Temporal and Spatial Changes in Bromine Incorporation into Drinking Water–Disinfection By-Products in Pennsylvania

Data from the ICR were used as a baseline to compare temporal shifts in THM species in Pennsylvania. The ICR was a national effort to collect and analyze data on disinfection by-products in drinking water systems; its sampling period ran from July 1997 to the end of December 1998 (18 months) and included drinking water systems that served more than 100,000 people. Samples were analyzed and collected through USEPA-approved laboratories using standard methods (Wysock et al. 2002). The data collected for the ICR have been used extensively in previous literature (e.g., Regli et al. 2015; Li et al. 2014; Francis et al. 2010; Obolensky et al. 2007). THM measurements were collected quarterly. Nationally, 500 treatment plants representing 296 public water systems conducted monitoring. In Pennsylvania, 14 drinking water systems (all using surface water) participated in the ICR (McGuire et al. 2002). Microsoft Access was used to access the ICR database (USEPA 2000a) and extract relevant data. Species-specific THM data (for chloroform, BDCM, DBCM, and bromoform) for the maximum residence points measured were selected. The maximum residence points are the points in a drinking water system where the water age is greatest and thus THM concentrations are expected to be the highest.

More recent data were accessed through the Pennsylvania Department of Environmental Protection (PADEP) Drinking Water Reporting System (DWRS) (PADEP 2016). The THM species-specific data in the DWRS are used for compliance monitoring, which requires that samples be collected quarterly from locations in the drinking water system where THM formation is expected to be the greatest. Samples are collected by drinking water personnel and analyzed at PADEP-approved laboratories (PADEP 2017a).

Recent species-specific THM data are available for only 2012–2016; prior to 2012, utilities reported only the sum of the species as total THM. For the comparative analysis, the available recent data were divided into two sampling periods. These sampling periods were selected to replicate the duration and seasonal composition of the ICR data collection effort. Thus, each sampling period started in July (3rd quarter) and ran through the entirety of the following year, for a duration of 18 months. The first sampling period (SP 1) selected aligns with the beginning of species-specific THM data availability in the PADEP DWRS: 3rd quarter 2012 to 4th quarter 2013. The second sampling period (SP 2) aligns with the most recent data available at the time of this analysis: 3rd quarter 2015 to 4th quarter 2016. Sample sizes by distributions system and sampling period are shown in Table 1. Raw data by region and sampling period can be found in Table S1.

The ICR data set is the most comprehensive national analysis of source and finished water available; however, concern has been noted about the representativeness of the data set due to the unusually warm and wet climatic conditions nationally during 1998 (Roberson 2002). Further, flow conditions are critical to understanding bromide concentration changes because similar loads are associated with different concentrations when flow conditions vary. Major river flow conditions for sampling years were compared with the period of record to assess whether sampling years were representative. Daily average flow data for sampling periods were retrieved from the USGS National Water Information System (NWIS) for rivers of interest (USGS 2017). The gauges used were USGS 01576000 at Marietta, Pennsylvania (Susquehanna River), USGS 03075070 at Elizabeth, Pennsylvania (Monongahela River), USGS 03049500 at Natrona, Pennsylvania (Allegheny River), and USGS 03086000 at Sewickley, Pennsylvania (Ohio River).

Some Pennsylvania systems that were included in the ICR were removed from the analysis after the initial data identification. The analysis was limited to drinking water systems that used chlorine as a disinfectant since the speciation and quantity of DBPs is highly dependent on the disinfectant used. Two systems were removed because they no longer used chlorine for disinfection and thus comparisons between ICR and more recent TTHM data would be confounded by this significant operational change. One system was removed because recent data were not available from the PADEP DWRS. Two additional systems were removed because they could not be grouped regionally and thus would not contribute to the regional trend analysis. The remaining systems included in the ICR were grouped as using source waters either in the southwestern or southeastern portions of the state, corresponding with large population centers in the Philadelphia and Pittsburgh regions. These regions also have different hydrologic and climatic conditions. The eastern part of the state is part of the mid-Atlantic coastal plain [Hydrologic Unit Code (HUC) Region 2]; the state contains large parts of the Susquehanna (0205) and Delaware (0204) river basins. Western Pennsylvania is the headwaters of the Ohio River watershed (HUC Region 5), which is formed by the Allegheny River (0501) and the Monongahela River (0502). The final analysis included nine drinking water systems: five in the southwest and four in the southeast.

