Long-Term Performance and Life-Cycle Costs of Stormwater Best Management Practices (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

43 C H a P T E r 4 Long-term BMP performance represents the average annual performance over the life of the BMP, which depends greatly on various BMP unit treatment processes. Pollutant removal mechanisms in BMPs are based on unit operations and pro- cesses (UOPs) and the BMP system components (e.g., fore- bay, vegetation, media, outlet structure) that improve or enhance those processes. UOPs can be divided according to four fundamental process categories: (1) hydrologic opera- tions, (2) physical operations, (3) biological processes, and (4) chemical processes (Strecker et al., 2005). • Hydrologic operations, which are essentially a subset of physical operations, include the principles of flow attenu- ation (e.g., peak shaving, detention) and volume reduction (e.g., infiltration, evapotranspiration). • Physical operations include the principles of size separation and exclusion (e.g., screening, filtration), density separa- tion (e.g., sedimentation, flotation), aeration and volatil- ization, and physical agent disinfection (e.g., ultraviolet light, heat). • Biological processes include the principles of microbially mediated transformations (e.g., redox reactions resulting from microbial respiration) and uptake and storage (e.g., bioaccumulation). • Chemical processes include the principles of sorption (e.g., ion exchange, surface complexation), coagulation and flocculation (e.g., particle agglomeration, precipitation), and chemical agent disinfection (e.g., chlorine, ozone). Biological and chemical unit processes cannot be easily mod- eled due to the complex interaction of these processes with environmental variables. Thus, empirical methods based on measured data were used to evaluate these processes. Conse- quently, a BMP modeling approach was developed that uses a combination of long-term hydrologic simulation as described in Chapter 3 and summarized empirical data (described herein) to predict average annual load reductions for a selected suite of structural BMPs and pollutants of concern applicable to the highway environment. 4.1 BMPs and Constituents Analyzed BMPs selected for the tool were those that are typically used by DOTs for runoff from highways that can operate passively, with extended maintenance intervals. All BMPs are proven, recognized BMPs that have had substantial study to assess pol- lutant removal effectiveness and whole life costs. The BMPs analyzed were: • Vegetated swale, • Filter strip, • Dry detention basin, • Bioretention, • Wet pond, • Sand filter, and • PFC. The pollutants of concern selected for the calculations in the tool were based on the types of pollutants commonly monitored and observed in highway runoff and identified in NPDES permits and other regulatory publications. Selected pollutants additionally required adequate BMP performance data for analysis. The pollutants analyzed were: • Total zinc (TZn) • Total lead (TPb) • Total copper (TCu) • Total nitrogen (TN); estimated as the sum of NO3 and TKN • Total phosphorus (TP) • Nitrate (NO3) • TKN • Dissolved phosphorus (DP) • Orthophosphate (OP) as a surrogate for DP when needed Water Quality Estimation Methods and Data Sources

44 • TSS • Fecal coliform (FC) • Escherichia coli (E. coli) The sources of data used and the details of the BMP mod- eling approach are provided in the following sections. Addi- tional information on BMP performance estimation methods for a variety of UOPs can be found in Strecker et al. (2005), Huber et al. (2006), and Leisenring et al. (2013). 4.2 Highway Runoff Water Quality Data Water quality data that were used to develop long-term BMP performance estimates included highway runoff water quality and BMP influent and effluent concentrations. The sources of the data used are described in the following. Highway runoff quality data were obtained from the Highway-Runoff Database (HRDB) (Granato and Cazenas, 2009; Smith and Granato, 2010) and the National Stormwater Quality Database (NSQD) (Pitt, 2008). Tables 4-1 and 4-2 summarize the data available for the two databases after 1986. Data before this date were excluded in the analysis because of the use of leaded gasoline that caused an unrepresentative sample of modern conditions. The HRDB provides nearly three times as much highway runoff data as the NSQD. 4.3 BMP Influent and Effluent Concentrations Paired BMP influent and effluent concentration data were obtained from the International Stormwater BMPDB Version 03 24 2013. Nearly all of the selected BMPs had more than three distinct studies and 20 distinct influent/effluent measurement pairs per pollutant. PFC was the only BMP with only one study in the BMPDB, and it also had no bacteria data. Filter strips had no data pairs for DP or E. coli, so OP and FC were used as surrogates, respectively. Similarly, sand filters had no E. coli data, so FC was used as the surrogate. A summary of the num- ber of inflow/outflow data pairs per BMP and constituent is in Table 4-3. 