Weighting of Field Monitoring Data With Probability Distributions of Daily Rainfall

Transcription

1 Weighting of Field Monitoring Data With Probability Distributions of Daily Rainfall Abstract James H. Lenhart, PE, D.WRE and Scott A. de Ridder CONTECH Stormwater Solutions, NE Glenn Widing Drive, Portland, OR, PH (800) ; FAX (503) ; Many BMPs undergo field monitoring to assess the ability of the BMP to meet water quality goals. This is typically accomplished using protocols that outline data quality objectives. Some of these objectives include a minimum number of storms, minimum storm depth, and flow rates or water quality volumes that span the operating range of the BMP. Typically during the data analysis, mass weighting is used to calculate performance efficiency on a load basis. However, considerations for the weighting of the storms, relative to the frequency of occurrence is typically not done. This can be problematic in that the collection of too many small storms may not reflect the ability of the BMP to perform for larger storms, or the collection of too many large storms could underestimate the relative performance of the BMP on a longterm basis. This paper investigates a method to weight the storms collected relative to the frequency distribution of the average annual daily rainfall. This method could be used in the development of sampling protocols and methods currently being established by committee through EWRI. Introduction Many BMPs undergo field monitoring to assess the ability of the BMP to meet water quality treatment goals. This is typically accomplished using protocols that outline data quality objectives (DQOs). DQOs typically include a minimum number of storms, minimum storm depth, antecedent conditions, and flow rates or water quality volumes that span the operating range of the BMP. Many BMPs such as ponds, swales, wetlands, and hydrodynamic separators have their performance coupled with the magnitude and frequency of flow. In general removal rates tend to be more efficient with lower flows since there is more opportunity time for settling of particulates and less turbulence. In addition, longer inter-event periods can allow for long term settling and biological processes to occur. Frequently, for the data analysis, mass weighting is used to calculate performance efficiency on a load basis. However, considerations for the weighting of the storms, relative to the frequency of occurrence is typically not done. This can be problematic in that the collection of too many

2 small storms may not reflect the ability of the BMP to perform (or not perform) for larger storms, or the collection of too many large storms relative to the frequency of small storms could underestimate the relative performance of the BMP on a long-term basis. A few protocols have recognized this issue and set DQOs to allow for the weighting of storms but not in accordance with the frequency distribution of storms. For example the TEIR II protocol (NJDEP, 2006) requires a minimum of 15 storms of which two need to be in excess of 75% of the design storm. The Washington State Tape protocol (Ecology, 2002) suggests that data be colleted to represent seasonal differences, which really is intended to represent pollutant characteristics with seasons but to some extent implies the need for a range of storm depth. What can also be at issue is that BMPs are sized to treat a certain water quality flow or water quality volume (WQV). However, the likelihood that a storm captured will be right at the design is small. In the event that the storms captured and analyzed are significantly smaller that the design storm, little information is conveyed as to how the BMP would actually operate at full design. This suggests that unless data are available at full design that approvals should only be granted up to the validated design flows or WQVs. Another consideration is how the BMP reacts to flows in excess of design. Theoretically, if the BMP is designed with a bypass that allows flows in excess of design to shunt around the system then the data would reflect operation at full design flow. Some BMPs are design with internal bypassing such that flows in excess of design flow through the system, which can dramatically reduce effectiveness or even cause resuspension of captured pollutants. This raises the question of how data colleted from storms in excess of design should be handled. Should these data be excluded from the data set and be used solely for the analysis of how it behaves in extreme conditions? Method The basic analytical methods involves a few steps 1. Obtain daily rainfall data for a period of record near the monitoring site. Use normal data with a period of record of 30 years 2. Create a frequency distribution histogram for the data 3. Truncate the data set to the level of the design storm 4. Develop a probability distribution function (PDF) from the truncated data set 5. Develop a PDF using the rainfall data for the storms actually collected 6. Develop a scatter plot and overlay both PDFs. 7. Based on the outcome of the scatter plot, select appropriate ranges to group the monitoring data 8. Determine the mass weighted mean for each range from the monitoring data 9. Determine the mean equivalent depth of rainfall for each range 10. Determine the rainfall frequency weighting factor for each range 11. Weight each range 12. Calculate the frequency mass weighted means. 2

