined the importance of OWTSs on stream NO3–N load in the urban watershed. • The SWAT model sufficiently predicted the effect of OWTSs on NO3–N load. • The NO3–N load was the main component of the total N load in the watershed.
N. Hoghooghi and D.E. Radcliffe, Dep. of Crop and Soil Sciences, Univ. of Georgia, 3111 Carlton St., Athens, GA 30602; M.Y. Habteselassie, Dep. of Crop and Soil Sciences, Univ. of Georgia, 264 Redding Bldg., Griffin, GA 30223; J. Jeong, Biological and Agricultural Engineering Dep., Texas A&M Univ., 720 E. Blackland Rd., Temple, TX 76502. Assigned to Associate Editor Cole Machado. Abbreviations: 95PPU, 95% prediction uncertainty; CDN, denitrification exponential rate coefficient; GIS, geographic information system; HRU, hydrologic response unit; NLCD, National Land Cover Dataset; NPERCO, NO3 percolation coefficient; NSE, Nash–Sutcliffe efficiency; OWTS, onsite wastewater treatment system; SDNCO, denitrification threshold water content; SWAT, Soil and Water Assessment Tool; SUFI-2, Sequential Uncertainty Fitting algorithm version 2; TMDL, total maximum daily load; TN, total nitrogen; TRNSRCH, the fraction of transmission losses from main channel that enter the deep aquifer; WWTF, wastewater treatment facility.
Copyright © American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. 5585 Guilford Rd., Madison, WI 53711 USA. All rights reserved. J. Environ. Qual. 46:632–640 (2017) doi:10.2134/jeq2016.08.0322 Received 16 Nov. 2016. Accepted 21 Mar. 2017. *Corresponding author (
[email protected]).
632
The calibrated model performed well in predicting groundwater NO3–N concentrations (R2 = 0.76), and OWTSs contributed 25% of total N inflow to groundwater. Onsite wastewater treatment systems may fail because of poor maintenance, system design, or unsuitable soils, and they release nutrients and pathogens into the environment (USEPA, 2002). No state has directly measured its failure rate, and definitions of failure vary (Nelson et al., 1999). The USEPA reported estimated failure rates for each state and for Georgia; the percentage was a relatively low rate of 1.7% (USEPA, 2002). There are three types of OWTS failure. The first type occurs when the septic tank is not pumped regularly and fills with solids. This can cause solids to overflow into the drain field, clog the biomat that forms at the infiltrative surface in the drain field, and create ponding of effluent at the soil surface (USEPA, 2002). The second type occurs when the soil is unsuited for the OWTS application rate and/or the water table is too close to the infiltrative surface. This type of failure can also result in effluent coming to the soil surface (USEPA, 2002). If the failure is not severe enough to cause effluent to back up into the house and the effluent ponding is intermittent (only during wet periods), we suspect that this type of failure may not be repaired because the homeowners cannot afford the repair or installation of a new drain field. The third type of failure occurs when the drain field is not properly sited (unsuitable soil or the water table is too high) or the system is overloaded (due to an increase in the number of people in the home beyond that for which the system was designed) (USEPA, 2002). In this case, the effluent may not back up into the home or even surface, but proper treatment of the effluent does not occur within the unsaturated zone. By the time of this study, there was no publication on modeling the failing effect of OWTSs on stream flow and NO3–N load using the SWAT version 2012 model. In a previous article, Oliver et al. (2014) developed a SWAT version 2012 model simulation using the new OWTS algorithm to predict the effect of OWTSs on stream flow in the Big Haynes Creek Watershed in metropolitan Atlanta (Fig. 1). They found that total water yield increased by ~3% with the presence of OWTSs. The impact of OWTSs on NO3–N load was not included in their model, and neither were failing OWTSs. The objective of this study is to extend our previous work and estimate the effect of OWTSs on stream NO3–N load in the Big Haynes Creek Watershed using a SWAT version 2012 model that includes failing OWTSs.
Materials and Methods Watershed Description
The Big Haynes Creek Watershed has a drainage area of 44 km2, a mean elevation of 297 m, and is located in the Southern Piedmont physiographic region east of Atlanta in Gwinnett County (Fig. 1). The average annual precipitation is 1270 mm. The main land uses are 67% urban, 27% forest, and 6% hay and pasture. Most of the homes in this region are on OWTSs for domestic wastewater treatment. The daily stream-flow data and NO3–N concentrations (daily NO3–N load was calculated by multiplying daily flow and NO3–N concentration) were obtained from the USGS station number 02207385 from 2003 to 2014 on Big Haynes Creek, which became the outlet of the watershed we delineated.
