Effective first flush volumes in experimental household-scale rainwater catchment systems

Roof-harvested rainwater, also known as rainwater harvesting (RWH), has been long practiced around the world by rural and urban communities (WHO 2016, 1997), and has been of increasing interest globally due to its potential for improving environmental sustainability, alleviating water scarcity, and controlling stormwater runoff (Helmreich & Horn 2009; Hamilton et al. 2019). RWH systems consist of a catchment surface such as an impervious rooftop, a collection system of gutters and downspouts, a quality control system (which could include a first flush diverter, debris screens, and/or filters), a collection tank, and piping for water use (Campisano et al. 2017). Although RWH is widely encouraged in many places, there is high variability in system designs and the resultant water quality based on climate, system design, and collection location (canopy cover, proximity to pollution) (de Kwaadsteniet et al. 2013; Hamilton et al. 2019). Additionally, there are few regulations or universal recommendations on building, maintaining, and treating water from the RWH systems. Rainwater in RWH can be contaminated from (1) washout; (2) roof wash-off; and (3) the collection systems. As rain travels through the atmosphere, it can washout gas and airborne particles (e.g. ash, air pollution) (Martin 1984; Guo et al. 2016). Roof wash-off can transport both dry and wet deposition from the roof surface, including pathogens from animal droppings, decomposing organic matter from nearby trees and plants, and leaching of catchment material metals. Rainwater can become contaminated through the process of RWH collection by accumulating contaminants through roof wash-off and debris or biofilms in the gutters and the tank (Meera & Ahammed 2006; de Kwaadsteniet et al. 2013; Ghernaout & Elboughdiri 2020).

First flush systems, designed as a part of a rainwater harvesting system, divert the first wash of more polluted water away from the storage tank, thereby improving the water quality of the collected water and reducing tank maintenance needs (de Kwaadsteniet et al. 2013). Prior studies recommend that the first flush systems are designed to divert the first 1–2 mm of runoff, as pollutants from roof deposition are easily disturbed early in the rain event (Kus et al. 2010; de Kwaadsteniet et al. 2013; Campisano et al. 2017). An analysis of the first flush volumes in Sydney, Australia, demonstrated that the first 2 mm of rainfall allowed the final collection water to meet all Australian Drinking Water Guidelines other than turbidity and lead, while a 5 mm first flush resulted in meeting all water quality guidelines (Kus et al. 2010). The same study also found decreasing natural organic matter (NOM) concentrations as the first flush volumes increased (Kus et al. 2010). NOM, which can originate from organics in the atmosphere such as airborne soil and dust, vapor-phase plant metabolites, and those from plants via canopy drip, is of concern in the RWH systems when chlorination is used to treat the collected water due to the potential of disinfection by-product (DBP) formation (Reckhow et al. 1990; Richardson et al. 2007). It is common to couple the first flush systems with water treatment, such as chlorination (de Kwaadsteniet et al. 2013). In conventional drinking water treatment, options for minimizing DBPs in consumed water include controlling disinfection and/or the DBP precursors in water. In rainwater, removing NOM in the first flush can decrease DBP formation potential in the collection tank.

The goal of a first flush system is to minimize the volume of diverted first flush water while protecting the quality of water in the final collection tank. Previous studies evaluating the rainwater harvesting first flush systems tend to focus on the removal of microbiological and chemical contaminants within a defined volume to decrease the potential negative health effects of the use of water for potable purposes (Kus et al. 2010; Amin et al. 2013; Gikas & Tsihrintzis 2017). As chlorination is a common treatment method for the inactivation of pathogens, there is a gap in understanding how well the first flush systems remove NOM and the parameters affecting the first flush volume required for contaminant removal.

The aim of this study was to evaluate the first flush volumes that effectively improve stored water quality in a rainwater harvesting system. The effects of rain intensity and tree canopy cover on the first flush volumes needed to remove contaminants were evaluated with the goal of improving recommendations for the design of the first flush systems.

