Performance evaluation of a full-scale natural treatment

Environ Monit Assess (2009) 157:391–406DOI 10.1007/s10661-008-0544-7

Performance evaluation of a full-scale naturaltreatment system for nonpoint source and pointsource pollution removal

Chi-Feng Chen · Jen-Yang Lin ·Chih-Hong Huang · Way-Ling Chen ·Nai-Ling Chueh

Received: 14 March 2008 / Accepted: 11 September 2008 / Published online: 9 October 2008© Springer Science + Business Media B.V. 2008

Abstract This study presents a full-scale per-formance of a natural treatment system (NTS)facility in Taiwan with nearly 2 years of ob-servations. The study site, composed of severaltreatment ponds in series, was designed primarilyto reduce polluted stormwater runoff from teagardens and partially to untreated domestic waste-water from nearby villages. Thus, both nonpointsource and point source pollution are treated inthis system. From 28 field samplings in 2006–2007,the NTS site shows satisfactory treatment perfor-mance and the effluent water quality is signifi-cantly improved. Seven of the 28 sampling eventsare storm events (nonpoint source pollution) andthe remainder are from regular monitoring(point source pollution). The average volume of

C.-F. ChenGraduate Institute of Environmental Engineering,National Taiwan University, 71 Chou-Shan Road,Taipei, 106, Taiwan

J.-Y. Lin (B) · W.-L. Chen · N.-L. ChuehDepartment of Civil Engineering and WaterEnvironment Research Center, National TaipeiUniversity of Technology, 1, Sec. 3,Chung-Hsiao E. Rd., Taipei, Taiwane-mail: [email protected]

C.-H. HuangGraduate Institute of Architecture and Urban Design,National Taipei University of Technology,1, Sec. 3, Chung-Hsiao E. Rd., Taipei, Taiwan

influent and effluent is 533 CMD and 196 CMD,respectively. In order to determine the removalefficiency, several assessment measures are em-ployed in an attempt to obtain unbiased con-clusions. They are removal rate (RR), efficiencyrate (ER), summation of loads (SOL), flux rate(FR), and effluent probability method (EPM).The average percent removal efficiency of NH3-Nis 53.5–75.2% and of TP is 59.0–84.7%, in whichthe highest result is calculated by SOL methodand the lowest rate is obtained from RR. In FRevaluation, larger treatment capacity for NH3-Nthan for TP is provided in the site and the averageFR is respectively 0.230 g/m2 day and 0.017 g/m2

day. Of the methods examined, EPM is the onlymethod capable of illustrating data distribution.Finally, recommendations on the usefulness ofthese measures are summarized to facilitate theunderstandings of NTS performance evaluations.

Keywords Natural treatment systems (NTS) ·Performance assessment · Stormwater runoff ·Removal efficiency

Introduction

Natural Treatment System (NTS) is an alterna-tive onsite treatment using natural attenuationability to purifying polluted flow. Natural treat-ment system, with advantages of cost effectivity

392 Environ Monit Assess (2009) 157:391–406

and minimum operation requirement, is widelyapplied in wastewater treatment and is consid-ered as the main technology in best managementpractices (BMP) for nonpoint source pollution-control (USEPA 2004). The NTS is not limited to-specific types; instead, many environment-friendlytreatment systems utilizing natural energy and-biodegradation as domain mechanisms can be-classified as a NTS facility. Leverenz et al. (2002)summarized several types of natural onsite treat-ment system including surface flow constructedwetlands, subsurface flow constructed wetlands,ecological systems, evapotranspiration systems,and lagoons. Among these NTS types, constructedwetlands (CW) have been widely applied as anefficient treatment system for wastewater, espe-cially in developing countries (Kivaisi 2001). Theflexibility of NTS application permits the treat-ment of various types of wastewater (Rousseauet al. 2004), such as sewage treatment plant efflu-ent (Toet et al. 2005), industrial wastewater (Chenet al. 2006), stormwater runoff (Carleton et al.2000), and domestic wastewater (Belmont et al.2004). Recently, a vertical flow CW was provedto remove pharmaceuticals and personal careproducts (Matamoros et al. 2007).

Feitsui Reservoir supplies drinking water foralmost five millions people in Taipei City (thecapital city) and Taipei County and is the mostimportant water source in Taiwan. Accordingto the water quality monitoring annual reportof Taipei Feitsui Reservoir Administration, theCarlson Trophic Status Index (CTSI) indicatedthat the water quality in Feitsui Reservoir is reach-ing near-eutrophic state. Several impact factorscontribute to the eutrophication, including exces-sive amounts of nutrient input and heavy rain-falls in upstream watersheds. Non-point sources(NPS) are concluded as the dominant pollu-tion source and require appropriate BMPs (Linand Hsieh 2003). Responding to this require-ment, several pollutions control strategies havebeen implemented, including the establishmentof NTSs to reduce polluted runoff and to con-trol alga propagation (Chou et al. 2007). TheNTS facility constructed in DuNan site is an

exceptions it treats both nonpoint source andpoint source pollution. Not only the storm runofffrom tea gardens is treated in DuNan NTS systembut also the domestic wastewater from nearbyvillages is directed into the system. The systemwas composed of four treatment units, includingfree water surface wetlands and subsurface flowwetlands. After nearly 2 years of monitoring, theDuNan site shows satisfactory performance onpollution removal and the water quality in FeitsuiReservoir is improved.

