Modeling of constructed wetland performance in ...

Environ Monit Assess (2017) 189:164 DOI 10.1007/s10661-017-5857-y

Modeling of constructed wetland performance in BOD5 removal for domestic wastewater under changes in relative humidity using genetic programming Vanitha Sankararajan & Nampoothiri Neelakandhan & Sivapragasam Chandrasekaran

Received: 18 September 2016 / Accepted: 15 February 2017 # Springer International Publishing Switzerland 2017

Abstract Despite the extensive use of constructed wetland (CW) as an effective method for domestic wastewater treatment, there is lack of clarity in arriving at well-defined design guidelines. This is particularly due to the fact that the design of CW is dependent on many inter-connected parameters which interact in a complex manner. Consequently, different researchers in the past have tried to address different aspects of this complexity. In this study, an attempt is made to model the influence of relative humidity (RH) in the effectiveness of BOD5 removal. Since it is an accepted fact that plants respond to change in humidity, it is necessary to take this parameter into consideration particularly when the CW is to be designed involving changes in relative humidity over a shorter time horizon (say a couple of months). This study reveals that BOD5out depends on the ratio of BOD5in and relative humidity. An attempt is also made to model the outlet BOD5 using genetic programming with inlet BOD 5 and relative humidity as input parameters.

Keywords Constructed wetland . Relative humidity . BOD5 removal . Genetic programming . Apparent temperature V. Sankararajan (*) : N. Neelakandhan : S. Chandrasekaran Center for Water Technology, Department of Civil Engineering, Kalasalingam University, Krishnankovil, Srivilliputtur, Virudhunagar District, Tamil Nadu, India e-mail: [email protected]

Introduction Wastewater treatment by constructed wetlands (CWs) has gained popularity and is emerging as a potential alternative to conventional treatment methods such as activated-sludge systems (Shelef et al. 2013). The quality of treated water is judged based on the BOD5 value of the effluent. The amount of oxygen consumed by bacteria to decompose active organic matter into simple compounds is called biochemical oxygen demand (BOD). In standard test, a 300-ml BOD bottle is used, and sample is incubated at 20 °C for 5 days; hence, it is cited as BOD5. Light shall be excluded from the incubator to prevent algal growth that may produce oxygen in the bottle (Priya et al. 2012). Many studies confirmed that CWs effectively reduce BOD5 (Pan et al. 2012; Vymazal and Kropfelova 2011; Steer et al. 2002; Villar et al. 2012; Christopher et al. 2012; Calheiros et al. 2007; Mina et al. 2011; Rai et al. 2013). However, full scale applications are still limited due to a number of process related challenges such as (a) the complex interaction of plants, microorganisms, soil matrix, and substances in the wastewater; (b) changes in climatic conditions and its effect on pollutant removal efficiency of the CW system; and (c) the loading rate/retention time (Zhang et al. 2010; Haiming Wu et al. 2015, Stottmeister et al. 2003; Tomenkoa et al. 2007). As observed by Langergraber (2008), CWs can be considered to be like figurative black boxes which treat water. Though many experimental studies have been carried out, there is lack of consensus on influence of operational parameters, unit parameters, and meteorological

4.8 to15.6 m3/day

Chang et al. (2007) Only slight decreases were found in BOD removal when HLR increased from 200 to 1200 L/m2/day. 5

HSSF SF Four serially connected FWS Integrated vertical flow constructed wetland 4

7.5 m × 20 m 7.5 m × 20 m Each 6.5 m × 8m 9 m × 9 m (81 m2 each) and 1.3 m depth

6 wetland units with different plants Subsurface Horizontal flow CW 3

6.8 days 3.4 days

Four beds each having size of 20 m × 2 m 2

3 cm/day 6 cm/day

200 to 1200 L/m2/day

Ayaz (2008)

Calheiros et al. (2007)

Solano et al. (2004)

No significant BOD5 removal is observed when HLR is increased by 3 times. (i) Significant relationship between percentage of removal and hydraulic application rate. (ii) Longer retention time removal is higher. No significant BOD removal is observed when HLR is increased by 2 times. If hydraulic load increases, BOD removal decreases. 1 day 3 days 1.5 days 3 days 18 wetlands each 22 m2 1