4.3.1 Estimating Influent Concentrations To provide representative highway runoff quality inflows for BMP treatment analysis, highway runoff mean concen- trations used in the tool were developed through statistical analyses of data within the HRDB and NSQD. To assess the impact of average annual daily traffic (AADT) on constituent concentration, five AADT categories were created: 0–25,000, 25,000–50,000, 50,000–100,000, >100,000, and unknown. To improve the representativeness of the statistics generated, only data after 1986 were included in the analysis since data prior to this time are influenced by leaded gasoline and less- stringent emission control requirements on vehicles. As shown in Table 4-4, these categories provide a reasonable division of the data, with a fairly balanced distribution of the data between categories. In general, the 25,000–50,000 category has the least data. Values for TN are sparse, and TKN values were used where there was no data. Fecal coliform data are sparse in all categories, and no categorization was possible with the E. coli data. Tables 4-5 and 4-6 summarize the medians and arithme- tic means with 90% confidence intervals, respectively, for the pooled data sets for each AADT bin. A median is defined as the concentration where approximately 50% of the data are above and 50% of the data are below. The mean is the sum of the data divided by the number of data points. While both metrics provide an indication of the central tendency, the median is resistant to the effects of outliers. However, since the median is not a weighted metric, it can result in an NSQD HRDB Combined No. of sites 43 93 136 No. of events 669 1,537 2,206 No. of sample results 3,027 8,813 11,184 No. of non-detects 41 458 499 Table 4-1. Summary of available highway runoff quality data. Constituent Non-Detects/Total Samples TSS 11/1,713 NO3 92/1,047 TN 0/122 TKN 49/1,408 DP 32/217 TP 120/2,022 TCu 72/1,808 TPb 102/1,683 TZn 12/2,099 FC 0/65 E. coli 0/13 Table 4-2. Summary of non-detects and total samples for each constituent from the HRDB and NSQD combined.

45 underestimate of pollutant loads when data are skewed to the right (typical of water quality data). Therefore, the arithmetic means are recommended when computing pollutant loads, and the medians are recommended when comparing concen- tration benchmarks or thresholds. To handle non-detects, a robust regression-on-order statis- tics (ROS) method as described by Helsel and Cohn (1988) was used to provide probabilistic estimates of non-detects before computing descriptive statistics. As compared to simple substi- tution methods [e.g., ½ detection limit (DL), DL, or zero], the ROS method reduces the potential bias caused by the presence of non-detects. Confidence intervals were generated using the bias corrected and accelerated (BCa) bootstrap method described by Efron and Tibishirani (1993). This method for computing confidence intervals is resistant to outliers and does not require any restrictive distributional assumptions common with parametric confidence intervals. Because the data were pooled for all sites, the analysis accounts for the variability at individual sites (temporal variability) as well as between sites (spatial variability). However, it is acknowledged that sites with more data points will have a larger influence on the pooled summary statistics. As indicated in Tables 4-5 and 4-6, there does not appear to be a clear relationship between AADT and pollutant concentration Constituent BMP Type Bioretention Grass Swale Filter Strip Wet Pond Detention Basin Sand Filter PFC* TSS 171 195 526 621 265 296 22 NO3 19 77 414 122 105 158 22 TN 160 92 122 300 59 127 0 TKN 167 151 512 406 176 270 22 DP 21 52 16 236 117 65 22 OP 123 26 435 361 34 99 0 TP 214 191 518 586 245 286 22 TCu 67 119 382 425 191 267 22 TPb 54 138 403 465 193 248 22 TZn 110 152 412 522 209 293 22 FC 26 79 20 100 109 121 0 E. coli 54 39 0 50 32 0 0 *PFC pairs are based on paired watershed data since the influent cannot be directly sampled for this type of BMP. Table 4-3. Summary of available data pairs per BMP and constituent from the BMPDB. Constituent AADT Bin 0–25k 25k–50k 50k–100k 100k + Unknown All TSS 388 198 301 563 263 1,713 NO3 355 151 191 350 0 1,047 TN 0 0 3 0 119 122 TKN 336 146 176 412 338 1,408 DP 46 38 28 73 32 217 TP 428 264 332 508 490 2,022 TCu 426 243 304 555 280 1,808 TPb 402 240 264 492 285 1,683 TZn 424 253 323 569 530 2,099 FC 3 0 4 19 39 65 E. coli 0 0 0 0 13 13 Table 4-4. Count of sample results by constituents by average annual daily traffic.