The improved water quality in effluent is eas-ily observed by sampling analysis; however, it isdifficult to conclude a deterministic evaluation forthe system performance. Many assessment mea-sures are suggested to evaluate constructed BMPsand NTS. For example, USEPA (1999) reportedEvent Mean Concentration (EMC) or Summationof Loading (SOL) is proper to assess removal rateof suspended solid, total phosphorous, and ammo-nia nitrogen. USEPA (2002) recommended thatEffluent Probability Method (EPM) and Box Plotto analyze sampling data since they are capableof showing extreme events and distribution types.Chen (2006) compared the methods of RemovalRate (RR) and Efficiency Ratio (ER) and con-cluded that the variability of stormwater runoffmight affect the assessment results. In order toevaluate the performance of the DuNan site, themeasures, ER, RR, SOL, and EPM are employedto give a whole view of the pollution removal effi-ciency. Besides, a new assessment measure withflux idea is proposed, named Flux Rate (FR).The FR method considers exposure pathway thattreats pollutants not considered in other assess-ment methods. The general calculation and com-parison of the measures are demonstrated in thisstudy to facilitate the efficiency assessment.

In “Material and methodology”, the dimen-sion of the case study structure is demonstratedand the approaches of efficiency assessment arepresented. The sampling observations of the siteand the evaluation of pollution removal rate arediscussed in “Results and discussions”. Finally,a brief summary of this study is concluded in“Conclusions”.

Environ Monit Assess (2009) 157:391–406 393

Material and methodology

Site conditions and features

The study site is located in Feitsui Reservoir wa-tershed to investigate the control of the reser-voir eutrophication. The two major land use inFeitsui Reservoir watershed are forest (85.9%)and agricultural (6.6%) use by the tea indus-try (EPA of ROC 2005). Thus, the main non-point source is from tea gardens surrounding thewatershed. This study site, nestled near DuNanBridge, was constructed to purify water quality ofPashin Creek that flows into the Feitsui Reser-voir (Fig. 1). The study site is therefore calledDuNan site in this study. The DuNan site wasdesigned to treat storm runoff from tea gardens.In addition, small quantity of domestic wastewaterfrom neighboring villages is also collected into thisNTS facility. Because both nonpoint source andpoint source pollution are treated in the same site,the sampling scheme was scheduled separately forstorm events and regular monitorings. Although

two influents flowed in, only one effluent is sam-pled for the overall evaluations. Figure 2 depictsthe layout of the NTS facility and a photo ofthe site.

At the beginning, the site was particularly de-signed for the storm water from nonpoint source.However, the storm water is the only influent anda satisfied water level in treatment ponds cannotbe maintained when no rain falls. Moreover, theinfiltration in the site is very strong. Without otherinflow the site consequently dried up in the mosttime. Therefore, new water source was needed.The untreated domestic wastewater from neigh-bor village or Pashin Creek water was consideredto supplement the water quantity. But the quan-tity of wastewater is too small to increase waterlevel in either retention ponds or eco-pond. Thus,sufficient water quantity of upstream PashinCreek was pumped into the treatment system. Thewater quality in upstream Pashin Creek does notbe contaminated by nonpoint sources and couldalso serve as clean water to dilute polluted stormrunoff.

Fig. 1 The location ofDuNan site and FeitsuiReservoir


Fig. 2 (a) Layout of theDuNan NTS site and(b) the field photo

(a)

(b)

Effluent

Domestic wastewater Runoff from tea gardens

Pumping water from Pashin CreekRetention pond (A1)

Limestone filter bed

Gravel filter bed

Vegetative buffer zone

Retention pond (A2)Eco-pond

Pumping Pashin Creek water as an additionalwater source was disagreed by some experts. Theysuggested that no new water flow was needed.The water level in ponds could be controlled byproper device, such as a control valve. Even nowater in the most time, the storm water from teagardens in rainy day is still able to be retainedand reduced. The water pumping from PashinCreek is not automatically and continually andthe additional maintenance and operation of thepump and pipe are required. After the arguments,the administration agency finally decided to pumpwater and to rebuild a new ecological system.

The drainage area of the DuNan site is 3 ha cov-ered mostly by tea gardens. The design flow rate inthis CW system is 600 CMD, collecting occasionalstormwater, domestic wastewater, and PashinCreek pumping water. The site is composed of

five treatment units, including two retentionponds, one eco-pond, and two filter beds (Fig. 2).Two retention ponds are settled to retain stormwater and to treat polluted water with naturaltreatment mechanism, such as oxidization, set-tling, and filtration. Two serial retention pondsare expected to offer sufficient time for treatmentprocess. The retention ponds provide mostly phys-ical treatment mechanism; meanwhile, the eco-pond is placed to strengthen chemical treatment,where many local plants are cultivated for pollu-tion absorption. Due to the plants hinder waterflow in the eco-pond, the small flow rate mightcause insect growing. The full discussions on re-moval mechanism can be found in literatures, suchas Urbonas (1994) and USEPA (1999).The detailunit dimensions are listed in Table 1. The totaltreatment area is 2,953 m2 and the total hydraulic


Table 1 The design dimensions of the DuNan site

Treatment Length Width Surface Water Storage Hydraulic retention Hydraulic loadingsunits (m) (m) area (m2) depth (m) volume (m3) time (day) (m3/m2day)

Retention 32 20 640 1.2 659.59 1.10 0.94pond (A2)

Retention 25 24 600 1.2 621.30 1.04 1.00pond (A1)

Eco-pond 95 15 1,425 1.4 1,812.95 3.02 0.42Limestone 12 12 144 0.4 23.04 0.04 4.17

filter bedGravel filter 12 12 144 0.4 23.04 0.04 4.17

bedTotal – – 2,953 3,140 5.24

retention time is 5.24 days. Local vegetations arecultivated, including over 30 aquatic species andten terrestrial species and the total amount is over5,000.

The monitoring plan began at February, 2006.Two influents and one effluent are sampled. Theinfluent from domestic wastewater and pumpingwater from Pashin Creek are monitored monthlyas dry weather flow. The storm water is sampledwhen cumulative rainfall is over 30 mm within a24 h interval. The rainfall data was used based onthe official records, which was announced immedi-ately by administration agency on internet. Oncethe rainfall reached the sampling standard, wesampled the two influents and one effluence in siteand analyzed them in laboratory. The samplingswere not made by automatic machine. We wentto the field to sample water and measure the flow.In addition to the storm events, we measured thewater quality monthly in regular time.

