Module 3a: Sources and Characteristics of Stormwater Pollutants in Urban Areas Robert Pitt Department of Civil and Environmental Engineering The University of Alabama Tuscaloosa, AL and Alex Maestre Stormwater Management Authority Birmingham, AL Urban runoff takes many forms and has highly variable characteristics Stormwater Snowmelt Stormwater NPDES Data Collection and Evaluation Project The University of Alabama and the Center for Watershed Protection were awarded an EPA 104(b)3 grant in 2001 to collect, review, and analyze selected Phase 1 NPDES stormwater permit data. Samples collected from the same site during the same storm We created a national database, the National Stormwater Quality Database (NSQD), that is available on the Internet. 1
Data Reviews We contacted Phase 1 communities throughout the nation to request their monitoring data and program descriptions. Chesapeake Bay area and the southeast were emphasized during this initial project activity, but we have acquired data from most areas of the U.S. to enable some basic statistical tests examining geographical influences. Data Reviews (cont.) Quality assurance/quality control reviews of the submitted data were a major project effort and were based on: data trends, relationships between constituents, analytical methods, reasonableness of data (comparisons with historical benchmarks), detection limits, sampling methods, sampling locations, extreme values, completeness of descriptions, etc. Current Database Content Stormwater data from about 4,000 events from 64 agencies and 17 states have been obtained, entered in our database, and undergone QA/QC reviews. This represents about 1/4 of the total NPDES Phase 1 MS4 data that has been collected by municipalities at a very high cost. Many of the Phase 1 communities spent about $1M each just for the application information and initial monitoring. About 125 site parameters and water quality constituents are recorded, but most data are available for about 35 pollutants. Available Stormwater Data Much stormwater information has been collected over the past 35 years, but it is not easily accessible. Main historical US nationwide database is the Nationwide Urban Runoff Program (NURP) information (EPA 1983). NPDES Phase 1 stormwater permit monitoring data collected since early 1990s, but not readily accessible, and some have questioned its quality control. 2
Comparison of National Stormwater Quality Database with NURP Data Database Years of Data Collection Number of Monitoring Stations Number of Metropolitan Areas Total Station-Storm Events NURP 1978-1983 81 28 2,300 NSQD Ver 1.1 1992-2003 369 64 3,770 Comparison of NURP with Phase 1 Data (median and COV) TSS COD Pb Zn NURP NSQD 1.1 NURP NSQD 1.1 NURP NSQD 1.1 NURP NSQD 1.1 Resid. 101 (1) 49 (1.8) 73 (0.6) 55 (0.93) 144 (0.8) 12 (1.9) 135 (0.8) 73 (1.3) Commer. 69 (0.9) 42 (2) 57 (0.4) (1) 104 (0.7) 18 (1.6) 226 (1) 150 (1.2) Open Space 70 (3) 49 (1.5) 40 (0.5) 42 (1.5) 30 (2) 10 (1.7) 195 (0.7) 40 (1.3) NURP data mostly from the upper midwest and northeastern areas of the US WI DNR data and slide 3
NSQD Database few data from the northern areas of the US, but better represents the southern tier Therefore, differences associated with sampled areas, not trends with time Experimental Design - Number of Samples Needed The number of needed to characterize stormwater conditions for a specific site is dependent on the COV and allowable error. For most constituents and conditions, about 20 to 30 may be sufficient for most objectives. Most Phase 1 sites only have about 10 events, but each stratification category usually has much more. Suspended Solids Concentrations (mg/l) for Different Rain Zones and Land Uses (number of ) Open Space Commercial Residential Industrial Freeways 1 (NE and mid west) (1) (25) (3) 25 (13) 2 (mid Atlantic) 40 (28) 40 (631) 43 (287) (212) 215 (3) 3 (SE) (57) 33 (24) (41) 4 (S central) 139 (13) 84 (89) 90 (54) 103 (62) 5 (Texas) (87) 30 (21) 147 (43) 6 (SW) 92 (25) 120 (11) 206 (18) 86 (104) 7 (coastal NW) 35 (51) (40) 121 (24) 99 (26) 8 (N mtns) (7) 9 (N central) (7) 152 (9) (9) Seasonal data are well distributed at each site, and did not statistically influence TSS Stormwater NPDES Data Collection and Evaluation Project The database is supplemented with detailed descriptions of individual agency monitoring activities and sites. The final report includes a comprehensive statistical analysis of this data and provides recommendations for improving the quality and management of future NPDES stormwater monitoring efforts. 4
