CRWR Online Report 10-07 
 
 
Evaluation of Sand Filter Performance 
 
 
by 
 
 
Michael E. Barrett, Ph.D., P.E., D.WRE 
 
 
 
 
 
October 18, 2010 
 
 
 
 
Center for Research in Water Resources 
Bureau of Engineering Research 
The University of Texas at Austin 
 
 
i 
 
 
Table of Contents 
 
 
1 Introduction ............................................................................................................................. 1 
1.1 Mathematical Modeling of Sand Filter Performance ....................................................... 4 
1.2 Statistical Analysis ........................................................................................................... 5 
2 Total Suspended Solids (TSS) Performance ........................................................................... 8 
3 Volatile Suspended Solids (VSS) Performance..................................................................... 21 
4 Total Phosphorus Performance .............................................................................................. 26 
5 Dissolved Phosphorus Performance ...................................................................................... 34 
6 Total Kjeldahl Nitrogen (TKN) Performance ....................................................................... 40 
7 Nitrate plus Nitrite Performance ............................................................................................ 48 
8 Fecal Coliform Performance ................................................................................................. 55 
9 Fecal Streptococcus Performance .......................................................................................... 61 
10 Biochemical Oxygen Demand (BOD) Performance .......................................................... 69 
11 Chemical Oxygen Demand (COD) Performance .............................................................. 77 
12 Zinc Performance ............................................................................................................... 84 
13 Copper Performance .......................................................................................................... 92 
14 Lead Performance (Pb) ...................................................................................................... 99 
15 Overall Conclusions ......................................................................................................... 104 
16 References ........................................................................................................................ 105 
 
 
 
 
 
ii 
 
 
List of Figures 
 
Figure 1 Typical Austin Sand Filter Bed ........................................................................................ 2 
Figure 2 Highwood Filter Bed ........................................................................................................ 2 
Figure 3 Mechanisms of Particle Capture in Filters ....................................................................... 4 
Figure 4 TSS Influent and Effluent Probability Plots for all Sites Combined ................................ 8 
Figure 5 Boxplot Influent TSS Concentrations .............................................................................. 9 
Figure 6 Boxplot of TSS Discharge Concentrations (all data) ..................................................... 10 
Figure 7 (A) Particle Size Distribution Function and (B) Volume Distribution ........................... 12 
Figure 8 Comparison of Maximum Discharge Rate and Average TSS Discharge Concentration at 
Brodie Oaks .................................................................................................................................. 13 
Figure 9 Influent and Effluent TSS concentrations for all Sites ................................................... 14 
Figure 10 Influent and Effluent TSS concentrations for Barton Ridge ........................................ 15 
Figure 11 Influent and Effluent TSS concentrations for Jollyville ............................................... 15 
Figure 12 Jollyville Sand Filter TSS Discharge Time Series ....................................................... 17 
Figure 13 Relationship between TSS Discharge Quality and Time for Barton Creek Square Mall
....................................................................................................................................................... 18 
Figure 14 Comparison of TSS removal to Hydraulic Residence Time ........................................ 19 
Figure 15 Probability Plots of VSS Influent and Effluent ............................................................ 21 
Figure 16 Boxplot of Influent VSS concentrations....................................................................... 22 
Figure 17 Boxplot of Effluent VSS concentrations ...................................................................... 22 
Figure 18 Relationship Between VSS Influent and Effluent Concentrations ............................... 23 
Figure 19 Temporal Pattern of VSS Discharge Concentrations at Jollyville ............................... 24 
Figure 20 Probability Plots of Total P Influent and Effluent ........................................................ 26 
Figure 21 Boxplot of Total P Influent Concentrations ................................................................. 27 
Figure 22 Boxplot of Total P Effluent Concentrations ................................................................. 27 
Figure 23 Relationship between P Influent and Effluent Concentrations All Sites ...................... 29 
Figure 24 Relationship between P Influent and Effluent Concentrations for Highwood ............. 29 
Figure 25 Relationship between P Influent and Effluent Concentrations for Barton Ridge ........ 30 
Figure 26 Relationship between P Influent and Effluent Concentrations for Barton Mall .......... 30 
Figure 27 Jollyville Total Phosphorus Time Series Discharge Concentration ............................. 31 
 
 
iii 
 
Figure 28 Relationship between HRT and Total P Removal (Jollyville) ..................................... 32 
Figure 29 Probability Plots of Dissolved P Influent and Effluent ................................................ 34 
Figure 30 Boxplot of Dissolved P Influent Concentrations .......................................................... 35 
Figure 31 Boxplot of Dissolved P Effluent Concentrations ......................................................... 35 
Figure 32 Relationship between Dissolved P Influent and Effluent Concentrations.................... 36 
Figure 33 Jollyville Dissolved Phosphorus Time Series Discharge Concentration ..................... 38 
Figure 34 Probability Plot of TKN Influent and Effluent ............................................................. 41 
Figure 35 Boxplots of TKN Influent Concentrations ................................................................... 42 
Figure 36 Boxplot of TKN Effluent Concentrations .................................................................... 42 
Figure 37 Relationship between Influent and Effluent TKN Concentrations (all sites) ............... 43 
Figure 38 Relationship between Influent and Effluent TKN Concentrations Barton Ridge ........ 44 
Figure 39 Relationship between Influent and Effluent TKN Concentrations Highwood ............. 44 
Figure 40 Jollyville TKN Time Series Discharge Concentrations ............................................... 45 
Figure 41 Relationship between HRT and TKN Removal (Jollyville)......................................... 46 
Figure 42 Probability Plots of NO2+3 Influent and Effluent Concentrations ................................ 48 
Figure 43 Boxplot of Nitrate + Nitrite Influent Concentrations ................................................... 49 
Figure 44 Boxplot of Nitrate + Nitrite Effluent Concentrations ................................................... 49 
Figure 45 Relationship Between Influent and Effluent NO23 Concentrations (all sites) ............. 51 
Figure 46 Relationship Between Influent and Effluent NO23 Concentrations Barton Ridge ...... 51 
Figure 47 Relationship Between Influent and Effluent NO23 Concentrations Jollyville ............ 52 
Figure 48 Relationship Between Influent and Effluent NO23 Concentrations Highwood .......... 52 
Figure 49 Temporal Trends for NO2+3 for Jollyville .................................................................... 53 
Figure 50 Effect of Antecedent Dry Period on Initial NO23 Concentration ................................ 54 
Figure 51 Probability Plots of Fecal Coliform Influent and Effluent ........................................... 55 
Figure 52 Boxplot of Fecal Coliform Influent Concentrations ..................................................... 56 
Figure 53 Boxplot of Fecal Coliform Discharge Concentrations ................................................. 56 
Figure 54 Relationship between FC Influent and Effluent Concentrations all Sites .................... 57 
Figure 55 Relationship between FC Influent and Effluent Concentrations Jollyville .................. 58 
Figure 56 Relationship between FC Influent and Effluent Concentrations Highwood ................ 58 
Figure 57 Temporal Trends for Fecal Coliform for Jollyville ...................................................... 59 
Figure 58 Relationship between Fecal Coliform Effluent Concentration and HRT at Jollyville . 60 
 
 
iv 
 
Figure 59 Probability Plots of Fecal Strep Influent and Effluent ................................................. 61 
Figure 60 Boxplot of Fecal Strep Influent Concentrations ........................................................... 62 
Figure 61 Boxplot of Fecal Strep Effluent Concentrations .......................................................... 62 
Figure 62 Relationship between Fecal Strep Concentrations (all sites) ....................................... 63 
Figure 63 Relationship between Fecal Strep Concentrations for Barton Ridge ........................... 64 
Figure 64 Relationship between Fecal Strep Concentrations Barton Mall ................................... 64 
Figure 65 Relationship between Fecal Strep Concentrations Jollyville ....................................... 65 
Figure 66 Relationship between Fecal Strep Concentrations Highwood ..................................... 65 
Figure 67 Relationship between Fecal Strep Concentrations Brodie Oaks .................................. 66 
Figure 68 Temporal Pattern of Fecal Strep Discharge Concentrations at Jollyville ..................... 67 
Figure 69 Probability Plots of BOD Influent and Effluent ........................................................... 69 
Figure 70 Boxplot of BOD Influent Concentrations..................................................................... 70 
Figure 71 Boxplot of Effluent BOD Concentrations .................................................................... 70 
Figure 72 Relationship Between Influent and Effluent BOD Concentrations .............................. 71 
Figure 73 Relationship Between Influent and Effluent BOD Concentrations Barton Ridge ....... 72 
Figure 74 Relationship Between Influent and Effluent BOD Concentrations Barton Mall ......... 72 
Figure 75 Relationship Between Influent and Effluent BOD Concentrations Jollyville .............. 73 
Figure 76 Relationship Between Influent and Effluent BOD Concentrations Highwood ............ 73 
Figure 77 Relationship Between Influent and Effluent BOD Concentrations Brodie Oaks......... 74 
Figure 78 Temporal Pattern of BOD Discharge Concentrations at Jollyville .............................. 75 
Figure 79 Relationship between BOD Effluent Concentration and HRT..................................... 76 
Figure 80 Probability Plot of COD Influent and Effluent Concentrations ................................... 77 
Figure 81 Boxplot of COD Influent Concentrations..................................................................... 78 
Figure 82 Boxplot of COD Effluent Concentrations .................................................................... 78 
Figure 83 Relationship between COD Influent and Effluent Concentrations all Sites ................. 79 
Figure 84 Relationship between COD Influent and Effluent Concentrations Barton Ridge ........ 80 
Figure 85 Relationship between COD Influent and Effluent Concentrations Highwood ............ 80 
Figure 86 Relationship between COD Influent and Effluent Concentrations Brodie Oaks ......... 81 
Figure 87 Temporal Pattern of COD Discharge Concentrations at Jollyville .............................. 82 
Figure 88 Relationship between HRT and COD Removal ........................................................... 83 
Figure 89 Zn Influent and Effluent Probability Plots ................................................................... 84 
 
 
v 
 
Figure 90 Boxplot of Zinc Influent Concentrations ...................................................................... 85 
Figure 91 Boxplot of Zn Discharge Concentrations ..................................................................... 85 
Figure 92 Relationship between Zn Influent and Effluent Concentrations .................................. 87 
Figure 93 Photograph of the Barton Ridge Sand Filter ................................................................ 87 
Figure 94 Jollyville Zn Time Series Discharge Concentrations ................................................... 89 
Figure 95 Relationship (or not) between Zn Discharge Concentration and Discharge Rate ........ 90 
Figure 96 Relationship between HRT and Zn Discharge Concentration (Jollyville) ................... 90 
Figure 97 Probability Plots of Total Copper Influent and Effluent .............................................. 92 
Figure 98 Boxplot of Cu Influent Concentrations ........................................................................ 93 
Figure 99 Boxplot of Copper Effluent Concentrations ................................................................. 93 
Figure 100 Relationship between Total Copper Influent and Effluent Concentrations ................ 94 
Figure 101 Relationship between Total Cu Concentrations Barton Mall ..................................... 95 
Figure 102 Relationship between Total Cu Concentrations Jollyville ......................................... 95 
Figure 103 Relationship between Total Cu Concentrations Highwood ....................................... 96 
Figure 104 Relationship between Total Cu Concentrations Brodie Oaks .................................... 96 
Figure 105 Temporal Pattern of Copper Discharge Concentrations at Jollyville ......................... 97 
Figure 106 Probability Plots of Total Pb Influent and Effluent Concentrations........................... 99 
Figure 107 Boxplot of Lead Influent Concentrations ................................................................. 100 
Figure 108 Boxplot of Effluent Lead Concentrations................................................................. 100 
Figure 109 Relationship between Pb Influent and Effluent Concentrations all Sites ................. 101 
Figure 110 Relationship between Pb Influent and Effluent Concentrations Highwood ............. 102 
Figure 111 Temporal Pattern of Lead Effluent Concentrations at Jollyville .............................. 103 
 
 
 
 
 
 
 
vi 
 
 
List of Tables 
 
Table 1 Description of Stormwater Filtration Facilities ................................................................. 1 
Table 2 Coefficient of Variation for Monitoring Data.................................................................... 6 
Table 3 Statistical Distribution of TSS Data for Each Site ............................................................. 8 
Table 4 TSS Concentrations for Individual Sites based on Paired Data....................................... 10 
Table 5 Statistical Distribution of VSS Data for Each Site .......................................................... 21 
Table 6 VSS Concentrations for Individual Sites based on Paired Data ...................................... 23 
Table 7 Statistical Distribution of Total Phosphorus Data for Each Site ..................................... 26 
Table 8 Total Phosphorus Concentrations for Individual Sites based on Paired Data ................. 28 
Table 9 Statistical Distribution of Dissolved Phosphorus Data for Each Site .............................. 34 
Table 10 Dissolved P Concentrations for Individual Sites based on Paired Data ........................ 36 
Table 11 Ratio of Effluent/Influent Volume for Events with Dissolved P Data .......................... 37 
Table 12 Statistical Distribution of TKN Data for Each Site ....................................................... 40 
Table 13 TKN Concentrations for Individual Sites based on Paired Data ................................... 43 
Table 14 Statistical Distribution of Nitrate Data for Each Site .................................................... 48 
Table 15 Nitrate Concentrations for Individual Sites based on Paired Data ................................ 50 
Table 16 Statistical Distribution of Fecal Coliform Data for Each Site ....................................... 55 
Table 17 Fecal Coliform Concentrations for Individual Sites based on Paired Data ................... 57 
Table 18 Statistical Distribution of Fecal Strep Data for Each Site ............................................. 61 
Table 19 Fecal Strep Concentrations for Individual Sites based on Paired Data ......................... 63 
Table 20 Statistical Distribution of BOD Data for Each Site ....................................................... 69 
Table 21 BOD Concentrations for Individual Sites based on Paired Data ................................... 71 
Table 22 Statistical Distribution of COD Data for Each Site ....................................................... 77 
 
 
vii 
 
Table 23 COD Concentrations for Individual Sites based on Paired Data ................................... 79 
Table 24 Statistical Distribution of Zinc Data for Each Site ........................................................ 84 
Table 25 Zinc Concentrations for Individual Sites based on Paired Data .................................... 86 
Table 26 Statistical Distribution of Copper Data for Each Site .................................................... 92 
Table 27 Copper Concentrations for Individual Sites based on Paired Data ................................ 94 
Table 28 Statistical Distribution of Lead Data for Each Site ........................................................ 99 
Table 29 Lead Concentrations for Individual Sites based on Paired Data ................................. 101 
 
 
1 
 
1 Introduction 
 
This report presents an analysis of water quality data collected by the City of Austin between 
1985 and 1997 from various stormwater filtration facilities located in the Austin area. The 
objective of the analysis is to determine how facility design and storm characteristics affect 
pollutant removal. Of particular interest are residence time, water depth over the filter, 
pretreatment, and temporal patterns of pollutant discharge. 
The facilities where the water quality data were collected include Jollyville, Highwood, Barton 
Creek Square Mall, Brodie Oaks, and Barton Ridge. The designs of these facilities vary 
significantly and their distinguishing characteristics are shown in Table 1. 
Table 1 Description of Stormwater Filtration Facilities 
 Jollyville Highwood Barton Creek 
Square Mall 
Barton 
Ridge 
Brodie 
Oaks 
Off/On-Line Off-line On-Line On-Line Off-Line On-Line 
Pretreatment None None None Sedimentation Basin  
Wet Pond 
(used for 
irrigation) 
Water Quality 
Volume WQV (cu.ft.) 17,000 2,970 143,000 7,000 NA 
Water Quality 
Volume WQV (in.) 0.5 0.23 0.5 0.65 NA 
Design Drawdown 
Time DDT (hr) 24 NA 24 40 NA 
Filtration Media 
Surface Area Af 
(sq.ft.) 
2,600 2,750 21,780 390 ~5,500 
Depth of Filtration 
Media D (ft.) 2.5 Varies 2.5 1.5 NA 
Maximum Ponding 
Depth over Sand 
Media H (ft) 
3.9 ~1.0 4 2.0 ~15 
 
The five sand filters analyzed in this study are all believed to have essentially the same media, 
which is a quartz and feldspar sand that has a size distribution consistent with fine aggregate as 
defined in ASTM C-33. However, there are substantial differences in the maximum water level 
over the filter media, which range from roughly one foot (Highwood) to approximately 15 feet 
(Brodie Oaks). One potentially significant difference is that most of the sites have a filter profile 
 
 
2 
 
like that shown in Figure 1. Highwood, on the other hand has a filter constructed as shown in 
Figure 2, which provides much less filter volume. 
 