Concentrations below detection are reported as below detection limit or nondetect, resulting in a left-censored data set common for water quality data. Samples that returned a nondetect for any DBP species were treated as having a concentration of zero for that species [40 C.F.R. 141.24 (2013)]. For the ICR data, nondetects were reported as $-999$; these values were converted to zeros for the statistical analysis. In the more recent PADEP THM data, nondetects are reported in the database as zero.

For sampling events that did not have an entry for all four THM species (incomplete), the data were removed from the analysis for metrics relying on measures from all species, but were included in the individual species analysis (BDCM and DBCM). For instances in which there were replicate entries, all but one of the entries were removed. There were a total of 11 incomplete sample entries and one instance where a single entry was replicated 16 times.

The retrieved and the calculated data metrics were compared across regions and sampling periods. Since the data were left censored and not normally distributed, the median value was taken as the central tendency point of comparison for the statistically significant difference. The Mann-Whitney test with an alpha of 0.05 and a null hypothesis of no difference in the medians was used (Mann and Whitney 1947). The Kolmogorov-Smirnov (KS) two-sample test was also used to test for statistically significant differences in empirical distributions (Smirnov 1948). An alpha of 0.05 was used. This test served as an indicator for similarity of the full spread of the data, rather than just the central tendency.

THM formation and bromine incorporation were examined through TTHM, BDCM, and DBCM concentrations, bromine substitution factor (BSF), and percent bromination. TTHM is the current metric monitored for compliance (80 ppb) [71 FR 387 (2006)]. BSF, a normalized molar bromine incorporation factor, was developed by Obolensky and Singer (2005). BSF and related incorporation metrics have been widely used to characterize the extent of DBP bromination (Wang et al. 2017; Tian et al. 2013; Hua and Reckhow 2012; Francis et al. 2010; Hua et al. 2006; Rathbun 1996). An alternative to this molar-based metric is to consider the incorporation on a mass basis (consistent with the mass-based regulatory standard) (Mao et al. 2014; States et al. 2013; Zhang et al. 2011; Sohn et al. 2006). Percent bromination represents the mass fraction of THM species that contain at least one bromine atom.

Epidemiological studies report a relationship between chlorinated drinking water and bladder cancer occurrence (Salas et al. 2013; Cantor et al. 2010; Villanueva et al. 2007; USEPA 2005b; Villanueva et al. 2004), and regulatory limits have been based on these studies. In writing the D/DBPR, the USEPA used available data to set a standard for TTHM [71 FR 387 (2006)] as one of two indicator groups [THMs and haloacetic acids (HAAs)] intended to protect consumers from the wide variety of DBPs formed in chlorinated drinking water. Recently, the World Health Organization (WHO) reported that THM data are equivocal with respect to carcinogenicity, based on recent studies with negative results (WHO 2017). Because of this, the guidelines put forth by WHO regarding THMs consider the species to be noncarcinogenic and thus indicate that much higher levels of THMs are acceptable.

In the present work, four methods were used to estimate changes in THM-associated risk in Pennsylvania. The first approach utilizes species-specific cancer slope factors (CSFs), taken from the USEPA Integrated Risk Information System (IRIS) (USEPA 2016), which were established in the early 1990s. More recently developed non-IRIS CSFs have been described by the USEPA (2005a); however, both sets of CSFs were derived from the same primary research studies (NTP 1989, 1987, 1985), with the difference stemming from a change to assumptions regarding exposure. While the present work uses the IRIS values, minimal difference in the relative risk changes described subsequently would be expected through the use of the modified values. The species-specific CSF method assumes response-additive carcinogenic behavior, following the approach the USEPA uses for the risk characterization of other chemical mixtures (Hrudey and Charrois 2012; Wang et al. 2007; USEPA 2000b, c). Zhang et al. (2018), Wang et al. (2017), and Kolb et al. (2017) previously applied this method to estimate risk associated with changing THM speciation. This species-specific approach to estimating risk captures the widely reported differences between chlorinated and brominated DBPs (Yang et al. 2014; Richardson et al. 2007; Echigo et al. 2004). However, it is limited by its consideration only of THM species and not other DBPs that may be drivers of carcinogenic and noncarcinogenic risk in chlorinated drinking waters.