Detailed narrative descriptions, including maps and aerial photographs, have also been prepared for most of the sites. This information, along with the database, is available on the Internet. Submitted by Chesterfield County. Office of Water Quality Factors Potentially Affecting Data We conducted numerous statistical tests of the data to identify important factors that may affect stormwater characteristics. We examined the data for such effects as: Sampling method effects (manual vs. automated sample collection, etc.) Land use and geographical location effects Seasonal effects Long-term trends Effects due to storm size (and initial sample vs. composite sample) The Anderson- Darling test was used to test for normality. Most of the parameters are log-normally distributed. The data were therefore transformed before many of the statistical analyses were conducted. Comparison of first 30 minute with 3 hour composite More than 400 paired available for comparison. The Fligner-Policello (symmetrical about the medians) and the Mann- Whitney (symmetrical and same variance) nonparametric comparison tests were used to compare the paired 30 min. concentrations with the whole storm composite concentrations. 3 hr comp. First 30 minutes 5
Initial High Concentrations COD, BOD 5, TDS, TKN, and Zn all were significantly higher during the first 30 minutes compared to the 3 hour sample for all land uses (except for open space). Comparison of Stormwater Control Practices in Residential Land Uses, EPA Rain Zone 2 The ratio of these higher 30 min. concentrations to the composite concentrations ranged from 1.3 to 1.7 for these constituents. These statistical evaluations were only for concentration differences, not for mass. Turbidity, ph, fecal coliforms, fecal strep., total N, dissolved P, and orthophosphate did not have significantly higher concentrations during the first 30 minutes compared to the 3 hour for most of the separate land uses. open space, and only a few institutional data sets had significantly higher initial concentrations. Land Use Comparisons Kruskal-Wallis analyses indicate that all constituents have at least one significantly different category from the others. Heavy metal differences are most obvious. Freeways, industrial, and commercial areas the highest. Residential Land Use by Geographical Area Zones 1-4 are east half of country, zones 5-9 are western half of country. Zones 3 and 7 are the wettest zones. 6
Residential Land Use by Season The most obvious difference is shown for fecal coliforms (a similar conclusion was found during NURP, EPA 1983). Plots of concentrations vs. rain depth typically show random patterns. Residential Land Use Lead Concentrations Model Building to Predict Stormwater Concentrations Major objective of database project. Concentration plots vs. time indicate possible trends. Lead has historically dropped significantly from the earliest stormwater studies to the present due to increased use of unleaded gasoline (simple regression trend line shown). Multivariate statistical analyses used to identify most important factors. Statistical models developed for separate land uses and regions, based on watershed and rain characteristics. 7
Source Area Monitoring to Predict Sources of Runoff Pollutants Controlled washoff tests Small area sheetflow sampling Large area sheetflow sampling Outfall monitoring Examples of Wisconsin source area monitoring locations Examples of Alabama critical source area sampling sites 8
Commercial home maintenance and lumber supply store critical source areas (high nutrients, high fertilizers, high copper, chromium, and arsenic). Grab sampling of source area flows Semi-automatic source area samplers in paved areas 9
Semi-automatic source area samplers in unpaved areas Semi-automatic source area samplers for roof runoff Need to understand where pollutants originate in a complex city The Source Loading and Management Model (SLAMM) Developed during past 30 years during EPA, state, and Canadian funded research. Identifies pollutant sources during different rain and climatic conditions. Prioritizes subwatersheds and critical source areas. Evaluates alternative development scenarios, pollution prevention, and combinations of source area and outfall control options. 10
SLAMM Inputs and Outputs Soil Type Landuse Area Rainfall Development Characteristics Description of Practices SLAMM Volume and Pollutant Load Mass Balance Medium Density Residential Development Characteristics Source % Area Roofs 15% Driveways 8% Sidewalks 2% Streets 13% Lawns 62% Other Characteristics Roofs: Pitched; % connected = 30%; Sandy Soils. Driveways: % connected = 70%; Low/Med. Density Streets: Texture = smooth; Length = 2 mi.; Dirt Accumulation = default value. Calibration of WinSLAMM by Comparing Observed and Predicted Runoff Predicted Runoff (in) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 Observed vs. Predicted Runoff at Monroe Outfall - - 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Observed Runoff (in) 11