Figure 1 Typical Austin Sand Filter Bed 
 
Figure 2 Highwood Filter Bed 
 
 
3 
 
One item of interest is whether there are pollutants for which reaction kinetics limit their 
removal. This is primarily of interest for dissolved constituents that might adsorb to, or otherwise 
react with, the filter media. Dissolved concentrations are available for nitrogen; however, there is 
no information for or any of the metals and only limited data for dissolved phosphorus. 
Consequently, variation in removal for metals could be the result of either changes in the 
removal or variation in fraction of the dissolved phase in the influent runoff. Nevertheless 
residence times were calculated for a number of events and compared to both effluent quality 
and removal efficiency. The residence time was calculated as the time difference between the 
centroids of the influent and effluent hydrographs. 
Another consideration in analyzing attributing changes in effluent water quality to physical 
processes occurring in the filter is differentiating between variation resulting from physical 
processes and those relating to sample collection and analysis. Substantial variation in reported 
constituent concentrations when identical samples are sent to multiple laboratories is not 
uncommon, even among certified laboratories. This is even more of an issue when the reported 
concentrations are near the method detection limit. In these cases very small absolute differences 
might still represent a substantial percentage difference from the reported amounts. This occurs 
most frequently for metals (especially copper in this dataset). 
Sample collection methodology can also result in differences between actual and estimated storm 
concentrations. Most of the water quality data consists of multiple individual samples taken 
during an event which are later averaged with a weight based on the volume assigned to each 
sample. Two issues have been observed in the data that suggests significant errors might have 
occurred. For many storm events at the facilities there is not a consistent relationship between 
influent and effluent volumes; consequently there are substantial mass balance errors. One might 
expect effluent volumes to be about the same as influent volumes since the basins are lined or 
one might think that the effluent volumes should be somewhat less because of runoff retained in 
the pores of the filter media and subsequently lost to evaporation between events. What is 
actually observed is that the ratio of influent to effluent volumes ranges from 3.45 to 0.14, which 
means that some error in the calculated concentrations is likely. 
The second issue related to sample collections is the event mean concentration (EMC) is 
calculated based on as few as three individual samples. Many of the constituents exhibit a 
pronounced first flush effect, which is particularly evident in the storms where at least six 
individual samples were collected. In the cases where only three samples were collected the 
initial sample concentration measured at the beginning of the event, when concentrations are 
briefly elevated, are assigned to a larger volume of runoff than is likely warranted. This results in 
the calculated EMC being larger than the actual concentration. 
 
 
4 
 
1.1 Mathematical Modeling of Sand Filter Performance 
Particle removal in sand filters is conventionally modeled as a combination of three attachment 
mechanisms, which are illustrated in Figure 3. These mechanisms include capture by 
sedimentation (particle is moving faster than the fluid due to gravity), interception (particle 
momentum causes collision), and diffusion (Brownian motion results in particle collision).  
 
Figure 3 Mechanisms of Particle Capture in Filters 
The classic model consists of quantifying the removal associated with a single filter media 
particle and then integrating over the entire volume of the filter. This results in the classic 
formulation (Yao, et al. 1971): 
?????? ???????? ???? LdT ce ??? )1(23e x p1  
Where: 
Te = Trapping efficiency 
????Filter porosity 
??= Collision frequency 
??= Attachment efficiency 
dc = Characteristic diameter of the filter media particles 
L = filter thickness 
 
 
 
5 
 
This formulation has two major empirical factors ?? and??. The first of these refers to the rate at 
which particles in the fluid strike a collector in the filter, while the second is the rate at which 
particles that strike the collector become attached. Consequently, it is fairly easy to make the 
model agree with measured data for an individual event.  
There are several issues with the use of this model for analyzing performance of stormwater sand 
filters. First of all, the model gives the same results for all events since it does not include 
information related to changes in particle size distribution in the runoff and it does not account 
changes in performance resulting from accumulation of particles within the filter. Perhaps the 
biggest shortcoming is that this conceptual model does not address particle removal via straining, 
which is generally avoided in engineered systems to reduce headloss. 
Straining is believed to be an important removal process when: 
? Flow rates are low (<300 ft/d) and solids flux is high 
? Ratio of particle size to media size is >0.2 
? Particles in the fluid have a diameter greater than 100 ?m. 
Sand filters used for stormwater treatment are considered to be slow sand filters and the typical 
flow rates are about 1/10 of the critical value for straining. In addition, many of the particles in 
stormwater are relatively large; consequently, straining may account of between 50-80% of 
particle removal when the filter is clean. As material accumulates on the surface even smaller 
particles would be subject to removal by straining, so that eventually straining would likely 
account for virtually all particle removal. Consequently, conventional mathematical formulations 
that described filter performance are not likely to be relevant in this application. 
1.2 Statistical Analysis 
One question that inevitably arises in evaluating BMP performance is the appropriate way to 
calculate mean influent and effluent concentrations. Since the data are often lognormally 
distributed, one might choose a method suitable for this type of data. Gilbert (1987) provides two 
methods to calculated the mean values for lognormal data. One of these provides the precise 
value, but involves an infinite series. The second method, which has been used by a number of 
authors, involves the simpler estimating method shown below. 
a = e (? + s2/2) 
where a is the mean of the Event Mean Concentrations (EMCs), ? is the mean and s2 the 
variance of the log transformed EMCs. 
 
 
6 
 
Gilbert also recommends that in cases where the data are not highly skewed (Coefficient of 
variation (COV) less than 1.2) that the arithmetic mean is the preferred measure of central 
tendency. The COV for all the measured data are presented in Table 2 and it is evident that the 
majority of the data have a COV of much less than 1.2. Consequently, the arithmetic mean is 
used to calculate average influent and effluent concentrations for the sand filters. 
Table 2 Coefficient of Variation for Monitoring Data 
Constituent Influent 
COV 
Effluent 
COV  
TSS  1.26 0.99 
Total Zn  0.79 0.68 
Total Cu  0.79 0.74 
Pb  0.82 0.71 
Total P  0.89 0.71 
Diss P 0.79 0.82 
TKN  0.70 0.70 
NO2+3  0.59 0.62 
BOD  0.87 1.20 
COD  0.79 0.72 
Fecal C.  1.53 1.47 
Fecal Strep  1.35 1.70 
 
A variety of statistical techniques are available to determining whether differences in influent 
and effluent concentrations are significant. A popular method for independent data is the t-test, 
which compares the means of two sample groups. This is probably the least discriminating test 
and an assumption of the test is that the underlying data are normally distributed. Given that 
many water quality datasets are lognormal, the test can be applied more correctly to the 
transformed data. Nevertheless, the test is not very sensitive to the distribution of the data. 
Another test that works very well for BMP performance data is the paired t-test, which is very 
powerful for identifying differences. This test determines if the difference between the influent 
and effluent concentrations of paired samples is significantly different than zero. This test makes 
no assumptions about the distribution of the actual data, but assumes that the differences between 
the paired samples are normally distributed; however, like the t-test, it is not very sensitive to this 
assumption. It is not uncommon to have effluent concentrations for some events exceed the 
influent concentrations, resulting in negative numbers, so a log transformation is not possible in 
this case.  
 
 
7 
 
When the differences in the paired samples are highly non-normal, it may be appropriate to use 
the nonparametric equivalent of the paired t-test, which is the Wilcoxon Signed Rank Test 
(SRT). This test determines whether the median of the paired differences is significantly 
different than zero. Both the paired t-test and Wilcoxon test were used to determine whether the 
influent and effluent concentrations were significantly different.   
The influent and effluent quality data from each of the sites was analyzed individually and 
pooled to determine removal efficiency. Because of the variety of distributions, the Wilcoxon 
SRT was the primary tool to determine whether removals were significant.  
The following sections describe the detailed performance of the filtrations systems for each of 
the constituents individually. In these sections, the performance for each constituent is calculated 
individually. In addition, data are provided to: 
? Summarize the influent and effluent concentrations with boxplots 
? Determine the distribution of the data 
? Identify performance for each facility 
? Display how concentrations vary during a storm event 
? Relate discharge concentrations to influent concentrations 
?  Related discharge concentrations to residence time  
 
 
 
8 
 
2 Total Suspended Solids (TSS) Performance 
 
The TSS data for each site were first analyzed to determine whether their distribution was 
normal or lognormal. Table 3 presents the results of this analysis. Where both distributions are 
shown, neither could be rejected by the null hypothesis. Question marks indicate that the data did 
not follow either of these common distributions.  
Table 3 Statistical Distribution of TSS Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie Oaks Highwood Jollyville All 
sites 
Influent Normal ? Normal/Lognormal Lognormal ? ? 
Effluent Normal/Lognormal Lognormal Lognormal Lognormal ? ? 
 
Figure 4 presents the cumulative probability plot for both influent and effluent concentrations for 
TSS for all the filters combined using only paired data. Note that the TSS influent concentration 
distribution is statistically different from the lognormal distribution (and the normal distribution 
as well).  
1 0 0 0 0 . 01 0 0 0 . 01 0 0 . 01 0 . 01 . 00 . 1
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
T S S  ( m g / L )
P
e
r
c
e
n
t
4 . 8 2 1 1 . 1 1 7 4 7 0 . 7 6 3 0 . 0 4 4
2 . 2 6 6 1 . 2 8 8 4 7 0 . 6 0 8 0 . 1 0 8
L o c S c a l e N A D P
T S S  I n
T S S  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  T S S  I n ,  T S S  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 4 TSS Influent and Effluent Probability Plots for all Sites Combined 
 
 
9 
 
A boxplot of influent TSS concentrations for the five sand filters is presented in Figure 5 and 
ANOVA indicates that these concentrations are significantly different (p < 0.000).  Figure 6 
presents a boxplot of TSS discharge concentrations for the five sand filters. An ANOVA analysis 
indicates there is no significant differences in the average discharge concentrations (p = 0.394), 
despite very different influent concentrations, facility design (pretreatment or not), maximum 
water depth, and other factors. The mean concentrations at Barton Ridge and Brodie Oaks are 
slightly higher than the other sites; however, median concentrations are almost identical. These 
sites have very different maximum water depths, only a couple of feet at Barton Ridge compared 
with about 15 feet at Brodie Oaks. We can, therefore, conclude that water depth has little effect 
on average particle removal.  
 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 0 0 0
8 0 0
6 0 0
4 0 0
2 0 0
0
B M P
T
S
S
 
I
n
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 5 Boxplot Influent TSS Concentrations 
 
 
10 
 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
0
B M P
T
S
S
 
E
f
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 6 Boxplot of TSS Discharge Concentrations (all data) 
Table 4 presents arithmetic mean concentrations for the individual sites based on storms where 
data were available for both the influent and effluent. TSS removal was statistically significant at 
all locations; however the efficiency ratio varied substantially, despite the very similar effluent 
quality produced. It is apparent from the table how influent concentrations strongly affect this 
ratio, with the cleanest sites have apparently the worst performance and the dirtiest sites the best. 
Consequently, efficiency ratio is not recommended for comparing BMP performance. 
Table 4 TSS Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Mall 74 13 82 0.002 
Barton Ridge 286 25 91 0.070 
Brodie Oaks 69 18 74 0.125 
Highwood 101 18 82 0.022 
Jollyville 304 16 95 <0.000 
All Sites 198 17  91 <0.000 
 
 
 
11 
 
Filtration is primarily a particle removal process and its efficiency is a function of several 
variables. These include the properties of the filtration media, including size distribution and 
surface chemistry. Similarly, characteristics for the particles in runoff play an equally important 
role. Finally, flow rates through the filters can often have a substantial impact on particle 
retention. These variables determine how well particles are removed within the filter either 
through straining or surface attachment.  
A previous study of particle removal in sand filters (TxDOT full sedimentation filter at Loop 360 
and Barton Creek) found that virtually all the mass associated with particles with diameters about 
3 ?m were removed (Figure 7). These particles account for the vast majority of TSS, which 
suggests that we should observe relatively constant effluent quality regardless of the 
concentrations in untreated runoff (Karamalegos et al., 2005). Pretreatment (either in a dry 
sedimentation basin or wet pond) would be expected to primarily reduce the maintenance 
requirements of the filters rather than affecting the quality of the system discharge. 
Consequently, one would expect that any differences observed in discharge quality would be 
associated with the maximum water depth, which controls flow rates through the filter. 
This conclusion is reinforced when one looks at a plot of discharge (an indirect measurement of 
water depth and loading rate) versus TSS effluent concentration for Brodie Oaks, which has the 
highest potential water level of all the filters (Figure 8). This plot shows that the two storms with 
the highest TSS discharge concentrations were produced by storms that had the two lowest 
hydraulic loading rates. It is important to note that the hydraulic loading rate for stormwater 
filters (3 ft/d or 0.016 gal/min*ft2) is far lower than that used in applications such as water 
treatment plants (3 gal/min*ft2 or 576 ft/d) 
 
 
 
 
 
 
 
 
 
 
 
 
12 
 
 
 
 
 
Figure 7 (A) Particle Size Distribution Function and (B) Volume Distribution 
 
 
 
B. 
A. 
0
1
2
3
4
5
6
7
H i g h w ay Ap pr o ach
BM P
P
a
r
t
ic
le Si
z
e D
ist
ri
b
u
ti
o
n
 F
u
n
ct
io
n
lo
g
(
?
N/
?
d
p
) 
(
?
N/
?
d
p
 in
 c
m
-3
?
m
-1
) 
SS C = 24 9 m g / L
SSC = 2 . 9 m g/ L
0
1 10
8
2 10
8
3 10
8
-0 .5 0 0.5 1 1.5 2
V
o
lu
m
e
 D
is
t
ri
b
u
t
io
n
?
V/
?
 lo
g
 d
p
 (V
 i
n
 
?
m
3
cm
-
3
) 
lo g  d
p
 ( d
p
 i n  ? m)
 
 
13 
 
 
Figure 8 Comparison of Maximum Discharge Rate and Average TSS Discharge 
Concentration at Brodie Oaks 
The two storms in Figure 8 with the elevated concentrations were also the smallest events, which 
suggests that what is observed is a first flush effect that is not diluted by runoff that would have 
occurred subsequently in a larger event.   
For many BMPs, the effluent quality is affected by the influent concentration. Consequently, a 
plot of influent versus effluent TSS concentrations using data from all the sites was prepared and 
linear regression performed (Figure 9). The R2 for the regression is 0.44, which is relatively high 
for stormwater analyses; consequently, discharge concentration can be predicted based on the 
influent concentration. This figure indicates that discharge concentrations tend to increase 
gradually with higher influent concentrations. The relatively mild slope suggests that higher 
runoff concentrations are mostly associated with an increase in the larger particles, which are all 
effectively removed by the filter. This is consistent with the results of a study of the effect of 
BMPs on particle size distributions of highway runoff (Karamalegos, 2005).  A previous study of 
sand filter performance (Barrett, 2003) found no statistical relationship between influent and 
effluent concentration; however, there was only one influent concentration in that study that was 
relatively high (420 mg/L). If one considers only the concentrations of less than 400 mg/L in the 
COA data, then one reaches the same conclusion that effluent concentration is independent of 
influent concentration.   
 