3 Data Collection and Processing For the purposes of analysis and discussion a sample site was selected to analyze data. Daily rainfall data were collected from the National Climatic Data Center (NCDC) for the Newark airport, which is near a monitoring site for a StormFilter located at Greenville Yards (CONTECH, 2006). Daily rainfall data for a period of record of 30 years from January 1973 to January 2008 were processed by spreadsheet to develop a frequency distribution for rainfall depths up to 1.30 inches, where 1.25 inches is the defined New Jersey water quality storm. Figure 1 shows the outcome in increments of 0.05 inches. Clearly a large proportion of the events are smaller storms. The frequency distribution was then put into a Cumulative Probability Distribution Function (PDF), illustrated in Figure 2 (upper line). Figure1 :Daily Rainfall Frequency Distribution Newark Airport, NJ ( ) Number of Observations Depth of Daily Rainfall The PDF was then weighted by event depth to create a depth weighted frequency curve. This was accomplished by multiplying the frequency of occurrence by the rainfall depth and reconstructing the PDF. The Depth Weighted PDF (lower line) now shows the cumulative distribution of rainfall depth (as in volume) as a function of the daily rainfall. As one would expect, days with more rain carry more weight and hence contribute more volume. 3

4 Figure 2: Cumulative Probability Distribution Newark Airport, NJ ( ) Cumulative Probability of x<=x Depth Weighted Frequency Frequency of Occurance Depth of Daily Rainfall (in) As an example, using Figure 2, it can be surmised from the past 30 years of rainfall data that 60 percent of the total annual rainfall depth occurs from days with 1.0 inches or less of rainfall. Monitoring Data Greenville Yards was monitored in 2004 and 2005 using the TARP Tier II protocol (CONTECH, 2006). Summary data are presented in Table 1. A total of 16 qualified events are used for the sake of the analysis. Since protocol event criteria are not based on daily rainfall totals, event rainfall depths have a degree of independence from daily rainfall. For the purposes of this analysis, it was assumed that the whole population of storms, independent of what hour of the day they occurred, would be essentially the same as that measured as daily rainfall totals. Table 1: Summary Data from StormFilter Monitoring at Greenville Yards ( ) Event ID Treated Volume (liters) TSS in (mg/l) TSS out (mg/l) Mass in (KG) Mass Out (KG) Removal Site Rainfall (in) GYS % 0.13 GYS % 0.19 GYS % 0.39 GYS % 0.46 GYS % 0.62 GYS % 0.68 GYS % 0.72 GYS % 0.72 GYS % 0.79 GYS % 0.98 GYS % 0.99 GYS % 1.73 GYS % 1.83 GYS % 2.06 GYS %

5 For this site a total of 5 storms exceeded the design storm of 1.25 inches. However, this system was designed with a bypass such that it treated up to the design flow and routed the remaining flow, untreated, around the system. Therefore the boundary data, i.e. storms greater than 1.25 inches, we assumed to be 1.25 storms. These data were then processed to determine the frequency distribution in 0.25 inch intervals. The result is shown by Table 2. The mean precipitation was calculated for each range. From there a cumulative probability distribution function can then be constructed and overlaid onto the cumulative probability distribution above. Table 2: Frequency Distribution for Monitoring Event Data Range Low Range High Frequency Probability Total in Range Mean within Range > > > > > n= 16 Total : All storms > 1.25 will be adjusted to 1.25 Figure 3 shows the result of the calculations. The individual points represent the observed distribution vs. what would be expected based upon the last 30 years of rainfall data. Two observations can be made given this data set. The first is that data are available within all of the ranges and second is that there is a reasonably good fit except that the large storms occur more frequently that expected and hence over represent what would happen over a long period of time. Cumulative Probability of x<=x Figure 3: Cumulative Probability Distribution Newark Airport, NJ ( ) Depth of Daily Rainfall (in) Depth Weighted Frequency Frequency of Occurance Monitoring Data Since the distribution is depth weighted and related to runoff volume, the mass removal data set can be adjusted to bring the data points from the observed to the expected line by use of 5