46 Constituent Medians (90% Confidence Intervals) by AADT Bin 0–25k 25k–50k 50k–100k 100k+ Unknown All TSS 42.05 61.49 69.07 100.2 39.64 69.2 (mg/L) (33.00–45.00) (51.00–71.00) (57.00–72.00) (92.00–106.0) (34.30–44.00) (66.00–73.44) NO3 0.2 0.82 0.59 1.07 No data 0.6 (mg/L) (0.20–0.24) (0.71–0.89) (0.49–0.66) (0.86–1.16) (0.52–0.61) TN No data No data 3.01 No data 3.00 3.00 (mg/L) (2.30–5.52) (2.59–3.27) (2.63–3.29) TKN 1.00 1.80 1.55 2.16 1.64 1.64 (mg/L) (0.84–1.10) (1.60–2.00) (1.42–1.67) (2.00–2.30) (1.50–1.73) (1.56–1.70) DP 0.07 0.10 0.08 0.18 0.09 0.10 (mg/L) (0.07–0.08) (0.05–0.13) (0.03–0.12) (0.15–0.20) (0.05–0.10) (0.07–0.10) TP 0.12 0.16 0.20 0.24 0.25 0.20 (mg/L) (0.10–0.13) (0.14–0.17) (0.17–0.22) (0.22–0.25) (0.21–0.26) (0.19–0.20) TCu 7.95 18.2 23.48 48.73 10.89 22.37 (ug/L) (6.80–9.00) (15.00–20.00) (21.30–25.00) (43.92–51.60) (9.79–12.50) (20.59–23.00) TPb 3.92 10.51 7.82 30.7 56.52 16.02 (ug/L) (3.00–4.50) (8.82–12.00) (6.20–8.80) (26.00–34.00) (45.00–67.00) (14.00–17.00) TZn 51.64 90.74 123.8 220.0 86.04 116.8 (ug/L) (43.00–56.65) (79.00–100.0) (110.0–131.0) (200.0–238.0) (76.37–93.33) (110.0–120.0) FC 5000 No data 5147 1626 2064 1986 (colonies /100mL) (300.0–13,000) (1,100–9,500) (1,300–1,700) (492.9–2,200) (1300–2,300) E. coli No data No data No data No data 1971 1977 (colonies /100mL) (727.2–2,300) (680.0–2,300) Table 4-5. Medians and confidence intervals for combined NSQD and HRDB data. Constituent Means (90% Confidence Intervals) by AADT Bin 0–25k 25k–50k 50k–100k 100k+ Unknown All TSS 162.8 178.3 120.1 143.6 85.2 138.8 (mg/L) (136.1–190.4) (127.1–233.8) (95.1–150.6) (130.6–157.1) (72.84–98.43) (127.4–150.3) NO3 0.48 1.12 0.82 1.74 No data 1.06 (mg/L) (0.42–0.53) (0.94–1.32) (0.73–0.92) (1.51–2.02) (0.96–1.16) TN No data No data 3.61 No data 3.59 3.59 (mg/L) (2.30–4.68) (3.17–4.03) (3.18–4.02) TKN 1.62 2.5 1.9 3.18 2.11 2.32 (mg/L) (1.45–1.81) (2.23–2.76) (1.72–2.09) (2.84–3.50) (1.94–2.28) (2.20–2.44) DP 0.09 0.14 0.12 0.54 0.09 0.25 (mg/L) (0.08–0.10) (0.11–0.17) (0.09–0.15) (0.32–0.81) (0.07–0.11) (0.17–0.34) TP 0.38 0.46 0.25 0.39 0.68 0.44 (mg/L) (0.27–0.49) (0.29–0.63) (0.23–0.28) (0.34–0.44) (0.47–0.99) (0.37–0.52) TCu 14.92 26.83 30.79 82.11 27.11 41.76 (ug/L) (13.50–16.44) (24.18–29.42) (28.23–33.32) (60.65–114.6) (20.29–35.10) (34.68–51.86) TPb 18.26 31.29 26.24 61.6 77.63 44.08 (ug/L) (10.17–30.10) (26.36–36.73) (21.38–31.64) (53.81–70.28) (70.32–85.98) (40.37–48.32) TZn 98.0 152.1 172.7 329.6 143.0 189.9 (ug/L) (87.7–108.0) (133.1–170.6) (157.6–188.2) (287.0–382.6) (128.1–157.6) (176.8–205.7) FC 6148 No data 5625 8702 9215 8700 (colonies/ 100mL) (300.0–10,333) (1,700–8,575) (1,795–15,786) (3,520–16,607) (4,519–13,557) E. coli No data No data No data No data 5948 6025 (colonies/ 100mL) (1,717–12642) (1,714–12,654) Table 4-6. Means and confidence intervals for combined NSQD and HRDB data.