 
 
14 
 
 
 
 
Figure 9 Influent and Effluent TSS concentrations for all Sites 
 
When the sites are analyzed individually only two the facilities have significant relationships 
between influent and effluent TSS concentrations, Barton Ridge and Jollyville. Graphs for these 
two sites are presented in Figure 10 and Figure 11, respectively. The other sites do not appear to 
have the range of influent concentrations necessary to observe a statistically significant 
relationship. 
 
 
 
15 
 
 
Figure 10 Influent and Effluent TSS concentrations for Barton Ridge 
 
Figure 11 Influent and Effluent TSS concentrations for Jollyville 
 
 
 
16 
 
One interesting phenomenon related to particle removal in filters is the pattern of concentrations 
of individual aliquots taken during a single event. Figure 12 presents the individual 
concentrations observed at Jollyville. There is an unmistakable ?first flush? type pattern evident, 
which is also observed at every other site. There are two potential explanations for this pattern. 
One is that subsequent events mobilize material retained in and on the filter bed from previous 
events and transport that material through the filter. This seems unlikely since visual observation 
of the filter media profile consistently indicates that most material is retained in the top few 
inches of the filter. Consequently, it seems unlikely that an initial storm surge could transport 
concentrations as high as 200 mg/L through the filter. 
A second explanation is that some amount of the fine material that is transported through the 
filter, settles out in the underdrain during the waning hours of the previous event. This material 
could then be remobilized during the high flows in the initial part of the subsequent event 
producing the first flush pattern that we observe. This is easily testable by flushing out the 
underdrain through the cleanout ports and observing whether the following event produces this 
same temporal pattern in concentration. 
 
 
 
17 
 
0
50
100
150
200
250
0 6 12 18 24 30 36 42 48
T
S
S
 
(mg
/
L
)
T i me  S i n c e  F i r s t  S a mp l e  (h o u r s )
5/29/1988
6/2/1988
4/19/1989
5/5/1989
5/10/1989
6/11/1989
6/13/1989
8/1/1989
8/15/1989
10/7/1989
11/22/1989
2/18/1990
3/14/1990
4/26/1990
6/4/1990
1/12/1995
3/13/1995
4/4/1995
5/8/1995
5/18/1995
6/11/1995
7/30/1995
11/17/1995
4/22/1996
4/25/1997
5/9/1997
 
Figure 12 Jollyville Sand Filter TSS Discharge Time Series
 
 
18 
 
An accumulation of fine grained material often occurs within the upper few inches of filters and 
on the surface (the Schmutzdecke).  One might conclude that this finer grained material would 
increase the TSS removal by providing additional sites where particles could attach and reducing 
the size of the pore throats increasing the efficiency of particle straining. To test this hypothesis, 
a multiple regression analysis was performed for selected facilities. The regression used date and 
TSS influent concentrations as independent variables. It was found that time was not a 
significant predictor of Jollyville (p = 0.931) or Barton Ridge (p = 0.494) effluent concentrations. 
Unexpectedly, the TSS concentration in the discharge of Barton Creek Square Mall grew 
significantly worse through time (p = 0.005) and unlike the other systems was independent of 
influent concentration (Figure 13). Similar statistical tests were also performed for TP and Zn to 
try to assess whether any change in dissolved concentrations occurred; however, neither of these 
constituents exhibited a significant trend with time (p = 0.914 and p = 0.290).  
1 0 / 1 / 1 9 8 67 / 1 / 1 9 8 64 / 1 / 1 9 8 61 / 1 / 1 9 8 61 0 / 1 / 1 9 8 57 / 1 / 1 9 8 54 / 1 / 1 9 8 5
3 0
2 5
2 0
1 5
1 0
5
0
D a t e
T
S
S
 
E
f
f
l
u
e
n
t
 
(
m
g
/
L
)
B C S M  
v 
Figure 13 Relationship between TSS Discharge Quality and Time for Barton Creek Square 
Mall 
The observation that TSS removal does not improve as a filter ages and straining becomes more 
dominant suggests that the removal processes change through time. Early in the life of the filter a 
substantial amount of particle removal likely occurs through deep bed filtration where 
sedimentation, interception, and Brownian motion are the dominant processes. As the filter 
 
 
19 
 
accumulates material near the surface, straining becomes more important and near the end of the 
filter life is probably the dominant particle removal mechanism. Consequently, we infer a shift in 
the processes for particle removal through time, but little change in overall efficiency. 
Hydraulic residence time (HRT) is another variable that might be expected to affect the removal 
of selected pollutants. HRT was calculated for events at the Jollyville site as the difference 
between influent and effluent hydrographs. During many events there were substantial 
differences between inflow and discharge volumes which add some noise to the analysis. Figure 
14 presents a graph of TSS removal versus HRT. It is apparent that HRT has no discernable 
effect on performance. A regression analysis was also performed to determine if TSS discharge 
concentration was a function of HRT (controlling for influent concentration); however, HRT was 
not a significant predictor (p = 0.156). 
 
Figure 14 Comparison of TSS removal to Hydraulic Residence Time 
Barton Ridge is the only one of the sites that includes samples collected from the discharge of 
the sedimentation basin, which allows a calculation of the TSS mass removed by sedimentation 
as opposed to filtration. Using the mean of the individual storm removals, the data indicate that 
about 68% of the TSS is removed by sedimentation, while the filter basin removes an additional 
75% of the remainder for a total removal by the system of approximately 90%. It should be noted 
 
 
20 
 
that material coarse enough to be removed by sedimentation, would also likely be completely 
removed by the filtration unit even if no pretreatment were provided.  
TSS Conclusions: 
1. Discharge TSS concentrations were similar for all facilities, so design factors such as 
pretreatment, maximum water depth, and filter area apparently have little effect on 
particle removal.  
2. The upper bound of diameter for particles that can pass through the filter is about 30 ?m. 
3. All the facilities evaluated (even Brodie Oaks, which includes a wet pond for 
pretreatment) had a distinct first flush that might be attributed to the accumulation of 
sediment in the underdrain system at the end of storm events. 
4. TSS removal was not a function of time, indicating that the accumulation of material on 
and within the filter had little impact on particle removal, but suggests that the removal 
processes change with time. 
5. Hydraulic residence time did not affect either removal efficiency or discharge 
concentration. 
6. Pretreatment reduces the total sediment load to the filter by about 65-70%, but may not 
material extend the life of the filter since this sediment likely is fairly coarse, which 
would result in little loss of permeability if it accumulated on the surface of the filter. 
 
 
 
21 
 
3 Volatile Suspended Solids (VSS) Performance 
 
There is very little data on the removal of VSS in sand filters, since data are available for only 
two sites, Barton Ridge and Jollyville. The data were initially examined to determine their 
distribution, which is mostly lognormal as shown in Table 5.  
Table 5 Statistical Distribution of VSS Data for Each Site 
 Barton 
Ridge 
Jollyville All sites 
Influent Lognormal Lognormal Lognormal 
Effluent Lognormal Normal/ Lognormal Lognormal 
 
Figure 15 presents the cumulative probability plots of VSS influent and effluent concentrations 
for the pooled data from both sites. The distributions are very distinct, which confirms that 
substantial reduction of VSS does, in fact, occur. 
1 0 0 0 . 01 0 0 . 01 0 . 01 . 00 . 1
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
V S S  ( m g / L )
P
e
r
c
e
n
t
3 . 1 2 5 0 . 9 0 0 5 2 0 0 . 5 5 9 0 . 1 2 9
0 . 5 2 3 9 0 . 9 8 8 9 2 0 0 . 4 8 5 0 . 2 0 2
L o c S c a l e N A D P
V S S  i n
V S S  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  V S S  i n ,  V S S  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 15 Probability Plots of VSS Influent and Effluent 
 
 
22 
 
Boxplots of influent and effluent concentrations at those sites are presented in Figure 16 and 
Figure 17, respectively. The influent concentrations at the two sites are not significantly different 
(p = 0.674); however, the effluent concentrations are significantly lower at Jollyville (p = 0.003). 
This somewhat unexpected since the Barton Ridge site includes pretreatment as well as 
infiltration, while Jollyville does not. 
J o l l y v i l l eB a r t o n  R i d g e
1 4 0
1 2 0
1 0 0
8 0
6 0
4 0
2 0
0
B M P
V
S
S
 
I
n
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 16 Boxplot of Influent VSS concentrations 
J o l l y v i l l eB a r t o n  R i d g e
1 0
8
6
4
2
0
B M P
V
S
S
 
E
f
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 17 Boxplot of Effluent VSS concentrations 
 
 
23 
 
Table 6 summarizes the average concentrations observed based on paired data. Performance of 
the two systems is similar and both exhibit statistically significant removal. 
Table 6 VSS Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Ridge 32 3.6 89 0.001 
Jollyville 30 1.2 96 0.004 
All Sites 31.5 2.5 92 <0.000 
 
Figure 18 presents the ?relationship? between influent and effluent concentrations, which is not 
statistically significant (p = 0.214). This is a difficult figure to interpret since the points for the 
two sand filters seem to fall in distinctly different areas, making a prediction of the effluent 
concentration very uncertain. 
 
Figure 18 Relationship Between VSS Influent and Effluent Concentrations 
Figure 19 presents the time series of VSS concentrations at Jollyville. There is a distinct first 
flush effect; however, the concentrations are fairly low ? normally less than 2 mg/L. 
 
 
24 
 
0
2
4
6
8
10
12
14
0 . 0 0 6 . 0 0 1 2 . 0 0 1 8 . 0 0 2 4 . 0 0 3 0 . 0 0 3 6 . 0 0 4 2 . 0 0 4 8 . 0 0
V
ol
a
t
i
l
e
 
S
us
pe
nde
d 
S
oi
l
ds
 
(
m
g
/
L
)
T i m e  s i nc e  f i r s t  s a m pl e  ( ho ur s )
1 /1 2 /9 5
3 /1 3 /9 5
4 /4 /9 5
5 /8 /9 5
5 /1 8 /9 5
6 /1 1 /9 5
7 /3 0 /9 5
1 1 /1 7 /9 5
3 /2 6 /9 6
4 /2 2 /9 6
4 /2 5 /9 7
4 /2 6 /9 7
5 /9 /9 7
 
Figure 19 Temporal Pattern of VSS Discharge Concentrations at Jollyville
 
 
25 
 
VSS Conclusions 
1. VSS concentrations tend to follow a lognormal distribution. 
2. Removal is similar to that observed for TSS and is statistically significant at both sites. 
3. VSS discharge concentrations tend to have a first flush pattern, although it is somewhat 
muted because of the generally low concentrations. 
 
 
26 
 
4 Total Phosphorus Performance 
 
The total P data were analyzed to determine whether they fit a normal or lognormal distribution. 
At most sites there was no clear cut answer with neither distribution being rejected. Figure 20 
presents the cumulative probability plots for both the influent and effluent total phosphorus 
concentrations based on the pooled and paired data. The effluent distribution is significantly 
different from the lognormal distribution, but this may be the result of detection limited values at 
the low end 
Table 7 Statistical Distribution of Total Phosphorus Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie Oaks Highwood Jollyville All sites 
Influent Normal/Lognormal Lognormal Normal/Lognormal ? Normal/Lognormal Lognormal 
Effluent Normal Lognormal Normal/Lognormal ? Normal/Lognormal ? 
 