The second approach to characterize risk for the spatial and temporal comparisons is based on species-specific guideline values derived by WHO (2017). These guideline values are considered to be protective of any added toxicity risk over a lifetime of consumption and consider THM4 to be primarily noncarcinogenic. The WHO-recommended approach consists of summing the fractions computed by dividing the observed concentration of each THM4 species by its guideline value. The four summed fractions should remain less than or equivalent to 1 for human health protection (WHO 2017). The guideline values are 300, 60, 100, and 100 ppb for chloroform, BDCM, DBCM, and bromoform, respectively. This approach is limited by its consideration of only THM species and by its focus on noncancer outcomes, neglecting possible correlations between specific THM species and other DBPs and cancer outcomes.

The third approach is based on the results of a mammalian cell cytotoxicity assay (Plewa et al. 2008), as used in Salas et al. (2013). In this case, each species is weighted based on its respective ${\mathrm{LC}}_{50}$, the concentration necessary to kill half of the cells in a bioassay. The approach is entirely dependent on the cell toxicity of the THM4 species, and makes no assumption of carcinogenicity.

Finally, a fourth approach is based on the odds ratio (OR) method, which was originally incorporated into the economic analysis for the Stage 2 D/DBP rule (USEPA 2005b), and was recently modified and applied to consider effects of bromide by Regli et al. (2015). This method is intended to estimate increases in bladder cancer risk from chlorinated drinking water, using TTHM concentration as a surrogate. It does not incorporate species-specific considerations because the epidemiological studies on which it is based measured only TTHM and not the individual species. The relationship used is shown in Eq. (1) (Regli et al. 2015)

The OR approach, while unable to directly account for effects of changing bromine incorporation, does weight bromine-containing THMs more than chloroform since the brominated THMs contribute more per mole to the mass-based concentration of TTHM.

The four methods represent different interpretations of THM4 species association with negative health outcomes. Specifically, the OR and CSF approaches assume cancer as a health end point, while the WHO approach assumes noncancer end points and the cytotoxicity approach assumes no specific health end point. The focus of this work is not on the quantification of toxicity or risk based on any of these methods, but rather on the relative changes in these computed indicator values spatially and temporally in Pennsylvania. The fact that each approach assigns different weights to each of the THM4 species, and thus leads to a different interpretation of the relative impacts of the THM4 species, allows a broader consideration of the potential for temporal or spatial changes than relying on a single term with associated limitations. The relative weights for different species in the approaches are the cause for differences in perceived trends across space and time. The mass-based relative toxicity weights (normalized to BDCM toxicity) considered here are shown in Table 2.

The CSF method places all the weight on the brominated species of THM4 (DBCM > BDCM > bromoform; chloroform zero). The WHO method also weights the brominated forms more heavily, but chloroform receives a nonzero value (BDCM > DBCM = bromoform > chloroform). The cytotoxicity approach weights DBCM and bromoform heavily, but weights chloroform more heavily than BDCM (bromoform > DBCM > chloroform > BDCM). The OR approach is based on TTHM (a mass measurement) and thus gives each THM4 species an equivalent mass-based weight; this weights the brominated species higher than chlorinated on a molar basis due to the higher mass of bromine relative to chlorine. These approaches cannot be directly compared since they have different end points and target levels. In the present work each metric was normalized (by dividing it by its median value in the southeast during the ICR). This allows relative comparisons across space and time for each of the metrics.