SWAT Input Data The ArcSWAT 2012 interface was used to set up the model. To build the SWAT model, a digital elevation model (DEM), Soil Survey Spatial Tabular (SSURGO 2.2) soils data, National Land Cover Dataset (NLCD) land use data, OWTSs drain field locations from the Gwinnett County geographic information system (GIS) database, and Precipitation Regression on Independent Slopes Model (PRISM) weather data were used. The soils in the Big Haynes Watershed were typically high in clay content (on average 35–60%) (NRCS, 2007). There were four small reservoirs within the watershed, and a dam shape file for each reservoir was downloaded from the Georgia GIS Clearinghouse and added to the model (Georgia GIS Clearinghouse, 2013). The Big Haynes Watershed was divided into 29 subbasins, which provided sufficient detail to separate subbasins with high and low OWTS density. Each subbasin was divided into hydrologic response units (HRUs) with unique combinations of land use, soil, and slope (Arnold and Fohrer, 2005). Threshold values of 0, 10, and 30% of the subbasin area were applied for soil, land uses, and slope classes, respectively. The GIS layer of OWTS drain fields was merged with the NLCD land use map to create new OWTSs representing individual drain fields (100 m2 drain field−1). In our model, N inputs from OWTSs, fertilizer applied to lawns, poultry manure added to hay fields and pasture, and atmospheric deposition were included. The recommended annual rate of N fertilizer for bermudagrass [Cynodon dactylon (L.) Pers.] in Georgia is 170 kg N ha−1 (Georgia Turf, 2013), and we assumed that 70% of homeowners in this region applied fertilizer to their lawns according to a study by Osmond and Hardy (2004) on lawns in North Carolina in a region of the Piedmont with similar land use to the Atlanta area. It is a common practice in this area to apply poultry litter (a combination of manure and dry bedding material) to hay or pasture land. According to Watts et al. (2010), each broiler produces 1.5 kg litter yr−1. To calculate the broiler litter application rate in the Big Haynes Watershed, we assumed that all of the litter produced in the county would be applied to hay and pasture fields within the county. It is common to apply litter to nearby land because it is an inexpensive source of nutrients (Marshall et al., 1998). The amount of litter produced was estimated by multiplying the number of broilers by 1.5 kg. The application rate of 1584 kg of litter ha−1 yr−1 was estimated by dividing the total production by the total area of hay and pasture in Gwinnett County. The average annual rates of atmospheric deposition for NH4 and NO3 in precipitation (0.24 and 0.74 mg L−1, respectively) and dry deposition (0.59 kg NH4 ha−1 and 0.17 kg NO3 ha−1) were obtained from Clean Air Status and Trends Network data for Georgia (CASTNET, 2015). There were no point sources in the watershed; the only wastewater treatment plant closed in February 2003 (Georgia Environmental Protection Division, personal communication, 2014). In SWAT version 2012, the effect of OWTS is simulated using a biozone algorithm (Jeong et al., 2011). The biozone layer is a biologically active layer of microorganisms feeding on the organic matter of the septic effluent in the absorption system. In this algorithm, biozone clogging is the mechanism causing OWTS failure, and this results in surface ponding of OWTS effluent in a drain field (Jeong et al., 2011). The time required for a biozone layer to clog and fail typically takes >10 yr. Many of the homes in the Big
Journal of Environmental Quality 633
Fig. 1. The Big Haynes Creek Watershed boundary, monitoring site, and distribution of onsite wastewater treatment systems (OWTSs) in Gwinnett County, GA.
Haynes Watershed were >20 yr old, and records of when OWTSs were installed were available. However, it would be difficult to assign different ages to each OWTS in the model; the best that could be done would be to assign an age for each subbasin, and that would cause all of the systems in that subbasin to fail at the same time. As an alternative, we assumed that all OWTSs in soil hydrologic group D (NRCS, 2009), ~1% of the total number of OWTSs in the watershed, were in a state of constant failure. This percentage is consistent with the reported failure percentage for the Metropolitan North Georgia Water Planning District (MNGWPD, 2006). This approach had the advantage that failing systems were distributed across the watershed on the basis of soil hydrologic group and location of OWTSs. The number of permanent residents in each house was set to 3.0 according to data for Gwinnett County (US Census Bureau, 2015). The Big Haynes Watershed model was simulated from 1 Jan. 1999 to 31 Dec. 2014 on a daily time step. The first 4 yr were used as an equilibration period (or model warmup period) to reduce the dependence on initial conditions, and they were not included in the model analysis. This was followed by 6-yr 634
calibration (2003–2008) and validation (2009–2014) periods. The average precipitation was 1219 and 1256 mm in calibration and validation periods, respectively. The years 2007 and 2012 were the driest, with 798 and 965 mm precipitation, respectively. The USGS data included daily stream discharge and 71 measurements of NO3–N load during the calibration period and 49 measurements of NO3–N load during the validation period. For the USGS NO3–N load measurements, 33% were collected under baseflow conditions in the calibration period, and 35% under baseflow in the validation period. We used a baseflow separation technique to find an initial value of the baseflow recession constant (ALPHA_BF) (Arnold et al., 1995), and then relative adjustments were made during model calibration.