1 . 0 00 . 1 00 . 0 1
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
T o t a l  P  ( m g / L )
P
e
r
c
e
n
t
- 1 . 7 3 5 0 . 8 4 2 0 4 1 0 . 5 6 1 0 . 1 3 8
- 2 . 6 5 4 0 . 6 1 8 2 4 1 0 . 9 8 6 0 . 0 1 2
L o c S c a l e N A D P
T P  i n
T P  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  T P  i n ,  T P  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 20 Probability Plots of Total P Influent and Effluent 
 
 
27 
 
Much of the phosphorus in runoff is associated with particulate material (50-80%); consequently, 
one would expect that the removal mechanism (particle retention) for the majority of phosphorus 
in sand filters would be similar to that observed for TSS. The influent and effluent concentrations 
at the five sites are compared in Figure 21 and Figure 22 respectively. The influent 
concentrations are significantly different (p < 0.000); however, discharge concentrations are not 
significantly different (p = 0.378), which is similar to what was observed for TSS.  
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 . 0
0 . 8
0 . 6
0 . 4
0 . 2
0 . 0
B M P
T
o
t
a
l 
P
 I
n
f
lu
e
n
t
 (
m
g
/
L
)
 
Figure 21 Boxplot of Total P Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 . 0
0 . 8
0 . 6
4
0 . 2
0 . 0
B M P
T
o
t
a
l 
P
 
E
f
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 22 Boxplot of Total P Effluent Concentrations 
 
 
28 
 
Concentrations for the individual sites are presented in Table 8. There is substantially variability 
in the efficiency ratio; however, it is clearly related to influent concentration. The two sites with 
apparent negative removal (although not statistically significant) both have influent 
concentrations that are similar to the effluent concentrations at all the other sites. 
Table 8 Total Phosphorus Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Paired t-test 
Barton Mall 0.15 0.08 47 0.002 
Barton Ridge 0.32 0.10 69 0.022 
Brodie Oaks 0.07 0.08 -14 1.000 
Highwood 0.11 0.12 -9 0.508 
Jollyville 0.24 0.08 67 <0.000 
All Sites 0.22 0.08  61 <0.000 
 
The relationship between influent and effluent phosphorus concentration is presented in Figure 
23 for the pooled data. Of particular interest is the elevated discharge concentrations at 
Highwood compared to the other locations based on paired data. (Non - paired data include many 
storms with low discharge concentrations, which results in the effluent concentrations not being 
significantly different on average than the other sites.) Highwood is a manicured vegetated area 
within an apartment complex, so it would not be surprising if the landscape management 
company applied fertilizer. There is no particular trend of concentration with time; however, 
many of the higher readings occurred for storms in the fall of 1986.  
The regression line shown for these data in Figure 23 do not include the Highwood values. The 
regression line is much steeper than that observed for TSS, which means that removal is less. 
This result is likely related to the fact that a substantial amount of the phosphorus is dissolved, 
but another factor could be that the P is more associated with the finest clay particles that are not 
removed as effectively. The regression analysis was also performed for the individual sites and 
was only statistically significant at three of the facilities: Highwood, Barton Ridge, and Barton 
Creek Square Mall. The plots for these locations are presented in Figure 24, Figure 25, and 
Figure 26, respectively. 
 
 
29 
 
y  =  0 . 2 0 8 2 x  +  0 . 0 3 4 8
R?  =  0 . 4 7 1
0
0 . 0 5
0 . 1
0 . 1 5
0 . 2
0 . 2 5
0 . 3
0 . 3 5
0 . 4
0 0 . 2 0 . 4 0 . 6 0 . 8 1 1 . 2
T
o
t
a
l
 P
h
o
s
p
h
o
r
u
s
 O
u
t
 (mg
/
L
)
T o t a l  P h o s p h o r u s  I n f l u e n t  (mg / L )
B a r t on R i dg e
B a r t on M a l l
J ol l y v i l l e
H i g hw ood
B r odi e  O a k s
 
Figure 23 Relationship between P Influent and Effluent Concentrations All Sites 
 
Figure 24 Relationship between P Influent and Effluent Concentrations for Highwood 
 
 
 
30 
 
 
Figure 25 Relationship between P Influent and Effluent Concentrations for Barton Ridge 
 
Figure 26 Relationship between P Influent and Effluent Concentrations for Barton Mall 
Figure 27 presents the temporal trend of total P concentrations in individual events at Jollyville, 
with time zero set at the time the first sample of the event was collected. The trend is very similar 
to that observed for TSS, with an obvious first flush type of pattern, but relatively constant 
concentrations of less than 0.10 mg/L beyond that. 
 
 
31 
 
0
0 . 1
0 . 2
0 . 3
0 . 4
0 . 5
0 . 6
0 . 7
0 . 8
0 . 9
1
0 6 12 18 24 30 36 42 48
T
o
t
a
l
 
P
h
o
s
p
h
o
r
u
s
 
(mg
/
L
)
T i me  S i n c e  F i r s t  S a mp l e  (h o u r s )
5/20/1988
5/29/1988
6/1/1988
6/2/1988
8/11/1988
10/31/1988
4/13/1989
4/19/1989
5/5/1989
5/10/1989
6/11/1989
6/13/1989
8/1/1989
8/15/1989
10/7/1989
11/22/1989
2/18/1990
3/14/1990
4/26/1990
6/4/1990
1/12/1995
3/13/1995
4/4/1995
5/8/1995
5/18/1995
6/11/1995
7/30/1995
11/17/1995
4/22/1996
4/25/1997
4/26/1997
5/9/1997
 
Figure 27 Jollyville Total Phosphorus Time Series Discharge Concentration
 
 
32 
 
Since a substantial amount of phosphorus is in the dissolved form, it is possible that removal 
might be affected by residence time. Figure 28 presents the relationship between removal 
efficiency and hydraulic residence time for the Jollyville data. There is no statistically significant 
relationship between either removal efficiency or discharge concentration and HRT. 
 
Figure 28 Relationship between HRT and Total P Removal (Jollyville) 
 
Total Phosphorus Conclusions: 
1. The distribution of the data is not clearly either normal or lognormal. 
2. Effluent concentrations at the five sites are not statistically different, indicating that 
facility design has little impact on performance. 
3. The temporal pattern of discharge concentrations exhibits a first flush effect similar to 
that observed for TSS. 
4. The single vegetated site tends to have higher discharge concentrations, suggesting that 
fertilizer has been applied by landscape crews. 
5. The relationship between influent and effluent phosphorus concentrations is much steeper 
that that observed for TSS indicating less removal, probably the result of a substantial 
dissolved fraction.  
 
 
33 
 
6. Removal efficiency and discharge concentrations are independent of HRT based on the 
range of data available currently. 
 
 
 
34 
 
5 Dissolved Phosphorus Performance 
 
Information on dissolved phosphorus is limited to two sites, Barton Ridge and Jollyville. The 
Jollyville data is from the more recent monitoring conducted in the late 90?s. An analysis was 
performed to determine the statistical distribution of the data. Results of this analysis are 
presented in Table 9 and Figure 29. Individually the dissolved phosphorus data has no clear 
distribution, mostly due to the small sample size; however, when the data are pooled they clearly 
fit a lognormal distribution. 
Table 9 Statistical Distribution of Dissolved Phosphorus Data for Each Site 
 Barton Ridge Jollyville All sites 
Influent Normal/Lognormal Normal/Lognormal Lognormal 
Effluent Normal/Lognormal Normal/Lognormal Lognormal 
 
1 0 . 0 0 01 . 0 0 00 . 1 0 00 . 0 1 00 . 0 0 1
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
D a t a
P
e
r
c
e
n
t
- 2 . 2 7 7 1 . 0 2 3 1 2 0 . 2 8 3 0 . 5 6 8
- 2 . 7 5 6 0 . 7 4 3 2 1 2 0 . 2 5 0 0 . 6 7 9
L o c S c a l e N A D P
D P  i n
D P  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  D P  i n ,  D P  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 29 Probability Plots of Dissolved P Influent and Effluent 
 
 
35 
 
Boxplots of the influent and effluent concentrations are presented in Figure 30 and Figure 31. 
ANOVA indicates that neither the influent (p = 0.393) nor the effluent (p = 0.576) are 
significantly different at the two sites. 
J o l l y v i l l eB a r t o n  R i d g e
0 . 4
0 . 3
0 . 2
0 . 1
0 . 0
B M P
D
is
s
o
lv
e
d
 
P
 
I
n
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 30 Boxplot of Dissolved P Influent Concentrations 
J o l l y v i l l eB a r t o n  R i d g e
0 . 3 0
0 . 2 5
0 . 2 0
0 . 1 5
0 . 1 0
0 . 0 5
0 . 0 0
B M P
D
is
s
o
lv
e
d
 
P
 
E
f
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 31 Boxplot of Dissolved P Effluent Concentrations 
 
 
36 
 
Average influent and effluent concentrations for each of the sites is presented in Table 10. The 
data indicate significant removal at each site. The efficiency ratio is less for Jollyville, but this 
appears to be primarily the result of the relatively low influent concentrations. 
Table 10 Dissolved P Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Ridge 0.22 0.11 52 0.625 
Jollyville 0.11 0.07 39 0.727 
All sites 0.15 0.08 45 0.023 
 
Figure 32 presents the relationship between influent and effluent dissolved P concentrations. 
There is a relatively strong correlation evident and substantial removal is indicated, which 
suggests that adsorption or precipitation is occurring. The regression is not significant when the 
sites are analyzed individually. 
 
Figure 32 Relationship between Dissolved P Influent and Effluent Concentrations 
 
 
 
37 
 
Figure 33 presents the temporal trend of dissolved P concentrations in individual events at 
Jollyville, with time zero set at the time the first sample of the event was collected. The trend is 
in contrast to that of total P and TSS, with no evident first flush. The data also tend to bounce 
around with samples collected only minutes apart having substantially different concentrations 
(e.g., event on 4/4/95). This suggests that much of the variability evidenced in the figure may be 
due to laboratory issues and the fact that many of the measurements are close to the detection 
limit. 
An attempt was made to determine whether the residence time, contact time, or some other 
variable could explain differences in performance. Unfortunately, there are substantial mass 
balance errors in the runoff volumes that preclude making this analysis. Table 11 presents the 
magnitude of the effluent volume to the influent volume for events with dissolved P data, which 
demonstrate the inconsistency in the measured volumes. 
 
Table 11 Ratio of Effluent/Influent Volume for Events with Dissolved P Data 
Jollyville 
Event 
Volume  Ratio Barton Ridge 
Event 
Volume Ratio 
6-April 2.3 31-Aug 0.14 
11-Jun 0.75 2-Nov 3.45 
30-Jul 0.93 11-Jun 0.31 
17-Nov 1.36 20-Sep 0.84 
26-Mar 2.8   
25-Apr 0.55   
26-Apr 0.43   
9-May 0.32   
 
 
38 
 
0
0 . 0 2
0 . 0 4
0 . 0 6
0 . 0 8
0 . 1
0 . 1 2
0 . 1 4
0 . 1 6
0 . 1 8
0 . 2
0 . 0 0 6 . 0 0 1 2 . 0 0 1 8 . 0 0 2 4 . 0 0 3 0 . 0 0 3 6 . 0 0 4 2 . 0 0 4 8 . 0 0
D
i
s
s
o
l
v
e
d
 
P
 
(mg
/
L
)
T i me  s i n c e  F i r s t  S a mp l e  (h r s )
1/12/1995
3/13/1995
4/4/1995
5/18/1995
5/18/1995b
6/11/1995
7/30/1995
7/31/1995
11/17/1995
3/26/1996
4/22/1996
4/25/1997
4/26/1997
5/9/1997
 
Figure 33 Jollyville Dissolved Phosphorus Time Series Discharge Concentration 
 
 
39 
 
 
Dissolved Phosphorus Conclusions: 
1. Effluent concentrations at the two sites are not statistically different, indicating that 
facility design has little impact on performance. 
2. There is no consistent temporal pattern of discharge concentrations and the variability 
observed is more likely the result of laboratory lack of accuracy than real changes in 
concentration. 
3. Substantial reductions in the concentrations of dissolved phosphorus are evident; 
however, poor mass balance between influent and effluent volumes makes analysis of 
rate based processes infeasible. 
4. Both influent and effluent concentrations are lognormally distributed. 
 
 
 
40 
 
6 Total Kjeldahl Nitrogen (TKN) Performance  
 
An important consideration when evaluating the performance of sand filters for nitrogen as well 
as phosphorus removal is the degree to which the influent samples represent the total loading of 
nutrients to the filter. Manufacturers of proprietary BMPs have frequently complained that 
influent samples are biased against the larger, heavier particles that are not effectively captured 
by automatic samplers. The bias also extends to lighter, larger material that is either too large to 
enter the sampler intake or which floats on the surface like a substantial amount of litter as well 
as organic matter. On the other hand, only the smallest material can be transported through the 
filter media, so the effluent sample general is an accurate representation of the overall water 
quality discharged. 
A substantial amount of organic matter, including leaves and grass clippings, are commonly 
transported in stormwater; however, they are rarely present in stormwater samples or would 
probably be ignored during laboratory analysis. In addition, many of the facilities are located in 
landscaped areas with trees in the vicinity. It would be expected that a substantial amount of 
leaves would be deposited directly in the filtration system during leaf drop (fall for most trees 
except live oaks). Currently, the Jollyville sand filter has almost a foot of this decaying organic 
matter resting on the surface of the filter media. This organic material will eventually breakdown 
and some portion will be nitrified, driving up the total load of nitrate leached from the filter. 
Consequently, the removal of TKN, nitrate, and possibly phosphorus is almost certainly 
underestimated. 
The monitoring data were analyzed initially to determine their distribution and the results are 
presented in Table 12. Data from many of the sampling locations were not clearly either 
normally or lognormally distributed. 
Table 12 Statistical Distribution of TKN Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie Oaks Highwood Jollyville All sites 
Influent Normal/lognormal Normal/lognormal Normal/lognormal Normal/lognormal Lognormal ? 
Effluent lognormal lognormal Normal/lognormal Normal/lognormal Lognormal Lognormal 
 
Figure 34 presents the cumulative probability plot of TKN influent and effluent concentrations 
using paired data from all the sites. The distributions are distinctly different from each other and 
the influent distribution is also significantly different from the lognormal distribution. 
 
 
41 
 
1 0 . 01 . 00 . 1
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
T K N  ( m g / L )
P
e
r
c
e
n
t
- 0 . 3 5 3 7 0 . 8 1 8 3 5 7 1 . 0 2 4 0 . 0 1 0
- 0 . 9 6 9 1 0 . 6 3 6 2 5 7 0 . 4 3 8 0 . 2 8 6
L o c S c a l e N A D P
T K N  I n
T K N  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  T K N  I n ,  T K N  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 34 Probability Plot of TKN Influent and Effluent 
Boxplots of influent and effluent TKN concentrations are presented in Figure 35 and Figure 36. 
Differences in concentrations are significant for both the influent (p < 0.000) and effluent (p = 
0.023), which is due to the elevated concentrations at Barton Ridge. However, when effluent 
concentrations are plotted against influents, the Barton Ridge data does not appear to be 
substantially higher (Figure 37). In fact, ANOVA indicates that the differences are not 
significant (p = 0.410) when only storms with paired data are used to make the comparison. The 
likely explanation for the differences in discharge quality is the substantial dissolved component 
which is not effectively removed by the filter. 
Mean influent and effluent concentrations, along with their efficiency ratio and Wilcoxon SRT 
comparison, are presented in Table 13. Discharge concentrations at all the sites are very similar; 
however, the statistical tests indicates that the removal at Barton Creek Square Mall and 
Highwood is not statistically significant. The apparent poor result at Highwood is due to three 
events where the influent TKN concentration was less than 0.1 mg/L, which is very low in 
comparison to most events monitored there and at other sites. These three events had negative 
removals, which resulted in the overall performance appearing worse than the other sites. 
 
