Impacts of Hurricane Harvey on drinking water quality in two Texas cities

Hurricane Harvey devastated large parts of the US Gulf Coast in 2017, and its floodwaters posed a number of threats to the environment and human health. In particular, an estimated 375 000 Texas residents experienced issues related to the provision of safe drinking water at the peak of the hurricane. In this study, physical, chemical, and biological water quality was monitored in two drinking water systems in Texas following Hurricane Harvey to understand the relationship between water quality parameters and changes in the drinking water microbiota. Results show initial surges in total organic carbon, trihalomethanes, and bacterial concentrations in finished water immediately following Hurricane Harvey. Microbial community analyses highlight the dependence of the distribution system microbiota on distribution system characteristics (i.e. water age), raw water quality, and disinfectant residual, among other factors. While both systems had problems maintaining disinfectant residual in the weeks following the hurricane, stabilization of water quality occurred over time. Overall, this study provides an understanding of the challenges associated with maintaining drinking water quality in the wake of a natural disaster and can be used to better prepare drinking water managers and engineers to combat changing weather patterns in the future.


Introduction
On 25 August 2017, Hurricane Harvey made landfall in Rockport, Texas, as a Category 4 hurricane [1]. Over the following week, it delivered over 127 cm of rain to the Gulf Coast and caused over $125 billion in damage, making it the costliest hurricane to hit the US mainland (tied with Hurricane Katrina in 2005) [2,3]. The floodwaters posed several threats to the environment and human health, one of which was the provision of high-quality drinking water. The Texas Commission on Environmental Quality (TCEQ) reported that 61 public-water systems (PWS) were rendered inoperable at the height of the storm, and more than 200 systems had to issue boil-water notices (BWNs) [2]. Smith et al [4] reported that potable water shortages in Texas following Hurricane Harvey resulted in instances of bottled water being sold for as much as $99. Under normal conditions, PWS face several challenges in their ability to consistently provide safe drinking water to their customers, including limited financial resources, aging infrastructure, and high operator turnover [5]. Moreover, sudden changes in the quality of source water, in conjunction with flooding, can be detrimental to plant performance. Widespread flooding and power outages at treatment plants and pump stations were among the many challenges faced by water treatment plant operators in the aftermath of Hurricane Harvey.
For these reasons, TCEQ suspended the enforcement of several water treatment requirements set forth in Texas Administrative Code, chapter 30, on 2 September 2017 [6]. Suspensions included the requirement for a 14 d public notice prior to changing between free chlorine and chloramine and for flushing dead-end mains on a monthly basis as long as minimum disinfectant residuals (0.2 mg l −1 free chlorine or 0.5 mg l −1 chloramine) were met; these suspensions were implemented to assist PWS in recovering from Hurricane Harvey without compromising human health.
As extreme weather events are predicted to occur with increased frequency and severity in the future [7], PWS are expected to face extensive challenges in terms of water quality management in the coming years. Changes in drinking water quality due to pollutant cycling during extreme weather events include impacts to sediment loading, chemical composition, total organic carbon (TOC) content, and microbial quality [8]. Kapoor et al [9] reported impairment of surface water quality by human fecal contamination in the Guadalupe River, Texas, following Hurricane Harvey, and similarly, Yu et al [10] found elevated levels of pathogenic indicator bacteria in Houston floodwaters. While these studies documented pollutants in natural water bodies as a result of the hurricane, it is the subsequent impact of such pollutants in PWS that is of the utmost importance to human health. However, outside of the efforts from Water Research Australia [11][12][13][14], little work has explored the impacts of extreme weather events on PWS performance during and after such events. Therefore, a more thorough understanding of drinking water quality following extreme weather events is necessary to expedite recovery efforts.
The objectives of this research were to monitor physical, chemical, and biological water quality in two drinking water systems in Texas following Hurricane Harvey and to understand the relationship between water quality parameters and changes in the drinking water microbiota throughout the recovery period. This study provides a holistic view of the storm's impacts on drinking water quality over an extended period of time, taking into account utility responses to immediate concerns, such as changes in type and dosage of disinfectant, as well as lingering effects of the hurricane. By studying the microbial community of drinking water samples during the recovery period, this study aimed to assess the impact of source water quality (especially natural organic matter (NOM)) and disinfectant residuals on microbial community structure (i.e. composition and abundance). Ultimately, this study can serve as a guide to water treatment plant operators and design engineers to assist them in preparing for future weather events.