The water flow rate was measured in differentways. The pumping water flow from Pashin Creekwas measured by the pump rate and the volumeof domestic wastewater and tea gardens runoffwas measured by weirs. The flow rate was calcu-lated from the dimension of weir and the waterlevel of overflow. In monthly regular monitoring,the water sample and water flow measure wasmade once each time because the flow is stable.However, more than one sample was measuredin storm events and the sample interval is about10 to 20 min. Water quality of pH, temperature,dissolved oxygen (DO), chemical oxygen demand(COD), suspended solid (SS), ammonia nitrogen(NH3-N), and total phosphorous (TP) are ana-

lyzed. The flow rate is calculated by containerestimation according to administration method(NIEA W020.51C).

Assessment methods of NTS performances

To give an unbiased assessment, five evaluationapproaches are employed to evaluate the pollu-tion removal performance. They are removal rate(RR), efficient ratio (ER), summation of loads(SOL), flux rate (FR), and effluent probabilitymethod (EPM). The brief descriptions of thesemeasures are summarized as follows.

Removal rate

Pollution concentration removal rate is the mostcommonly-used measure. Removal Rate (RR)compares directly water concentration in inlet andoutlet samplings. The expression of RR is as Eq. 1,where Cin and Cout is the concentration of influentand effluent, respectively.

RR (%) =(

Cin − Cout

Cin

)× 100% (1)

Efficiency ratio

Efficiency Ratio (ER) is similar to RR, but theconcentration is replaced by the Event Mean Con-centration (EMC; Eq. 2). EMC is the ratio of thetotal pollution mass to total runoff volume. Ina single stormwater event, the average pollutionconcentration is used in the computation. Distri-bution of pollution concentration is depended on


storm intensity and duration time; thus, the EMCis usually used to describe the average pollutionconcentration for stormwater events.

ER (%) =(

average EMCin − average EMCout

average EMCin

)

×100% (2)

where EMCin is EMC of influent and EMCout is ofeffluent.

EMC =

n∑i=1

CiVi

n∑i=1

Vi

(3)

where, Ci is the average pollution concentration ina sampling period, Vi refers to the runoff volumein duration time i, and n is the sampling numbers.

Summation of loads (SOL)

SOL is also a percent measure, but unlike RRand ER, SOL uses pollution mass change insteadof concentration. The expression of SOL is asEq. 4. The detailed assumptions and commentsabout SOL method was given in the technicalmemorandum of ASCE and USEPA (1999).

SOL (%) =(

Lin − Lout

Lin

)× 100% (4)

where

L =n∑

i=1

CiVi

Flux rate

The notion of pool(s) is a simplified assumptionto describe transition of a substance in organismor nonorganism compartments. For example, inaquatic system, phosphorous is distributed in ei-ther waterbody or plankton. The waterbody andplankton can be regarded as pools of phospho-rous and the transition of phosphorous can becalculated between pools. The transition quantityis denoted as flux rate, the number of substance

transisting from one pool to others by unit area orvolume in particular time period. The flux rate isalso used for the mass balance between systems.Unlike RR and ER, the traditional methods forevaluating efficiency of pollution removal alwaysfocus on the change of pollution concentration (ormass) but ignore the life time inside the treatmentsystem. The total treatment area and retentiontime are impact factors of pollution removal butare not considered in conventional approaches. Itis inappropriate to compare the performance of asmall treatment system to a large one. Flux rateconsidering exposure path of pollutants providesa new measure to evaluate efficiency assessmentfor BMPs.

flux rate = Cin Qin − Cout Qout

A(5)

where A is the lateral area of the treatment system

Effluent probability method

Effluent Probability Method (EPM) incorporatingprobability distribution is recommended to assessBMP performances (ASCE and USEPA 1999;USEPA 2004). The use of the EPM is preferredover other methods as it measures BMP efficiencywith various pollutant concentrations and rainfallintensities (Strecker et al. 2000). All influent andeffluent samplings are ranked according to pol-lution concentration and depicted in a cumula-tive probability graph. Thus, two parallel curvesof cumulative probability and the treatment effi-ciency of CWs are revealed. A log transformationof both influent and effluent EMCs are neededto elicit the probability graph since log normaldistribution is generally better for water qualitysamples. The distance between the two parallelcurves indicates the level of removal rate and dif-ferent removal efficiency are found for differentinfluent concentrations. The distributions of waterquality of samplings have to be verified by statisti-cal test to ensure the rationality of distribution as-sumption. Chi-Square statistic and Kolmogorov–Smirnov test was suggested by Burton and Pitt(2001) for EPM methods.


Results and discussions

Water quality in DuNan site

The water quality monitoring plan started at Feb-ruary 2006 and is still ongoing. A total of 28 eventswere sampled, including seven stormwater eventsand 21 regular events (dry weather). The averageinflow rate is 533 CMD and the outflow rate is 196CMD. The loss of influent volume might be infil-trated into soil or be retained in treatment ponds.Infiltration is one of the major removal mecha-nisms in natural treatment system. Infiltration canaffect both particle transport and dissolved runoff;especially for high removal efficiency of particlepollutants. The performance of infiltration is sig-nificantly relied on texture of soil. A finer soil caneven result in a better removal efficiency of dis-solved pollutants. In addition, hydraulic diffusionand chemical diffusion might control the dissolvedpollutants in infiltration (Wiertz and Marinkovic2005). In DuNan site, the nutrient pollutions areconcerned. Due to the dissolved forms of nu-trient pollutants are less than particle ones, thedissolved pollutants did not analyzed. However,infiltration can help to remove not only parti-cle but also dissolved pollutants. For example, inKrutz et al. (2003) study, more infiltration volumeturned out more trapping efficiency of dissolvedpollutant (atrazine). Therefore, infiltration func-tion was taken place in the DuNan site and waslikely benefit to its performance. The flow rateof influent and effluent at the study site is shownin Fig. 3. The variability in the influent flow rate

occurs because (1) part of the influent was fromthe pumping of water from Pashin Creek and thepumping time varies from day to day, and (2) theoccurrence of the occasional storm. At the initialperiod, the difference of inflow and outflow rateis not significant. After September 2006, due toevaporation and high infiltration, the effluent wa-ter quantity is significantly less than the influent.In addition, the pipe block and hydraulic retentionmight also affect the variability.