 
42 
 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
7
6
5
4
3
2
1
0
B M P
T
K
N
 
I
n
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 35 Boxplots of TKN Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
2 . 0
1 . 5
1 . 0
0 . 5
0 . 0
B M P
T
K
N
 
E
f
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 36 Boxplot of TKN Effluent Concentrations 
 
 
43 
 
Table 13 TKN Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Mall 0.87 0.40 54 0.386 
Barton Ridge 1.35 0.64 53 0.001 
Brodie Oaks 0.59 0.41 30 0.063 
Highwood 0.59 0.43 27 0.344 
Jollyville 0.96 0.43 55 <0.000 
All Sites 0.92 0.46  50 <0.000 
 
Figure 37 shows a relatively strong correlation between influent and effluent concentrations of 
TKN at all the sites, which is consistent with the relationships exhibited by many other 
constituents. When sites are analyzed individually only Barton Ridge and Highwood have 
statistically significant relationships. These are shown in Figure 38 and Figure 39, respectively. 
 
Figure 37 Relationship between Influent and Effluent TKN Concentrations (all sites) 
 
 
44 
 
 
Figure 38 Relationship between Influent and Effluent TKN Concentrations Barton Ridge 
 
Figure 39 Relationship between Influent and Effluent TKN Concentrations Highwood 
Figure 40 presents a time series of the discharge of TKN at Jollyville. Like TSS, there is a strong 
first flush phenomenon. This is likely also the result of resuspension of material in the underdrain 
or breakdown of accumulated organic matter during the inter-event period into particles small 
enough to be transported through the filter media. 
 
 
45 
 
0
0 . 5
1
1 . 5
2
2 . 5
3
3 . 5
4
0 6
12 18 24 30 36 42 48
T
K
N
 
(m
g
/
L
)
T i me  S i n c e  F i r s t  S a mp l e  (h o u r s )
J o l l yvi l l e  T K N
5/29/1988
6/2/1988
10/31/1988
4/19/1989
5/5/1989
5/10/1989
6/11/1989
8/1/1989
8/15/1989
10/7/1989
11/22/1989
2/18/1990
3/14/1990
4/26/1990
6/4/1990
1/12/1995
3/13/1995
4/4/1995
5/8/1995
5/18/1995
6/11/1995
7/30/1995
11/17/1995
4/22/1996
4/26/1997
5/9/1997
 
Figure 40 Jollyville TKN Time Series Discharge Concentrations
 
 
46 
 
It has long been observed that sand filters are an exporter of nitrate, which is probably the result 
of oxidation of ammonia and organic nitrogen. Consequently, it would not be unexpected to find 
that lower TKN concentrations would be produced by storms with a longer HRT. Figure 41 
presents this relationship for the Jollyville site and the regression equation demonstrates the high 
correlation between these variables. Unfortunately, the effect is the opposite that one would 
expect with higher discharge concentrations associated with longer hydraulic residence times. 
Therefore this pattern is more likely related to factors other than processes that occur on and 
within the filter.  
 
Figure 41 Relationship between HRT and TKN Removal (Jollyville) 
A regression analysis was also performed on the Jollyville data to determine if the discharge 
concentration could be predicted based on influent concentration and HRT. The regression was 
statistically significant (p = 0.013), with both influent concentration (p = 0.005) and HRT (p = 
0.052) being significant. The R2 for this relationship was 0.75, which means that virtually all the 
variability in the discharge concentrations are explained by just these two variables. Note that in 
the multiple regression the coefficient for HRT is negative which indicates that longer residence 
times do reduce the TKN discharge, all other factors being equal. 
TKN out = 0.343 - 0.0225 HRT + 0.363 TKN in 
 
 
 
 
47 
 
Conclusions for TKN 
1. Performance of all the facilities for TKN removal is very similar, except that discharge 
concentrations at Barton Ridge tend to be higher than other sites, which is the result of 
significantly higher influent concentrations. 
2. A first flush phenomenon in the discharge from the filtration systems is also evident for 
TKN. 
3. TKN is one of the few constituents whose removal is correlated with hydraulic residence 
time; however the effect is only apparent when influent concentration is included in the 
regression. 
 
 
48 
 
7 Nitrate plus Nitrite Performance 
 
Sand and other media filters have been routinely shown to be exporters of nitrate, with effluent 
concentrations substantially above influent concentrations. As mentioned previously, this may be 
due in part to unsampled contributions from leaves, grass clippings, and other organic material 
washed into the basins during storms or otherwise deposited in the filter in the interevent period.  
As shown in Table 14, the concentration data at the individual does not have a clear statistical 
distribution. Figure 42 presents the cumulative probability plot of nitrate influent and effluent 
concentrations using paired data from all the sites. The distributions are distinctly different from 
each other and the influent distribution is also significantly different from the lognormal 
distribution. 
Table 14 Statistical Distribution of Nitrate Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie Oaks Highwood Jollyville All sites 
Influent Normal/ lognormal Normal/ lognormal normal lognormal ? ? 
Effluent Normal/ lognormal lognormal lognormal Normal/ lognormal ? lognormal 
 
1 . 00 . 1
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
N O 3  ( m g / L )
P
e
r
c
e
n
t
- 1 . 1 3 7 0 . 6 0 4 4 5 8 1 . 0 7 7 0 . 0 0 7
- 0 . 5 4 3 7 0 . 5 3 5 8 5 8 0 . 5 6 4 0 . 1 3 8
L o c S c a l e N A D P
N O 3  i n
N O 3  o u t
V a r ia b le
P r o b a b i l i t y  P l o t  o f  N O 3  i n ,  N O 3  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 42 Probability Plots of NO2+3 Influent and Effluent Concentrations 
 
 
49 
 
Boxplots of NO2+3 concentrations indicate that both influent (p = 0.002) and effluent (p = 0.019) 
concentrations, Figure 43 and Figure 44 respectively, are significantly different at the five sites. 
The differences in discharge concentrations are the result of particularly low concentrations 
observed at Highwood. Excluding that site differences in discharge concentrations are not 
significant (p = 0.544).  
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
2 . 5
2 . 0
1 . 5
1 . 0
0 . 5
0 . 0
B M P
N
O
2
3
 
I
n
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 43 Boxplot of Nitrate + Nitrite Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
4
3
2
1
0
B M P
N
O
3
+
2
 
E
f
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 44 Boxplot of Nitrate + Nitrite Effluent Concentrations 
 
 
50 
 
One reason for the low concentrations at Highwood may be the particularly low filtration 
volume. The filtration component of the system is confined to some 10?s of feet of linear trench, 
so the volume available for nitrification is very limited compared to the other locations. In 
addition, the system currently has a bad connection between the overflow outlet and discharge 
pipe that may allow a substantial amount of the runoff to discharge without passing through the 
filter media. It is unknown if the same condition existed when the monitoring was conducted. 
Like previous studies these data also indicate about a 75% increase in discharge concentrations 
compared to influent concentrations. Mean EMCs for each of the sites are presented in Table 15. 
Table 15 Nitrate Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Mall 0.37 0.96 -160 0.006 
Barton Ridge 0.56 0.68 -21 0.549 
Brodie Oaks 0.24 0.48 -100 0.062 
Highwood 0.26 0.40 -54 0.006 
Jollyville 0.41 0.72 -76 <0.000 
All Sites 0.38 0.67  -75 <0.000 
 
Figure 45 presents the relationship between influent and effluent concentrations. The effluent 
concentrations tend to be higher than the influent, which is a common finding in every sand filter 
monitoring study. Some of the poorest nitrate removal shown in Figure 45 apparently occurs at 
Barton Creek Square Mall, which has some of the highest reported discharge concentrations 
especially in relation to the influent concentration for the event. A closer look reveals that all of 
these higher discharge concentrations were calculated based on only three samples. As is evident 
in Figure 49 almost all storms have very elevated NO23 concentrations for the first sample, but 
within a very brief period of time the concentrations drop dramatically. Because of the few 
discrete samples from many events at BCSM, these higher concentrations are applied to a much 
larger volume of runoff than is likely warranted.  
For the individual sites, the relationship between influent and effluent concentrations was 
significant for Barton Ridge Jollyville, and Highwood. The individual regressions for these sites 
area presented in Figure 46, Figure 47, and Figure 48, respectively. 
 
 
 
51 
 
y  = 0 . 9 3 2 x  + 0 . 3 2 2
R?  = 0 . 2 6 0 2
0
0 . 5
1
1 . 5
2
2 . 5
0 0 . 2 0 . 4 0 . 6 0 . 8 1 1 . 2 1 . 4
N
O
2
3
 E
f
f
lu
e
n
t
 (mg
/
L
)
N O 2 3  i n f l u e n t  (mg / L )
BR
B C S M
JV
HI
B  O a k s
 
Figure 45 Relationship Between Influent and Effluent NO23 Concentrations (all sites) 
 
 
Figure 46 Relationship Between Influent and Effluent NO23 Concentrations Barton Ridge 
 
 
52 
 
 
Figure 47 Relationship Between Influent and Effluent NO23 Concentrations Jollyville 
 
Figure 48 Relationship Between Influent and Effluent NO23 Concentrations Highwood 
Like many other constituents there is a pronounced first flush in the filter discharge (Figure 49). 
Since this is a dissolved constituent, it is likely that the nitrate has formed during the inter-event 
period from oxidation of organic nitrogen and ammonia (TKN); however, antecedent dry period 
is not a significant predictor of average discharge concentrations. 
 
 
53 
 
0
0 . 5
1
1 . 5
2
2 . 5
3
3 . 5
4
0
.
0
0
6
.
0
0
1
2
.
0
0
1
8
.
0
0
2
4
.
0
0
3
0
.
0
0
3
6
.
0
0
4
2
.
0
0
4
8
.
0
0
N
O
3
 
(mg
/
L
)
T i me  S i n c e  F i r s t  S a mp l e  (h o u r s )
J o l l yvi l l e  N O 3 5/29/1988
6/2/1988
4/19/1989
5/5/1989
5/10/1989
6/11/1989
6/13/1989
8/2/1989
8/15/1989
10/7/1989
11/22/1989
2/18/1990
3/14/1990
4/26/1990
6/4/1990
1/12/1995
3/13/1995
4/4/1995
7/30/1995
11/17/1995
4/22/1996
4/25/1997
5/9/1997
 
Figure 49 Temporal Trends for NO2+3 for Jollyville 
 
 
54 
 
On the other hand, antecedent dry period appears to have a substantial impact on the initial 
concentration observed in the effluent (Figure 50). All the storms with initial discharge 
concentrations of less than 0.5 mg/L have antecedent dry periods of less than 3 days. This 
suggests that most nitrification occurs during the inter-event period. 
 
Figure 50 Effect of Antecedent Dry Period on Initial NO23 Concentration 
Removal efficiency and discharge concentrations of TKN improve with increased residence 
time, so one would expect the opposite relationship for nitrate; however, HRT was not a 
predictor of removal efficiency or effluent concentration, even when normalized for influent 
concentration (p = 0.588).  
Nitrate Conclusions 
1. Sand filters tend to export nitrate; however, no correlation between TKN removal and 
nitrate export could be documented. 
2. All of the filters except Highwood performed similarly, which may be the result of the 
very small filter area (and volume) in comparison to the other designs.   
3. Nitrate exhibits a pronounced first flush effect that is similar to that observed for other 
constituents. 
4. Initial nitrate concentration increases with increasing antecedent dry period, which 
suggests that most of the nitrification occurs in the inter-event period. 
 
 
55 
 
8 Fecal Coliform Performance 
 
The fecal coliform data at all the sites were analyzed to determine their statistical distribution. 
Neither the normal or lognormal distributions were dominant as shown in Table 16. Figure 51 
presents the cumulative probability plot of fecal coliform influent and effluent concentrations 
using paired data from all the sites. The distributions show a lot of overlap (marginal differences) 
and the influent distribution is also significantly different from the lognormal distribution. 
Table 16 Statistical Distribution of Fecal Coliform Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie Oaks Highwood Jollyville All sites 
Influent Lognormal Lognormal Lognormal /Normal ? Lognormal ? 
Effluent Lognormal Lognormal/Normal Lognormal Normal Lognormal Lognormal 
 
1
0
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
0
0
0
1
0
0
1
01
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
F e c a l  C o l i f o r m  ( C F U / 1 0 0  m g / L )
P
e
r
c
e
n
t
7 . 9 7 9 2 . 2 1 2 4 3 1 . 0 2 6 0 . 0 1 0
7 . 7 1 6 2 . 1 0 4 4 3 0 . 5 4 7 0 . 1 5 0
L o c S c a l e N A D P
F C O L  i n
F C O L  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  F C O L  i n ,  F C O L  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 51 Probability Plots of Fecal Coliform Influent and Effluent 
Boxplots of the influent and effluent fecal coliform concentrations are shown in Figure 52 and 
Figure 53, respectively. Both influent and effluent concentrations are significantly different 
 
 
56 
 
among the sites (p = 0.016 and p = 0.008), with the difference essentially the result of the low 
concentrations observed at Jollyville. 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 0 0 0 0 0
8 0 0 0 0
6 0 0 0 0
4 0 0 0 0
2 0 0 0 0
0
B M P
F
e
c
a
l 
C
o
li
f
o
r
m
 
(
C
F
U
/
1
0
0
m
L
)
 
Figure 52 Boxplot of Fecal Coliform Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 0 0 0 0 0
8 0 0 0 0
6 0 0 0 0
4 0 0 0 0
2 0 0 0 0
0
B M P
F
e
c
a
l 
C
o
li
f
o
r
m
 
E
f
f
lu
e
n
t
 
(
C
F
U
/
1
0
0
m
L
)
 
Figure 53 Boxplot of Fecal Coliform Discharge Concentrations 
The performance of sand filters for bacteria removal has become a higher priority with the 
adoption in the Austin area of the TMDL for indicator bacteria in Gilleland Creek. When the 
paired data from all the sites is used to estimate bacteria removal, the removal is modest and the 
 
 
57 
 
statistics indicate marginal certainty that removal has occurred. The performance for the 
individual sites is even more problematic, as shown in Table 17. The site with the least 
significant removal is Brodie Oaks; however, there are only four paired samples for analysis. 
None of the sites appear to have significant removal when all data are used (non-paired test), so 
it is critical to conduct the analysis with paired data only. 
Table 17 Fecal Coliform Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Mall 14,000 11,050 21 0.180 
Barton Ridge 17650 18400 -4 0.375 
Brodie Oaks 11,500 19,500 -70 1.000 
Highwood 25500 11,700 54 0.754 
Jollyville 3750 2100 44 0.607 
All Sites 13,300 9700  27 0.007 
 
A plot of influent versus effluent fecal coliform concentrations is presented in Figure 54. There is 
a substantial amount of scatter in the data, do it is difficult to draw any firm conclusions about 
the results. Influent and effluent concentrations are significantly related at both Jollyville and 
Highwood and these graphs are presented in Figure 55and Figure 56, respectively. 
 
Figure 54 Relationship between FC Influent and Effluent Concentrations all Sites 
 
 
58 
 
 
Figure 55 Relationship between FC Influent and Effluent Concentrations Jollyville 
 
Figure 56 Relationship between FC Influent and Effluent Concentrations Highwood 
A plot of time series of fecal coliform discharge concentrations at Jollyville are presented in 
Figure 57. There are only two events which exhibit a strong first flush effect, which suggests that 
re-growth or re-suspension of previously collected bacteria did not occur. The natural conclusion 
is that the bacteria are preferentially attached to the smallest size fraction, which is conveyed 
through the media with only modest removal. 
 