21 l/min 61 l/min 3 m3/day 6 m3/day

Reference Observation HLR HRT Tank dimension S. No.

parameters on effective functioning of CWs. For instance, while Ayaz (2008) found that BOD removal is directly proportional to increase in hydraulic residence time (HRT), the works reported by Calheiros et al. (2007) indicate there is no significant direct correlation. The work of others (Trang et al. 2010, for instance) indicates that with increase in organic loading, there is a greater efficiency in BOD5 removal. Table 1 summarizes the wide variations reported in the past in selection of the size of CWs, the loading rates, retention time, etc. Similarly, Table 2 presents the wide variations adopted in the choice of plants, operating horizon of the CW, the frequency of sampling, and percentage of BOD removal. Villar et al. (2012) attempted vertical flow constructed wetland (VFCW) because it possesses greater oxygen transfer capacity. Similarly, size and shapes of constructed wetland also vary. Babatunde et al. (2010) adopted circular constructed wetland, whereas others report adopting rectangular constructed wetland (Zhang et al. 2010; Christopher et al. 2012; Solano et al. 2004), and yet others discussed only the area of constructed wetland without being specific about the shape (Villar et al. 2012; Mina et al. 2011). Studies have been conducted with operating horizon as short as 4 weeks to 1 year or more. BOD5 removal is also seen to vary from about 30 to over 90% with inlet BOD5 as low as 30 mg/l to as high as 1088 mg/l. The foregoing discussions clearly indicate the difficulties in arriving at well-defined guidelines on design and operation of CWs. This calls for more extensive studies to address the existing issues as well as issues that are yet to receive sufficient attention. Many studies concerning influence of weather on the removal of BOD5 in CWs have been carried out (Karathanasis et al. 2003, Steer et al. 2002, Solano et al. 2004). Karathanasis et al. (2003) conducted experiments with different macrophytes (single as well as polyculture) over a temperature range of almost 0 to 30 °C covering an entire year. The treatment efficiency is found to be the best during summer months while winter period showed poor performance. Changes in temperature also affect the microbial efficiency. Steer et al. (2002) observed a decrease in microbial activities during spring season. Most of the studies relating the influence of weather on treatment efficiency of CWs focused on variation in temperature. To the authors’ knowledge, there is no specific study on influence of other climatic parameters such as relative humidity (RH), sunshine, and wind speed. In this study, an

Bastviken et al. (2009)

Environ Monit Assess (2017) 189:164

Page 2 of 10

Table 1 Effect of HLR and HRT in BOD5 removal

164

Four stage novel CW

Subsurface flow CW

Subsurface flow CW

Subsurface horizontal flow CW

VFCW

3

4

5

6

7

2

Aerated planted CW Planted CW Aerated unplanted CW Non aerated unplanted HSSF

Area = 20 m2 Depth = 0.8 m

4 beds each 20 m × 2m 6 wetland units each Area = 1.2 m2 Depth = 0.6 m

0.59 × 0.33 × 0.37 m

Diameter-9.5 cm Depth = 0.6 m

Cattail Reed Canna indica Phragmites australis Stenophrum secundatum Iris pseudacorus Unvegetated control Scirpus alternifolios

Cyperus haspan

12 week

17 months

18 months

4 weeks

Phragmites austraulis 91 days

Phragmites austraulis 1 year

462 m2

1 year

Operation of CW

Not mentioned

Plant

3 m × 0.3 m ×1.2 m

Types of constructed wetland Dimension of CW

1

S. no.

Table 2 Frequency of sampling in different types of constructed wetland

Weekly study

Not mentioned

Monthly sample

10 no of sample

30 no of readings

Monthly sampling

Monthly 2 samples

Frequency of sampling

30

1000 ± 88

327

686

392.1

90 ± 33

285.2 ± 40.2

BOD value (mg/L)

84.9

93 in summer season 41–58

29.8–53.8

90.6 ± 7.5

61

Up to 94.4 in aerated planted filter

Villar et al. (2012)