SWAT Model Calibration, Validation, and Uncertainty Analysis Calibration, validation, and uncertainty analysis of the model were conducted with SWAT Calibration and Uncertainty Program (SWAT-CUP) version 5 using Sequential Uncertainty Fitting algorithm version 2 (SUFI-2) (Abbaspour, 2013). The Journal of Environmental Quality
SUFI-2 tool combines optimization with uncertainty analysis. Uncertainty in parameters in SUFI-2 accounts for all sources of uncertainty, such as conceptual model, input parameters, driving variables, and measured data. Uncertainties in the parameters result in the model output uncertainties, which are quantified by the 95% prediction uncertainty (95PPU) band between the 2.5 and 97.5% levels of the cumulative distribution of an output variable using Latin hypercube sampling. At the beginning of the calibration process, SUFI-2 evaluates a large number of parameters with wide ranges in values and then decreases the parameter number and value ranges in steps while monitoring parameter sensitivities and the p- and r-factors. The p-factor is the fraction of measured data bracketed by the 95PPU band and ranges from zero to one. A p-factor of one indicates 100% bracketing of the measured data within model prediction uncertainty. The r-factor is the ratio of the average width of 95PPU band and the standard deviation of the measured data and ranges between zero and infinity. The strength of the model calibration and validation is judged on the basis of these two indices. A perfect fit between measured and simulated data can be achieved with a p-factor of one and an r-factor of zero (Abbaspour, 2013). Goodness of fit and model uncertainty are assessed by the objective function. The SUFI-2 allows usage of 10 different objective functions, including the commonly used Nash– Sutcliffe efficiency (NSE) coefficient, which emphasizes peak flows (Nash and Sutcliffe, 1970). The SWAT model parameters used in our calibration procedures were identified through a literature review (Santhi et al., 2001; White and Chaubey, 2005; Abbaspour et al., 2007; Bekele and Nicklow, 2007; Geza et al., 2010; Lam et al., 2010; Moriasi et al., 2012; Oliver et al., 2014). Parameters controlling stream flow were calibrated first, and N parameters were calibrated second. Stream-flow parameters and their calibrated ranges were included in the second step, but their ranges were not changed to include “equifinality” of stream flow in the NO3–N load (Beven, 2006).
Results and Discussion
Model Calibration, Validation, and Uncertainty Analysis The calibration process for the Big Haynes Watershed started with 22 hydrologic parameters. The parameters, fitted values, and maximum and minimum values are listed in Table 1. The values of runoff curve no. 2 (CN2) (SCS, 1972), available water capacity of the soil layers (SOL_AWC), the exponential decay factor for groundwater flow to the stream (ALPHA_BF), the soil saturated hydraulic conductivity (SOL_K), and reservoir parameters (RES_RR, RES_ESA, and RES_EVOL) are expressed as a percentage change from default values. After 1000 simulations, the most sensitive flow parameters were identified and are listed in Table 2 in order of decreasing sensitivity. The most sensitive flow parameters were runoff curve no. 2(CN2), the fraction of transmission losses from main channel that enter the deep aquifer (TRNSRCH), effective hydraulic conductivity of the alluvium in the main channel (CH_K2), the Manning’s roughness coefficient of the main channel (CH_N2), the groundwater reevaporation coefficient (GW_REVAP), the effective hydraulic conductivity of the alluvium in the tributary channel (CH_K1), the threshold depth of water in the shallow aquifer required for return flow to occur (GWQMN). Parameter values must be considered after a
Table 1. Stream-flow and nitrogen parameters and fitted, minimum, and maximum values used for calibration period using the Soil and Water Assessment Tool Calibration and Uncertainty Program (SWAT-CUP). Parameter† Flow parameters CN2 ALPHA_BF GW_DELAY GWQMN ALPHA_BNK CH_K1 CH_K2 CH_N1 CH_N2 EPCO ESCO GW_REVAP GW_SPYLD RCHRG_DP REVAPMN SOL_AWC () SOL_K() SURLAG TRNSRCH RES_RR RES_EVOL RES_ESA NO3–N parameters CDN SDNCO NPERCO RS1 RS3 RS4 HLIFE_NGW COEFF_NITR COEFF_DENITR BC1_BSN BC2_BSN BC3_BSN
