Data to define the scenarios were compiled from effluent discharges from CWTPs in western Pennsylvania, receiving river flows at the point of discharge, and data on flow accretion with downstream travel distance. The reported plant effluent discharges and USGS gauge information on flows in the Allegheny River and Blacklick Creek are given in Table
1. The frequency distribution of flows during the average low-flow and high-flow months (August and March) were used in the scenarios that follow.
Commercial Treatment Plant Effluent
Five CWTPs voluntarily submitted effluent data, which included the discharge rate and bromide concentration from six outfalls to surface waters in western Pennsylvania for the period from February to December 2011 (
U.S. EPA 2011e), and are denoted A–F in the following discussion. Three of the plants also submitted additional data in response to information requests issued by EPA pursuant to Section 308 of the Clean Water Act (
U.S. EPA 2011c,
b,
d). The NPDES permits for these plants required monitoring and reporting of bromide concentrations, but did not impose an effluent limit. After September 2011, the plants stopped treating Marcellus Shale produced water, although they continued to accept other oil and gas produced water (
Hart Resources 2011). Subsequent administrative settlements with EPA require the installation of controls what will reduce effluent concentrations from these plants (
U.S. EPA 2013); therefore the bromide frequency distributions described subsequently represent historic practices.
Laboratory analysis reports were included with the data and identified the analytical methods, detection limits, dilution factors, and other information. Taken together this information was used to assess the quality of the data. Generally one sample was collected and analyzed per month, however, in some cases up to four samples were collected and analyzed.
The bromide concentrations from the outfalls ranged from 3.4 to
. As no single outfall matched this span of observed bromide concentrations, frequency distribution curves were developed for the individual outfalls (Fig.
1). Plant A had the three lowest effluent bromide concentrations in the dataset, but supplied no information on the volumetric discharge (Fig.
1). One of the plants used separate treatment processes for flowback water (Outfall C) and other produced water (Outfall E) and monitored each effluent separately (
U.S. EPA 2011d). This plant then combined Outfalls C and E with a stipulated maximum amount of stormwater prior to discharging to the receiving creek. For this paper, the concentration of the combined discharge was estimated as the flow-weighted average for the bounding cases of no stormflow and the maximum permitted stormflow (
or 15,196 gal. per day). The resulting estimated concentrations of bromide from the combined outfalls range from 325 to
with no contribution from stormflow, and
with the maximum permitted stormflow (Fig.
1). The concentrations from this plant’s two outfalls and literature data on flowback and produced water were used as described in a subsequent section to define characteristic effluent concentrations for flowback, other produced water and mixed Marcellus Shale and non-Marcellus Shale produced water.
The produced water outfall (E) had reported effluent bromide concentrations ranging from 766 to
. The highest and lowest concentrations, however, occurred after the plant stopped accepting Marcellus Shale produced water in September 2011. Outfall B bromide concentrations covered a similar range, but the two highest values (4,620 and
) for this outfall also were reported after September 2011. Fig.
1 shows the bromide frequency distribution curves for data collected prior to voluntary removal of Marcellus wastewater, while the entire data set is shown in Fig.
S1. The bromide concentration from the flowback water treatment train (Outfall C) ranged from 63 to
and was considered to represent flowback concentrations. The range of bromide concentrations from Outfall D (
) indicated treatment of a mixture of flowback and other produced water, based on comparison with the estimated combined effluent from Outfalls C and E (Fig.
1).
From these data, five categories of effluent bromide concentrations were developed and used in the following simulations (Table
2). Because of the use of hydraulic fracturing in the Marcellus Shale, continuing conventional oil and gas production in Pennsylvania, and the similarity of produced water from conventional and unconventional production wells, CWTP effluent may represent a mixture of produced water from conventional and unconventional hydraulically fractured formations. This situation was simulated for the complete produced water effluent, denoted by
Produced-C (Table
2). Given the sources of data available for this study, differentiating between conventional and unconventional produced water is not generally possible. For produced water treated after the exclusion of the Marcellus Shale produced water, however, the effluent was denoted
Produced-ME. The range of concentration for the mixed flowback and other produced water, as based on the combined Outfall C and E data, overlaps the range for Marcellus Shale flowback reported by Hayes (
2009) (
). Thus the concentration range denoted
Mixed/Flowback represents the mixed flowback and other produced water of the outfalls and the flowback reported by Hayes (
2009).
Lower Flowback denotes effluent concentration from Outfall C which was designed to treat flowback (
).
Lower Bromide denotes the bromide concentrations reported for Outfall A (
).
Scenarios
Construction of the steady-state scenarios began with the monthly discharge data. Based on the entire period of record, the median flow in Blacklick Creek was
, and the median flow in the Allegheny River was
. These were chosen for scenarios because they have two orders-of-magnitude difference in median flow, and both had permitted discharges from the outfalls (B and D) discussed previously. Variation in river/creek flow was included by simulating the entire period of record for the months with the highest (March) and lowest mean flows (August). Comparing the results for these two months gives the expected bounding bromide concentration values. The incremental flow accretion frequency curve (Fig.
2) was applied to each simulation so that all reaches gained or lost flow, based on the random selection from the flow accretion frequency distribution. The CWTP data were used to generate operational definitions for types of water (Table
2). Concentration distributions for these waters and the permitted discharge rates were used in the scenarios. Because the effluent discharge for each outfall was fixed at its permitted level, a separate set of simulations were performed in which the bromide concentration
Mixed/Flowback with its range of bromide concentration from 82 to
, was used along with 100, 50, 33, and 25% of the permitted discharge rate to vary the mass loading. A summary of all scenarios is given in Table
S3.
To augment the steady-state results, transient scenarios were constructed for releases that lasted for 24, 12, 8, and 4 h per day during a five-day working week. The effluent discharge rate was again fixed at the level permitted for the outfall, and the effluent bromide concentrations were chosen to represent Produced-ME, Mixed/Flowback, and Lower Flowback using the same concentration frequency distributions as the steady-state simulations. The scenarios were developed for releases to the Allegheny River and transport of 127 km from Franklin to Kittanning, Pennsylvania (Tables
3,
S4, and
S5). Simulations were made for both the 5th and 95th percentile bromide concentrations determined from the produced water input concentration data (Produced-ME). USGS gauge data on mean monthly flows were used to characterize the flow at Franklin, Parker, and Kittanning, Pennsylvania. Flows were estimated for a former station with incomplete data at Rimer, Pennsylvania, by linear interpolation of the frequency distributions at Parker and Kittanning. Similar to the steady-state simulations, variability in river flow, travel time, and dispersion were directly simulated by Monte Carlo-Latin Hypercube simulation for the low-flow and high-flow months (August and March). The flows at each location along the length of the river were drawn from their respective frequency distributions, based on a randomly selected cumulative probability from 0.0 to 1.0 applied to each station.