Calheiros et al. (2007)

Solano et al. (2004)

Christopher et al. (2012)

Babatunde et al. (2010)

Mina et al. (2013)

Zhang et al. (2010)

BOD removal (%) References

Environ Monit Assess (2017) 189:164 Page 3 of 10 164

164

Page 4 of 10


attempt is made to study the influence of change in RH on BOD5 removal. In order to exclusively study the effect of RH change, the experimental period is so chosen that the temperature variation during the period is negligible. It is also desired to mathematically model the relationship between RH and outlet BOD5. Many works have been reported in the past on modeling of CW process using regression techniques including Artificial Intelligence (Brix 1994; Knight et al. 1993; Reed and Brown 1995; Vymazal 1988; Akratos et al. 2008; Yalcuk 2013, and Tomenkoa et al. 2007). In the last one decade, the success of evolutionary algorithm-based regression models in modeling complex natural processes has been demonstrated (Sivapragasam et al. 2015; Park et al. 2008; Sivapragasam et al. 2012). In this study, genetic programming (GP) is chosen as the modeling tool based on its ability to evolve physically more meaningful models when compared to other conventional regressions models.

to possess greater oxygen transport ability than horizontal subsurface flow beds (Pan et al. 2012, Villar et al. 2012). Consequently, for this study, a VFCW of size 2 m × 1 m × 1 m is constructed (Fig. 1). The treatment system consist of (from bottom to top): 30-cm-thick gravel bed (with about 10–20 cm size coarse gravel), about 50-cm-thick sand layer (particle size between 75 μ and 4.75 mm) and about 10-cm-thick local soil. Non woven geo-textile layer is placed between gravel layer and sand layer. The purpose of geo-textile layer is to prevent clogging of gravel layer with sand. Uniformity coefficient (Cu) for native soil and sand is found to be 2.8 and 3.94, respectively. Phragmites austraulis and Typha latifolia are particularly found to be effective in treating human and animal derived wastewater (Shutes 2001). Phragmites austraulis is taken for this study as it is one of the invasive plants in the study area and is planted at every 0.2 m distance in the CW system.

Study area description

Genetic programming

The study area is located at the foothills of the Western Ghats in Virudhunagar District of the State of Tamil Nadu, India (latitude 9.5733676 N and longitude 77.689401 E). Being situated in a semiarid region, the temperature in the study area remains more or less constant throughout the year. In order to better simulate the influence of RH alone, the study period is chosen from 20.09.2014 to 20.10.2014 during which temperature varies in the range of 32.8–35.8 °C while RH varies between 40 and 70%. The study area receives scanty rainfall with an average of 811 mm annually, the bulk of which is received during the North East monsoon in the months of October, November, and December. Since the time constants of certain microbial and physiochemical reactions range between seconds and hours Rousseau et al. (2002), a total of 11 samples are collected during the study period in order to better capture the effect of variations in RH.

Genetic programming (GP) is very similar to genetic algorithm (GA), being an evolutionary algorithm based on Darwinian theories of natural selection and survival of the fittest. However, GP operates on parse trees, rather than on bit strings as in a GA, to approximate the equation (in symbolic form) that best describes how the output relates to the input variables. The algorithm considers an initial population of randomly generated programs (equations), derived from the random combination of input variables, random numbers, and functions. The functions can include arithmetic operators (plus, minus, multiply, divide), mathematical functions (sin, cos, exp., log), logical/comparison functions (OR/ AND), etc., which have to be appropriately chosen based on some understanding of the process. This population of potential solutions is then subjected to an evolutionary process and the Bfitness^ (a measure of how well they solve the problem) of the evolved programs are evaluated. Individual programs that best fit the data are then selected from the initial population. The programs that best fit are selected to exchange part of the information between them to produce better programs through Bcrossover^ and Bmutation^, as used in GAs (to mimic the natural reproduction process). Here, exchanging the parts of best programs with each other is called crossover, copied exactly into the next