Sampling sites
Water samples were collected at the treatment plants and throughout the distribution systems (i.e. with varying water age (WA)) of two drinking water systems in southeastern Texas. City A is located along the Gulf Coast and has a seasonally dynamic population of approximately 10 000 people. City A purchases its finished drinking water from a local water district where the treatment includes alum coagulation prior to filtration, chlorine dioxide oxidation and disinfection in the plant, and residual chloramination in the distribution system. Details of treatment and the distribution system are provided in the supplementary materials. In City A, three sites were tested: the chloramine booster station (designated as Low WA), a water tower (Medium WA), and one of the farthest points in the distribution system (High WA). Lacking a detailed model of the distribution system, the WAs could not be estimated but the relative designations low, medium and high WA are clear from the system geometry.
City B is located inland along the Guadalupe River, which serves as its raw water source; the city's treatment plant is located within the city limits and delivers water to a population of approximately 60 000. The treatment is conventional flocculation/sedimentation/filtration using polyaluminum chloride, and like City A, City B uses chlorine dioxide for pre-oxidation and disinfection and chloramine as the residual. In City B, four sites were tested: the treatment plant raw and finished waters and two sites in the distribution system: Low WA, which is located in close proximity to one of the water towers, and High WA, which is located in the far reaches of the distribution system. Again, the WAs could not be estimated, but the relative designations of low and high WA are clear from the system geometry.

Sample collection and analyses
Water samples were collected approximately monthly for four months following Hurricane Harvey, and samples from both cities were collected in May 2018 (approximately 9 months after the hurricane) to assess water quality long after the initial recovery period. In addition to the samples collected in City A's distribution system, raw and finished water samples were collected from the water district that supplies City A on 22 November 2017.
Physical, chemical, and biological water quality analyses, including pH, conductivity, turbidity, total dissolved solids (TDS), disinfectant residual, trihalomethane (THM) concentrations, TOC, dissolved organic carbon (DOC), specific ultraviolet absorbance (SUVA 254 ), anion and cation concentrations, heterotrophic plate counts (HPCs), and fecal indicator bacteria concentrations, were performed in accordance with Standard Methods [15], with the exception of THM analyses, which followed US Environmental Protection Agency (USEPA) Method 551.1 [16]. Microbial community analysis was completed using the process previously described in Maestre et al [17]. Further details on water sample collection and analyses are provided in the supplementary materials.

Results and discussion
3.1. Regional effects of Hurricane Harvey on water quality in two cities Each city faced a unique set of challenges in the aftermath of Hurricane Harvey. These challenges, along with distinct water treatment processes and distribution system characteristics, led to noteworthy differences in drinking water quality between the two cities.
3.1.1. City A City A suffered extensive structural damage during the height of the storm, but the raw water that supplies City A remained unaffected. However, due to high flows and pressure losses in the distribution system, which were predominantly due to a large number (approximately 5000) of broken connections between the distribution system and mobile homes that were dislodged during the storm, the water district lowered the pressure in the distribution system immediately following Hurricane Harvey and was forced to post a BWN on 26 August 2017. This BWN was resolved the next day, and another BWN was posted on 10 September 2017, lasting for 13 d. Aside from these two BWNs, City A did not implement any other measures or experience any other major issues during the sampling period.
As mentioned previously, the population in City A fluctuates throughout the year; this fluctuation in population creates a seasonally dynamic water demand in City A (figure S1 is available online at stacks.iop.org/ERL/14/124046/mmedia). With the destruction brought about by Hurricane Harvey, many homes in City A were abandoned not only before and during the storm but for an extended period afterwards. In turn, the decrease in the average monthly water demand from the summer to fall months was more drastic in 2017 than in 2016 and the peak water demand in the summer of 2018 was lower than that of 2017, suggesting that a large number of people did not return to City A following Hurricane Harvey. This decrease in water demand led to increased WA, which presented City A with difficulties in maintaining a disinfectant residual within the distribution system.