An experimental RWH system on a rooftop in Amherst, MA (USA) was used to conduct a series of experiments to study the changes in water quality throughout the first flush period. First, the designed volume of the fractioned first flush system was based on the results of a tracer study (Experiment 1); subsequent experiments evaluated pH, conductivity, UV 254, total and dissolved organic carbon (DOC), and, in a subset of buckets, total coliform and Escherichia coli throughout the first flush under real precipitation events (Experiment 2) and under two sides of the roof that had different tree canopy cover (Experiment 3).

Experiments were conducted from June to October 2020 (Supplemental material, Fig. S1). Amherst, MA, received an average annual precipitation of 45.2 in. from 2015 to 2020 (PRISM Climate Group 2021). Notably, Amherst experienced a moderate drought from June to September 2020 and an extreme drought in October 2020, as classified by the U.S. Drought Monitor (NIDIS 2020). The highest total monthly precipitation in 2020 occurred in October, at 5.8 in. On-site precipitation data were collected from a rain gauge (Rainwise RainLogger 2.0, forestry-suppliers.com) placed 10 feet from the study roof under the open sky, which collected 0.01″ in a tipping bucket at a time, and then, the total number of tips per rain event (RE) was counted and logged.

The first study was designed to model the roof's flow characteristics using a tracer (NaCl) of known concentration to predict the first flush volume needed for the test area roof size based on the volume required to wash-off the tracer. The experimental setup (Supplemental material, Fig. S2) consisted of a manifold on the study roof, a pump, and two 55-gallon drums connected to the same pump with a t-connection to switch the pump flow between them. A 10 ft manifold system was constructed to simulate rainfall on one side of the roof, creating a test area of 125 ft². The manifold consisted of a nominal 1-inch inner diameter Schedule 40 PVC pipe, 10 downspouts created by T connections placed 1 ft apart to center flow, down each channel, and 90° elbow connectors (Supplemental material, Fig. S2). The test area was cleaned prior to testing using a power washer. The two 55-gallon drums contained: (1) simulated rain (Amherst tap water) and (2) synthetic contaminated rainwater (mixed Amherst tap water with varying doses of salt tracer (NaCl) with a pump for mixing). Flow reducers were used to simulate different rain intensities. Salt concentration at the downspout of the test area was measured continuously through the proxy of conductivity through a sample port (Supplemental material, Fig. S2).

Flow rates were varied between experiments to simulate different rain intensities (4, 5, and 10 gallons per minute (gpm)) and salt doses to represent changes in contaminant concentrations (52, 500, and 1,000 g to the 55-gallon drum; 0.25, 2.40, and 5.28 mg/L) (Table 1). The flow was first equalized from each manifold downspout and then from the measured time to fill a 1,000-mL Erlenmeyer flask for each spout, three times (Supplemental material, Fig. S2). The mean flow rate of all 10 downspouts was assumed to represent the mean flow rate over the roof surface, which was then converted to an average rain intensity (in/h) by dividing by the test area (125 ft²). Initial conductivity measurements were taken in an influent tap water tank (C₀) and a saltwater tank (C). Saltwater was then pumped to the roof, and through the manifold, with sample water channeled into the downspout. Effluent conductivity was measured every 30 s in an overflow sample port open to the atmosphere (Thermo Scientific 4-cell Conductivity). Once the effluent concentration reached influent conductivity (C), the valve on the influent tank was switched instead to pump tap water onto the roof, marking the beginning of flushing. Effluent conductivity was then manually measured and recorded every 30 s until the conductivity reached C₀, with this time recorded as the tracer flushing time.

Both the estimated first flush volumes from the tracer study and previously published work (Campisano et al. 2017) were used to inform the experimental first flush capture system design. The test area for fractionating first flush water was one entire side of the WET Center roof (total area of 800 ft²). Using the average first flush volume result from Experiment 1 and the ‘rule of thumb’ of 2 mm runoff (Kus et al. 2010), the design value of 40 gallons was calculated for the first flush system (calculated as the product of the 800 ft² roof area and a 2 mm runoff depth) which fractionated the first flush water to create a profile of the first washout over both time and volume.

