Focus and OrganizationFirst-flush characterization of pollutants has been monitored from highway and other road surface runoff by several investigators (Bertrand-Krajewski, Chebbo, and Saget 1998; Charbeneau and Barrett 1998; Deletic 1998; Geiger 1987; Gupta and Saul 1996; Larsen, Broch, and Andersen 1998; Legret and Pagotto 1999; Saget, Chebbo, and Bertrand-Krajewski 1995; Sansalone and Buchberger 1997; Thornton and Saul 1987). In most cases, monitoring was performed for one season or for selective chemical constituents. The review of first-flush monitoring among these investigators revealed that there is no standard protocol to collect samples, present the results and interpret them, and utilize the results for potential best management practice (BMP) treatment application. The most comprehensive first-flush highway runoff characterization study performed to date is the study commissioned by the California Department of Transportation (Caltrans) Division of Environmental Analysis (DEA). The DEA first-flush highway runoff characterization study was performed through a collaborative effort between the Departments of Civil and Environmental Engineering at the University of California–Los Angeles (UCLA) and the University of California–Davis (UCD) (Stenstrom and Kayhanian 2005). The result of this study is presented in a final report (CTSW-RT-05-073.02.6) and is currently available at Caltrans’s Web site:
http://www.dot.ca.gov/hq/env/stormwater/special/newsetup/index.htm.
The monitoring was performed at three highly urbanized highway sites in Los Angeles, CA, over five years. Sharing with the public the experience that we have gained and the information we gathered through our study is the main aim of this article. The information presented here may serve as a basis for improving first-flush characterization studies on a national or an international basis. To serve this purpose, this article is organized into the following five topics: (1) strategies to collect first-flush sampling, (2) the meaningful definition of first flush, (3) computation of event mean concentration and mass first flush from grab samples, (4) first-flush data presentation (concentration and mass first flush), and (5) practical application of first-flush results for BMP selection and design.
Strategies to Collect First-Flush Samples
Strategies to collect samples for first-flush stormwater runoff characterization are extremely important. First, the sampling teams must be at the sites before runoff begins. Weather forecasting is important to avoid time-consuming and frustrating mobilizations for storms that do not occur, as well as to ensure that the teams are prepared for the real events. Highway sites and other sites that are highly impervious are “flashy,” and runoff occurs within minutes of the onset of rainfall. Generally, when forecasts suggest that a storm probability is greater than 50%, the sampling teams should mobilize to the sampling site in advance of storm event.
Second, to properly detect and quantify the first flush, discrete samples must be collected not only in the early part of the storm but also at the end of the storm. This requirement means that sampling to characterize the first flush will be more resource-intensive than ordinary stormwater sampling. Grab samples can be collected manually or with automated samplers having multiple bottles. We selected manual sampling for several reasons, including (1) collecting representative samples from water column outfall, (2) allowing larger sample volumes collection, and (3) providing greater flexibility for collecting special samples using different bottles or preservation techniques.
We recommend the following sampling method for first-flush characterization studies, as shown in Figure 1. Five grab samples are collected in the first hour, with the first grab sample being collected as soon as adequate runoff volume reaches the sampling point. The additional four samples are collected during 15-minute intervals. After the first hour, one grab sample is collected per hour for the next seven hours, providing a total of 12 grab samples. For storms lasting fewer than eight hours, fewer grab samples will be collected. For storms lasting longer than eight hours, an additional one or two grab samples can be collected in the period from eight hours to the end of the storm. The runoff volume must be continuously monitored and recorded during the entire storm. This sampling strategy was successful in our five-year study to characterize the initial runoff as well as the later runoff and especially for long storms with lengthy periods of light rainfall.
Runoff samples can be collected by a polypropylene scoop and then transferred to 4-L amber glass bottles. Whenever possible, all samples should be collected from a free waterfall. When collecting samples, the entire water column from waterfall should be represented, especially when the particle size distribution (PSD) and particle-bound pollutant characterization are desired. The sample volume will depend on the types of analysis being performed. To avoid problems with holding time, the collected samples should be delivered to the lab, and appropriate preservation and filtration should be implemented on time. This became important in our first-flush study, when PSD was being measured. Changes in the PSD were observable after 10 to 12 hours of storage. Therefore, a holding time of six hours was established for particle size distribution analysis (Li et al. 2005; listed in the bibliography).