Experimental setup: pilot scale constructed wetland Generally, poor removals are attributed to microbial breakdown of carbonaceous compounds being limited by low oxygen availability (Solano et al. 2004). In treating many types of wastewater, a VFCW is reported


Page 5 of 10 164

Fig. 1 Schematic diagram of VFCW Primary Treated Sewage Local Soil

Sand Geotextile Membrane Gravel

generation is called reproduction and randomly changing programs to create new programs is called mutation (Koza 1992). The user must decide a number of GP parameters before applying the algorithm to model the data, such as population size, number of generations, crossover, and mutation probability. The programs that fitted the data less well are discarded. This evolution process is repeated over successive generations and is driven towards finding symbolic expressions describing the data, which can be scientifically interpreted to derive knowledge about the process being modeled (Sivapragasam et al. 2010). Discipulus tool is used to implement GP.

Methodology (a) Pumping of sewage: The domestic sewage after primary treatment is pumped from sewage treatment plant to the VFCW. (b) Sampling and analysis: The plants are allowed to grow from the month of August to the mid of September. Sampling and analysis are started after 40 days of plant growth (when a height of 62 cm is reached) and the study is completed when the plant height reached about 100 cm. Fig. 2 Block diagram for GP case study 1

Outlet

(c) BOD5 analysis: BOD5in and BOD5out have been measured in VFCW during experimental period. (d) Collection of meteorological parameters: Meteorological parameters such as RH and air temperature (T) are collected from the nearby meteorological station for the study period. In order to get a clear picture about the influence of RH, experiments are conducted with other operational parameters such as [hydraulic loading rate (HLR), HRT, wastewater temperature] kept almost constant. (e) Modeling studies: GP is applied for two different case studies (Figs. 2, and 3). The first case study uses BOD5in and RH as the inputs to the model while the second case study uses apparent temperature (Ta) and BOD5 in as the inputs. Ta is the perceived temperature equivalence caused by the combined effects of RH and T and can be estimated using standard charts and tables. Table 3 shows the training data used for modeling of BOD5out.The percentage of training, testing, and validation data in GP modeling is kept as about 50, 25, and 25%, respectively. (f) Training of GP: Mathematical functions such as addition, subtraction, multiplication, division, comparison, and trigonometric operators are used for evolving suitable mathematical model. Exponential function is not included in the modeling as the BOD

RH

VFCW

164


Page 6 of 10

Fig. 3 Block diagram for GP case study 2

Ta

VFCW

removal process is not expected to be exponentially dependent on the input parameters. It is to be noted that the selection of proper mathematical functions affect the quality of modeling.

Results and discussions Table 4 shows experimental results of CW. An analysis is presented below on the BOD5 removal efficiency with change in RH followed by modeling of BOD5. (a) Influence of relative humidity variation in BOD5 removal efficiency

Performance measure RMSE value is taken as the performance measure to check the performance of GP as shown in Eq. (1). sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 1 n RMSE ¼ ∑ ð X m Þ i −ð X s Þ i n i¼1

ð1Þ

where X is any variable that is being modeled; the subscripts m and s represent the observed and simulated values.

As seen from the Table 4, on an average, RH varies in the range 45 to 55% during the first phase of the study period whereas it increases to 55 to 65% during the next phase. This increase in RH with almost constant temperature is mainly because of precipitation during that period. A closer insight reveals that there are two distinct spell of almost constant humidity viz., during 22.09.14 to 25.09.14 with an average RH of 50% and during 10.10.14 to 14.10.14 with an average RH of 60% with a maximum of 65%. Clearly, there is a reduction in the BOD5 removal with this increase in RH. The spells of intermediate rise and fall in the RH value may not be given absolute significance due to the fact that the plant

Table 3 Training data used for BOD5out model Data

Case study 1 BOD5in (mg/l)

Training

Testing

Validation

Case study 2 RH (%)

BOD5out (mg/l)

BOD5in (mg/l)

Apparent temperature (Ta°C)

BOD5out (mg/l)