For each measured rain event (RE), rainwater flowed into the collection system with an overflow on the 55-gallon collection tank. 250 mL samples were collected in amber jars from each fractionation bucket by removing the bucket lid, mixing the bucket with a 1-inch piece of pipe that had been cleaned with deionized water (DI), and then taking the sample. If the collection tank was full, average values from samples collected from the tank's top and bottom overflow valves were reported. A sample was also collected from the control bucket located on the roof (‘raw’ rainwater). System maintenance after sampling included dumping and draining excess collected water from the collection system and cleaning the fractionation buckets, atmospheric sample buckets, and collection tank with DI water. The gutter was cleaned periodically by removing clumps of leaves and tree deposits from the gutter guard.

Four REs were studied with measured samples from atmospheric rain, fractionation buckets, and collection tank for pH, conductivity, UV 254, as well as total and dissolved organic carbon. Total coliform and E. coli were measured in a subset of buckets.

The effect of the collection environment on dry and wet deposition in collected rainwater was investigated by splitting the gutter system into two downspouts: one at the center of the roof and the other directly under a white pine tree. For this experiment, the gutter was converted to a PVC pipe using the downspout 2″ PVC adapter and directing each downspout into a 55-gallon drum with an overflow port (Supplemental material, Fig. S3). Data were collected during the three precipitation events by collecting two 250 mL composite samples from both collection tanks and one of ‘raw’ rainwater and pH, conductivity, UV254, as well as total and DOC were measured.

pH and conductivity were measured in samples immediately after the sample collection in 250-mL amber bottles (Orion Star A215 pH/Conductivity Benchtop Multiparameter Meter with Atlas Scientific Lab Grade pH probe and the Thermo Scientific 4-cell Conductivity probe). Ultraviolet absorbance at 254 nm (UV254), used as a surrogate measurement of NOM concentration in water, was measured using a Hach DR6000 Laboratory Spectrophotometer after filtering each sample through a 0.4 μm pore size syringe filter. Total organic carbon (TOC) and DOC were measured following the Standard Methods Method 5310 (APHA/AWWA/WERF 1998). The DOC samples were first filtered through a 0.4 μm syringe filter and then run on the Shimadzu TOC-VCPH Total Organic Carbon Analyzer. Calibration of the instrument was performed using standard solutions of potassium hydrogen phthalate at concentrations of 0.5, 1, 2, 5, and 10 mg-C/L. Milli-Q ultrapure water was used for dilutions and blanks. Samples were analyzed in duplicate, with each sample having multiple injections and the mean value was reported with a standard deviation <0.05. Specific ultraviolet absorbance (SUVA254, L/mg-m) was calculated by dividing UV254 (m⁻¹) by DOC (mg/L), and was used to represent the nature of NOM (a higher SUVA representing hydrophobic NOM (SUVA >4), while a lower SUVA (<2) demonstrates hydrophilic organics) (USEPA 2009). Total coliform and E. coli were measured using IDEXX Colilert Quanti-Trays 2000, to quantify the most probable number (MPN) of cells per 100 mL of the sample. These samples were incubated at 35 °C for 24 h.

A mass balance of NaCl was calculated to determine if the system resulted in a complete flush-out of the tracer. The initial salt concentration in the salt tank was calculated using the recorded mass of NaCl added to the 55-gallon tank and was converted to concentration (mg/L). The amount of salt recovered in harvested water was calculated by integrating the area under the curve in the plot of conductivity versus time for each test (Supplemental material, Fig. S4). For all six simulated rain tests, 90% or more of the dosed salt was recovered (Supplemental material, Fig. S4). In test 6, 110 more grams were recovered than dosed, potentially due to the salt left from the prior test runs (Supplemental material, Fig. S4).