City B
In City B, changes in surface water quality and quantity led to different water quality challenges. Flooding in the Guadalupe River reached nearly 30 feet above the flood stage during the storm [1], and the storage reservoirs for raw water in City B were inundated with floodwater. The floodwaters had elevated turbidity and TOC levels, so the influent water quality to the treatment plant was affected for several months as the storage reservoirs were used. Due to the immense flooding, the treatment plant was offline for two days before the emergency generators could be accessed by the plant operators. After coming back online on 28 August 2017, City B remained on a BWN for nine days. Six days after the BWN was resolved, on 12 September 2017, another BWN was posted due to pressure losses in the system. This BWN lasted for eight days. City B increased their coagulant dosage during their initial recovery period. At the end of October 2017, City B implemented an emergency chlorine burn (4.0 mg l −1 free chlorine as the water entered the distribution system) to prevent nitrification within the distribution system. An annual monthlong chlorine burn (4.0 mg l −1 free chlorine) was performed in April 2018, as well. While structural damage in City B was not as drastic as in City A, a decline in population before, during, and after the hurricane led to increased WA in City B's distribution system. As a result, City B also had difficulty maintaining a disinfectant residual in the distribution system following Hurricane Harvey.
3.2. Impacts of raw water quality on distribution system water chemistry 3.2.1. City A Since City A's raw water was not subject to drastic changes following Hurricane Harvey, the water quality in treated samples taken throughout City A was consistent throughout the sampling campaign. The pH of treated water samples taken in City A was in the range of 7.8-8.5, while TDS measurements were between 500 and 750 mg l −1 (table S1). Primary Maximum Contaminant Levels (MCLs) of bromide, nitrate, and nitrite, and secondary MCLs of chloride, iron, fluoride, sulfate, and zinc were not exceeded at any point during sampling (tables S2, S4). Additionally, the USEPA action levels for lead (15 μg l −1 ) and copper (1.3 mg l −1 ) were not reached (table S4). USEPA regulates turbidity with a maximum level of 1 nephelometric turbidity unit (NTU) and a maximum monthly average of 0.3 NTU in 95 percent of samples; only one sample (Medium WA on 22 November 2017) was found to exceed the maximum monthly-average level (table S1).
TOC concentrations were stable at approximately 4 mg l −1 in City A's distribution system during sampling ( figure 1(a)). The comparison of TOC concentrations in the months following Hurricane Harvey to those from previous years (figure S2) suggest that Hurricane Harvey did not impact concentrations of NOM in raw and finished water samples in City A. Variability in NOM composition over time is evident through changes in SUVA 254 measurements. This measurement is a good proxy for NOM content and composition; an increase in SUVA 254 values indicates a shift from hydrophilic to hydrophobic NOM [18,19]. In City A, SUVA 254 increased from September 2017 to November 2017 and decreased from December 2017 to May 2018 (table S7). These variations might be seasonal (and not specifically hurricane-related), as other researchers (e.g. [20,21]) have observed similar seasonal changes in the relative proportions of hydrophilic and hydrophobic NOM in natural waters. The stable NOM concentrations in City A led to stable formation of total trihalomethanes (TTHMs) in the distribution system, and the TTHM levels remained below the USEPA MCL (80 μg l −1 ) throughout the entire sampling campaign ( figure 1(a)). Values reported to TCEQ from 2016 to 2019 show similar TTHM concentrations as those collected throughout the sampling campaign (table S8).