To provide a national context, TTHM data collected during the ICR in 1997–1998 from southwestern and southeastern Pennsylvania were first compared with the national ICR TTHM data. The national median was not significantly different from the median in the southeast ($p=0.706$), while southwestern Pennsylvania represented an area with elevated TTHM during the ICR data collection. The median TTHM for southwestern Pennsylvania (66.5 ppb) was significantly higher ($p<0.001$) than the national (36.5 ppb) and nearly double the southeastern Pennsylvania median (34.2 ppb) during the 1997–1998 sampling. Elevated TTHM in the southwest during the ICR may have stemmed from differences in source water bromide and natural organic matter (NOM) or from different treatment choices at the utilities.

Fig. 2 presents results for TTHM for the ICR and the two recent time periods for the southwest and southeast. The southwest had a statistically higher TTHM median and different distribution than the southeast for all sampling periods. In the southwest, the median TTHM decreased significantly ($p=0.012$) from 1997–1998 (66.5 ppb) to 2012–2013 (SP 1 = 47.9 ppb), and then remained steady in 2015–2016 (SP 2 = 47.4 ppb). TTHM was unchanged between the ICR and more recent samples in the southeast (ICR = 34.2 ppb; SP 1 = 32.8 ppb, $p=0.222$; SP 2 = 35.0 ppb, $p=0.536$).

Examination of the full distribution yielded differences in the number of samples that exceeded the maximum contaminant level (80 ppb). In the 1997–1998 sampling, in the southeast, no samples exceeded 80 ppb, while 40% of the samples in the southwest exceeded this threshold. The decrease in TTHM observed for the southwest also led to a decrease in exceedance frequency to 15% (SP 1) and 18% (SP 2). The southeast showed a small increase in exceedance frequency (from 0% to 7% for both SP 1 and SP 2). These results provide insight into how the distribution of TTHM has changed in these systems; however, they represent point measurements, not locational running annual averages (LRAAs), and thus do not indicate compliance violations.

The decrease in TTHM in the southwest from the ICR to SP 1 and SP 2 is not surprising. It is likely that the implementation of the D/DBPR motivated plant operators to reduce TTHM from the elevated values observed during the ICR. In the southeast, where TTHM was already relatively low, operational changes may not have been needed.

The BSF was used as the primary metric of comparison for THM bromine incorporation. BSF is the mole-based metric representing the fraction of the THM4 halogen atoms that are bromine. This metric was of particular interest given observations of increased surface water bromide concentrations in southwestern Pennsylvania (States et al. 2013; Wilson and VanBriesen 2013). The results for all sampling periods and both regions are shown in Fig. 3.

The data collected in 1997–1998 do not indicate a significant difference in regional medians for BSF (southwest = 0.13, southeast = 0.07, $p=0.154$) or empirical distribution of BSF; this was also true for percent bromination by mass (data not shown). Thus, while the regions showed significant differences in TTHM during the ICR (with the southwest much higher), the regions had a similar extent of bromination within the THM. Considering temporal changes, in the southeast there was no significant difference in median or empirical distribution for either of the two subsequent sampling periods (SP 1 = 0.07, SP 2 = 0.08). However, in the southwest, SP 1 and SP 2 had statistically significantly higher BSF medians, 0.19 ($p=0.001$) and 0.17 ($p=0.039$), respectively, than in the southwest during the ICR and in the southeast during any time period. BSF in the southwest peaked during SP 1 and dropped slightly, although not to ICR levels, by SP 2. SP 1 had a statistically significantly different distribution when compared with the ICR and the SP 2 in the southwest.

Elevated bromine incorporation is associated with increased source water bromide, and thus was expected in the southwest, where recent reports point to increasing anthropogenic bromide loads altering concentrations of bromide in surface waters used as sources for these utilities (Good and VanBriesen 2016; States et al. 2013; Wilson and VanBriesen 2012). The decline in TTHM in the southwest suggests operational changes designed to remove total organic carbon (TOC) (e.g., enhanced coagulation). This would be expected to reduce THM formation (Edzwald and Tobiason 1999; Vrijenhoek et al. 1998); however, bromide is known to increase the rate and extent of formation of DBPs (Heeb et al. 2014; Hua and Reckhow 2012; Amy et al. 1994), and thus its presence may limit the usefulness of operational interventions such as enhanced coagulation.