When litter first-flush monitoring is desired, we recommend using a large, reusable bag with a maximum of 6-millimeter openings. A drawstring bag can be placed over the discharge pipe to capture the entire flow but still allow the grab samples to be collected from a free waterfall. A monitoring site with a litterbag is shown in Figure 2. We recommend a minimum of three bags to collect litter during each storm event. The first bag is installed prior to the storm event and changed one hour after the start of the runoff, the second bag is installed to collect the litter for next two hours, and the third bag is used to capture litter until the end of the storm event. The third bag can be left in place until the next day, well after the end of the storm. The bags can be cleaned and reused for next storm event, and the drawstring arrangement allows the bag to break away if it becomes clogged, thus avoiding flooding. Litter characteristics can be evaluated using the procedures specified in Ruby and Kayhanian (2003) and Kim et al. (2004) and listed in the bibliography.
Meaningful Definition of First Flush
The first-flush phenomenon is generally assumed for single rainfall events and can be described as a concentration first flush or a mass first flush. A concentration first flush occurs when the first runoff has high concentration relative to runoff later in the storm event. A mass first flush (concentration times flow rate) is flow-dependent, and it occurs when both concentration and the initial runoff is high relative to mass emission rate in the later runoff. Concentration first flushes have been frequently reported, but mass first flushes have rarely been quantified. For example, most of the water-quality parameters monitored for all the events in our first-flush characterization study had higher concentrations at the beginning of the runoff than later in the runoff. Mass first flushes were usually observed but with lower magnitudes. This is due to the nature of the runoff, which generally has lower flow rate at the beginning of the storm than in the middle of the storm. Therefore, the mass emission rate in the middle of the storm event may be greater than at the beginning of the storm event, in spite of lower concentrations in the middle of the storm. The concept can be applied to any particular constituent or water-quality parameter. Therefore, a first flush in oil and grease (O&G), for example, can be called O&G first flush. A definition sketch of concentration first flush is shown in Figure 3. As shown, the concentration of chemical constituent in early runoff can be 10 times higher than the concentration of runoff at the end of storm event.
The concept of first flush can also be applied to a rainfall season. In California and many other areas of the world, rainfall occurs over distinct periods. For example, the bulk of the rainfall in southern California occurs from approximately November to March, with the months of January and February usually having the greatest rainfall. The long dry period from April or May to October allows contaminants to build up. The first large rainfall of the season, occurring anytime from October to January, generally mobilizes the built-up contaminants, creating a larger discharge. This phenomenon is called a seasonal first flush. In this article, the term first flush is used as follows:
First flush: The discharge of a larger mass or higher concentration in the early part of a storm relative to the later part of the storm. The term can be applied to any water-quality parameter or constituent, such as metals, litter, particles, toxicity, or turbidity, and both terms can be used to describe a mass first flush or a concentration first flush. The magnitude of the first flush will depend on site-specific conditions, but the term first flush is applicable.
Seasonal first flush: The discharge of a larger mass or higher concentration of the first storm or first few storms of a rainy season, relative to storms later in the season
Various ways have been previously proposed to quantify mass first flush, and absolute quantitative definitions have been offered. An early definition offered by Bertrand-Krajewski, Chebbo, and Saget (1998) is typical and suggested the existence of a first flush if 80% of the pollutant mass is emitted in the first 30% of the runoff. Other definitions and observations have been offered and will be discussed in greater detail later (Thornton and Saul 1987; Geiger 1987; Vorreiter and Hickey 1994; Saget, Chebbo, and Bertrand-Krajewski 1995; Gupta and Saul 1996; Sansalone and Buchberger 1997; Larsen, Broch, and Andersen 1998; Sansalone et al. 1998; Deletic 1998). They all in some way suggest a higher pollutant mass emission rate in the early part of the storm than in the later part, and the early part is generally considered the first 20% to 40% of the runoff volume. Under our study, we have proposed a mass first-flush ratio, or MFF, which quantitatively describes the mass first flush and is sufficiently broad to apply to any initial portion of the storm. A definition sketch of mass first-flush ratio for oil and grease is shown in Figure 4. Based on this definition sketch, we can determine the mass first flush for any constituent at any runoff volume.
It is possible to have a concentration seasonal first flush as well as a mass seasonal first flush. The techniques used to describe a mass first flush can also be used to describe a mass seasonal first flush. Occasionally, when investigators are describing both the first flush of a single storm and an entire season, they may use the term storm first flush to emphasize that the first flush is for a single storm event. In this article, the term storm first flush is not used.
The term first flush always refers to a single storm event, and seasonal first flush will always be used for an entire season.