City B
In City B, extensive changes in source water quality led to shifts in a number of water quality parameters in drinking water samples. Since the storage reservoirs are replenished with water from the Guadalupe River, the water quality of the influent water gradually shifted from (nearly) that of the floodwaters to normal river water. Increases in TDS were noted over time (table S1) and associated with the stabilization of source water in the Guadalupe River. Alkalinity values determined from raw water samples collected in this study (table S6) and reported to TCEQ from 2016 to 2019 (table S9) support this claim, as a decrease in raw water alkalinity in the fall of 2017 (i.e. immediately following Hurricane Harvey) was more drastic than that of other years, indicating water quality fluctuations in the Guadalupe River after the hurricane. All treated water samples taken in City B had turbidity levels below the USEPA maximum monthly-average level. The pH of treated water samples taken in City B varied between 6.8 and 8.6, and an increase in pH was noted over time (table S1). As was the case in City A, primary and secondary MCLs and the action levels of all measured contaminants were not exceeded at any point in City B (tables S3, S5).
Because of the inundation of the raw water storage reservoirs in the aftermath of Hurricane Harvey, surface water came in contact with a variety of biotic and abiotic sources (e.g. decomposing organisms, plants, waste products), which exposed City B's raw water to a plethora of chemicals, including high concentrations of NOM. As shown in figure 1(b), raw water TOC peaked at approximately 8 mg l −1 during September 2017 and declined to approximately 4 mg l −1 in December 2017. Plant operating data show a similar trend (figure S2) [22]. The surge in raw water TOC concentrations following Hurricane Harvey was accompanied by a surge in finished water TOC concentrations ( figure 1(b)). TOC concentrations in raw and finished waters in City B were stable prior to Hurricane Harvey, with the exception of a surge in raw water TOC concentrations in June 2016, which coincided with heavy rainfall and flooding in the Guadalupe River (figure S2).
For several months following Hurricane Harvey (i.e. the first four samples), SUVA 254 measurements in the distribution system in City B were approximately 2 l 1 mg −1 m −1 , representing hydrophilic NOM (table S7). In December 2017, SUVA 254 values increased at all treated water sampling points, and then decreased from December 2017 to May 2018. The fluctuation in SUVA 254 values in City B was more substantial than that of City A; therefore, while this trend may be seasonal, it is also possible that microbial degradation of hydrophilic NOM, which tends to be more biodegradable than hydrophobic NOM [23], may have impacted the NOM composition in the distribution system.
Raw water samples had high SUVA 254 values immediately following Hurricane Harvey, and a decrease in SUVA 254 was noted over the subsequent months (table S7). This initial surge may be due to the presence of soil NOM in City B's raw water, as soil organic matter is generally more hydrophobic than is aquatic NOM [24]. The decrease in SUVA 254 from raw to finished water indicated that hydrophobic NOM was removed during treatment; coagulation has been identified as the primary removal mechanism of hydrophobic NOM in water treatment [18,25].
TTHM concentrations in City B fluctuated in accordance with the variation in TOC concentrations. TTHM concentrations exceeded the USEPA MCL at all sampling points on 9 November 2017 ( figure 1(b)). The two sets of samples from September 2017 were not analyzed for TTHMs, although it is reasonable to speculate that TTHMs were high on those dates. Three months after the hurricane, TTHM concentrations had stabilized below USEPA's TTHM MCL. TTHM concentrations in water samples collected quarterly in City B and reported to TCEQ from 2016 to 2019 (table S9) are below the USEPA TTHM MCL, indicating that Hurricane Harvey led to increased TTHM concentrations in finished water samples collected in City B immediately following the hurricane.
3.3. Impact of disinfectant residuals on distribution system microbiota 3.3.1. City A In City A, measured levels of total chlorine decreased from Low WA to Medium WA and from Medium WA to High WA; the decrease from Medium WA to High WA was significantly more drastic. In late September 2017, just one month following Hurricane Harvey, HPCs at High WA were over 200 MPN ml −1 ( figure 2(a)). This HPC value is relatively high compared to other dates (in line with the low disinfectant residual); however, heterotrophs are not directly correlated with human health risks. Two months after the BWNs in City A, on 22 November 2017, HPC were significantly reduced in all samples compared to the first set of samples. The assessed fecal indicator bacteria (E. coli and Enterococci) were never detected in City A throughout the entire sampling campaign (table S10).
While HPCs in City A fluctuated based on disinfectant residuals, the dominant phyla of City A's distribution system microbiota did not change drastically over time. As seen in figure 3, five phyla (Acidobacteria, Actinobacteria, Bacteroidetes, Cyanobacteria, and Proteobacteria) dominate the microbial community at all points in City A's distribution system. These phyla have been commonly reported in other drinking water distribution systems [26][27][28][29]. Still, changes in bacterial taxa occurred over time, including an increase in Actinobacteria from September 2017 to December 2017 and a subsequent decrease from December 2017 to May 2018, together with inverse changes in Proteobacteria (figure 3). Seasonal variation in Actinobacteria and Proteobacteria in drinking water samples has previously been reported by Li et al [27]. Due to the stability of raw water quality at the water district, the distribution system water microbiota in City A remained spatially consistent following Hurricane Harvey.