The water quality is summarized in Table 2.The average concentrations of NH3-N, TP, COD,and SS of storm water are 2.30, 0.14, 20.91, and25.89 mg/L, respectively. The average water qual-ity of the influent in dry weather (0.25, 0.06,20.52, and 7.59 mg/L), is much better than thatof stormwater flow, especially in terms of nutri-ent pollutant and SS. The concentration of NH3-N and TP in stormwater (as nonpoint source) isalmost ten times and two times higher than in dryweather (as point source). This finding suggeststhat the runoff through tea gardens carried morenutrients and might cause eutrophication in re-ceiving waterbody if not treated. The concentra-tion of SS is higher in storm runoff because of soilerosion. The COD concentrations are similar instormwater and dry weather flow.

Box-and-Whisker Plot (Box Plot) displayingdata distribution and extreme value in the samegraph is commonly used for data analysis (Tukey1977). The Box Plot of DuNan site is shown inFig. 4. The NH3-N and TP concentration in efflu-ent are lower than in influent, implying that theremoval of nutrient is efficient in DuNan site. The

Fig. 3 The flow rate ofinfluent and effluent inDuNan site

0

100

200

300

400

500

600

700

800

900

2006/2/16 2006/4/17 2006/11/21 2007/1/25 2007/3/6 2007/4/17 2007/6/5 2007/7/10

Flo

w (

CM

D)

Influent Effluent


Tab

le2

Res

ults

ofw

ater

qual

ity

sam

plin

gan

dR

R

Eve

nts

Sam

plin

gda

teN

H3-N

(mg/

L)

TP

(mg/

L)

CO

D(m

g/L

)SS

(mg/

L)

(mm

/dd/

yy)

Influ

ent

Effl

uent

RR

(%)

Influ

ent

Effl

uent

RR

(%)

Influ

ent

Effl

uent

RR

(%)

Influ

ent

Effl

uent

RR

(%)

Stor

mw

ater

04/0

6/20

061.

180.

4462

.70.

160.

0382

.04.

08.

7/

13.8

03.

5074

.6(n

onpo

ints

ourc

e)04

/10/

2006

13.9

05.

0863

.50.

470.

0393

.011

.72.

876

.13.

003.

000.

003

/05/

2007

0.31

0.24

22.6

0.01

0.01

0.0

26.8

10.0

62.7

48.6

56.

0087

.703

/06/

2007

0.34

0.20

41.2

0.02

0.15

/15

.05.

066

.745

.36

23.0

049

.304

/10/

2007

0.14

0.09

35.7

0.01

0.01

0.0

33.2

10.0

69.9

<1.

00.

00/

06/0

5/20

070.

110.

0646

.50.

030.

0240

.90

––

–2.

510.

8068

.208

/07/

2007

0.09

0.03

68.1

0.25

0.05

81.8

34.7

810

.00

71.2

42.0

29.

0078

.6A

vera

gea

2.30

0.88

48.6

10.

140.

0474

.43

20.9

17.

7569

.33

25.8

96.

4759

.73

Dry

wea

ther

02/1

6/20

060.

260.

1734

.60.

100.

0637

.59.

83.

366

.36.

804.

5033

.8(p

oint

sour

ce)

03/2

8/20

060.

200.

0765

.00.

110.

0554

.32.

87.

2/

10.8

025

.50

/04

/14/

2006

0.40

0.16

60.0

0.06

0.04

35.5

4.2

11.2

/2.

003.

00/

04/2

0/20

060.

740.

0889

.20.

080.

0358

.72.

86.

4/

4.00

2.00

50.0

09/1

2/20

060.

110.

0918

.20.

050.

0420

.050

.720

.060

.63.

392.

0041

.010

/03/

2006

0.23

0.08

65.2

0.03

0.02

33.3

10.0

3.0

70.0

0.00

0.00

0.0

11/0

7/20

060.

270.

60/

0.03

0.01

66.7

60.7

50.0

17.6

<1.

0<

1.0

/12

/05/

2006

0.10

0.05

50.0

0.01

0.01

0.0

3.1

20.0

/<

1.0

<1.

0/

01/0

9/20

070.

170.

22/

0.02

0.01

50.0

73.8

50.0

32.2

<1.

0<

1.0

/01

/11/

2007

0.28

0.01

96.4

0.23

0.01

94.3

12.1

4.8

60.3

<1.

0<

1.0

/01

/25/

2007

0.18

0.01

94.4

0.05

0.02

71.2

2.8

2.8

0.0

<1.

0<

1.0

/02

/06/

2007

0.11

0.06

45.5

0.01

0.01

0.0

17.4

10.0

42.5

<1.

0<

1.0

/02

/26/

2007

0.42

0.04

90.5

0.07

0.01

87.3

8.8

8.4

4.5

1.5

<1.

0/

05/0

2/20

070.

260.

0580

.80.

180.

1043

.77.

916

.2/

<1.

0<

1.0

/05

/09/

2007

0.23

0.08

65.2

0.04

0.02

50.0

14.4

20.0

/0.