Often one sees or reads of an investigator describing a very large watershed and noting that a first flush of watershed was not observed. Such conclusions are naïve, because in a large watershed, stormwater must be transported a great distance to a single discharge point or mouth of the watershed. Therefore, the time of travel of the runoff from various places in the watershed to the monitoring point is different (time of travel is the elapsed time for a quantity of stormwater to flow from the point of generation to the monitoring point). In this case, the first flush from each small area in the watershed arrives at the mouth of the watershed at different times, which mixes the smaller first flushes of each area into a broad discharge pattern. Therefore, the first flush from one area is mixed with runoff from other areas that occurred much later in the storm. The definition of large watershed for this context is a function of the time of travel. The first flush of pollutants observed in our study was generally within the first few minutes to the first hour after observable runoff. More important, most BMPs are designed to collect and treat smaller drainage areas rather than big watersheds.
First flushes are much less likely to occur in large watersheds.
Computation of Event Mean Concentrations (EMCs) and Mass First-Flush Ratio From Grab Samples
Procedure to Calculate EMC
Mathematically, EMCs can be defined as total pollutant mass (M) discharged during an event divided by total volume (V) discharge of the storm event.
In Equation 1, C(t) is a smooth real-valued function of time that represents the pollutant concentration curve, and Q(t) is also a smooth real-valued function of time that represents the stormwater flow rate curve. However, in practice, the integrals are not continuous functions of Q(t) and C(t) but approximations created by discrete measurements of Q(t) and C(t). If we assume we measure the concentration and the flow rate based on equal time interval in a storm event, the EMC can be estimated as
where qi and ci are the measurements for the discharge rate and pollutant concentration in the ith interval. From the point of view of approximating the continuous functions in Equation 2, the more measurements we take, the more accurate approximation we can obtain by Equation 2. When we view the measurements of the flow rate as the weights, Equation 2 becomes the discharge-weighted average throughout the storm event, as follows:
where wi is the flow weight, and
In practice, one common situation is the number of concentration measurements does not match the number of flow measurements. Generally, there are many fewer concentration measurements, because concentration measurements are much more expensive and time consuming; flow measurements can be easily and automatically obtained by most autosamplers with velocity probes. For most situations, the weights must be adjusted for each concentration measurement in Equation 3. One of the reasonable ways to adjust the weights is to use the discharge volume. One approach (Charbeneau and Barrett 1998) splits the discharge volume from the midpoint between two consecutive concentration measurements.
Figure 5 shows this approach, and the adjusted weight can be written as
where Vi is the corresponding discharge volume for the ith concentration measurement. This mid-discharge splitting method can also be applied for measurements at unequal time-interval bases. Alternatively, if the concentration measurements are based on constant discharge volume, the weighted average of wici form is reduced to the arithmetic average. Ideally, automated samplers collect samples in proportion to discharge volume. Additionally there are always slight errors (noise) in sample volume and pace that change the equal weights. Thus, an EMC can be calculated using a series of flow-weighted grab samples.
Procedure to Calculate MFF
To compute mass first-flush ratio, plot cumulative normalized mass (y-axis) versus normalized cumulative volume (x-axis) similar to that shown in Figure 6. This plot is known as load-graph. The mass first-flush ratio for n fraction runoff volume (MFFn) can easily be produced from load-graph. For example, for the mass first-flush ratio for 10% of runoff volume (MFF10), determine the normalized mass from the plot and then divide the normalized mass by normalized volume. The calculation for MFF10 and MFF30 is shown on this plot. The higher MFF ratio represents a larger mass first-flush effect.
First-Flush Data Presentation
Visual Observation of First Flush
First flush can qualitatively be evaluated through visual observation. Visual observation can be used to determine the first-flush effect of litter, turbidity, and other organics such as oil and grease. Figure 7 is an example showing the color and turbidity of water samples as a storm event progress. Clearly, the first few samples are more turbid, and the color is darker. However, the darker color and higher turbidity by itself is not indicative of higher organic and inorganic pollutant concentration. Therefore, the quantitative assessment of first flush is needed.
Concentration First-Flush Reporting
Pollutograph: Pollutographs are representations of the variability of water-quality-parameter concentrations throughout storm events. A pollutograph is a plot showing both the water-quality-parameter concentration and the hydrograph on the same plot. A higher concentration in an early storm event compared with the later period is an indicative of concentration first flush. An example pollutograph is shown in Figure 8. As shown, a pollutograph can show a number of water-quality parameters in a single plot, which can be helpful in visualizing relationships among parameters. Pollutographs can be produced for all water-quality parameters, including those with units other than volumetric concentrations (e.g., turbidity, conductivity). For these pollutographs, a correlation relationship with other water-quality parameters having volumetric concentration will be used.