130

50

82.4

130

39.9

82.4

122

52

71.0

122

38.9

71.0

120

51

68.0

120

39.2

68.0

126

46

68.0

126

38.4

68.0

128

53

82.0

128

37.7

82.0

128

65

82.4

128

45.1

82.4

127

62

83.2

127

44.2

83.2

132

60

79.5

132

42.0

79.5

132

53

80.0

132

41.2

80.0

118

46

72.0

118

40.2

72.0

125

52

70.4

125

38.9

70.4


Page 7 of 10 164

Table 4 Experimental results of constructed wetland and its surrounding environment BOD5out (mg/L)

BOD5 removal Plant height (%) (cm)

Increase of plant height (%)

Relative humidity (%)

Temperature (°C)

20.09.2014 118

72.0

39.3

62. 0

1.6

46

35.3

22.09.2014 122

71.0

41.8

65. 0

6.6

52

33.7

3

24.09.2014 125

70.4

43.7

67. 0

9.8

52

33.7

4

25.09.2014 120

68.0

43.3

67. 0

9.8

51

34.0

5

30.09.2014 126

68.0

46.0

72. 0

18.0

46

34.1

6

08.10.2014 128

82.0

35.9

81. 0

32.8

53

32.8

7

10.10.2014 128

82.4

35.6

84. 0

37.7

65

36.0

8

11.10.2014 127

83.2

34.5

86. 0

41.0

62

35.8

9

14.10.2014 132

79.5

39.8

89. 0

45.9

60

34.7

10

18.10.2014 130

82.4

36.6

94. 0

54.1

50

34.6

11

20.10.2014 132

80.0

39.4

97. 0

59.0

53

35.1

S. no.

Date

1 2

BOD5in (mg/L)

metabolism takes some time to adjust to the variations in weather conditions. It may be argued that during these two spells of investigation, the inlet BOD5 also increased by 6 mg/L in the second spell when compared to the first spell (on an average), and hence, this variation may affect the result. However, it can be explained that the variation in inlet BOD5 is very small. Further, the reported works of the previous researchers (Mina et al. 2011; Trang et al. 2010) indicate that BOD5 removal is likely to increase with increase of inlet BOD5 concentration. This apparent contradiction can be attributed to change in RH influencing the BOD5 removal. The previously reported studies do not give any detail about the prevailing weather conditions. It has to be noted that the sampling was done after every few days. The RH values used in the study are the

average RH on the particular day of study. As mentioned in earlier, since the time constants of certain microbial and physiochemical reactions range between seconds and hours, it is assumed that the microbial and physiochemical reactions are complete within 24 h and the RH used in the study corresponds directly to the day in which BOD5in was measured. Hence, experimental observations due to time lag in the biological response to changes in RH will not introduce any significant error. As observed from Table 4, it can also be concluded that there is no correlation between the plant growth stage (in terms of plant height) and the BOD5 removal after the plant height has reached about half its total height. (b) Modeling of BOD5out with GP: case I Akratos et al. (2008) have critically examined the limitations of first-order model and proposed an ANN

Table 5 Modeling equation for different types constructed treatment wetlands according to different authors References

Types of CW

Input

Output

Equation

Brix (1994)

HSSF

Operational parameter (influent BOD)

Cout = (0.11 × Cin) + 1.87

Knight et al. (1993)

HSSF

Operational parameter (influent BOD)

Griffin et al. (1999)

HSSF

Vymazal (1988)

HSSF

Operational parameter (waste water temperature T) Operational parameter (influent BOD)

Akratos et al. (2008)

HSSF

Operational parameter (HRT, T)

Effluent BOD Effluent BOD Effluent BOD Effluent BOD Effluent BOD

Cin and Cout: influent and effluent concentrations (mg BOD/L)

Cout = (0.33 × Cin) + 1.4 Cout = 502.20 × exp.(−0.111 × T) Cout = (0.099 × Cin) + 3.24