47 except for possibly TSS, total phosphorus, total copper, and total zinc, particularly when comparing the low traffic AADT (<25k) against the high traffic AADT (>50k). For consis- tency across all pollutants, the mean concentrations for all of the data combined (rightmost column in Table 4-6) are used in the tool as the default highway runoff concentrations regardless of AADT, but these defaults may be overridden if desired. 4.3.2 Estimating Effluent Concentrations Effluent concentrations were estimated based on regres- sion analysis of influent and effluent water quality data, when appropriate, or were simply summarized effluent data when regression analysis was not possible for the available data. BMPs are assumed to not be a source of pollutants, and thus effluent concentrations will not exceed the influent concentra- tions or load. While some BMPs can contribute to increased constituent concentrations, quantifying export in excess of the incoming load introduces mass balance errors that cannot be reconciled without quantifying the pollutant mass available within the BMP at the time of installation. Since this infor- mation is not typically available, the default assumption is no concentration reduction for pollutants that may in time be exported by a BMP. The BMPDB is a repository of influent and effluent water quality data from over 500 BMP studies (as of March 2013). This database provides an avenue for a data-driven analysis of the relationship between influent concentration (Cinf) and effluent concentration (Ceff) for a wide range of BMP-pollutant combinations. Data from the BMPDB were analyzed using a multistep process. This process is shown in Figure 4-1 and consists of five steps: 1. Determine if sufficient paired data for analysis exist in the BMPDB. 2. Determine if there is a statistical difference between Cinf and Ceff. 3. Determine if a monotonic relationship exists between Cinf and Ceff. 4. Conduct linear and log-linear regression between Cinf and Ceff and develop functional relationship. 5. Ensure that results do not show logical inconsistencies (e.g., dissolved fraction is greater than total). Since water quality data are often highly variable and posi- tively skewed, nonparametric statistics were selected over parametric statistics for this analysis. The Wilcoxon signed- rank test was used to evaluate whether the influent and effluent concentrations are statistically different, and the Spearman’s rho correlation coefficient was used to evaluate whether a monotonic relationship exists (Helsel and Hirsch, 2002). The Wilcoxon signed-rank test assumes that the distribution of the paired differences is symmetric, so the data were log- transformed prior to conducting the test. No transformation Figure 4-1. Analysis process for influent-effluent regression. Note: KTRL = Kendall-Theil robust line. BMP-pollutant pair Yes Yes Yes No KTRL regression on (1) Ceff vs. Cinf, (2) Ceff vs. ln(Cinf), (3) ln(Ceff) vs. ln(Cinf). Select best fit. Monotonic relaonship (Spearman’s rho test)? Stascal difference between influent and effluent (Wilcoxon test)? Sufficient data for regression (≥3 disnct studies and ≥20 pollutant data pairs)? No No removal assumed for BMP-pollutant No Ceff = mean effluent from BMP database Use relaonship based on similar pollutant (DP uses OP data, E. coli uses FC data)

48 was needed for the Spearman’s rho computation because the correlation analysis uses the ranks of the data. If the Wilcoxon test found a statistically significant differ- ence between the influent and effluent concentrations, and the Spearman’s rho test found that a monotonic relationship exists, regression equations were developed using the Kendall- Theil robust line procedure described by Granato (2006). Linear and log-linear relationships were evaluated, and the best-fit equation was used based on the median absolute differ- ence. Statistical significance for all analyses was determined at a level of a = 0.10. The analysis results are presented and discussed in the following. 4.3.3 Statistical Difference Between Influent and Effluent Quality While some pollutants, such as TSS, are easily removed by a wide variety of BMPs, others, such as NO3, are more difficult to remove. The nonparametric Wilcoxon signed- rank test was used to verify a statistical difference between influent and effluent quality for each BMP-pollutant pair to determine if removal of a pollutant was occurring in a BMP. Because this test requires a symmetric distribution, the data were log-transformed prior to performing the analysis. As shown in Table 4-7, most BMP-pollutant combinations involving nutrients and bacteria indicators show statistically significant concentration reductions (p < 0.1; bolded). For BMP-pollutant combinations that failed to show statisti- cal significance (p > 0.1), no removal due to concentration changes would be assumed. Wet ponds are the only BMP type that show statistically significant removal for all ana- lyzed pollutants. 4.3.3.1 Monotonic Relationship Between Influent and Effluent The next step in this process required establishing the pres- ence of a monotonic relationship between influent and efflu- ent quality. To do this, the Spearman’s rho test was applied to each BMP-pollutant combination. Those combinations showing a statistically significant difference between Cinf and Ceff generally exhibited a monotonic relationship between the two. The only exceptions were the swale-DP combina- tion, the filter strip-FC combination, and all available pollut- ant data for PFC, where a statistically significant monotonic relationship between Cinf and Ceff was not observed. In these cases, a regression analysis was not performed. However, since the Wilcoxon test results indicated a statistically signifi- cant reduction in DP for swales and a statistically significant reduction in all pollutants except for NO3 and DP for PFC, the arithmetic estimate of the log mean of effluent concen- tration data from the BMPDB was selected as an appropriate estimate of Ceff for these BMP-pollutant combinations. Note that when implementing constant effluent concentrations, the BMPs are assumed to never be a source of pollutants. There- fore, if Cinf is estimated to be less than Ceff, no concentration reduction is assumed. As shown in Table 4-8, the correlation analysis for PFC indicates that the effluent concentrations for all available pol- lutants are not correlated with the influent concentrations. Viewing these results for PFCs with the Wilcoxon signed-rank test results, it is concluded that average effluent concentra- tions independent of influent concentrations are appropriate for all pollutants except for NO3 and DP. No removal will be assumed for these two pollutants, and no removal will also be assumed for E. coli due to lack of data. For other constituents, Pollutant Wilcoxon p-values by BMP Type (Bold values indicate statistically significant removals.) Bio- retention Grass Swale Filter Strip Wet Pond Detention Basin Sand Filter PFC TSS <0.001 0.023 <0.001 <0.001 <0.001 <0.001 <0.001 NO3 N/A <0.001 <0.001 <0.001 <0.001 <0.001 0.118 TKN 0.037 0.485 0.157 <0.001 <0.001 <0.001 <0.001 DP 0.035 <0.001 N/A <0.001 0.659 0.066 0.239 OP <0.001 <0.001 <0.001 <0.001 0.458 <0.001 N/A TP 0.984 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 TCu <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 TPb <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 TZn <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 FC <0.001 0.525 0.279 <0.001 0.007 <0.001 N/A E. coli 0.026 0.128 N/A <0.001 <0.001 N/A N/A Table 4-7. Wilcoxon signed-rank test results (p-values).