 
59 
 
0
2000
4000
6000
8000
10000
12000
14000
0 . 0 0 6 . 0 0 1 2 . 0 0 1 8 . 0 0 2 4 . 0 0 3 0 . 0 0 3 6 . 0 0 4 2 . 0 0 4 8 . 0 0
F
e
c
a
l
 
C
o
l
i
f
o
r
m 
(c
f
u
/
1
0
0
mL
)
T i me  S i n c e  F i r s t  S a mp l e  (h r s )
5/20/1988
5/29/1988
6/1/1988
6/2/1988
3/28/1989
4/19/1989
5/5/1989
5/10/1989
5/14/1989
6/11/1989
6/13/1989
8/15/1989
10/7/1989
11/22/1989
3/14/1990
4/26/1990
6/4/1990
 
Figure 57 Temporal Trends for Fecal Coliform for Jollyville
 
 
60 
 
Finally, Figure 58 presents a graph relating fecal coliform discharge concentration to hydraulic 
residence time at Jollyville. The relationship is significant (p = 0.052) and influent concentration 
does not seem to affect the discharge concentration. An attempt was made to develop this same 
relationship for several other sites without success. At Barton Creek Square Mall, the HRT for 
all the events fell in a narrow range from about 7 to 9 hours, with no obvious relationship 
apparent. The residence times at Barton Ridge are probably more related to mass balance issues 
than actual residence times. At that site effluent volumes ranges from 14% to 344% of influent 
volumes. In addition, as will be described later, fecal strep concentrations at Jollyville were not 
related to HRT at all. 
 
Figure 58 Relationship between Fecal Coliform Effluent Concentration and HRT at 
Jollyville 
 
Fecal Coliform Conclusions: 
1. Fecal coliform removal in the sand filters is modest and marginally significant 
statistically. 
2. Discharge concentrations at Jollyville decreased with increasing residence times, 
although this could not be confirmed using data at the other sites. 
3. There was only a subdued first flush for most events, which suggests that bacteria 
regrowth in the media and underdrain was not an important phenomenon. 
 
 
61 
 
9 Fecal Streptococcus Performance 
 
Fecal Streptococcus is another bacteria indicator organism that has been used historically to 
assess the potential human health risks associated with stormwater.  
As shown in Table 18, most of the fecal strep data are lognormally distributed. Figure 59 
presents the cumulative probability plot of fecal strep influent and effluent concentrations using 
paired data from all the sites. The distributions are clearly more distinctly different than the fecal 
coliform data, which confirms the higher confidence that removal of fecal strep actually occurs 
than that observed for fecal coliform. Both the influent and effluent distributions are 
indistinguishable from the lognormal distribution. 
Table 18 Statistical Distribution of Fecal Strep Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie 
Oaks 
Highwood Jollyville All sites 
Influent ? Normal/lognormal Normal/ Lognormal Lognormal Lognormal Lognormal 
Effluent Lognormal Normal/lognormal Lognormal Normal/lognormal Lognormal Lognormal 
 
1 0 0 0 0 0 01 0 0 0 0 01 0 0 0 01 0 0 01 0 01 0
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
F e c a l  S T r e p  ( C F U / 1 0 0 m L )
P
e
r
c
e
n
t
9 . 0 2 8 1 . 3 1 2 5 0 0 . 1 5 3 0 . 9 5 6
8 . 2 2 9 1 . 6 0 0 5 0 0 . 2 8 3 0 . 6 1 9
L o c S c a l e N A D P
F S T R  i n
F S T R  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  F S T R  i n ,  F S T R  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 59 Probability Plots of Fecal Strep Influent and Effluent 
 
 
62 
 
Boxplots of influent and effluent concentrations are presented in Figure 60 and Figure 61, 
respectively. Influent concentrations at the various sites are not significantly different (p = 
0.144); however, discharge concentrations are (p = 0.029). As can be seen in the two figures this 
difference is due primarily to a narrowing of the observed ranges in the effluent, since the 
relative relationships between the many sites remain the same (i.e., the high influent sites are the 
high effluent concentration sites). 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 2 0 0 0 0
1 0 0 0 0 0
8 0 0 0 0
6 0 0 0 0
4 0 0 0 0
2 0 0 0 0
0
B M P
F
e
c
a
l 
S
t
r
e
p
 
I
n
f
lu
e
n
t
 
(
C
F
U
/
1
0
0
m
L
)
 
Figure 60 Boxplot of Fecal Strep Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 0 0 0 0 0
8 0 0 0 0
6 0 0 0 0
4 0 0 0 0
2 0 0 0 0
0
B M P
F
e
c
a
l 
S
t
r
e
p
 
(
C
F
U
/
1
0
0
m
L
)
 
Figure 61 Boxplot of Fecal Strep Effluent Concentrations 
 
 
63 
 
On average, the concentrations of fecal strep are modestly reduced; however, the variability is 
much less than that observed for fecal coliform, which results in a more statistically significant 
relationship. The performance for the individual sites is presented below in Table 19. 
Table 19 Fecal Strep Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Mall 18,900 15,700 17 0.022 
Barton Ridge 4,650 2,150 54 0.375 
Brodie Oaks 21,900 19,500 11 0.375 
Highwood 30,800 21,900 29 0.774 
Jollyville 11,000 3,550 68 <0.000 
All Sites 17,780 11,825  33 <0.000 
 
This suggests that effluent concentrations should be correlated with influent concentrations. 
Figure 62 presents this relationship for all the sites. As can be seen the two variables are highly 
correlated (r2 approximately 0.70). Given the other vagaries in bacteria analysis, the implication 
that 70% of the variability in effluent concentrations is directly correlated (p = 0.001) with 
influent concentrations is rather surprising. The correlation is also significant when all the sites 
are analyzed individually; however, the result at Brodie Oaks is unlikely to be real since the 
equation predicts that discharge quality improves with decreasing influent quality. 
 
Figure 62 Relationship between Fecal Strep Concentrations (all sites) 
 
 
64 
 
 
Figure 63 Relationship between Fecal Strep Concentrations for Barton Ridge 
 
 
Figure 64 Relationship between Fecal Strep Concentrations Barton Mall 
 
 
65 
 
 
Figure 65 Relationship between Fecal Strep Concentrations Jollyville 
 
Figure 66 Relationship between Fecal Strep Concentrations Highwood 
 
 
66 
 
 
Figure 67 Relationship between Fecal Strep Concentrations Brodie Oaks 
A multiple regression was also performed to determine if HRT as well as influent concentration 
affected discharge concentrations; however, HRT was not a significant predictor (p = 0.342). 
This suggests that die-off and predation on the bacteria in the filter is not very substantial either 
because the bacteria within the media are protected from sunlight or because the rather sterile 
media does not support a very large community of bacteria predators. 
The temporal pattern of fecal strep discharge concentrations at Jollyville are presented in Figure 
68. There are a couple of events with very high and erratic concentrations reported, but most of 
the events have only a modest first flush type response. 
 
 
 
 
67 
 
0
5000
10000
15000
20000
25000
30000
35000
40000
0 . 0 0 6 . 0 0 1 2 . 0 0 1 8 . 0 0 2 4 . 0 0 3 0 . 0 0 3 6 . 0 0 4 2 . 0 0 4 8 . 0 0
Fe
c
a
l
 
S
t
r
e
pt
oc
c
c
i
 
(
C
FU
/
1
0
0
 m
L
)
T i m e  s i nc e  f i r s t  s a m pl e  ( hour s )
5 / 2 0 / 8 8
5 / 2 9 / 8 8
6 / 1 / 8 8
6 / 2 / 8 8
8 / 1 1 / 8 8
3 / 2 8 / 8 9
4 / 1 3 / 8 9
4 / 1 9 / 8 9
5 / 5 / 8 9
5 / 1 0 / 8 9
5 / 1 3 / 8 9
6 / 1 1 / 8 9
6 / 1 4 / 8 9
8 / 1 / 8 9
8 / 1 5 / 8 9
1 1 / 2 2 / 8 9
2 / 1 8 / 9 0
3 / 1 4 / 9 0
4 / 2 6 / 9 0
6 / 4 / 9 0
3 / 1 3 / 9 5
5 / 1 8 / 9 5
7 / 3 0 / 9 5
4 / 2 2 / 9 6
4 / 2 5 / 9 7
4 / 2 6 / 9 7
5 / 9 / 9 7
 
Figure 68 Temporal Pattern of Fecal Strep Discharge Concentrations at Jollyville
 
 
68 
 
Fecal Strep Conclusions: 
1. Statistically significant reduction in concentrations occur; however, the reduction is not 
large (roughly 30%). 
2. The discharge concentrations at the various sites are significantly different, but this is 
primarily the result of different influent concentrations. 
3. HRT is not a significant predictor of discharge quality for fecal strep. 
 
 
 
69 
 
10 Biochemical Oxygen Demand (BOD) Performance 
 
The distribution of measured BOD concentrations tends to be highly lognormal at the individual 
sites, as shown in Table 20. Figure 69 presents the cumulative probability plots of influent and 
effluent concentrations for the pooled data. The plots are distinctly different which supports the 
earlier finding that significant BOD reduction occurs in sand filters. Note that the effluent 
concentrations are not lognormally distributed. 
Table 20 Statistical Distribution of BOD Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie 
Oaks 
Highwood Jollyville All sites 
Influent Lognormal Lognormal ? Lognormal Lognormal Lognormal 
Effluent Lognormal ? Lognormal Lognormal ? ? 
 
1 0 0 . 01 0 . 01 . 00 . 1
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
B O D  ( m g / L )
P
e
r
c
e
n
t
1 . 8 3 1 0 . 6 7 0 5 5 3 0 . 4 3 4 0 . 2 9 1
1 . 2 9 9 0 . 8 7 6 8 5 3 0 . 6 4 6 0 . 0 8 7
L o c S c a l e N A D P
B O D  i n
B O D  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  B O D  i n ,  B O D  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 69 Probability Plots of BOD Influent and Effluent 
Boxplots of BOD influent and effluent concentrations are presented in Figure 70 and Figure 71, 
respectively. BOD Influent concentrations not significantly different (p = 0.345), but effluent 
concentrations are different (p = 0.002), mostly due to low concentrations observed at Barton 
 
 
70 
 
Ridge and Jollyville. It?s not apparent what causes the difference in performance for these two 
systems, since their designs have almost nothing in common. One potential explanation is that 
the BOD in the influent is more associated with the solid phase and, consequently, is more easily 
removed. 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
4 0
3 0
2 0
1 0
0
B M P
B
O
D
 
I
n
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 70 Boxplot of BOD Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
4 0
3 0
2 0
1 0
0
B M P
B
O
D
 
E
f
f
lu
e
n
t
 
(
m
g
/
L
)
 
Figure 71 Boxplot of Effluent BOD Concentrations 
 
 
71 
 
Mean influent and effluent BOD concentrations for the five sites are presented in Table 21. 
Performance is quite variable among the sites, with the worst performance occurring at Brodie 
Oaks, although there are only 5 storms at that site with paired data.  
Table 21 BOD Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT  
Barton Mall 9.2 7.3 18 0.065 
Barton Ridge 8.5 3.8 55 0.039 
Brodie Oaks 10.1 12.8 -27 1.000 
Highwood 8.5 6.3 26 0.344 
Jollyville 6.0 2.7 55 <0.000 
All Sites 8.0 5.6 30 <0.000 
 
Figure 72 presents a regression analysis of influent and effluent concentrations for all the sites 
pooled together. The regressions for all the sites individually are also statistically significant and 
those figures are presented in Figure 73 through Figure 77. It should be noted that although the 
regression is significant, the analysis of paired data indicate that substantial removal of BOD 
does not occur at either Brodie Oaks or Highwood. 
 
Figure 72 Relationship Between Influent and Effluent BOD Concentrations 
 
 
 
72 
 
 
Figure 73 Relationship Between Influent and Effluent BOD Concentrations Barton Ridge 
 
Figure 74 Relationship Between Influent and Effluent BOD Concentrations Barton Mall 
 
 
73 
 
 
Figure 75 Relationship Between Influent and Effluent BOD Concentrations Jollyville 
 
Figure 76 Relationship Between Influent and Effluent BOD Concentrations Highwood 
 
 
74 
 
 
Figure 77 Relationship Between Influent and Effluent BOD Concentrations Brodie Oaks 
The temporal pattern of BOD discharge from a sand filter is illustrated in Figure 78, which 
shows a fairly consistent first flush pattern, much like that observed for TSS. Figure 72 presents 
a graph of influent versus effluent concentrations. The relationship between the two is fairly 
strong (r2 = 0.48), so this is a reasonably good way to estimate removal efficiency or discharge 
concentration. 
Using the data at Jollyville, an estimate was also made of the relationship between HRT and 
BOD discharge concentration. The data are presented in Figure 79, which show a marked 
decrease in discharge concentration with increasing residence time. This suggests that some 
oxidation of organic material does occur during the storm event. A multiple regression was also 
performed which indicated that both HRT (p = 0.024) and influent concentration (p = 0.064) 
were significant predictors, with overall p value equal to 0.012. The following equation results: 
H R TB O DB O D ie ???? 227.0308.0  
A nonlinear formulation can also be developed, which relates the BOD discharge concentration 
to a power of the hydraulic residence time, which results in a p value of 0.006. In this 
formulation presented below, influent concentration is not a significant variable. This simple 
relationship did not hold for the data from Barton Creek Square Mall or Highwood, however. 
 
H R Te eB OD 103.048.7 ???  
 
 
75 
 
0
5
10
15
20
25
0 . 0 0 6 . 0 0 1 2 . 0 0 1 8 . 0 0 2 4 . 0 0 3 0 . 0 0 3 6 . 0 0 4 2 . 0 0 4 8 . 0 0
B
O
D
 
E
f
f
l
u
e
n
t
 
(
m
g
/
L
)
T i m e  S i nc e  Fi r s t  S a m pl e  ( hour s )
5 / 2 0 / 8 8
5 / 2 1 / 8 8
5 / 2 9 / 8 8
6 / 3 / 8 8
4 / 1 9 / 8 9
5 / 5 / 8 9
6 / 1 1 / 8 9
6 / 1 4 / 8 9
8 / 3 / 8 9
8 / 3 / 8 9
1 0 / 7 / 8 9
2 / 1 9 / 9 0
8 / 1 5 / 8 9
1 0 / 7 / 8 9
1 1 / 2 2 / 8 9
2 / 1 8 / 9 0
4 / 2 6 / 9 0
6 / 4 / 9 0
1 / 1 3 / 9 5
3 / 1 3 / 9 5
4 / 4 / 9 5
5 / 8 / 9 5
5 / 1 8 / 9 5
6 / 1 1 / 9 5
7 / 3 0 / 9 5
7 / 3 1 / 9 5
1 1 / 1 7 / 9 5
4 / 2 2 / 9 6
4 / 2 5 / 9 7
4 / 2 6 / 9 7
5 / 9 / 9 7
 
Figure 78 Temporal Pattern of BOD Discharge Concentrations at Jollyville 
 
 
76 
 
 
 
 
Figure 79 Relationship between BOD Effluent Concentration and HRT 
 
BOD Conclusions 
1. Removal of BOD in sand filters is statistically significant. 
2. BOD discharge concentrations are significantly different, mostly due to low 
concentrations at Jollyville and Barton Ridge; however, it is not clear what is responsible 
for the higher removal unless it is related to influent concentrations at those two sites that 
are more particle associated. 
3. BOD effluent concentrations at Jollyville were significantly related to HRT, but this was 
not observed at the other sites. In addition, it seems unlikely that substantial oxidation of 
the organic matter would occur in a matter of hours, since it often takes several days for 
substantial oxygen demand to be exerted in the laboratory. 
 
