Chapter VI: Conclusion remark
Chapter 6. Developing a Hydrological Simulation Tool to Design Bioretention in a watershed
ABSTRACT
Continuous urbanization has negatively impacted the ecological and hydrological environments at the global, regional, and local scales. This issue was addressed by developing Low Impact Development (LID) practices to deliver better hydrologic function and improve the environmental, economic, social and cultural outcomes. This study developed a modeling software to simulate and optimize bioretentions among LID in a given watershed. The model calculated a detailed soil infiltration process in bioretention with hydrological conditions (e.g. unsaturated and saturated soil) and hydraulic facilities (e.g. riser and underdrain) and also generated an optimized plan using Flow Duration Curve (FDC). The optimization result from the simulation demonstrated that the location and size of bioretention, as well as the soil texture, are important elements for an efficient bioretention.
We hope that this developed software will aid in establishing effective LID strategies for improving urban water sustainability and management.
150
6.1 Introduction
Continuous urbanization, such as the removal of vegetation and the replacement of pervious area with impervious area, provokes a change in the characteristics of surface runoff hydrograph thereby increasing the potential for floods and drought (Davis and McCuen, 2005; Kayhanian et al., 2012).
The increase of impervious area can also impact the ecological and hydrological environments at the global, regional, and local scales (Gao et al., 2003; Shi et al., 2016). Hence, the urbanization of a watershed needs careful assessment and planning to attain sustainability (Randhir and Raposa, 2014).
In the last two decades, low impact development (LID) practices have been developed to deliver better hydrologic functions and also improve the environmental, economic, social and cultural outcomes (Elliott and Trowsdale, 2007). These LID facilities have been largely applied as two different methods; centralized (conventional method) and distributed LID. Conventional LID facilities have been employed in a centralized strategy adjacent to a stream channel, focusing on mitigating peak discharge and minimizing hydrologic alterations. Recently, however, distributed treatment in the urban areas mentioned runoff management in the watershed and located nearby source (i.e., rainwater runoff spot) with an emphasis on infiltration (Davis, 2005; Roy et al., 2008;
Loperfido et al., 2014). The appropriate combination of centralized LID and distributed LID is practical and significant due to the different hydrological functions of each type of LID (Davis, 2005;
Loperfido et al., 2014; US EPA 2000).
Albeit there are increasing needs and awareness of LID and urbanization, the transition to a more sustainable urban design has been slow (Elliott and Trowsdale, 2007). Since it is difficult to generate an adequate guideline for an effective LID performance that considers the characteristics of both LID and watershed (Ahiablame et al., 2012a), many researchers have made efforts to produce an appropriate guideline for the monitoring and modeling of LID (Baek et al., 2016;Davis et al., 2001;
Elliot et al., 2009). Numerical modeling was used in LID modeling because it has significant potential strength that is applicable for diverse experimental scenarios while, LID monitoring needs considerable cost and labor for watershed-scale experiments (Elliott and Trowsdale, 2007; Prez- Pedini et al., 2005). However, previous LID modeling researches largely focused on simple applications that simulated the effects of LID on a study site and mostly targeting short-term rainfall events (Lee et al., 2012a; Kwak et al., 2016). There are few studies about long-term LID simulations and the optimization of the size and position of LID in the watershed, which some researchers consider as an important part of LID installation (Ahiablame et al., 2012a). The existing LID modeling softwares (e.g, SWMM (Rossman, 2010), L-THIA-LID (Ahiblame et al., 2012b),
151
SUSTAIN (Lee et al., 2012b)) are still having issues on reasonable LID simulations, such as infiltration under the unsaturated soil condition. This problem leads to deviations of simulated infiltration from the actual process (Herrada et al., 2014) and significant assumptions with regards to simulating infiltration in homogeneous soil with uniform initial moisture (Ali et al., 2016).
Here, we developed a modeling software that was able to simulate and optimize bioretentions in a given watershed. The software was employed in three different hydrological strategies based on critical exceedance percentiles of streamflow to provide improved urban water management: 1) flood mitigation, used as an index to alleviate high flow; 2) base flow restoration, used as an index to enhance low flow; and 3) hydrological properties improvement, used as an index for stabilizing middle flow variation. The aims of this study were: 1) to develop the watershed model to predict hydrological effect; 2) to analyze model performance; and 3) to develop the optimization system of bioretention and determine the optimal bioretention strategy for the watershed.
152
6.2 Development of K-LIDM model
The Korea - Low Impact Development Model (K-LIDM) is a decision tool based on the World Wide Hydrology Model version 4 (WWHM4) (Beyerlein, 2011) and the Hydrological Simulation Program - Fortran (HSPF) (Bicknel et al., 2001). It is used to design LID on a target watershed. The model can be used for long-term simulations to derive practical strategies that can easily calculate the hydrologic processes related to bioretention. K-LIDM essentially comprises two components, the watershed and LID modules, which simulate bioretention. The watershed module of K-LIDM is to use a hydrological simulation in a watershed. The LID module of K-LIDM can also consider water infiltration throughout the bioretention media under either unsaturated or saturated soil conditions.