49 Pollutant Spearman’s Rho p-values by BMP Type (Bold values indicate statistically significant influent/effluent correlation.) Bioretention Grass Swale Filter Strip Wet Pond Detention Basin Sand Filter PFC TSS 0.30 (<0.001) 0.46 (<0.001) 0.46 (<0.001) 0.46 (<0.001) 0.55 (<0.001) 0.41 (<0.001) 0.2 (0.286) NO3 N/A 0.89 (<0.001) 0.65 (<0.001) 0.53 (<0.001) 0.79 (<0.001) 0.75 (<0.001) 0 (0.636) TKN 0.57 (<0.001) 0.73 (<0.001) 0.57 (<0.001) 0.59 (<0.001) 0.70 (<0.001) 0.71 (<0.001) 0.07 (0.389) DP -0.06 (0.786) 0.68 (<0.001) N/A 0.52 (<0.001) 0.67 (<0.001) 0.69 (<0.001) 0.05 (0.416) OP 0.46 (<0.001) 0.80 (<0.001) 0.58 (<0.001) 0.57 (<0.001) 0.67 (<0.001) 0.65 (<0.001) N/A TP 0.38 (<0.001) 0.63 (<0.001) 0.46 (<0.001) 0.63 (<0.001) 0.66 (<0.001) 0.71 (<0.001) 0.36 (0.207) TCu 0.41 (<0.001) 0.81 (<0.001) 0.70 (<0.001) 0.58 (<0.001) 0.87 (<0.001) 0.61 (<0.001) 0.27 (0.245) TPb N/A N/A 0.78 (<0.001) 0.55 (<0.001) 0.90 (<0.001) 0.71 (<0.001) 0.29 (0.236) TZn 0.49 (<0.001) 0.82 (<0.001) 0.63 (<0.001) 0.50 (<0.001) 0.72 (<0.001) 0.43 (<0.001) 0.19 (0.291) FC 0.70 (<0.001) 0.83 (<0.001) 0.31 (0.177) 0.78 (<0.001) 0.65 (<0.001) 0.70 (<0.001) N/A E. coli 0.34 (0.012) 0.83 (<0.001) N/A 0.78 (<0.001) 0.58 (<0.001) N/A N/A Table 4-8. Spearman’s rho test results (p-values). influent and effluent relationships that did not fail the Spear- man’s rho test (p < 0.1; bold) are shown in Table 4-8. 4.3.3.2 Regression Analysis of the Relationship Between Influent and Effluent Based on the results of the Wilcoxon and Spearman’s rho tests, several BMPs appear to provide statistically significant reductions in pollutant concentrations along with mono- tonic influent/effluent relationships. These results together indicate that regression analyses can be conducted to develop functional relationships that can be used to predict BMP performance. Given the prevalence of outliers in environmental data and the strong influence these outliers can have on standard lin- ear regression techniques, the nonparametric Kendall-Theil robust line (KTRL) regression approach was used (Granato, 2006). The KTRL was applied to the influent and effluent data in original units and after log-transformation. The regression plots are provided in Appendix C: International Stormwater BMP Database Performance Information. The KTRL method computes the median of all possible pairwise slopes between two data sets. A y-intercept is then calculated according to Equation 2. (Eq. 2)Intercept median y median slope median x( ) ( ) ( )= − p Similar to linear regression, the calculation of slope (m) and intercept (b) creates a line of the form y = mx + b that can be used as a generalized relationship between x and y. Kendall- Theil robust lines were calculated for three possible relation- ships between influent and effluent, as shown in Table 4-9. The median absolute deviation (MAD) was used to select the best regression equation for each BMP-pollutant combi- nation. This statistic is defined by Equation 3. MAD median C for all values ofCeff predicted= − Ceff( ) ( . )Eq 3 Where Cpredicted is the value of the Ceff predicted by the Kendall- Theil regression line. Best-fit regression results and plots are provided in Appen- dix C: International Stormwater BMP Database Performance Information. Table 4-9. KTRL equations used for nonparametric regression. Data Pairs Plotted for KTRL Calculations KTRL Equation Derived Ceff, Cinf Ceff = m * Cinf + b Ceff, ln(Cinf) Ceff = m * ln(Cinf) + b ln(Ceff), ln(Cinf) ln(Ceff) = m * ln(Cinf) + b