City B
Similarly, in City B, HPC increased with WA throughout the entire sampling campaign, probably due to disinfectant decay at high WA ( figure 2(b)). In comparing total chlorine measurements for water samples collected at Low WA and High WA (table S1) to monthly measurements collected by plant personnel from 2016 to 2019 (table S9), it is clear that samples taken during the initial recovery period have lower disinfectant residuals than that of normal operating conditions. E. coli and Enterococci levels in source water samples were found to decrease over time (table S10). Human fecal contamination in the Guadalupe River, as reported by Kapoor et al [9] and confirmed by TCEQ records (personal communication), accounts for the introduction of fecal indicator bacteria into source waters by the large number of combined sewage overflows and high stormwater runoff during and immediately following the storm. E. coli was detected once in treated samples in City B, while Enterococci was never detected (table S10). Disinfection at the treatment plant and throughout the distribution system was effective at controlling fecal indicator bacteria in finished water samples.
Similar to City A, the dominant phyla in treated samples in City B include Actinobacteria, Proteobacteria, and Cyanobacteria (figure 3). Firmicutes also were found in some treated samples; this phylum of Gram-positive bacteria has been found in drinking water samples due to its chlorine resistance [27,30]. Despite the presence of a few dominant phyla in samples from City B, their relative abundances fluctuate over time and with WA; specifically, increases in proportions of Proteobacteria are noted with increased WA ( figure 3). While Proteobacteria are dominant in many natural and engineered systems, including drinking water distribution pipes [26][27][28], their increased presence in samples taken in City B following Hurricane Harvey suggests that they may have had a competitive advantage over other bacteria during the recovery period.
Plant-and soil-associated bacteria were present in water samples from City B following Hurricane Harvey. For example, the finished water sample collected on 8 September 2017 contained a high relative abundance of Stramenopiles (67% of the total operational taxonomic units (OTUs), table S12), which is commonly associated with algae [31]. A surge in Methylobacterium was also noted in samples taken from City B's distribution system on the first two sampling dates (table S12); this phylum is often associated with soil enviornments [32]. This surge in plant-and soil-associated bacteria coincides with a surge in TOC in City B's raw and finished waters. The presence of these bacteria in treated water samples indicates a shift in bacterial community composition within the treatment plant and the distribution system of City B following Hurricane Harvey.
Furthermore, the chlorine burn performed in City B in November 2017 had immense impacts on the microbiota in parts of the distribution system. During and following the chlorine burn, total chlorine concentrations were high (>1.5 mg l −1 , as expected) in Finished and Low WA samples, yet total chlorine concentrations at High WA were below 1 mg l −1 . It is likely that spatial variation in water demand influenced the extent of chlorine decay in samples. Consequently, an order of magnitude difference is noted between HPCs at Low WA and High WA on 9 November 2017 ( figure 2(b)). The water sample collected at High WA on 22 November 2017 contained a unique microbiota compared to other samples, with 31% of the OTUs belonging to the OP3 phyla ( figure 3). This phyla is comprised of anoxic bacteria that thrive in marine sediments and stratified freshwater lakes, among other anoxic environments [33]. As suggested by Kelly et al [34], excessive biofilm growth on the distribution pipes may have facilitated the growth of anoxic bacteria in lower layers of biofilm. Then, when the chlorine burn reached the far ends of the distribution system, these bacteria were mobilized, causing shifts in water quality at far reaches of the distribution system. In addition, Mycobacterium were present in samples taken following the emergency chlorine burn in City B. Mycobacteria is a genus of Actinobacteria that contains opportunistic pathogens known to cause tuberculosis and leprosy [35]. Due to its complex cell wall and hydrophobicity, Mycobacterium are quite resistant to chlorine and other disinfectants [36]. Their presence in drinking water systems occurs through surface attachment, as slow growth rates prevent their presence as free-living species in drinking water pipes [36]. Prior to the chlorine burn, the average relative abundance of Mycobacterium in the treated water samples in City B was 8%, while that of post-chlorine burn samples was 15% (table S12). During the chlorine burn, the average relative abundance was 17%. It is hypothesized that partial biofilm destruction by the chlorine burn released Mycobacterium into the water, but that the chlorine burn did not fully eliminate Mycobacterium fractions in protected levels of biofilm. While the levels of Mycobacterium in City B's distribution system under normal operating conditions are unknown, the elevated presence of Mycobacterium in City B's distribution system following the chlorine burn potentially exemplifies the lingering effects of Hurricane Harvey on drinking water quality.
3.4. Stabilization of water quality and distribution system microbiota over time Despite the challenges faced by the drinking water managers in Cities A and B, water quality in both systems recovered after approximately four months following Hurricane Harvey. The dynamics of microbial community structure over time are illustrated in a Principal Coordinate Analysis (PCoA) plot in figure 4. PCoA is an ordination technique that reduces the dimensionality of data to a few principal coordinates, each explaining a certain fraction of the variability observed in the data. The first two principal coordinates account for approximately 60% of variation in the microbiota data. A PCoA plot expressing points by sampling location is provided in the supplementary materials (figure S3).