The first flush volume based on salt breakthrough was specific to the test roof area of 125 ft². This was scaled up to the entire roof area (800 ft²) (Table 1). The first flush volume was determined as described above and it increased with increasing salt dose, and the lower intensity rainfall events resulted in smaller first flush volumes based on the same salt breakthrough criterion (Table 1).

For each measured RE, the effects of rain intensities and antecedent dry days were evaluated (defined as the number of days since the last rainstorm of greater than or equal to 0.1 in.) (Supplemental material, Fig. S5). RE1 and RE4 had an average rain intensity of 0.13 and 0.06 in./h, respectively, and relatively short dry periods. RE2 had the highest rain intensity, while RE3 had the lowest intensity of 0.06 in. an hour and an antecedent dry period of 18 days (Table 2).

The measured water quality parameters followed similar trends within each bucket within a given RE. For example, in RE3, buckets 3 and 5 show a drop in conductivity with corresponding drops in UV254 and DOC. Similar trends were observed with TOC. When comparing each of the rain events, raw water has the lowest concentrations, with relatively higher concentrations in the first flush buckets and lower concentrations in the collection tank (Figure 3).

The DOC and intensity values for bucket 1 were similar between every RE (approximately 11 mg/L), suggesting that the first bucket was most likely to receive DOC from a similar source (e.g. the section of gutter immediately next to the downspout). Comparison of each RE first flush DOC washout to rain intensity suggests that the rain intensity per bucket did not directly affect the DOC concentration washed into the bucket. However, in RE2, bucket 4 (at 6 min) had the lowest DOC concentration and lowest intensity, suggesting that high-intensity storms may be associated with more washout of deposited particulates. When looking at the effect of the overall RE intensity, the results demonstrate that RE2, the highest intensity storm at 0.27 in./h, provided collection tank water with DOC concentration closest to the raw water level (1.3 mg/L difference). Again, higher intensity rainfall may better washout particulate contaminants such as pine needles within the first flush.

UV 254 and SUVA values within each first flush bucket were investigated to characterize NOM in the collected rainwater (Figure 5). The overall SUVA values were within the range of 2–4 L/mg-m, with an outlier in RE3. Consistent SUVA values (approximately 2 L/mg-m) were observed during RE1, demonstrating that the NOM was likely atmospheric. During RE2, SUVA was in the mid-range (2–4 L/mg-m), which can be interpreted as NOM collected from a mixture of both atmospheric and roof deposition. For RE3 and RE4, SUVA were all <2 L/mg-m, with one outlier during RE3 in the fifth bucket (6 L/mg-m). NOM that accumulated during the preceding dry period likely came from both the atmosphere and the collection system. In RE2, the SUVA increased when the intensity dropped; however, this phenomenon was not observed in RE3 or RE4 potentially due to the irregular DOC concentrations in RE3 and the overall intensity of the storm event. SUVA and DOC data demonstrated that RE3 would have a higher potential for DBP formation if chlorinated, because of the quantity of DOC and the hydrophobic condition of the DBP precursors.

Indicator bacteria samples for fecal contamination were collected for RE 1 and 2 at the beginning of the first flush, in the middle, and at the collection tank. In both rain events, E. coli concentrations were higher in the collection tank than in the buckets, suggesting that the first flush was insufficient in removing the indicator bacteria from the final collected water.

To investigate the source of NOM in the collected rainwater, the collection system was split into two separate downspouts collected from two different roof areas (one under the white pine tree, and one not). Composite samples were taken from two collection tanks, one subject to canopy drip from the white pine tree, and the other, under the open sky. Three events were collected with varying rain intensities. The south side of the roof underneath the white pine (left) had visibly more pine needles on the roof surface and the gutter guard (Supplemental material, Fig. S3).