50 4.4 Influent Highway Runoff Water Quality Methods Influent highway runoff concentrations are calculated as described in Section 4.3.1 and are shown in the rightmost column of Table 4-6. Tool users have the option of overriding this default with a value from the table or from other moni- toring data. Runoff volumes and loads are calculated as: i i (Eq. 4)V R A Pw v w= i (Eq. 5)L V Cw w w= Where, Vw is the average annual runoff volume, Rv is the long-term, volumetric runoff coefficient, Aw is watershed area, P is average annual rainfall depth, Lw is the average annual load, and Cw is the characteristic runoff concentration. 4.5 BMP Effluent Quality Performance by Pollutant Effluent quality is estimated using the regression analysis approach described in this section. Regression equations were developed using all available storm event data pairs for each BMP-pollutant combination where both a statistically signifi- Pollutant Bioretention Grass Swale Filter Strip Wet Pond Detention Basin Sand Filter PFC TSS 3 3 3 3 3 3 8 NO3 4 1 1 1 1 1 4 TKN 2 4 4 2 1 1 8 TN 9 9 9 9 9 9 9 DP 8 1 4 1 4 1 4 TP 4 2 2 1 2 2 8 TCu 3 3 3 3 3 3 8 TPb 4 3 1 2 1 1 8 TZn 3 3 3 3 3 3 8 FC 3 4 4 3 3 3 4 E. coli 3 4 7 3 3 7 4 1 - KTRL regression of Ceff vs. Cinf. 2 - KTRL regression of Ceff vs. ln(Cinf). 3 - KTRL regression of ln(Ceff) vs. ln(Cinf). 4 - Failed Wilcoxon test or lack of data for analysis. No removal assumed. 5 - Insufficient data for DP analysis. KTRL line [Ceff vs. ln(Cinf)] based on OP data. 6 - Insufficient data for DP analysis. OP data failed Wilcoxon test. No removal assumed. 7 - Insufficient paired data for analysis. Used data for fecal coliform to develop equation parameters for this BMP. 8 - Failed Spearman’s test for monotonic relationship, but passed Wilcoxon test. Ceff = arithmetic estimate of log mean for all available effluent data in the BMP database using regression-on-order statistics for handling non-detects followed by bootstrapping as described in Geosyntec and Wright Water Engineers (2012). 9 - To be determined by addition of NO3 and TKN (nitrite assumed negligible). Table 4-10. Equation selection summary for BMP-pollutant combinations. cant reduction was observed (Wilcoxon) and a monotonic rela- tionship was found. Table 4-10 summarizes the form of equation selected for each BMP-pollutant combination based on the hypothesis test results and the best-fit regression equation. Based on the various possible influent–effluent relationships considered in Table 4-9, a generalized equation was developed: C C A B C C C D C eff E = + + ( ) + min , max ln inf inf inf inf i i i +          e DLi , ( . )Eq 6 where Ceff is the predicted effluent concentration; Cinf is the estimated influent concentration; A, B, C, D, and E are parameters of the equation; ei is the bias correction factor for Equation 3; and DL is the minimum detection limit reported for the avail- able data sets. This generalized equation allows for any regression equa- tion to be used as long as the correct parameters are used and the remaining parameters have a value of zero. This equation ensures that BMPs are not a source of pollutants (e.g., Ceff is never greater than Cinf) and predicted effluent concentration is never below a reported detection limit. Tables 4-11 through 4-17