Ratio of PEMC/EMC: Concentration first flush can also be reported based on the ratio of the partial event mean concentration (PEMC) to the entire event mean concentration. PEMC is calculated in the same fashion as EMC, except that Equation 2 or 3 is applied to only the first part of the storm. PEMC can be calculated for first 60, 90, or 120 minutes of rainfall and hence will be reported as PEMC60, PEMC90, or PEMC120. Table 1 presents the PEMC60/EMC for wide ranges of water-quality parameters. Evidence of first flush is present as long as the PEMC/EMC is larger than 1. The higher the ratio, the larger the concentration first flush. Hence, the PEMC/EMC ratio can be used in ranking the water-quality parameters based on their concentration first flush.
Mass First-Flush Reporting
As previously shown, the MFF ratio is computed through load-graph, which is the plot of normalized mass (y-axis) versus cumulative normalized runoff volume (x-axis). MFF ratio can usually be determined for 10%, 20%, 30%, 40%, and 50% of runoff volume. These MFF ratios are then reported as MFF10, MFF20, MFF30, MFF40, and MFF50, respectively. MFF results can be presented in different ways, including box plots and bar graphs. An example box plot showing the MFF ratio for total suspended solids (TSS) is presented in Figure 9.
Mass first flush of constituents with respect to 10%, 20%, 30%, 40%, or 50% of runoff can also be ranked and plotted. An example ranking of MMF20 based on their respective mass first-flush ratio is shown in Figure 10.
Toxicity First-Flush Reporting
A first-flush toxicity effect can be assessed when the frequency and magnitude of toxicity in the grab samples collected during the first period of each storm event is more apparent with greater magnitude than the toxicity observed in samples collected later during the storm. Using this definition, a first-flush effect was almost always observed in our previous toxicity evaluation with both species (P. promelas and C. dubia) for lethal and sub-lethal endpoints (Kayhanian and Stenstrom 2005). A typical hydrotoxicity graph (a plot that presents the level of toxicity during the entire storm event) is shown in Figure 11. As shown in this example, the average survival of P. promelas at the end of a seven-day exposure to the first five grab samples was near zero. The results presented in Figure 11 clearly indicate a first-flush effect. It is true that most early runoff is more toxic; however, a higher toxicity can occasionally occur in grab samples collected later during a storm. It is important to note that, even when a strong first-flush effect is observed, the composite samples can still be non-toxic.
The first-flush toxicity can also be reported through a visual plot showing relative toxicity on a mass basis normalized to flow volume over time. This evaluation removes the effect of flow rate on the interpretation of toxicity results. As an example, a first-flush toxicity effect on sea urchins for a representative storm event on February 11, 2003, from an urban highway site is shown in Figure 12. As shown, the first-flush effect for toxicity was evident, as the normalized proportion of toxicity to discharged runoff volume is greater for the initial stage of the storm event. For example, in Figure 12 the proportion of sea urchin toxicity discharged during the first 20% of storm duration is 80%.
Particle First-Flush Reporting
Particle concentrations have in the past usually been characterized by the mass of particles for specific size ranges, which is the most convenient method if particle counters are not available. With modern particle counters, it is possible to determine both the number of particles for specific size ranges as well as the mass of particles within these ranges using more traditional sizing methods (i.e., sieving, filtration, centrifugation, or settling). To distinguish between mass and numbers of particles, the term particle number is frequently used. Particle first flush can best be described by (1) hydrographic particle number or mass concentration (hydroparticle) graph, (2) change of particle size within certain size range at different storm-event duration, (3) partial particle event mean concentration (PPtEMC), and (4) particle number first-flush ratio (PNFF). The number of particles within a certain particle size range is defined as
where
N = particle concentration (#/mL) and
D1 and D2 = particle diameter (µm), D2>D1
PPtEMC is the accumulated number of particles at any time divided by the accumulated flow volume at the same point of time, as shown in Equation 7. Partial particle EMC is defined when it is integrated for runoff volume up to time t.
where
n(t) = particle number transported up to time t
v(t) = flow volume up to time t (m3)
ct = particle number concentration at time t (#/m3)
qt = flow rate at time t (m3/s)
t = time (s)
The PtNFF ratio is defined as the normalized number of particles divided by normalized volume fraction at any point of normalized runoff diagram. PtNFF ratio is calculated in exactly the same fashion as MFF ratios, except that particle numbers are used instead of chemical constituent concentrations. Therefore, PtNFF ratio for an x percent of runoff volume at time t1. is computed by Equation 8.