003.

33/

05/1

5/20

070.

110.

0462

.90.

030.

0160

.9–

––

–<

1.0

/06

/12/

2007

0.18

0.05

72.6

0.03

0.02

44.1

23.1

220

.00

13.5

2.51

2.50

0.5

06/2

0/20

070.

170.

0852

.10.

030.

0214

.9–

––

––

–07

/03/

2007

0.11

0.04

62.9

0.03

0.02

54.2

––

–<

1.0

–/

07/1

0/20

07–

––

0.03

0.02

26.2

––

–31

.20

2.00

93.6

09/0

4/20

07–

––

0.03

0.01

60.5

––

––

––

Ave

rage

a0.

250.

1358

.27

0.06

0.03

43.6

020

.52

15.1

445

.60

7.59

5.19

36.4

8

/ne

gati

veva

lue

isge

nera

ted,

−no

sam

ple

data

a The

nega

tive

valu

e(s)

isex

clud

edin

RR

calc

ulat

ion


0

4

1216

Con

cent

ratio

n (m

g/L)

Ammonium nitrogen Box-whisker Plot

Influent Effluent

0

0.1

0.2

0.3

0.4

0.5

Con

cent

ratio

n (m

g/L)

Phosphorus Box-whisker Plot

Influent Effluent

(a) NH3-N (b) TP

0

20

40

60

80

Con

cent

ratio

n (m

g/L)

COD Box-whisker Plot

Influent Effluent

0

10

20

30

40

50C

once

ntra

tion

(mg/

L)SS Box-whisker Plot

Influent Effluent

(c) COD (d) SS

Fig. 4 a–d The Box Plot of sampling water quality in DuNan site

data in influent has a wider range than effluent,especially for COD and SS. Positive skewness isobserved in influent of TP, COD and SS. Severaloutliers are appeared in raw data, especially in thenutrient matters. In these calculations, a negativevalue is regarded as an outlier and should beexcluded. Negative values indicate the pollutantsconcentration in effluent is higher than that ininfluent. It might be caused by several rationalreasons, such as the effects from lateral flows, thedisturbance of fishes, or errors in experimental

analysis. However, these values would signifi-cantly influence the average value in calculation.To the reason, negative values are treated as sta-tistic outlier and are not considered in the evalua-tion methods.

Assessment results of pollution removalefficiency

The performance efficiency of DuNan site is eval-uated with five measures. Three measures, RR,


ER, and SOL present percent removal assess-ment. Flux rate (FR) assesses site treatment ca-pacity and EPM helps to explain data distribution.

Removal rate

The removal rate (RR) is calculated from thewater quality in influent and effluent. The aver-age RR of NH3-N, TP, COD, and SS are 48.6%,74.4%, 69.3%, and 59.7% in storm events. Butthere is high variation of daily RR values causedby system operation and maintenance. For ex-ample, RR of NH3-N increased from 34.6% atFebruary 2006 to 89.2% in April 2006 when reg-ular maintenance occurred. However, as no main-tenance occurred during April to September, theRR at September dropped to 18.2%. After con-tinuous and effective maintenance, the rate in-creased to 96.4% in January 2007. The RR resultsof the system are drastically influenced by thequality of operation and maintenance. The otherexample is that many fish are put into ecopoolat early March and lead to confounding TP re-moval, the concentration in effluent is higher thanthat in influent at 6th, March, 2007. The purposeof putting fishes in to pond is to prevent mos-quitoes larvae from thriving. Although the fishesdisturb the stable water and brought out somesediment in the initial period, the overall treat-ment environment became better and the qualityof treated water was satisfied when the system wasrevived. Additionally, the influent concentrationsof pollutants may influence the RR results. Forexample, the RR of COD could be as much as to70% if the influent concentration is in the rangeof 10–50 mg/L. Otherwise, negative RR mightappeared when the influent concentration is lessthan 10 mg/L. Higher influent concentration ofNH3-N and TP leads to higher RR values.

The negative RR value implies that treatmentfacility produces more contamination instead ofpollution removal and the irrational data is ex-cluded in performance assessment. The unex-pected value might be caused by the time lag insampling, implying that the batch of effluent beingsampled is different from the “same” batch ofinfluent water. The major pitfall of RR methodis not considering the hydroretention time. Al-though RR is the most straightforward way to

assess the performance efficiency of any watertreatment system, it might produce misleadingresults such as the negative removal rates.

Efficiency ratio

In the ER method, flow data is required in orderto obtain EMC. The EMC is not properly for as-sessing regular samples since the influent flow rateis represented by only one dry weather flow unlikethe distributed flow curve is occurred in stormwa-ter events. Thus, the discussion of ER is confinedto the site performance in storm events. The ERof NH3-N, TP, COD, and SS are 62.4%, 83.4%,68.0%, and 70.9%, respectively. Comparing withthe average RR value (excluding negative ones),the results of ER are higher than the averageRR values with the exception of COD pollutant(Table 3). Since the COD concentration is similarin storm and dry weather events, the results of ERand RR are not much variant.

Summation of loads

The SOL is evaluated from the total summation ofinfluent mass and effluent mass. Table 4 displaysthe example of the calculation process of SOL.The average SOL of NH3-N, TP, COD, and SS are75.2%, 84.7%, 81.4%, and 88.8%. The summaryof percent removal efficiency of DuNan site isgiven in Table 5. The removal efficiency of TP isalways higher than NH3-N in storm events; on thecontrary, the removal rate of TP is less than ofNH3-N in dry weather events. Regarding to stormevents, the removal efficiency of NH3-N is 48.6%-64.8% and that of TP is 74.4–90.8%. The results ofSOL give the highest removal rate and the lowestrate is obtained from RR in both NH3-N and TP.The different measures lead to about 16% for theefficiency results.