Residential Particulate P Values Used in SLAMM total P, mg/kg 5000 4500 4000 3500 3000 2500 2000 1500 1000 500 0 lawn roof driveway sidewalk street Measured source area concentrations used in WInSLAMM Can determine sources of flows and pollutants for different rain conditions WinSLAMM used to predict phosphorus sources to urban lake 12
Lincoln Creek, Milwaukee (Wisconsin) toxicant source study Side-stream toxicity/bioassay test facility Bioassay tests to measure benefits of sedimentation controls to reduce toxicity 13
Sample Processing: Organo- Metallic Complexes Laboratory tests to understand characteristics of metallic toxicants Colloidal and Ionic Form Analyses The Chelex-100 resin (5 g of resin mixed with 100 ml of filtered stormwater sample and shaken for 1 hr) separates ionic forms of metals from metals strongly bound to the metal-ligand complexes or those strongly adsorbed to colloidal particles. After the Chelex exposure, the sample is filtered again to remove the newly bound material (the ionic forms) and the Chelex. The filtered sample is then irradiated with UV to separate the colloidal and organo-metallic associations. Particle size and toxicity measurements 14
Sample Processing: Metal Binding Strength Minocqua, WI, MCTT Installation Treatment device developed based on toxicant characteristics Pilot-Scale Test Results Wisconsin Full-Scale MCTT Test Results (median % reductions and median effluent quality) Milwaukee (15 events) Minocqua (7 events) Suspended Solids 98 (<5 mg/l) 85 (10 mg/l) Phosphorus 88 (0.02 mg/l) >80 (<0.1 mg/l) Copper 90 (3 µg/l) 65 (15 µg/l) Lead 96 (1.8 µg/l) nd (<3 µg/l) Zinc 91 (<20 µg/l) 90 (15 µg/l) Benzo (b) fluoranthene >95 (<0.1 µg/l) >75 <0.1 µg/l) Significant zinc reductions within MCTT Phenanthrene Pyrene 99 (<0.05 µg/l) 98 (<0.05 µg/l) >65 (<0.2 µg/l) >75 (<0.2 µg/l) 15
Development of other Control Devices Multiple treatment processes can be incorporated into other stormwater treatment units sized for various applications. Gross solids and floatables control (screening) Capture of fine solids (settling or filtration) Control of targeted dissolved pollutants (sorption/ion exchange) Upflow filter insert for catchbasins Upflow Filter TM patent pending Uses sedimentation (22), gross solids and floatables screening (28), moderate to fine solids capture (34 and 24), and sorption/ion exchange of targeted pollutants (24 and 26). Upflow significantly reduces clogging of media common to most stormwater filters. Conclusions Phase 1 data shows significant patterns for different land uses and geographical location for most constituents. Only bacteria showed significant differences by season. Similar to historical NURP data, but Phase 1 data may be somewhat lower in concentration (lead especially), and has greater variability (due to increased variety of conditions monitored and is more representative). NSQD lacks data from northern tier of US while NURP lacks data from southern tier of US. Conclusions - The NSQD is a useful tool to characterize stormwater constituents for different geographical areas and land uses. - The database can be used to evaluate the performance of stormwater controls, type of conveyance, sampling procedures, etc. - More data would be useful in underrepresented areas for more useful evaluations. 16
Conclusions Stormwater quality highly variable, but some of this variability can be explained by land use and source area (much variability also due to rain characteristics). Some source areas are critical and require special attention as they are responsible for disproportionate amount of pollutant discharges. Historical and recent research reports describe stormwater characteristics for critical source areas and treatability requirements. Possible to develop stormwater controls that provide treatment train approach considering unique characteristics of stormwater, either at source areas or at outfalls. Acknowledgements for NSQD Project Bryan Rittenhouse, the US EPA project officer for the Office of Water, is gratefully acknowledged. The many municipalities who worked with us to submit data and information were obviously crucial and the project could not be conducted without their help. A number of graduate students at the University of Alabama were active project participants and supplied critical project assistance. 17