Percent bromination by mass followed similar trends to BSF for the median values (data not shown); in the southwest all sampling periods were significantly different, while in the southeast no sampling periods were significantly different. During the ICR, the southwest (39.3%) and southeast (26.0%) had median percent brominations that were not significantly different ($p=0.196$). For SP 1, the southwest (51.9%) was significantly higher ($p<0.001$) than the southeast (25.0%). For SP 2, the southwest (48.0%) was also significantly higher ($p<0.001$) than the southeast (27.1%). In the southwest, all empirical distributions were significantly different from each other, while in the southeast no significant differences were seen. Percent bromination is not a commonly reported metric; however, it was recently suggested by PADEP as a useful indicator of concern for drinking water utilities. Handke (2009) suggested that percent bromination above 33% warrants concern. For the southeast, the percentage of samples exceeding this threshold was highest during SP 1 at 40%, and lowest during the ICR (33%), while for the southwest, increasing bromine incorporation has shifted this exceedance from its lowest during the ICR (56%) to the highest during SP 1 (84%).

Concentrations of two brominated species of THM, BDCM and DBCM, were also examined. The results are shown in Fig. 4. Also included in this plot are reference lines for concentrations of BDCM and DBCM associated with risk thresholds of ${10}^{-5}$ (6 and 4 ppb, respectively), based on the CSF values (USEPA 2017a, b).

For all sampling periods, BDCM and DBCM had significantly higher median concentrations in the southwest than the southeast. In the southeast, the median BDCM decreased for both SP 1 (6.0 ppb, $p=0.003$) and SP 2 (6.2 ppb, $p=0.010$) relative to the ICR (8.4 ppb); SP 1 and SP 2 were not different from each other ($p=0.512$). The distribution for BDCM from the ICR was significantly different from SP 1 and SP 2. DBCM did not change significantly in the southeast in any sampling period (ICR = 1.8 ppb, SP 1 = 1.7 ppb, SP 2 = 1.4 ppb) according to both central tendency and distribution tests. Thus, consistent with other metrics, these results indicate no important changes in the effect of bromide on THM formation and bromine incorporation in southeastern Pennsylvania.

In the southwest, there was no significant decrease for either SP 1 or SP 2 relative to the ICR medians for BDCM (15.4 ppb, SP 1 $p=0.588$, SP 2 $p=0.702$) or DBCM (7.2 ppb, SP 1 $p=0.577$, SP 2 $p=0.773$). For BDCM, SP 1 (14.5 ppb) was not significantly different ($p=0.807$) from SP 2 (14.0 ppb) by median or distribution test. For DBCM, SP 2 saw a significant decrease ($p=0.029$) in median relative to SP 1 (8 to 6.4 ppb). The lack of a significant decrease in BDCM and DBCM concentrations in the southwest is particularly notable given the decrease in TTHM over the same time periods (a nearly 20 ppb decrease in the median). Since bromoform generally forms in small quantities (and is frequently below detection), this indicates that the primary species contributing to the TTHM reduction in the southwest was chloroform, which is generally assigned the lowest toxicity weight of the THM4 species (WHO 2017; USEPA 2016). Its decrease, while relevant to ensuring the utilities are meeting the D/DBPR for TTHM, may not provide a good indication of changes in other DBP species of concern. As Francis et al. (2010) have noted previously, TTHM may be an inadequate surrogate when bromide concentrations are elevated in source waters.

Also, the observed concentrations for DBCM and BDCM for the southwest show greater variability (wider spread of data shown in Fig. 4) than in the southeast. These results may indicate more variability in bromide concentrations in southwestern surface waters or may reflect differences in treatment systems in drinking water plants in southwestern Pennsylvania.