 
77 
 
11 Chemical Oxygen Demand (COD) Performance 
 
As shown in Table 22, the data at most of the sites are lognormally distributed. Cumulative 
probability plots of influent and effluent COD concentrations for all the paired data from the 
pooled sites is presented in Figure 80. The distributions are very distinct, providing confirmation 
that the removal of COD is statistically significant. 
Table 22 Statistical Distribution of COD Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie 
Oaks 
Highwood Jollyville All sites 
Influent Lognormal Lognormal Lognormal/normal ? Lognormal/ normal Lognormal 
Effluent Lognormal Lognormal Lognormal/normal Lognormal/normal Lognormal Lognormal 
 
1 0 0 01 0 01 0
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
C O D  ( m g / L )
P
e
r
c
e
n
t
3 . 5 5 9 0 . 8 1 3 7 4 8 0 . 1 9 0 0 . 8 9 4
2 . 9 0 6 0 . 5 8 4 7 4 8 0 . 5 3 1 0 . 1 6 7
L o c S c a l e N A D P
C O D  i n
C O D  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  C O D  i n ,  C O D  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 80 Probability Plot of COD Influent and Effluent Concentrations 
Boxplots of COD influent and effluent concentrations are presented in Figure 81 and Figure 82, 
respectively. Influent and effluent concentrations are significantly different at the five sites (p < 
0.000, and p = 0.056). To a large extent the differences in effluent concentration mirror the 
 
 
78 
 
differences in influent concentration at all the sites except Barton Creek Square Mall, where the 
effluent concentrations are relatively high compared to the other sites. 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
2 0 0
1 5 0
1 0 0
5 0
0
B M P
C
O
D
 
I
n
f
lu
e
n
t
 
(
m
g
/
L
)
B o x p l o t  o f  E M C
 
Figure 81 Boxplot of COD Influent Concentrations  
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
0
B M P
C
O
D
 
(
m
g
/
L
)
 
Figure 82 Boxplot of COD Effluent Concentrations 
 
 
79 
 
Average COD concentrations for the five sites are presented in Table 23. The effect of influent 
concentration on the efficiency ratio is again apparent with the cleanest watershed (Highwood) 
resulting in the lowest removal and worst t-test result, and the dirtiest watershed (Jollyville) 
having the best apparent performance and strongest statistical result. 
Table 23 COD Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Paired t-test 
Barton Mall 59 26 57 0.227 
Barton Ridge 59 25 58 0.004 
Brodie Oaks 12 9 25 0.375 
Highwood 29 18 38 0.065 
Jollyville 78 25 68 <0.000 
All Sites 47 22 53 <0.000 
 
Figure 83 presents a linear regression of influent and effluent COD concentrations. There is a 
reasonable relationship between the two, although a number of values for Barton Creek Square 
Mall appear to be abnormally high. One factor that explains this is that those three events are 
represented by just three samples. Consequently, the initially high sample concentrations get 
used to represent a substantial percentage of the total storm event, driving up the calculated 
average concentration. Linear regressions for individual sites were also significant for Barton 
Ridge, Highwood, and Brodie Oaks. These regressions are presented in Figure 84 through Figure 
86. 
 
Figure 83 Relationship between COD Influent and Effluent Concentrations all Sites 
 
 
80 
 
 
 
Figure 84 Relationship between COD Influent and Effluent Concentrations Barton Ridge 
 
Figure 85 Relationship between COD Influent and Effluent Concentrations Highwood 
 
 
81 
 
 
Figure 86 Relationship between COD Influent and Effluent Concentrations Brodie Oaks 
 
The temporal pattern of discharge concentrations for Jollyville is presented in Figure 87. There is 
a substantial first flush effect, with discharge concentrations after about 6 hours exhibiting only a 
gradual decline through time. 
 
 
 
 
82 
 
0
50
100
150
200
250
300
0 . 0 6 . 0 1 2 . 0 1 8 . 0 2 4 . 0 3 0 . 0 3 6 . 0 4 2 . 0 4 8 . 0
C
O
D
 (
m
g
/
L
)
T i m e  S i nc e  F i r s t  S a m pl e  ( ho ur s )
5 / 2 0 / 1 9 8 8
5 / 2 9 / 1 9 8 8
6 / 1 / 1 9 8 8
6 / 2 / 1 9 8 8
6 / 3 / 1 9 8 8
8 / 1 1 / 1 9 8 8
1 0 / 3 1 / 1 9 8 8
4 / 1 3 / 1 9 8 9
4 / 1 9 / 1 9 8 9
5 / 5 / 1 9 8 9
6 / 1 1 / 1 9 8 9
6 / 1 3 / 1 9 8 9
8 / 2 / 1 9 8 9
8 / 1 5 / 1 9 8 9
1 0 / 7 / 1 9 8 9
1 1 / 2 2 / 1 9 8 9
2 / 1 8 / 1 9 9 0
3 / 1 4 / 1 9 9 0
4 / 2 6 / 1 9 9 0
6 / 4 / 1 9 9 0
1 / 1 2 / 1 9 9 5
3 / 1 3 / 1 9 9 5
4 / 4 / 1 9 9 5
5 / 8 / 1 9 9 5
5 / 1 8 / 1 9 9 5
6 / 1 1 / 1 9 9 5
7 / 3 0 / 1 9 9 5
8 / 1 / 1 9 9 5
1 1 / 1 7 / 1 9 9 5
4 / 2 2 / 1 9 9 6
4 / 2 5 / 1 9 9 7
4 / 2 6 / 1 9 9 7
5 / 9 / 1 9 9 7
 
Figure 87 Temporal Pattern of COD Discharge Concentrations at Jollyville
 
 
83 
 
 
Several different formulations were tried to determine the impact of HRT on removal efficiency 
and discharge concentration for Jollyville. It was found that removal efficiency increased with 
increasing residence time (p = 0.017); however, this was not confirmed at any of the other sites. 
Discharge concentration, on the other hand, was only a function of influent concentration and 
HRT was not statistically significant. 
 
Figure 88 Relationship between HRT and COD Removal 
 
COD Conclusions 
1. Removal of COD in sand filters is statistically significant. 
2. Unlike BOD discharge concentrations at Jollyville and Barton Ridge, COD 
concentrations tend to be higher than at the other sites, which is somewhat surprising 
since the two tests are generally considered to measure the same characteristic.  
 
 
 
 
 
84 
 
12 Zinc Performance 
 
The statistical distribution of the data at the individual sites is presented in Table 24. The 
distributions are not clearly either normal or lognormal. Figure 89 presents the influent and 
effluent probability plots for the pooled paired data from all the sites. The difference between the 
two distributions is substantial and neither are statistically distinguishable from a lognormal 
distribution. 
Table 24 Statistical Distribution of Zinc Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie 
Oaks 
Highwood Jollyville All sites 
Influent Lognormal/normal Lognormal/normal Lognormal/normal Lognormal/normal Lognormal Lognormal 
Effluent Lognormal Lognormal/normal Lognormal/normal Lognormal/normal Lognormal/ normal Lognormal 
 
1 0 0 01 0 01 01
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
T o t a l  Z n  ( u g / L )
P
e
r
c
e
n
t
4 . 0 7 1 0 . 7 7 3 0 5 1 0 . 4 3 1 0 . 2 9 5
2 . 8 5 2 0 . 7 0 6 7 5 1 0 . 2 6 9 0 . 6 6 8
L o c S c a l e N A D P
Z n  I n
Z n  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  Z n  I n ,  Z n  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 89 Zn Influent and Effluent Probability Plots 
Boxplots of Zn influent and effluent concentrations for the five sand filters are presented in 
Figure 90 and Figure 91 respectively. Statistical analysis (ANOVA) indicates that the influent 
concentrations are significantly different (p = 0.016), with the highest values being observed at 
 
 
85 
 
Jollyville. The discharge concentrations, however, are not statistically different (p = 0.566), 
which is what was observed for TSS.  
J o l l y v i l l eH I g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 0 0 0
8 0 0
6 0 0
4 0 0
2 0 0
0
B M P
Z
n
 
I
n
f
lu
e
n
t
 
(
u
g
/
L
)
 
Figure 90 Boxplot of Zinc Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
2 0 0
1 5 0
1 0 0
5 0
0
B M P
Z
n
 
E
f
f
lu
e
n
t
 
(
u
g
/
L
)
 
Figure 91 Boxplot of Zn Discharge Concentrations 
 
 
86 
 
Monitoring in the Austin area indicates that zinc in runoff is predominantly associated with 
particles, with the dissolved portion accounting for roughly 25% of the total (Barrett and Stanard, 
2008). Complicating the analysis of Zn removal in this dataset is the fact that only the 
concentrations for total Zn were measured. If there are changes in Zn removal from event to 
event, it could be due to two very different causes ? chemical reactions in the filter media or 
differences in the influent partitioning between the dissolved and particulate fraction. If the 
discharge from an event has a large concentration it could be the result of a particularly high 
proportion of dissolved Zn in the influent or the lack of time or other factors that would not allow 
the reaction to reach equilibrium. This cannot be resolved with these data. 
Since the majority of zinc is associated with the particulate phase, one would expect that the 
removal would be somewhat less than that observed for TSS, but still particle dominated unless 
there was substantial removal of the dissolved component through adsorption, precipitation, or 
complexation. Table 25 presents average influent and effluent zinc concentrations for the five 
sites. Like many other constituents the sites with the highest influent concentrations tend to have 
the higher efficiency ratio. The particularly poor performance at Barton Ridge will be discussed 
subsequently. 
Table 25 Zinc Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Paired t-test 
Barton Mall 127 17 87 0.002 
Barton Ridge 58 38 35 0.289 
Brodie Oaks 45 24 46 0.375 
Highwood 27 15 45 0.039 
Jollyville 107 21 81 <0.000 
All Sites 84 22  74 <0.000  
 
Figure 92 presents the relationship between influent and effluent concentrations for Zn. Unlike 
the graph for TSS, there is little increase in discharge concentrations at higher influent 
concentrations. It is interesting to note that Barton Ridge consistently has some of the higher 
discharge concentrations, although not significantly higher. A likely explanation is the amount of 
galvanized material used in the construction of the facility. As shown in Figure 93, there are 
galvanized grates, equipment boxes, hand rails, and access ladders that are potential sources of 
zinc within the BMP itself. In a study conducted for TxDOT (unpublished) samples of rainfall 
dripping from a galvanized bridge rail were collected and concentrations of Zn ranged from 9480 
?g/L to 3260 ?g/L; consequently, the construction materials at this site are likely responsible for 
the somewhat elevated concentrations observed. The effect of these materials should be 
 
 
87 
 
somewhat muted, since they only contribute zinc during the storm itself and not during the 
subsequent days required to fully drain the facility. 
 
Figure 92 Relationship between Zn Influent and Effluent Concentrations 
 
Figure 93 Photograph of the Barton Ridge Sand Filter 
 
 
88 
 
The time series of Zn discharge concentrations observed at the Jollyville are presented in Figure 
94. The first flush effect that was so dominant for TSS is much more muted for zinc. A working 
hypothesis based on the results from the TSS analysis is that zinc is not associated as strongly 
with the particles trapped in the underdrain. This could be because it is not associated strongly 
with that particle size class or that the particles zinc is associated with are less dense and not as 
prone to settling in the underdrain. One major source of zinc in stormwater runoff is rubber from 
tire wear, since zinc oxide constitutes approximately 1 - 2% of tires by mass (Shaheen and Boyd, 
1975). Consequently, the zinc may be attached more strongly to rubber particles that are less 
dense and therefore do not settle as readily or that are somewhat larger than can pass through the 
filter. 
A comparison can also be made of the removal of Zn that occurs in the sedimentation basin 
relative to the filtration basin at Barton Ridge. This comparison is strongly affected by two 
EMCs that seem to be far higher than any others observed. These measurements were for Events 
19940808A (single aliquot) and 19950907A (two aliquots). Eliminating these two events from 
consideration, the comparison indicates that effectively all of the Zn removal occurs in the 
sedimentation basin. The mean concentration in the discharge from the sand filter is actually 
slightly higher than the sedimentation basin discharge, although the difference is not significant. 
Consequently, we can conclude that the particulate fraction of zinc is associated with larger, 
denser particles that are effectively removed by gravity separation. Overall, removal was about 
50 percent for Zn. This suggests that the reason for the lack of a first flush phenomenon is the 
result of Zn being primarily associated with the particle size fraction that is too large to pass 
through the filter and collect in the underdrain system. 
Figure 95 presents a graph of zinc effluent concentrations versus maximum event discharge rate 
for the Brodie Oaks facility, which is a surrogate for loading rate. It is apparent from the figure 
that some of the lowest discharge concentrations were produced during events with the greatest 
water depth (resulting in the highest discharge rate), so loading rate does not seem to be a 
significant design factor for zinc removal. 
 
 
 
 
89 
 
0
50
100
150
200
250
300
0 6 12 18 24 30 36 42 48
Z
n
 
E
f
f
l
u
e
n
t
 
(u
g
/
L
)
T i me  S i n c e  F i r s t  S a mp l e  (H o u r s )
5/29/1988
6/2/1988
4/19/1989
5/5/1989
5/10/1989
6/11/1989
6/13/1989
8/1/1989
8/15/1989
10/7/1989
11/22/1989
2/18/1990
3/14/1990
4/26/1990
6/4/1990
1/12/1995
3/13/1995
4/4/1995
5/8/1995
5/18/1995
6/11/1995
7/30/1995
11/17/1995
4/22/1996
4/25/1997
5/9/1997
 
Figure 94 Jollyville Zn Time Series Discharge Concentrations
 
 
90 
 
 
Figure 95 Relationship (or not) between Zn Discharge Concentration and Discharge Rate 
HRT would be expected to player a more prominent role in zinc removal, since some fraction is 
in the dissolved phase and removal might be expected to improve with increased time for 
precipitation, complexation, or adsorption. Figure 96 presents a comparison of discharge 
concentrations and HRT; however, the relationships between those two variables or between 
HRT and removal efficiency are not statistically significant. 
 