Method‡ Fitted value Min. value Max. value r r v v v v v v v v v v v v v r r v v r r r
−0.16 0.017 220 6.5 0.89 290 29 0.46 0.054 0.18 0.17 0.069 0.31 0.067 42 0.27 −0.69 21 0.18 0.20 0.70 −0.11
v v v v v v v v v v v v
0.64 0.42 0.52 0.67 0.72 0.07 120 0.30 0.0065 0.99 1.4 0.22
−0.20 −0.10 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.00 0.00 0.02 0.00 0.00 0.00 −0.20 −0.80 1.00 0.00 −0.50 −0.20 −0.20 0.00 0.00 0.00 0.15 0.00 0.001 0.00 0.00 0.00 0.10 0.20 0.20
0.20 0.10 500.00 1000.00 1.00 300.00 30.00 0.50 0.15 1.00 1.00 0.20 0.40 0.20 500.00 0.40 0.80 24.00 1.00 1.00 1.00 1.00 3.00 1.00 1.00 1.82 1.00 0.10 200.00 3.00 1.00 1.00 2.00 0.30
† EPCO, plant uptake compensation factor; ESCO, soil evaporation compensation factor; GW_SPYLD, specific yield of shallow aquifer; RCHRG_DP, deep aquifier percolation fraction; SOL_AWC (), available water capacity of the soil layer; SOL_K, saturated hydraulic conductivity; SURLAG, surface runoff lag coefficient; RES_RR, average daily principal spillway release; RES_EVOL, volume of water needed to fill the reservoir to the emergency spillway; RES_ESA, reservoir surface area when the reservoir is fulled to the emergency spillway; RS1, local algal settling rate in the reach at 20°C; RS3, benthic source rate for NH4–N in the reach at 20°C; RS4, rate coefficient for organic N settling in the reach at 20°C; HLIFE_NGW, half-life of NO3–N in the shallow aquifer; COEFF_NITR, nitrification rate coefficient; COEFF_DENITR, denitrification rate coefficient; BC1_BSN, rate constant for biological oxidation of NH3; BC2_BSN, rate constant for biological oxidation of NO2 to NO3; BC3_BSN, rate constant for hydrolysis of organic N to NH3. See Table 2 for parameters not defined here. ‡ r means the existing parameter value is multiplied by (1 + a given value) and v means the existing parameter value is replaced by a given value.
calibration process because incorrect values may result in unrealistic simulations. For instance, deep aquifier percolation fraction (RCHRG_DP) and TRNSRCH are parameters that govern processes that result in a loss of water from the system to the deep
Journal of Environmental Quality 635
Table 2. The most sensitive flow and NO3–N load parameters and their P-values at the outlet of the Big Haynes Watershed from sensitivity analysis in the Soil and Water Assessment Tool Calibration and Uncertainty Program (SWAT-CUP), ranked from most sensitive to least sensitive. Parameter CN2 TRNSRCH CH_K2 CH_N2 GW_REVAP CH_K1 GWQMN GW_DELAY CH_N1 ALPHA_BF REVAPMN CDN NPERCO SDNCO
Description Parameters sensitive to flow
Rank
Runoff curve number II Fraction of transmission losses from main channel Effective hydraulic conductivity of the alluvium in the main channel Manning’s roughness coefficient of the main channel Groundwater reevaporation coefficient Effective hydraulic conductivity of the alluvium in the tributary channel Threshold depth of water in the shallow aquifer Groundwater delay time Manning’s roughness coefficient of tributary channels Exponential decay factor for groundwater flow to the stream Threshold depth of water in the shallow aquifer Parameters sensitive to NO3–N Denitrification exponential rate coefficient Nitrate percolation coefficient Denitrification threshold water content
aquifer. As presented in Table 1, the fitted values in our models for these parameters were small and appropriate for a “gaining stream.” Values for other sensitive flow parameters were reasonable and not at the upper or lower limit of recommended values. The model fit was satisfactory (Moriasi et al., 2015) for streamflow calibration, with a daily NSE of 0.61, R2 of 0.64, p-factor of 0.70, and r-factor of 0.64. For the validation period, the model fit was slightly better, with a NSE of 0.66, R2 of 0.68, p-factor of 0.80, and r-factor of 0.77 (Table 3). The log-scale plots of the observed stream flow, the model best fit, and the 95PPU band calibration and validation periods are shown in Fig. 2a and 2b. The wide range in stream flow that occurred from 2003 to 2014 is apparent, with a drought extending from 2006 to 2009. The higher stream flow in 2013 than in 2014 is apparent. For stream flow, the uncertainty band was narrow, and