6 weighting factors (Table 3). This can be accomplished by dividing the expected probability into the observed frequency and multiplying this number times the volume of water treated. The individual storm removal efficiencies will remain unchanged but on the total mass basis, total percent removal will change. This is illustrated in Table 4 Table 3 Determination of Weighting Factors Site Rainfall (in) Observed Expected FrequencyProbability Weighting Factor Table 4: Calculation of Weighted Storms Event ID Treated Volume (liters) Weighting Factor Weighted Volume TSS in TSS OUT Weighted Mass In Weighted Mass Mass Out Discrete Storm Removal GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % GYS % The aggregate removal efficiency was calculated to be 77.2% versus the non-weighted value of 76.7%. Discussion Though the outcome of this example resulted in a relatively small difference, it should be noted that there was a fairly good match of the observed frequency to the expected probability. Also, the larger storms produce low cumulative mass and hence thus less influence on the outcome. 6

7 It is important to recognize that the outcome of this analysis can reduce the aggregate mass load removal. One shortcoming of this method may be that monitoring projects often have limited data sets. Many times no more than ten storms are collected which reduces the confidence that BMP performance within any given range is representative. Though this analysis uses daily rainfall data, a similar method could be developed on the intensity distribution as well. Lehman and de Ridder (2005) found a correlation between storm intensity and TSS concentration. This could be a more precise method of analysis, however, intensity data are typically published on a 15-minute or hourly basis and may not be reflective of the actual peak flows for BMPs that have very short times of concentration. In addition, volume based BMPs may be more dependent on total rainfall depth rather than instantaneous peak flow rates. Another benefit of this procedure could be for project scoping and QAPP development. Knowing the probability distribution of rainfall can allow for more targeting collection of data. If in advance the expected PDF is developed and sectioned into ranges, then the researcher could determine at what point enough storms in a range are collected and what type of future storms to target when setting up the sampling instruments. Though this method could use some refinement it could be used as part of a standard sampling protocol development to increase the confidence that a BMP evaluation is congruent with long term rainfall characteristics. Recommendations The following recommendations are made relative to using this information for program development: Weighting of storms relative to the long term rainfall frequency of occurrence helps to increase the confidence that BMP performance is representative; BMP s should not be approved at flow or volume levels above that which they have been evaluated. For example if a BMP is designed to treat 2 CFS but never operates above 1 CFS during the monitoring period, the BMP should only be allowed to treat to the 1 CFS level, assuming the outcome for the lower flows was favorable and met the water quality goals. This method accommodates the elimination of minimum depth rainfall DQOs. These types of DQOs are frequently set in place to prevent a large population of small storms with higher performance over weighting larger storms with less performance. 7

8 References CONTECH Stormwater Solutions, Inc. (2006). Greenville Yards Stormwater Treatment System Field Evaluation: Stormwater Management StormFilter with Perlite Media at 57 L/m/cartridge. (PE-G080). Portland, OR: Author. Lehman, J. M. & de Ridder, S. A. (2005). Predicting Solids Pollutant Concentrations from Storm Event Variables. Proceedings of the North American Surface Water Quality Conference and Exposition (StormCon). Washington State Department of Ecology. (2002). Guidance for Evaluating Emerging Stormwater Treatment Technologies: Technology Assessment Protocol Ecology (TAPE). Publication Lacey, WA: Author. New Jersey Department of Environmental Protection (NJDEP). (2006). New Jersey Tier II Stormwater Test Requirements - Amendments to the TARP Tier II Protocol. Trenton, NJ: Author. Technology Assessment and Reciprocity Partnership (TARP). (2003). The Technology Acceptance Reciprocity Partnership Protocol for Stormwater Best Management Practice Demonstrations. Harrisburg, PA: Author. 8