Table 3 ER for storm events and the comparison with RR

Storm NH3-N (%) TP (%) COD (%) SS (%)events

ER 62.4 83.4 68.0 70.9RR 48.6 74.4 69.3 59.7


Table 4 The example of SOL calculation (NH3-N)

Events Sampling date Qi(m3/d) Qe(m3/d) Ci(mg/L) Ce(mg/L) Mi(g) Me(g) SOL (%)

Stormwater 2006/4/6 612.36 612.36 1.180 0.440 722.58 269.44(nonpoint source) 2006/4/10 629.58 629.58 13.900 5.080 8,751.16 3,198.27

2007/3/5 531.8 27.08 0.310 0.240 164.85 6.502007/3/6 543.5 91.14 0.340 0.200 184.80 18.232007/4/10 467.4 26.58 0.140 0.090 65.44 2.392007/6/5 625.0 426.00 0.112 0.060 70.14 25.562007/8/7 420.0 86.40 0.094 0.030 39.44 2.59

Total – – – – 9,998.41 3,522.98 64.76Dry weather 2006/2/16 608.57 608.57 0.260 0.170 158.23 103.46

(point source) 2006/3/28 611.42 611.42 0.200 0.070 122.28 42.802006/4/17 198.52 198.52 0.400 0.160 79.41 31.762006/4/20 192.31 192.31 0.740 0.080 142.31 15.382006/9/12 559.3 183.80 0.110 0.090 61.52 16.542006/10/3 587.4 118.69 0.230 0.080 135.10 9.502006/12/5 399.0 107.13 0.100 0.050 39.90 5.362007/1/11 576.0 60.40 0.280 0.010 161.28 0.602007/1/25 516.0 78.00 0.180 0.010 92.88 0.782007/2/6 462.8 86.00 0.110 0.060 50.91 5.162007/2/26 437.0 86.00 0.420 0.040 183.54 3.442007/3/20 641.0 95.00 0.211 0.040 135.28 3.802007/4/17 627.0 104.00 0.237 0.040 148.32 4.162007/5/2 653.0 102.00 0.260 0.050 169.78 5.102007/5/9 370.9 105.88 0.230 0.080 85.31 8.472007/5/15 654.0 85.00 0.108 0.040 70.50 3.402007/6/12 733.0 147.52 0.183 0.050 133.89 7.382007/6/20 650.0 548.00 0.167 0.080 108.50 43.842007/7/3 675.0 53.00 0.108 0.040 72.76 2.12

Total – – – – 2,151.71 313.05 85.45

Flux rate

The concept of pool system is applied to the CWsite in flux rate calculations. Flux rate has theadvantages of revealing the mass balance overtime scale. Instead of using surface loading rate,flux rate considers the entire exposure path ofpollutants inside the treatment system. Larger FR

value means more pollutants are being retained insystems and less likely to flow out. The FR alsogives information about the time required for pu-rification. The average FR value of NH3-N and TPin storm events is 0.433 0.025 g/m2 day, respec-tively. The FR value is relatively small in normalday events. The value is 0.026 g/m2 day for NH3-Nand 0.008 g/m2 day for TP. The intercomparison of

Table 5 The average removal efficiency of NH3-N and TP in DuNan site

Event NH3-N TPRR (%) ER (%) SOL (%) EPMa(%) RR (%) ER (%) SOL (%) EPMa(%)

Storm 48.6 62.4 64.8 – 74.4 83.4 90.8 –Dry weather 58.3 – 85.5 – 43.6 – 78.6 –Average 53.5 – 75.2 65.2 59.0 – 84.7 45.6aThe percent rate is calculated by the cumulative probability of 50% in EPM diagram


pollutant types shows that the treatment amountof NH3-N is much higher than TP as there wasmore NH3-N initial concentration. However, theFR values do not provide insights into the treat-ment efficiency of each pollutant type. The FRonly gives the information about the treatment ca-pacity of NTSs and is beneficial for site design butdoes not provide a relatively comparison betweenpollutants removal.

Effluent probability method

As shown in Fig. 5, the EPM graphical figureprovides probability information of efficiency interms of pollution concentration. The method isbased on the assumption that the rank of pollutionconcentration in influent is consistent with that ineffluent and a parallel curve is then formed. TheEPM results (excluded outlier samples) indicate

0.01 0.1 1 10Concentration(mg/L)

1

2

5

10

20

30

40

50

60

70

80

90

95

98

99

Pro

babi

lity(

%)

InfluentEffluentAmmonium nitrogen Probability Plot

0.001 0.01 0.1 1Concentration(mg/L)

1

2

5

10

20

30

40

50

60

70

80

90

95

98

99

Pro

babi

lity(

%)

InfluentEffluentPhosphorus Probability Plot

(a) NH3-N (b) TP

1 10 100Concentration(mg/L)

2

5

10

20

30

40

50

60

70

80

90

95

98

Pro

babi

lity(

%)

InfluentEffluent

COD Probability Plot

1 10 100Concentration(mg/L)

5

10

20

30

40

50

60

70

80

90

95

Pro

babi

lity(

%)

InfluentEffluent

SS Probability Plot

(c) COD (d) SS

Fig. 5 a–d The EPM of water quality observed in DuNan site (excluding outlier value)


Table 6 Comparisons ofthe efficiency assessmentapproach

Measures Flow data PS/NPS application Infiltration effect Illustration ofneeded data distribution

RR No Both Low NoER Yes NPS only Low NoSOL Yes Both High NoFR Yes Both High NoEPM No Both Low Yes

that the four pollutants were removed efficientlyfrom the study site. The effluent concentrationsare all lower than influent concentrations. TheNH3-N plot shows better removal efficiency forinfluent concentrations ranging from 0.1 to 1 mg/Land decreasing removal efficiency with increasinginfluent concentration (Fig. 5a). The removal effi-ciency of TP pollution is relatively consistent. Theremoval rates of TP in low influent concentrationsare less than those in high concentrations. Thesame phenomenon is observed in SS data as well.In addition, the average percent removal rate canbe calculated from the cumulative probability of50%. For example, the influent and effluent ofNH3-N is respectively 0.23 and 0.08 mg/L at cumu-lative probability of 50% and the removal rate is65.2%, which can be used as representative value.Similarly, the removal rate of TP is 45.6% in EPMresult. Table 5 gives the summary of the averageremoval efficiency results.