51 Pollutant A B C D E ei DL TSS (mg/L) 0.00 0.00 0.00 2.49 0.37 1.35 0.00 NO3 (mg/L) 0.00 1.00 0.00 0.00 0.00 0.00 0.00 TKN (mg/L) 0.83 0.00 0.50 0.00 0.00 0.71 0.04 TN (mg/L) ← TN = TKN + NO3 → DP (mg/L) -0.82 0.00 0.00 0.00 0.00 0.00 0.03 TP (mg/L) 0.00 1.00 0.00 0.00 0.00 0.00 0.01 TCu (ug/L) 0.00 0.00 0.00 2.77 0.44 1.26 0.50 TPb (ug/L) 0.00 1.00 0.00 0.00 0.00 0.00 1.00 TZn (ug/L) 0.00 0.00 0.00 1.11 0.68 1.26 0.01 FC (colonies/100mL) 0.00 0.00 0.00 0.01 1.06 7.29 100.00 E. coli (colonies/ 100mL) 0.00 0.00 0.00 2.40 0.51 24.48 1.00 Table 4-11. Equation parameters for predicting bioretention effluent concentrations. Pollutant A B C D E ei DL TSS (mg/L) 0.00 0.00 0.00 5.74 0.45 1.35 0.50 NO3 (mg/L) 0.02 1.07 0.00 0.00 0.00 0.00 0.10 TKN (mg/L) 0.00 1.00 0.00 0.00 0.00 0.00 0.10 TN (mg/L) ← TN = TKN + NO3 → DP (mg/L) -0.01 1.41 0.00 0.00 0.00 0.00 0.02 TP (mg/L) 0.44 0.00 0.12 0.00 0.00 0.00 0.01 TCu (ug/L) 0.00 0.00 0.00 0.85 0.88 0.92 6.00 TPb (ug/L) 0.00 0.00 0.00 0.66 0.92 0.87 3.00 TZn (ug/L) 0.00 0.00 0.00 2.99 0.56 1.21 0.01 FC (colonies/100mL) 0.00 1.00 0.00 0.00 0.00 0.00 1,000.00 E. coli (colonies/ 100mL) 0.00 1.00 0.00 0.00 0.00 0.00 0.00 Table 4-12. Equation parameters for predicting swale effluent concentrations. Pollutant A B C D E ei DL TSS (mg/L) 0.00 0.00 0.00 2.602685 0.526759 1.41 1.00 NO3 (mg/L) 0.107391 0.608696 0.00 0.00 0.00 0.00 0.01 TKN (mg/L) 0.00 1.00 0.00 0.00 0.00 0.00 0.1 TN (mg/L) ← TN = TKN + NO3 → DP (mg/L) 0.000 1.000 0.000 0.000 0.000 0.00 0.001 TP (mg/L) 0.224431 0.00 0.039056 0.00 0.00 0.00 0.001 TCu (ug/L) 0.00 0.00 0.00 1.224849 0.662004 1.29798 0.50 TPb (ug/L) 0.083685 0.300187 0.00 0.00 0.00 0.00 0.15 TZn (ug/L) 0.00 0.00 0.00 1.362558 0.679601 1.22 1.00 FC (colonies /100mL) 0.00 1.00 0.00 0.00 0.00 0.00 1.00 E. coli (colonies/ 100mL) 0.00 1.00 0.00 0.00 0.00 0.00 1.00 Table 4-13. Equation parameters for predicting filter strip effluent concentrations.

52 Pollutant A B C D E ei DL TSS (mg/L) 0.00 0.00 0.00 1.62 0.51 2.33 0.50 NO3 (mg/L) -0.02 0.50 0.00 0.00 0.00 0.00 0.06 TKN (mg/L) 0.88 0.00 0.51 0.00 0.00 0.00 0.06 TN (mg/L) ← TN = TKN + NO3 → DP (mg/L) 0.02 0.41 0.00 0.00 0.00 0.00 0.00 TP (mg/L) 0.05 0.34 0.00 0.00 0.00 0.00 0.00 TCu (ug/L) 0.00 0.00 0.00 1.41 0.55 1.22 0.01 TPb (ug/L) -0.12 0.00 0.92 0.00 0.00 0.00 0.00 TZn (ug/L) 0.00 0.00 0.00 1.94 0.60 1.26 0.01 FC (colonies/100mL) 0.00 0.00 0.00 0.26 1.05 1.45 2.00 E. coli (colonies/ 100mL) 0.00 0.00 0.00 0.04 1.12 3.45 2.00 Table 4-14. Equation parameters for predicting wet pond effluent concentrations. Pollutant A B C D E ei DL TSS (mg/L) 0.00 0.00 0.00 2.16 0.59 1.42 1.00 NO3 (mg/L) 0.13 0.73 0.00 0.00 0.00 0.00 0.10 TKN (mg/L) 0.32 0.68 0.00 0.00 0.00 0.00 0.02 TN (mg/L) ← TN = TKN + NO3 → DP (mg/L) 0.00 1.00 0.00 0.00 0.00 0.00 0.02 TP (mg/L) 0.41 0.00 0.14 0.00 0.00 0.00 0.02 TCu (ug/L) 0.00 0.00 0.00 0.94 0.84 1.10 0.10 TPb (ug/L) 0.60 0.36 0.00 0.00 0.00 0.00 0.10 TZn (ug/L) 0.00 0.00 0.00 1.87 0.71 1.06 0.01 FC (colonies/100mL) 0.00 0.00 0.00 11.37 0.66 2.60 1.00 E. coli (colonies/ 100mL) 0.00 0.00 0.00 2.84 0.65 2.89 1.00 Table 4-15. Equation parameters for predicting detention basin effluent concentrations. Pollutant A B C D E ei DL TSS (mg/L) 0.00 0.00 0.00 1.38 0.46 1.69 0.50 NO3 (mg/L) 0.11 1.21 0.00 0.00 0.00 0.00 0.01 TKN (mg/L) 0.19 0.35 0.00 0.00 0.00 0.00 0.10 TN (mg/L) ← TN = TKN + NO3 → DP (mg/L) 0.02 0.69 0.00 0.00 0.00 0.00 0.02 TP (mg/L) 0.20 0.00 0.05 0.00 0.00 0.00 0.00 TCu (ug/L) 0.00 0.00 0.00 1.16 0.73 1.10 0.40 TPb (ug/L) 0.20 0.11 0.00 0.00 0.00 0.00 0.12 TZn (ug/L) 0.00 0.00 0.00 2.26 0.46 1.37 0.01 FC (colonies/100mL) 0.00 0.00 0.00 0.89 0.87 2.85 2.00 E. coli (colonies/ 100mL) 0.00 0.00 0.00 0.89 0.87 2.85 2.00 Table 4-16. Equation parameters for predicting sand filter effluent concentrations.