RBOD ¼

HRT

ð22:8 T ÞþHRT

164


Page 8 of 10

response curve-based model mapping BOD5out to HRT and wastewater temperature. Table 5 summarizes the attempts made by previous researchers in developing BOD5out model as an alternative to first-order model. As seen from the Table 5, only a limited number of studies take weather factors into account in the modeling. In this study, GP is chosen in lieu of ANN due to its superiority, particularly in its ability to evolve mathematical models without pre-defined functional form unlike the conventional regression equations. BOD5out is modeled with inlet BOD5 and RH as inputs and is functionally represented as BOD5out ¼ f ðBOD5in ; RHÞ

exponential model, the exponential term does not appear in the equation. However, the presence of cosine function can be interpreted to represent the complex process of BOD5 removal in the presence of weather conditions. (c) Modeling of BOD5out with GP: case II Sometimes, the combined effect of temperature and RH is reflected by the term Bapparent temperature (Ta)^, which gives a better physical, feel of the climatic conditions. BOD5out is modeled with inlet BOD5in and Ta as inputs and is functionally represented as

ð2Þ

BOD5out ¼ f ðBOD5in ; Ta Þ

The GP is run with a population size of 500. The optimal cross over rate and mutation rate are arrived at as 95% and 0.02 respectively after trial and error. The final form of the evolved model is shown in Eq. (3). 2cos 0:38 BOD25 in BOD5out ¼ þ 0:61 *BOD5in ð3Þ RH

An attempt is further made to model BOD5out with this apparent temperature using GP. The best evolved GP model is as below: BOD5

where BOD5in = inlet BOD5 (mg/L), BOD5out = outlet BOD5 (mg/L), and RH = relative humidity (%). A comparison of measured BOD5out and that obtained from GP is given in Fig. 4 for the validation data. The RMSE is found to be 2.85 mg/l. As seen from Eq. (3), BOD5out is found to depend on the ratio BOD5in and RH. Apart from arithmetic operators, other functions such as exponential and trigonometric functions were also included in the GP modeling in order to allow full flexibility in the development of the model. Since the BOD5 removal process is not an Fig. 4 Comparison of predicted and observed BOD5out with RH

ð4Þ

out

¼

7:9cos½cosð1:75 T a Þ−BOD5 in − BOD5 in 1:67

ð5Þ

Figure 5 shows the Ta for the verification data set along with the predicted BOD5out. The RMSE for this model is found to be 1.8 mg/l which is less than that obtained from Eq. (3). The combined influence of RH and temperature gives the plant a feel of increased temperature which might be affecting the plant metabolism and the BOD5 removal process. Hence, it is recommended to include apparent temperature in modeling of BOD5out. It can be noted that the approach and technique adopted in this study can easily be replicated in any

90 80 70 60 50 Predicted BOD5out (mg/L)

40

Observed BOD5out (mg/L) 30 20 10 0 1

2

3

4

5

6

7

8

9

10

11

Environ Monit Assess (2017) 189:164 Fig. 5 Comparison of predicted and observed BOD5out with Ta

Page 9 of 10 164 90 80 70 60 50 40 Predicted BOD5out (mg/L)

30 20

Observed BOD5out (mg/L)

10 0 1

other study setting as described in the BMethodology^ section. However, since the weather conditions change significantly with the change in geographical locations and since BOD5 in will also vary significantly based on the quality of the raw sewage considered, GP will evolve different form of mathematical models than what has been given in this study (Eqs. (3) and (5)). In summary, it can be said that while the evolved mathematical models are site specific, the procedure to evolve the models is generic which can be adopted for any other study.

2

3

4

5

6

7

8

9

10

11

(d) GP seems to evolve meaningful models with small data sets for training. This can be explored more through further experimental studies. (e) The GP evolved models are all site specific and different models will be evolved for any change in input conditions such as geographical locations, weather conditions, and quality of the input sewage.

References Conclusions The treatment of wastewater in the CW system is difficult to understand due to complex physical, biological, and chemical processes operating simultaneously and influencing each other. The major conclusions from this study are summarized as follows: (a) BOD5 removal is found to be influenced by variation in the RH. It is found to decrease with the increase in RH while maintaining a steady temperature and fixed operational parameters. (b) In the BOD5out modeling, apparent temperature can be included in lieu of RH. Since the apparent temperature reflects the combined effect of RH and temperature, this can be a preferred choice. (c) CWs involve complex processes and in order to arrive at clear cut guidelines for the design, region specific studies need to be carried out to yield region specific guidelines.