where
Q(t) = runoff flow rate (L3/T)
t(t) = particle number concentration (L-3)
V = total runoff volume of an event (L3)
N = total number of particles in an event
Figure 13 shows an example hydroparticle graph showing the change of particle concentration over the duration of the storm event. Similarly, particle concentration can be plotted against cumulative runoff volume during storm events. This plot for multiple storm events is shown in Figure 14. Similar particle number concentration plots can be produced for individual storm events. In both cases, it is clear that the particle concentration in early storm duration and early runoff volume is much higher than the later time and runoff volume.
Practical Application of First Flush in Treatment BMPs
As previously shown, both concentration and mass first flushes have almost always occurred for smaller and impervious (e.g., paved) watersheds. The existence of a first flush, either a storm or a seasonal first flush, may present opportunities for managers and regulators to affect better pollutant-reduction programs. Treating early runoff that has higher contaminant concentrations or mass may be a better policy than treating a similar fraction of the entire runoff volume. This is true for two reasons. The first reason is the cost of treatment is generally more dependent on the volume of water to be treated than the contaminant concentration. The second reason relates to the way the stormwater BMPs function; removal efficiency is greater at higher concentrations. Treatment efficiency at low concentrations can be nearly zero, but significant removal can be obtained at higher concentrations. The emerging American Society of Civil Engineers (ASCE) database on BMP trials shows this effect (Strecker et al. 2001).
The MFF ratios presented earlier can be very useful in estimating potential removals of pollutant mass from BMPs. For example, the California State Water Quality Control Board regulations require that all constructed BMPs must capture or treat 80% of a storm. For our three first-flush sites, this requirement means that storms as large as 35 millimeters (approx. 1.4 inches) of rainfall must be treated, as shown in Figure 15.
For storms larger than 80%, some portion of the flow must be bypassed. For very large storms, only a portion of the flow can be treated. For example, if only 50% of the flow can be treated, the BMP, if it captures all the runoff until it bypasses, has an opportunity to remove not just 50% of the mass of pollutants but 50% times the MFF50 ratio of the pollutants. From our investigation, the MFF50 for total zinc from combined three highway sites was found to be is approximately 1.6. Therefore, a BMP that treats 50% of the flow would in fact treat 80% of the total zinc mass.
One possible way to take advantage of first-flush treatment is to divide a BMP such as a detention basin into two compartments. A flow diagram showing the two-compartment design concept is shown in Figure 16. The first compartment captures the initial runoff and, after filling, bypasses to a second compartment, which functions as a continuous flow clarifier. The two-compartment design takes advantage of the first flush as well as other factors such as higher initial concentrations. This design is especially beneficial for removing particles and the associated particle-bound pollutants (Li et al. 2006a; listed in bibliography).
The two-compartment design presented in Figure 16 can also be applied to a detention basin. Under the first-flush treatment concept, the first compartment of the detention basin will have lower overflow rate (i.e., longer detention) to remove smaller particle size and the associated contaminants. Under larger storm events, much cleaner water will be discharged from the second compartment. The treated water from both compartments can be discharged from the surface, which is usually much less contaminated.
Acknowledgements
The first-flush characterization study was founded by the Caltrans Division of Environmental Analysis under an interagency agreement, contract 43A0073, between the University of California and Caltrans. We gratefully acknowledge the continuous support of technical and management staff of the DEA. We are thankful to our graduate students and research staff for their valuable contributions, especially Dr. Sim-Lin Lau, Lee-Hyung Kim, Sunny Li, Joohyon Kang, Simon Ha, Sabbir Khan, Haejin Lee, Mike Ma, and Dr. Peter Green. Special thanks to Mr. Ali Abrishamchi, who reproduced the colorful drawing for all figures (except Figures 7 and 16) used in this article.
Mention of the names of equipment, products, or supplies in this article shall not be construed as an endorsement. Opinions, findings, and conclusions or recommendations expressed in this report are those of the authors and do not necessarily reflect the views of Caltrans or the Regents of the University of California.
The final first-flush characterization report is currently available at the following Caltrans Web site: http://www.dot.ca.gov/hq/env/stormwater/special/newsetup/index.htm.
For detail information on first-flush characterization and related study, the interested readers may refer to the following publications.