53 indicate the parameters for predicting effluent concentrations for each BMP. The regression equations are used to represent the average performance for each BMP type, not event-by-event concen- trations. For a particular site, the equations are used to produce an average effluent concentration given an average influent concentration. Example best-fit bioretention regression lines are shown in Figure 4-2 for copper and Figure 4-3 for zinc. The 95% confidence interval about the effluent median concentra- tions as reported in Geosyntec Consultants and Wright Water Engineers (2012) is also shown in the figures to illus- trate what the estimated concentrations would be if influ- ent concentrations were not considered in the performance estimate. Pollutant A B C D E DL TSS (mg/L) 13.70 0.00 0.00 0.00 0.00 1.00 NO3 (mg/L) 0.00 1.00 0.00 0.00 0.00 0.04 TKN (mg/L) 1.11 0.00 0.00 0.00 0.00 0.40 TN (mg/L) ← TN = TKN + NO3 → DP (mg/L) 0.00 1.00 0.00 0.00 0.00 0.02 TP (mg/L) 0.086 0.00 0.00 0.00 0.00 0.02 TCu (ug/L) 13.00 0.00 0.00 0.00 0.00 2.00 TPb (ug/L) 0.84 0.00 0.00 0.00 0.00 0.50 TZn (ug/L) 25.80 0.00 0.00 0.00 0.00 5.00 FC (colonies/100mL) 0.00 1.00 0.00 0.00 0.00 1.00 E. coli (colonies/ 100mL) 0.00 1.00 0.00 0.00 0.00 1.00 Table 4-17. Equation parameters for predicting PFC effluent concentrations. 4.6 Load Reduction Assessment Load reduction prediction in the BMP Evaluation Tool depends on three primary calculations that use the hydrologic simulation results to predict volume captured and volume reduced, as described in Chapter 3, and effluent concentra- tion analysis methodology to provide volume and pollut- ant estimations. The tool reports the following: (1) annual stormwater runoff volume to the BMP, (2) amount of runoff captured and reduced by the BMP, and (3) BMP influent and effluent concentrations. Thus, the tool computes runoff loads and load reductions in a sequence of steps based on a mass balance approach, as indicated in Figure 4-4. Runoff loads are estimated as the product of the average annual runoff volume (Vw) and the characteristic runoff Figure 4-2. Bioretention regression line for total copper.

54 Figure 4-3. Bioretention regression line for total zinc. concentration (Cw). The total estimated percent capture is used to determine the load bypassed (VbyCw) and influ- ent load (VInfCw). Concentration reductions by the BMP are determined using the influent–effluent relationships described in Section 4.3 using the equation parameters for each BMP-pollutant combination shown in Tables 4-11 through 4-17. The effluent volume (VEff) is computed as the difference between the influent volume (VInf) and volume reduction estimated from the nomographs (VRd). The effluent load is then the product of the effluent vol- ume and estimated effluent concentration (CEff). The com- bined discharge load and the load reductions are simply computed by applying a mass balance of the other terms. The percent average annual load reduction (%LR) is finally computed as: L L L L w Eff Dis w %LR (Eq. 7)= − − Where all terms are defined as previously and in Figure 4-4. Figure 4-4. General approach for computing BMP load reductions.