DOC concentrations measured in water from the collection tank under the tree resulted in up to six times the concentration of DOC (Figure 6). Additionally, the gutter under the white pine clogged, potentially leading to more stagnant water, and increasing the available time for leaching NOM, increasing the DOC concentration. Since the fractionation experiment was set up directly under the canopy, it was likely that the water collected in the first flush represented the concentrated NOM water under the white pine, as it was closest to the downspout. These results demonstrate that the location of a rainwater harvesting system should likely account for a canopy drip.

In this study of the volume of first flush water that would need to be diverted to avoid the entry of contaminants into the RWH system, the tracer study results led to an estimate that it would require 33 gallons of first flush to reach the quality of the water where key contaminant concentrations reached steady-state, minimum values. This volume was similar to the ‘2 mm rule’ for the first flush volumes (which, for the test roof, would have been 40 gallons). Increasing the concentration of the contaminant (i.e. tracer salt dose) increased the volume of the first flush that would be required to flush the salt from the roof surface, which aligns with a previous study in Australia that found decreased rainwater organic matter in the collection tank when the first flushed volume increased (Kus et al. 2010). The later experiments determined that most contaminants originated from roof deposition rather than the raw rainwater itself. Lower simulated rain intensities required smaller first flush volumes, potentially due to longer contact times between water and the contaminants on the roof.

The results from the four collected rain events demonstrated that the use of a first flush was successful in diverting water that contained concentrated contaminants, as concentrations generally decreased during flushing. Conductivity, UV254, and DOC all trended together during the measured rain events, demonstrating a uniform washout of roof deposition. However, the designed 40-gallon first flush volume was insufficient to decrease conductivity, UV254, and DOC to the low levels measured in raw rainwater. There were notably high concentrations of DOC detected in the samples, likely from canopy and pine needle depositions from the white pine tree, located directly above the downspout, which may have led to the need for a larger first flush volume. In the converse of the results from the salt tracer experiments, these results suggest that lower intensity storms or longer antecedent dry periods between rain events potentially increased the required first flush volume. These results may have been influenced by the charged particles in salt that would attach more readily to surfaces than the dissolved organics. One measured high-intensity storm (RE2) resulted in collection water with a DOC concentration closest to that of the raw rainwater.

The simplest and most common form of disinfection of rainwater is chlorination, either through adding chlorine tablets or bleach to the collection tank (Campisano et al. 2017). The risk of adding chlorine when NOM is present is the formation of DBPs, where the organic matter reacts with free chlorine to create chlorinated species such as THM and HAAs. When looking at the first flush volumes and the removal of DBP precursors, there were no observable reductions of UV 254 and DOC within the rainwater collection system's first flush. The risk of DBP formation with a canopy-covered rainwater harvesting environment was elevated due to the higher dry deposition. The collected rainwater had high DOC: up to 13 mg/L in the collection tank during RE1 and 25 mg/L in bucket 8 during RE3. Approximating rainwater as the surface water and a cumulative frequency of 50% (e.g. 50% of carbon in the water would react with chlorine to produce THMs), the specific THM concentration in the water would be 25 μg of THMs per 1 mg of carbon (assuming chlorine dosing was aligned with water in a distribution system) (Reckhow 2007). Using this estimate, THM concentrations could have been up to 300 μg/L in RE1 and 600 μg/L in RE3 if the water had been chlorinated, both exceeding the EPA regulation of 80 μg/L. While this is an approximation (rainwater differs from surface water and, notably, the average pH in measured rainwater was around 5.5, while a distribution system supplied with surface water would be 7.5–8.5), this is a high estimate of the potential of DBP formation. High concentrations of indicator bacteria were not detected, suggesting the need for treatment for potable use. Fecal contamination could originate from the collection system or bucket contamination.