Figure 96 Relationship between HRT and Zn Discharge Concentration (Jollyville) 
 
 
91 
 
Zinc Conclusions: 
1. Zinc discharge concentrations are relatively constant throughout the duration of an event 
and do not exhibit the first flush effect evident for TSS.  
2. Zinc discharge concentrations are independent of influent concentrations and average 
about 20 ?g/L. 
3. Zinc fixtures within a BMP result in a noticeably, but not significantly, higher discharge 
concentration. 
4. Discharge concentrations are not correlated with hydraulic loading rate. 
5. Removal efficiency and effluent concentrations are not significantly correlated with 
hydraulic residence time. 
 
 
 
92 
 
13 Copper Performance 
 
The statistical distribution of the copper concentrations for the individual sites are presented in 
Table 26. Cumulative probability plots were prepared using the paired data from all the sites. 
These are presented in Figure 97. The distributions are distinctly different, which reinforces the 
conclusion that significant removal does occur in sand filters, and they are lognormally 
distributed. 
Table 26 Statistical Distribution of Copper Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie 
Oaks 
Highwood Jollyville All 
sites 
Influent Lognormal Normal/lognormal Lognormal/normal Lognormal/normal Lognormal/ normal Lognormal 
Effluent Lognormal/normal ? Lognormal/normal Lognormal/normal Lognormal Lognormal 
 
1 0 01 01
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
T o t a l  C u  ( u g / L )
P
e
r
c
e
n
t
1 . 9 7 5 0 . 9 4 1 2 4 7 0 . 6 1 1 0 . 1 0 6
1 . 5 1 2 0 . 7 8 1 1 4 7 0 . 2 9 9 0 . 5 7 2
L o c S c a l e N A D P
T C u  I n
T C u  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  T C u  I n ,  T C u  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 97 Probability Plots of Total Copper Influent and Effluent 
Boxplots of total copper concentration for the five sites are presented in Figure 98 and Figure 99, 
respectively. The influent concentrations are distinctly different (p = 0.001); however, the 
analysis indicates that the effluent concentrations are only moderately different (p = 0.107). 
 
 
93 
 
Interestingly, if only the storms are used for which paired data are available the ANOVA 
indicates almost no difference at all (p = 0.936), which indicates that all the sites produce very 
similar effluent concentrations.  
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
1 0 0
8 0
6 0
4 0
2 0
0
B M P
C
u
 
I
n
f
lu
e
n
t
 
(
u
g
/
L
)
B o x p l o t  o f  E M C
 
Figure 98 Boxplot of Cu Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
3 5
3 0
2 5
2 0
1 5
1 0
5
0
B M P
C
u
 
E
f
f
lu
e
n
t
 
(
u
g
/
L
)
 
Figure 99 Boxplot of Copper Effluent Concentrations 
 
 
94 
 
 
Mean influent and effluent copper concentrations for the five sand filters are presented in Table 
27. In general, there is only modest removal, but the influent concentrations are very low. It 
appears that the average discharge concentration is approximately 6 ?g/L, which is very similar 
to the influent concentrations at many of the sites.  
Table 27 Copper Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Wilcoxon SRT 
Barton Mall 6.5 5.1 22 0.5078 
Barton Ridge 7.3 6.2 15 1.000 
Brodie Oaks 6.0 5.0 17 0.625 
Highwood 7.8 6.7 14 1.000 
Jollyville 15.3 6.3 59 <0.000 
All Sites 10.4 6.1 42 <0.000 
 
A regression analysis was performed on the paired data for all the sites to determine if there is a 
significant relationship between influent and effluent concentrations. This relationship was 
determined to be statistically significant (p < 0.000). The regression was also significant for all 
the individual sites except Barton Ridge, even though Jollyville was the only facility with 
statistically significant removal. These regressions are presented in Figure 101 through Figure 
104.In addition, HRT was also considered as a predictor for the Jollyville data; however, it was 
not significant (0.377). 
 
Figure 100 Relationship between Total Copper Influent and Effluent Concentrations 
 
 
95 
 
 
 
Figure 101 Relationship between Total Cu Concentrations Barton Mall 
 
Figure 102 Relationship between Total Cu Concentrations Jollyville 
 
 
96 
 
 
Figure 103 Relationship between Total Cu Concentrations Highwood 
 
Figure 104 Relationship between Total Cu Concentrations Brodie Oaks 
The temporal pattern of copper discharge concentrations at Jollyville, which are presented in 
Figure 105, are very similar to those observed for zinc. There is a subdued first flush effect, but 
most events have relatively constant discharge concentrations. 
 
 
97 
 
 
0
5
10
15
20
25
30
35
40
45
0 . 0 0 6 . 0 0 1 2 . 0 0 1 8 . 0 0 2 4 . 0 0 3 0 . 0 0 3 6 . 0 0 4 2 . 0 0 4 8 . 0 0
C
u 
(
?
g
/
L
)
T i m e  S i nc e  Fi r s t  S a m pl e  ( hour s )
5 / 2 0 / 8 8
5 / 2 9 / 8 8
6 / 1 / 8 8
6 / 2 / 8 8
8 / 1 1 / 8 8
4 / 1 9 / 8 9
5 / 5 / 8 9
5 / 1 0 / 8 9
5 / 1 3 / 8 9
6 / 1 1 / 8 9
6 / 1 3 / 8 9
8 / 1 / 8 9
8 / 1 5 / 8 9
1 0 / 7 / 8 9
1 1 / 2 2 / 8 9
2 / 1 8 / 9 0
1 / 0 / 0 0
4 / 2 6 / 9 0
6 / 4 / 9 0
1 / 1 2 / 9 5
3 / 1 3 / 9 5
4 / 4 / 9 5
5 / 8 / 9 5
5 / 1 8 / 9 5
6 / 1 1 / 9 5
7 / 3 0 / 9 5
1 1 / 1 7 / 9 5
4 / 2 2 / 9 6
4 / 2 5 / 9 7
4 / 2 6 / 9 7
5 / 9 / 9 7
 
 
Figure 105 Temporal Pattern of Copper Discharge Concentrations at Jollyville 
 
 
98 
 
Copper Conclusions: 
 
1. Little copper reduction was observed at most of the sand filters; however, average 
influent concentrations were very low. 
2. Behavior of copper appears to be very similar to zinc. 
 
 
 
99 
 
14 Lead Performance (Pb) 
 
Table 28 presents the results of the analysis of statistical distribution for each of the sites 
individually. The observed data are both normally and lognormally distributed. Cumulative 
probability plots for influent and effluent lead concentrations are presented in Figure 106. The 
distributions are very distinct, which supports the finding that significant reductions in 
concentration occur in sand filters. 
Table 28 Statistical Distribution of Lead Data for Each Site 
 Barton Mall Barton 
Ridge 
Brodie 
Oaks 
Highwood Jollyville All sites 
Influent Normal Lognormal/normal Lognormal/normal Lognormal/normal Normal ? 
Effluent Lognormal lognormal Lognormal/normal ? Lognormal/ normal Lognormal 
 
1 0 0 01 0 01 01
9 9
9 5
9 0
8 0
7 0
6 0
5 0
4 0
3 0
2 0
1 0
5
1
D a t a
P
e
r
c
e
n
t
2 . 5 3 3 1 . 1 1 3 5 2 1 . 1 3 7 0 . 0 0 5
1 . 0 6 5 0 . 6 4 4 7 5 2 0 . 6 1 7 0 . 1 0 3
L o c S c a l e N A D P
T P b  i n
T P b  o u t
V a r i a b l e
P r o b a b i l i t y  P l o t  o f  T P b  i n ,  T P b  o u t
L o g n o r m a l  -  9 5 %  C I
 
Figure 106 Probability Plots of Total Pb Influent and Effluent Concentrations 
Boxplots of influent and effluent concentrations are presented in Figure 107 and Figure 108, 
respectively. ANOVA indicates that influent concentrations are not significantly different (p = 
 
 
100 
 
0.149), while effluent concentrations are somewhat different (p = 0.087). Using only paired data 
the effluent concentrations are not statistically different (p = 0.188). 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
5 0 0
4 0 0
3 0 0
2 0 0
1 0 0
0
B M P
P
b
 
I
n
f
lu
e
n
t
 
(
u
g
/
L
)
B o x p l o t  o f  E M C
 
Figure 107 Boxplot of Lead Influent Concentrations 
J o l l y v i l l eH i g h w o o dB r o d i e  O a k sB a r t o n  R i d g eB a r t o n  M a l l
3 5
3 0
2 5
2 0
1 5
1 0
5
0
B M P
P
b
 
E
f
f
lu
e
n
t
 
(
u
g
/
L
)
 
Figure 108 Boxplot of Effluent Lead Concentrations 
 
 
101 
 
Lead in stormwater runoff is predominantly associated with the solid phase so its removal should 
be similar to what is observed for TSS. The average concentrations and efficiency ratios are 
presented in Table 29, and it is clear that removal is substantial at all sites, although not always 
statistically significant. 
Table 29 Lead Concentrations for Individual Sites based on Paired Data 
Site EMC in EMC out Efficiency Ratio Paired t-test 
Barton Mall 33 4.8 86 0.002 
Barton Ridge 11.8 2.8 76 0.070 
Brodie Oaks 8.1 1.9 76 0.125 
Highwood  7.5 2.9 61 0.215 
Jollyville 27 4 85 <0.000 
All Sites 20 3.6 82 <0.000 
 
A regression analysis was also performed to determine the effect of influent concentration on 
discharge quality. The results are presented in Figure 109, and they show little effect of influent 
concentrations. This is very similar to what was observed for TSS, which confirms our 
understanding that very little of the lead is in the dissolved form. On an individual basis, only 
Highwood showed a statistically significant relationship between influent and effluent and this 
relationship is presented in Figure 110. 
 
Figure 109 Relationship between Pb Influent and Effluent Concentrations all Sites 
 
 
102 
 
 
Figure 110 Relationship between Pb Influent and Effluent Concentrations Highwood 
 
The temporal pattern of lead discharge concentrations at Jollyville are presented in Figure 111. 
There is little evidence of a first flush phenomenon in contrast to what was observed for copper.  
 
 
 
103 
 
0
5
10
15
20
25
30
35
40
45
0 . 0 0 6 . 0 0 1 2 . 0 0 1 8 . 0 0 2 4 . 0 0 3 0 . 0 0 3 6 . 0 0 4 2 . 0 0 4 8 . 0 0
L
e
a
d 
(
?
g
/
L
)
T i m e  s i nc e  f i r s t  s a m pl e  (hou r s )
5 / 2 0 / 8 8
5 / 2 9 / 8 8
6 / 1 / 8 8
6 / 2 / 8 8
8 / 1 1 / 8 8
4 / 1 9 / 8 9
5 / 5 / 8 9
5 / 1 0 / 8 9
5 / 1 3 / 8 9
6 / 1 1 / 8 9
6 / 1 3 / 8 9
8 / 1 / 8 9
8 / 1 5 / 8 9
1 0 / 7 / 8 9
1 1 / 2 2 / 8 9
2 / 1 8 / 9 0
1 / 0 / 0 0
4 / 2 6 / 9 0
6 / 4 / 9 0
1 / 1 3 / 9 5
3 / 1 3 / 9 5
4 / 4 / 9 5
5 / 8 / 9 5
5 / 1 8 / 9 5
6 / 1 1 / 9 5
7 / 3 0 / 9 5
1 1 / 1 7 / 9 5
4 / 2 2 / 9 6
4 / 2 5 / 9 7
4 / 2 6 / 9 7
5 / 9 / 9 7
 
Figure 111 Temporal Pattern of Lead Effluent Concentrations at Jollyville
 
 
104 
 
15 Overall Conclusions 
 
1. Discharge concentrations for TSS, TP,TKN, Zn, Cu, and Pb were similar at all facilities, 
so design factors such as pretreatment, maximum water depth, and filter area apparently 
have little effect on pollutant removal.  
2. Discharge concentrations for fecal coliform, fecal strep, and COD were correlated with 
influent concentrations, so differed between the sites. This suggests that a substantial 
amount of these materials are associated with the finest particle fraction (or dissolved in 
the case of COD) that can pass through the filter media. 
3. Nitrate + Nitrite concentrations were significantly lower at Highwood, which might be 
related to the extremely small filter media volume that provided less opportunity for 
nitrification and nitrate export. 
4. BOD discharge concentrations were lower at Barton Ridge and Jollyville, but given the 
similar performance at all of the sites for the other constituents it seems likely the 
differences were due to influent characteristics.  
5. Pollutant removal was not a function of time, indicating that the accumulation of material 
on and within the filter had little impact on pollutant removal. 
6. Most of the constituents had a distinct first flush that might be attributed to the 
accumulation of sediment and associated pollutants in the underdrain system at the end of 
storm events. The exception was nitrate, which had a first flush that was correlated with 
the time since the last event, indicating nitrification was occurring in the filter.  
7. In general, removal efficiency and discharge concentration were not consistently related 
to the hydraulic residence time. Consequently reaction kinetics did not appear to be a 
limiting factor in pollutant removal. One caveat is that the calculated residence time is 
strongly affected by the influent and effluent volumes. In many cases, substantial mass 
balance errors likely resulted in poor estimates of HRT. 
8. Pretreatment reduces the total sediment load to the filter by about 65-70%, but may not 
material extend the life of the filter since much of this sediment likely is fairly coarse, 
which would result in little loss of permeability if it accumulated on the surface of the 
filter. 
 
 
 
105 
 
16 References 
 
Barrett, Michael and Stanard, Christina, 2008, Effects of the Permeable Friction Course (PFC) 
on Highway Runoff, in the proceedings of the 11th International Conference on Urban 
Drainage, Edinburgh, Scotland, September 1-5, 2008. 
Barrett, Michael E., 2003, ?Performance, Cost and Maintenance Requirements of Austin Sand 
Filters, Journal of Water Resources Planning and Management, Vol. 129, No. 3, pp. 234-
242. 
Karamalegos, A.,  Barrett, M.E. ,Lawler, D. F., Malina,  J. F. Jr., 2005, Particle Size Distribution 
of Highway Runoff and Modification Through Stormwater Treatment, CRWR Online 
Report 2005-10, University of Texas at Austin. 
Shaheen, J.D., and Boyd, G.B., 1975, Contributions of urban roadway usage to water pollution: 
U.S. Environmental Protection Agency Final Report EPA 600/2-75-004, 358 p. 
Yao, Habibian, and O?Melia, 1971, Water and Waste Water Filtration: Concepts and 
Applications, Environmental Science and Technology, Vol. 5, pp. 1105-1112.