The empirical distributions for response-additive CSF risk are shown in Fig. 5. For every sampling period, the response-additive risk calculated in the southwest was significantly higher than in the southeast (ICR $p=0.003$, SP 1 $p<0.001$, SP 2 $p<0.001$). In the southeast, the median risk in SP 2 represented a significant decrease from the ICR ($p=0.033$), which is notable since the TTHM levels did not change significantly. The median value from SP 1 was not significantly different from either of the other periods. None of the empirical distributions were significantly different from each other.

In the southwest, none of the median response-additive risks were significantly different (ICR–SP 1 $p=0.814$, ICR–SP 2 $p=0.769$, SP 1–SP 2 $p=0.118$), and none of the empirical distributions were significantly different from each other. This is particularly noteworthy because it demonstrates that, at least in areas with high surface water bromide, a substantial reduction in TTHM (20 ppb) does not necessarily mean that the species of most concern have been addressed. The current analysis determined that the reduction in TTHM was not proportionate across THM species, but rather was dominated by reductions in chloroform, with no change in BDCM and DBCM.

The World Health Organization describes a tolerable burden of disease of about ${10}^{-5}$ excess lifetime cancer risk (WHO 2017). The USEPA has also described ${10}^{-5}$ and ${10}^{-4}$ as cancer risk thresholds of interest [63 FR 69390 (1998)]. Assessment of national ICR data revealed that risk was found to mostly be between ${10}^{-5}$ and ${10}^{-4}$ [71 FR 387 (2006)]. Thus, percent exceedances for thresholds of ${10}^{-5}$ and ${10}^{-4}$ were also examined, shown as the vertical dashed lines in Fig. 5. By looking at these thresholds, we can assess whether target levels were exceeded more frequently, even when the overall distribution and its median do not show significant differences. At the ${10}^{-5}$ threshold, exceedance in the southeast dropped from the ICR (93%) to SP 1 (80%) and SP 2 (80%). No samples in the southeast exceeded the ${10}^{-4}$ threshold in the ICR and SP 1. In SP 2, 1% of the samples exceeded ${10}^{-4}$. In the southwest, exceedance of the ${10}^{-5}$ threshold increased from the ICR level of 86% to 100% in SP 1 and 99% in SP 2; however, exceedance of a target ${10}^{-4}$ risk threshold was highest during the ICR (14%) and dropped to 11% and 7% in SP 1 and SP 2, respectively. These results suggest increasing risk from brominated THM in the southwest during the more recent time periods.

For the approach described by the World Health Organization, the fraction sums were significantly higher in the southwest than in the southeast for every sampling period. In the southeast, the ICR median was significantly higher than the SP 1 median ($p=0.033$), but no statistically significant difference in empirical distribution was found. In the southwest, there was no significant difference found between medians of the ICR, SP 1, and SP 2 sampling periods for the WHO-based metric (ICR–SP 1 $p=0.371$, ICR–SP 2 $p=0.174$, SP 1–SP 2 $p=0.352$). None of the empirical distributions were different from each other.

The WHO approach states that the sum of each THM4 species concentration divided by its respective guideline value should remain at or below 1. The only sampling periods that had exceedance of this threshold were in the southwest during the ICR (7%) and SP 1 (5%). Thus, while the trends remained similar to the CSF approach (no significant improvement in the southwest since the ICR), the values seen in this case generally remained below the threshold of concern.

According to the cytotoxicity approach, which weighted chloroform more heavily than BDCM, the southwest was significantly higher than the southeast in each sampling period (ICR $p=0.001$, SP 1 $p<0.001$, SP 2 $p<0.001$). In the southeast, there was no significant difference between the medians of any of the sampling periods. In the southwest, the median in SP 2 was significantly lower than in the ICR ($p=0.025$). The median in SP 1 was not significantly different from either of the other sampling period medians.

Because the species-specific metrics previously described are not directly comparable due to their different interpretations of THM4 species toxicity, a normalized comparison of the metrics was completed. The OR approach was included here for comparison. OR uses only TTHM as an input and thus represents an even weight applied to all THM4 species. The comparison of the metrics normalized to their respective median values in the southeast during the ICR is shown in Fig. 6.