Akratos, C. S., Papaspyros, J. N. E., & Tsihrintzis, V. A. (2008). An artificial neural network model and design equations for BOD and COD removal prediction in horizontal subsurface flow constructed wetlands. Chemical Engineering Journal, 143, 96–110. Ayaz, S. C. (2008). Post treatment and reuse of tertiary treated waste water by constructed wetlands. Desalination, 226, 249–255. Babatunde, A. O., Zhao, Y. Q., & Zhao, X. H. (2010). Alum sludge based constructed wetland system for enhanced removal of P and OM from waste water: concept, design and performance analysis. Bioresource Technology, 101, 6576–6579. Bastviken, S. K., Weisner, S. E. B., Thiere, G., Svensson, J. M., Ehde, P. M., & Tonderski, K. S. (2009). Effects of vegetation and hydraulic load on seasonal nitrate removal in treatment wetlands. Ecological Engineering, 35, 946–952. Brix, H. (1994). Constructed wetlands for municipal wastewater treatment in Europe. In W. J. Mitsch (Ed.), Global wetlands: old world & new (pp. 325–333). Amsterdam: Elsevier [Chapter 20]. Calheiros, C. S. C., Rangel, A. O. S. S., & Castro, P. M. L. (2007). Constructed wetland systems vegetated with different plants to the treatment of tannery waste water. Water Research, 41, 1790–1798.

164

Page 10 of 10

Chang, J., Zhang, X., Perfler, R., Xu, Q.-S., Niu, X.-Y., & Ge, Y. (2007). Effect of hydraulic loading rate on the removal efficiency in a constructed wetland in subtropical China. Fresenius Environmental Bulletin, 16(9), 1082–1086. Christopher, O., Akinbile, M. S., Yusoff, A. Z., & Ahmad, Z. (2012). Landfill leachate treatment using subsurface flow c o n s t r u c t e d w e t l a n d b y C y p e r u s h a s p an . Wa s t e Management, 32, 1387–1393. Griffin, D. M. J., Bhattarai, R. R., & Xiang, H. (1999). The effect of temperature of biochemical oxygen demand removal in a subsurface flow wetland. Water Environment Research, 71(4), 475–482. Karathanasis, A. D., Potter, C. L., & Coyne, M. S. (2003). Vegetation effects on fecal bacteria, BOD, and suspended solid removal in constructed wetlands treating domestic wastewater. Ecological Engineering, 20, 157–169. Knight, R.L., Ruble, R., Kadlec, R.H., & Reed, S. (1993). North American Treatment Wetland Database-electronic database created for the US Environmental Protection Agency. Koza, J. R. (1992). Genetic programming: on the programming of computers by natural selection. Cambridge, MA: MIT Press. Langergraber, G. (2008). Modeling of processes in subsurface flow constructed wetlands: a review. Vadose Zone Journal, 7(2), 830–842. Mina, A.P. I., Costa, M., Matos, A., & Sousa Coutinho, C. (2011). Polishing domestic water on a subsurface constructed wetland: organic matter removal and microbial monitoring. International Journal of Phytoremediation, 13, 947–958. Pan, J., Zhang, H., & Ke, W. L. F. (2012). Full scale experiment on domestic wastewater treatment by combining artificial aeration vertical and horizontal flow constructed wetland system. Water Air Soil Pollution, 223, 5673–5683. Park, S., Bae, H., & Kim, C. (2008). Decision model for coagulant dosage using genetic programming and multivariate statistical analysis for coagulation/flocculation at water treatment process. Korean Journal of Chemical Engineering, 25(6), 1372–1376. Villar, P. M., Dominguez, E. R., Tack, F., Ruiz, H., Sanchez Morales, R., & Arteaga, L. E. (2012). Vertical subsurface wetland for waste water purification. Procedia Engineering, 42, 1960–1968. Priya, A., Avishek, K., & Pathak, G. (2012). Assessing the potentials of Lemna minor in the treatment of domestic wastewater at pilot scale. Environmental Monitoring and Assessment, 184, 4301–4307. Rai, U. N., Tripathi, R. D., Singh, N. K., Upadhyay, A. K., Dwivedi, S., Shukla, M. K., Mallick, S., Singh, S. N., & Nautiyal, C. S. (2013). Constructed wetland as an ecotechnological tool for pollution treatment for conservation of Ganga river. Bioresource Technology, 148, 535–541. Reed, S. C., & Brown, D. (1995). Subsurface flow wetlands—a performance evaluation. Water Environment research, 67(2), 244–248. Rousseau, D., Geenens, D., Vanrolleghem, P. A & Pauw, N.D. (2002). Short-term behaviour of constructed reed beds: pilot plant experiments under different temperature conditions. In:

Environ Monit Assess (2017) 189:164 Proceedings of the 8th International Conference on Wetlands Systems for Water Pollution Control. Arusha, Tanzania, September 16–19, 128–139. Shelef, O., Gross, A., & Rachmilevitch, S. (2013). Role of plants in a constructed wetland: current and new perspectives. Water, 5(2), 405–419. Shutes, R. B. E. (2001). Artificial wetlands and water quality improvement. Environment International, 26, 441–447. Sivapragasam, C., Muttil, N., Arun, V. M., & Jeyasudha, S. (2012). Suitability of universal function approximators in redesigning rainguage network for a river basin. International Journal of Engineering and Earth Sciences, 5(2), 255–267. Sivapragasam, C., Muttil, N., Catherin Jeselia, M., & Visweshwaran, S. (2015). Infilling of rainfall information using genetic programming. Aquatic Procedia, 4, 1016–1022. Sivapragasam, C., Muttil, N., Muthukumar, S., & Arun, V. M. (2010). Prediction of algal blooms using genetic programming. Marine Pollution Bulletin, 60, 1849–1855. Solano, M. L., Soriano, P., & Ciria, M. P. (2004). Constucted wetland as a sustainable solution for wastewater treatment in small villages. Biosystems Engineering, 87(1), 109–118. Steer, D., Fraser, L., Boddy, J., & Seibert, B. (2002). Efficiency of small constructed wetland for subsurface treatment of singlefamily domestic effluent. Ecological Engineering, 18, 429–440. Stottmeister, U., Wiener, A., Kuschk, P., Kappelmeyer, U., Kastner, M., Bederski, O., Muller, R. A., & Moormann, H. (2003). Effects of plants and microorganisms in constructed wetlands for wastewater treatment. Biotechnology Advances, 22, 93–117. Tomenkoa, V., Ahmedb, S., & Popova, V. (2007). Modelling constructed wetland treatment system performance. Ecological Modeling, 205, 355–364. Trang, N. T. D., Konnerup, D., Schierup, H. H., & HuuChiem, N. (2010). Kinetics of pollutant removal from domestic waste water in a trophical horizontal subsurface flow constructed wetland system effect of hydraulic loading rate. Ecological Engineering, 36(4), 527–535. Vymazal, J. (1988). Czech constructed wetlands database. Ecology and Use of Wetlands: Prague, Czech Republic [in Czech]. Vymazal, J., & Kropfelova, L. (2011). A three stage experimental constructed wetland for treatment of domestic sewage: first 2 years of operation. Ecological Engineering, 37, 90–98. Wu, H., Zhang, J., Ngo, H. H., Guo, W., Hu, Z., Liang, S., Fan, J., & Liu, H. (2015). A review on the sustainability of constructed wetlands for waste water treatment: design and operation. Bioresource Technology, 175, 594–601. Yalcuk, A. (2013). Modeling of different types of constructed wetlands for removing phenol from olive mill waste water using an artificial neural network. Ekoloji, 22(88), 28–35. Zhang, L. Y., Zhang, L., Liu, Y. D., Shen, Y. W., Liu, H., & Xiong, Y. (2010). Effect of limited artificial aeration on constructed wetland treatment of domestic wastewater. Desalination, 250, 915–920.