City A
For all sampling dates, clustering of City A data in figure 4 indicates minimal spatial variation in the bulk microbial community among water samples. However, temporal variability in the microbial community in City A's distribution system is noted. While the samples taken in September, November, and December 2017 cluster together, the 23 May 2018 data cluster separately. These results do not directly suggest that this shift in microbial community is a consequence of Hurricane Harvey because it is possible that this is a seasonal trend. All in all, this stability in the microbial community of City A's finished water samples supports the influence of stable raw water quality, including TDS and TOC concentrations, on drinking water quality following Hurricane Harvey. To this end, it is important to note that these samples do not fully represent the water quality everywhere in City A, and it is possible that water qualtiy at at other locations (e.g. the points of broken connections) was not as stable as the samples analyzed in this study.

City B
PCoA analyses indicate substantial variability in the microbial community of water samples taken in City B following Hurricane Harvey (figure 4). For each sampling location, PCoA points are not well clustered ( figure S3); however, in considering each date, it is clear that treated samples collected on earlier sampling dates (darker-colored points in figure 4) have more variability than do those collected towards the end of the sampling campaign (lighter-colored points in figure 4). Clustering of treated water samples over time indicates stabilization of the distribution system microbiota during the recovery period. The stabilization of City B's microbial community structure over time coincides with the stabilization of typical water quality parameters, such as TDS and TOC concentrations.
3.5. Considerations for future drinking water management following extreme weather events The primary impacts of Hurricane Harvey on drinking water in Cities A and B involved changes in raw water quality and a decreased water demand; PWS managers must be prepared for these changes in future storm events. To prevent excess water and pressure losses due to broken connections, it is recommended that cities allow their customers to close premise valves on the city side during emergency situations even though this action often comes with fines or penalties during normal operation, as the valves on the city side are often more-protected than the valves on the customer side. Managing WA in distribution systems with a decreased water demand might also involve opening fire hydrants to reduce stagnation; while many would consider this practice to be a waste of water, it is sometimes appropriate and necessary in situations such as in the recovery from Hurricane Harvey. Moreover, PWS managers should not be reluctant to post BWNs, as human health is of utmost importance and should not be compromised following an extreme event. As seen in this study, lingering effects can impact drinking water quality following an extreme weather event for a substantial period of time, and thus, PWS managers should develop long-term framework plans for drinking water management in future situations. Finally, future research efforts should consider collecting baseline samples in vulnerable areas prior to extreme weather events so that postevent samples can be compared to samples taken during normal operating conditions.

Conclusion
By measuring physical, chemical, and microbial water quality in two Texas cities following Hurricane Harvey, this study gives insight into the impacts of raw water quality, utility responses, and lingering effects on drinking water quality following a natural disaster. Both cities studied in this work had difficulty maintaining a disinfectant residual in their distribution systems, but stabilization of drinking water quality occurred over time. The timeframe associated with the recovery period depended on several considerations, including changes in source water quality and water demand. Source water quality is shown to correlate with the microbial community of the distribution system. Furthermore, disinfectant decay due to increased WA was shown to have profound impacts on drinking water quality and the distribution system microbiota. Intentional chlorine burns in City B were successful at controlling microbial growth in the distribution system and this approach caused the drinking water microbiota to re-establish community structure, possibly favoring chlorine-resistant bacteria such as Mycobacterium. Results of this study suggest that utilities must be prepared for major changes in water quality and water demand in the aftermath of an extreme weather event, not only immediately following the event, but for several months following a natural disaster.