Summary of the use of the different measures

Many approaches can be used to assess pollutionremoval efficiency. However, there is no “cook-book” approach to the use of these measures.

Selecting methods is often intuitively and arbitrar-ily, depending on limited available information.To facilitate the understandings of these mea-sures, their properties are summarized in Table 6.The properties are categorized into four terms,i.e., flow data, PS/NPS application, infiltration ef-fect, and illustration of data distribution. Flowdata indicates the necessary of flow data in thesite. For example, RR is calculated only by thewater quality of influent and effluent; therefore,flow data is not needed. PS/NPS application standsfor the application for either PS or NPS. All thesemeasures can be used in both PS and NPS removalassessment with the exception of ER, which isonly for NPS. Infiltration serves as one of theremoval mechanisms in NTS facility. Thus, theperformance assessment would be affected signif-icantly by the level of field infiltration, especiallyfor those measures accounting for the infiltrationinto calculation process, such as SOL and FR.Among the five measures, only EPM with graph-ical illustration is able to show data distributioninstead of an average deterministic rate. Withthe four indicators, based on available site in-formation, users can easily select the appropriate(or multiple) measure to implement performanceassessment.

Fig. 6 Removal mass ofNH3-N and TP inmonitoring period

0

50

100

150

200

250

300

2006/2/16 2006/4/17 2006/11/21 2007/1/25 2007/3/6 2007/4/17 2007/6/5 2007/7/10

TP

rem

oval

mas

s (g

/day

)

0

1000

2000

3000

4000

5000

6000

NH

3-N

rem

oval

mas

s (g

/day

)

TPNH3-N


Fig. 7 Water quality inFeitsui Reservoir inrecent 3 years

0

10

20

30

40

50

60

01/2005 06/2006 11/2005 04/2006 09/2006 02/2007 07/2007 12/2007time (mm/yy)

CT

SI

0

10

20

30

40

50

60

TP(

µg/L

)

CTSI TP

Water quality improvement in Feitsui Reservoir

Based on the actual observations in DuNan site,the removal nutrient mass is shown in Fig. 6.The scale on the left column of Fig. 6 shows theremoval mass for TP and the right scale for NH3-N. The average removal mass of NH3-N and TPin the monitoring period is 281.0 and 39.1 g/day,respectively. In other words, the average an-nual removal pollutions are estimated as 103 and14.3 kg/year for NH3-N and TP, respectively. Inthe first month, the RR of TP is 37.5% but in-creases to 46.5% because of regular maintenance.However, the system performance reduced to38.2% after 11 months. Due to the lack of goodmaintenance, the removal efficiency is unstablebetween the third to the 11th month. Once theDuNan site is reconditioned at the 12th month,the RR of TP recovered to 47.4% and the phos-phorous removal quantity increased from 16.57 to25.21 g/day.

Figure 7 shows the change of TP and CarlsonIndex (CTSI) of Feitsui Reservoir from the year of2005 to 2007. The eutrophication level improvedgradually throughout the 30 months. The averageof CTSI is from 45.18 in 2005 to 41.51 in 2007and the average of TP is from 24.34 to 14.64 μg/Lthrough the 3 years. According to Chou et al.(2007), the average annual TP loading into theFeitsui Reservoir is 18,910 kg/year. It implies that0.07% of total TP is removed from the DuNansystem.

The improvement of water quality in FeitsuiReservoir is definitely contributed by various ef-forts, such as expanding the sewage system and

best management practice for nonpoint sourcepollution control. The surface area of DuNan siteis about 0.3 ha (2,953 m2) and is not possibleto remove significant pollutants from the wholewatershed (the total area is 303 km2). Based onthe sampling data, the treatment efficiency pertreatment area is estimated about 343.3 kg/ha peryear of NH3-N and 47.7 kg/ha per year of TP.Therefore, if all TP pollution is desired to beremoved from Feitsui Reservoir by NTS systems,the treatment areas need about 400 ha. However,application of NTS is not the only way to controlpollutions. Limited by the available site and thecost, structural and non-structural BMP should beconsidered simultaneously to improve reservoirwater quality.

Conclusions

This study demonstrates a full-scale natural treat-ment system with a 2-years of monitoring. Thesystem was designed primarily to purify pol-luted stormwater runoff from tea gardens and toease the increasing eutrophication in the FeitsuiReservoir. In addition to nonpoint source pollu-tion, domestic wastewater is also treated in thesite. From field observation, the study site demon-strated nice treatment ability in either nonpointsource or point source pollution. However, dueto the different types of treatment ponds (withvarying dimensions), types of vegetations, influ-ent sources, storm conditions, etc. incorporatedinto this particular NTS, it is difficult to evaluate


its performance efficiency. Different measuresapplied to a single site would lead to varyingconclusions. In order to obtain an unbiased assess-ment, five measures are applied in this study. Theyare removal rate (RR), efficiency ratio (ER), sum-mation of loads (SOL), flux rate (FR), and effluentprobability method (EPM). The RR and ER areconcentration percent removal approach utilizingdirect pollution concentration and event meanconcentration (EMC). The SOL is a mass percentremoval measure. The FR is a new measure ex-plored in this study to consider exposure path ofpollutants throughout the treatment system, andEPM is strongly recommended for BMP evalua-tion by USEPA (2002). The different measuresresult in difference by as much as 25% (averageTP results).