Overall, the first flush volume that would be required to reduce UV254, DOC, and conductivity to levels of raw rainwater was variable and may not be a ‘one size fits all’ model as approximated by the ‘2 mm’ rule and collection roof size; in this study, the ‘2 mm’ rule resulted in an undersized first flush system. In this system, during some of the observed rain events, the designed ‘2 mm’ first flush system did not result in a sufficient decrease in contaminant levels, particularly during low-intensity storms and long antecedent dry periods. The first flush volume should account for potential increases in contamination from air washout, roof wash-off, and collection system contamination. Most of the contamination from the collection system originated from the roof and/or gutter collection system. This dry and wet deposition is dependent on environmental factors including the nearby sources of deposition, seasonal variation, and time since the previous rainfall. It is also important to consider the parameters that affect the hydraulics of the wash-off such as rain intensity and collection location. Accounting for these factors could decrease treatment needs, system maintenance, and concern about treatment by-products.

This study had several limitations. First, all tests were performed in the Northeast United States on the same sampling roof, leading to unique conclusions for these specific conditions and contaminants (e.g. the white pine tree and atmospheric levels). Sampling for these experiments took place from June to October 2020, with July–September classified as moderate drought months and October as an extreme drought month (NIDIS 2020). Due to the drought conditions, it was difficult to collect data during multiple rain events with similar conditions. Variability in the rain events collected complicated comparisons between events, and the collected rain events were primarily from one season. Also, the first flush volume was capped at the design estimate of 40 gallons (based on an ideal, dissolved contaminant with a freshly cleaned roof) and did not consider the variability of washout, roof wash-off, and build-up of deposition from antecedent dry periods.

Taking these results and limitations into consideration, future research could include performing a tracer study to model roof wash-off volumes by spreading the solid tracer on the roof or experimenting with smoother roof surfaces that deterred the ionic bonding of salt to the roof. Collecting more rain events using the fractionation system, with varying intensities and seasonal variation, could help to generalize the observed trends. Furthermore, fractionated sampling of rain events under non-canopy conditions informs the hydraulics of the first flush water. Increasing the total fractionation volume to more than 40 gallons allows for measuring whether the collected water quality concentrations decreased to the raw level. Based on the estimation from this study for the high potential of DBP formation, it would be insightful to measure THM and HAA concentrations after chlorination to understand the formation potential in acidic rainwater. Further research in these areas could lead to an equation that allows for the calculation of a variable first flush volume based on the parameters of rain intensity, antecedent dry period duration, location, and maintenance of the system. Ranking of these parameters may help optimize the volume of the first flush that needs to be diverted for each rain event or season.

The goal of rainwater first flush systems is to divert concentrated water from the final collection tank. This study aimed to evaluate the effects of rain intensity, antecedent dry period duration, and the collection location to determine if the first flush volumes can be matched to specific conditions. The results demonstrated that the predicted first flush for the roof (the ‘2 mm rule’) was insufficient to washout the deposited NOM from the collection surface and ensure that water collected for use resembled the water quality observed in atmospheric rain. Rain events with lower rain intensity and longer antecedent dry days since the last rain resulted in increased DOC concentrations in the first flush. This study demonstrates that the rainwater collection system design and the first flush volume calculations should be guided by specific characteristics of rain and the environment surrounding the system. Specifically, local precipitation patterns, storm intensity, and canopy conditions suggested that the ‘2 mm’ rule was insufficient, particularly when there was extensive contamination from the collection system from the roof and/or gutter collection systems. For example, the collection system of this system was located under a canopy drip, which required a first flush system focused on removing roof wash-off contamination. However, the rainwater systems in areas of high air pollution need to focus on designing for air washout and pH control. Understanding the mechanisms that affect contamination within collected rainwater could lead to optimized first flush systems, reducing the overall treatment needs and maintenance of this system, while providing quality potable water.

We would like to thank Christopher Watt, Isaac Reyes, and Amanda Isak for their assistance with the rainwater harvesting system setup and experiments. Funding was provided by the Department of Civil and Environmental Engineering at the University of Massachusetts Amherst.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.