In the southwest, for OR, SP 1 and SP 2 were significantly lower than the ICR ($p=0.012$ and 0.016, respectively). This improvement is in contrast with the WHO and CSF approaches, which indicated no significant improvement in the southwest since the ICR. The cytotoxicity approach agreed with the significant reduction in the median (between the ICR and SP 2), though it did not indicate a significant improvement for SP 1. This difference stems from different weights assigned to the THM4 species (Table 2), specifically, the valuation of chloroform relative to the brominated species. The CSF approach, which places the largest weights on the moderately brominated species (BDCM and DBCM), not only saw no significant improvement in the southwest, but median values in the southwest also remained greater than twice the median ICR value in the southeast. OR, on the other hand, which weights each species equally, indicated a large improvement, which was related to the chloroform reduction in the southwest (and corresponding reduction in TTHM). Further, as was shown with the CSF and WHO results, even if metrics demonstrate similar trends, they may call for varying levels of concern. The results are largely dependent on the interpretation of the THM4 species health effects and the use of THM4 as a surrogate for unmonitored and unregulated contaminants that may cause toxicity and carcinogenicity in drinking water. Considering the elevated toxicity of brominated THMs, use of TTHM as a risk metric may be inadequate for areas with substantial bromide concentrations in surface waters.

Since TTHM formation is driven by a number of source water factors [generally increasing with increasing dissolved organic carbon (DOC), temperature, and bromide] that can be affected by changes in climate and flow, it is important to assess these conditions for the sampling periods. It has previously been reported that 1998 was a warm and wet year across the United States (Roberson 2002). However, river flows within Pennsylvania indicate the median flows in the Allegheny, Susquehanna, and Ohio rivers were lower during the ICR (1997–1998) than during the more recent sampling (2012–2016). In the Monongahela, river flow during the ICR was significantly lower than during SP 1 (2012–2013) but not significantly different than during SP 2 (2015–2016). The empirical cumulative distributions for yearly (calendar) average flows during the period of record of major rivers in Pennsylvania are shown in Fig. 7; years associated with the previously described sampling periods are labeled.

The flow analysis indicates that climatic conditions for the major rivers in southwest Pennsylvania during the ICR (1997–1998) would not have diluted bromide loads in surface waters. Rather, if loads remained similar, increased flows in the more recent periods would have diluted bromide, leading to lower bromide concentrations and less bromine incorporation in THM for the more recent periods.

The majority of year and river combinations remained within the middle 60% (20th percentile to 80th percentile). The only two exceptions were for lower flow conditions in 1998 on the Allegheny (13th percentile) and 2016 in the Susquehanna (9th percentile). Such low-flow conditions would likely have led to higher organic carbon or bromide concentrations in the source waters.

The lower flow in the Allegheny River in 1998 could have led to elevated brominated DBPs for several drinking water systems in the ICR collection period (and thus elevated TTHM). This might have made the change observed in TTHM (decreases) less significant, and it might have masked an even greater increase in bromine incorporation between 1998 and the present. Thus, the increased bromine incorporation into DBPs in southwestern Pennsylvania cannot be explained due to low flow. As previously reported, increasing anthropogenic bromide loads are the likely cause of changing THM bromine incorporation in drinking waters using these rivers as sources (Wang et al. 2017; Wilson and VanBriesen 2013; States et al. 2013; Wilson et al. 2013).

The unusually low flow in the Susquehanna River in fall 2016 due to drought conditions (Hess 2016) was observed to lead to increased bromide (National Water Quality Monitoring Council 2017; PADEP 2017b) and could have altered DOC. These changes would be expected to increase the concentration and bromine incorporation of the formed TTHM. However, as reported previously, in southeastern Pennsylvania, no significant changes in TTHM or bromine incorporation were observed between the 1997–1998 and 2012–2016 periods. It is possible that these unusually low flows masked what could have been a decrease in TTHM or bromine incorporation if flows were more typical.