This study is not aimed to rank the preferabil-ity of the measures but to provide insight aboutthe use of these assessments, including the cal-culation, data input, sampling time, and the finalexpression. The selection of efficiency measuresare basically dependant on the site conditionsand project requirements. The percent removal ofRR is the simplest measure with minimum datarequirement. If a complete set of flow data isavailable, ER and SOL provide more objectivelyresults. If the mass retaining ability is considered,the FR provides supplement information for sitedesign or retrofit. However, regardless of the mea-sure used, regular maintenance is key factor for aNTS performance. A good operation and mainte-nance schedule would be very helpful to extendthe lifetime and the treatment capacity of a NTSfacility. The guidelines of a good maintenance andinspection can be found in USEPA website, e.g.the USEPA Stormwater Control Operation andMaintenance database (USEPA 2006).

References

ASCE and USEPA (1999). Development of performancemeasures task 3.1—Technical memorandum: Deter-mining urban stormwater best management practice(BMP) removal efficiencies.

Belmont, M. A., Cantellano, E., Thompson, S., Williamson,M., Sanchez, A., & Metcalfe, C. D. (2004). Treat-ment of domestic wastewater in a pilot-scale natural

treatment system in central Mexico. EcologicalEngineering, 23, 299–311.

Burton, A. G., & Pitt, R. (2001). Stomwater effects hand-book: A toolbox for watershed managers, scientists,and engineers. Boca Raton: Lewis.

Carleton, J. N., Grizzard, T. J., Godrej, A. N., Post, H.E., Lampe, L., & Kenel, P. P. (2000). Performanceof a constructed wetland in treating urban stormwa-ter runoff. Water Environment Research, 72(3), 295–304.

Chen, W. L. (2006). Study on the assessment of pollutionremoval efficiency of BMPs by different statistic meth-ods. Master Thesis of Institute of Environmental Engi-neering and Management, National Taipei Universityof Technology, Taipei, Taiwan. (in Chinese).

Chen, T. Y., Kao, C. M., Yeh, T. Y., Chien, H. Y., & Chao,A. C. (2006). Application of a constructed wetland forindustrial wastewater treatment: A pilot-scale study.Chemosphere, 64, 497–502.

Chou, W. S., Lee, T. C., Lin, J. Y., & Yu, S. L. (2007).Phosphorus load reduction goals for Feitsui ReservoirWatershed, Taiwan. Environmental Monitoring andAssessment, 131, 395–408.

EPA of Republic of China (2005). The program for reduc-tion of pollution sources in Fei-Tsuei Reservoir water-shed. Taipei, Taiwan. (in Chinese).

Kivaisi, A. K. (2001). The potential for constructed wet-lands for wastewater treatment and reuse in develop-ing countries: A review. Ecological Engineering, 16,545–560.

Krutz, L. J., Senseman, S. A., Dozier, M. C., Hoffman,D. W., & Tierney, D. P. (2003). Infiltration and ad-sorption of dissolved atrazine and atrazine metabolitesin buffalo grass filter strips. Journal of EnvironmentalQuality, 32, 2319–2324.

Leverenz, H., Tchobanoglous, G., & Darby, J. L. (2002).Review of technologies for the onsite treatment ofwastewater in California. California State WaterResources Control Board, Report No. 02-2.

Lin, J. Y., & Hsieh, C. D. (2003). A strategy for imple-menting BMPs for controlling nonpoint source pol-lution: The case of the Feitsui Reservoir Watershedin Taiwan. Journal of the American Water ResourcesAssociation, 39(2), 401–412.

Matamoros, V., Arias, C., Brix, H., & Bayona, J. M.(2007). Removal of pharmaceuticals and personal careproducts (PPCPs) from urban wastewater in a pi-lot vertical flow constructed wetland and a sand fil-ter. Environmental Science & Technology, 41, 8171–8177.

Rousseau, D. P. L., Vanrolleghem, P. A., & De Pauw, N.(2004). Model-based design of horizontal subsurfaceflow constructed treatment wetlands: A review. WaterResearch, 38, 1484–1493.

Strecker, E., Quigley, M. M., & Urbonas, B. R. (2000).Determining urban stormwater BMP effectiveness.In National conference on tools for urban waterresources management and protection proceedings.Chicago, Illinois. 7–10 Feb. 2000.

Toet, S., Van Logtestijn, R. S. P., Schreijer, M., Kampf, R.,& Verhoeven, J. T. A. (2005). The functioning of


a wetland system used for polishing effluent from asewage treatment plant. Ecological Engineering, 25,101–124.

Tukey, J. W. (1977). Exploratory data analysis, reading.Reading: Addison-Wesley.

Urbonas, B. (1994). Assessment of stormwater BMPsand their technology. Water Science and Technology,29(1–2), 347–53.

USEPA (1999). Preliminary data summary of urban stormwater best management practices. Washington DC,EPA-821-R-99-012.

USEPA (2002). Urban stormwater BMP performance mon-

itoring: Guidance manual for meeting the nationalstormwater BMP database requirements. WashingtonD.C., EPA-821-B-02-001.

USEPA (2004). Stormwater best management prac-tice design guide: Volume 1 general considerations.Washington D.C., EPA-600-R-04-121.

USEPA (2006). Stormwater control operation &maintenance database. http://www.epa.gov/owow/nps/ordinance/stormwater.htm.

Wiertz, J. V., & Marinkovic, F. A. (2005). Dissolvedpollutant transport in tailings ponds. EnvironmentalGeology, 47, 237–240.

http://www.epa.gov/owow/nps/ordinance/stormwater.htm

http://www.epa.gov/owow/nps/ordinance/stormwater.htm

Documents

Performance evaluation of a full-scale natural treatment