The main goal of K-LIDM is to provide a practical solution for optimizing bioretention performance in the watershed (Figure 6.1).
Figure 6.1. Schematic of Korea Low Impact Development Model (K-LIDM)
153
6.2.1 Watershed module
The watershed module of K-LIDM is based on the hydrologic module of the HSPF, where there exists a function that incorporates the meteorological data into the model. The module has three main sub-modules; PERLNDC, which simulates the pervious areas; IMPLAND for impervious areas; and RCHRES for the open channels combined with reservoirs (Singh et al., 2005). The parameters in the K-LIDM are defined in Table 4.1. The selected parameters are referred to the HSPF manual and literature (Al-Abed and Whiteley 2002; Bicknel et al., 2001; Mishra et al., 2007). In this model, each sub-watershed is connected to the reach, reservoir, and lake modules that utilize the FTABLE. The FTABLE can calculate discharge by considering the hydraulic characteristics of reach, reservoir and lake. These are grounded in the function table based on the relationship between the surface area, stream stage, volume, and discharge. (Zhang et al., 2009). In this study, we use FTABLE for simulating the performance of bioretention considering infiltration of both saturated and unsaturated statues.
6.2.2 LID module
Bioretention consists of surface and subsurface layers that include hydrologic (e.g., infiltration, evaporation, surface runoff) and hydraulic (e.g., riser, notch, orifice) processes (Figure 6.2). The surface layer calculates discharge from the riser and considers the evapotranspiration, however; the discharge of surface layer can be simulated through the riser if the water depth is above the riser. The subsurface layer simulates infiltration in the soil layer and water movement along the underdrain from the soil to orifice. However, each layer has a different soil characteristic, hence; the water movement simulation in each soil layer considered different hydrologic and hydraulic characteristics. The two processes can be simulated in different layers by undergoing the following processes (Figure 6.3): (1) The rate of the water movement through the top soil layer was determined by Van Genuchtenβs and Darcyβs equations, (2) The beginning of infiltration into the second layer was calculated after considering the metric head as the soil approaches field capacity, (3) Water enters the underdrain towards the orifice and riser (Clear Creek Solutions, 2014).
154
Figure 6.2. Schematic diagram of Bioretention.
Figure 6.3. Schematic overview of the bioretention process.
155
K can be calculated using the Van Genuchten approximation equation (Blum et al., 2001):
πΎ(π)
πΎπ = (πβππ
β βππ)1/2[1 β (1 β (πβππ
β βππ)
1 π)
π
]
2
(1)
where, K(π) = relative hydraulic conductivity, Ks = saturated hydraulic conductivity, π = water content, ππ = residual water content, and, β = porosity, and m = constant.
Storage and movement of water into soil layers is calculated using Darcyβs equation, as follows (Hillel, 1998):
π = βKββ
βπ§ (2)
where, q = Darcy flux (cm/hr), K = hydraulic conductivity of the porous medium (cm/hr), h = total hydraulic head (cm), and z = elevation (cm).
The total head, h, is the summation of the metric head, πΉ, and the gravity head, z:
β = πΉ + π§ (3)
Substituting h to Darcyβs equation in Eq. (2):
q = βKπ(πΉ+π§)
ππ§ (4) The Van Genuchten equation was then used to compute total head, h (Blum et al., 2001):
β = β1
π[ 1
ππΈ 1 π
β 1]
1
π+ π§ (5)
where, a = constant, SE = effective saturation, m = constant, n = constant, and z = elevation head.
Effective saturation (SE) can be calculated by the Van Genuchten equation (Blum et al., 2001):
πβππ
πβππ = [ 1
1+(ππΉ)π]π = SE (6)
156
where, π = water content, ππ= residual water content, a = constant (a = yb -1), n = constant (n = Ξ» + 1), m = constant (m = 1- 1
π+1), π = poro size distribution index, yb = bubbling pressure, πΉ = pressure head = h β z, h = total hydraulic head, z = elevation head, and SE = effective saturation.
The underdrain orifice can be calculated by the following this equation (Clear Creek Solutions, 2014):
π = 3.7892 β (πππππππ π·πππππ‘ππ)2β (π»πππππ)0.5 (7) where, Header = the water height over orifice bottom.
The bioretention used the riser as an outlet for simulating discharge from the facility wherein the user can set the specifications of the riser (e.g. riser height, diameter). K-LIDM used the riser equation provided below (Clear Creek Solutions, 2014):
q = 9.739 Γ Riser Diameter Γ H1.5 (8) where, H = water level above riser.
6.2.3 The decision method for optimal Bioretention
Numerous previous studies have suggested that centralized LID and distributed LID have different functionalities depending on the hydrologic landscape of pre-urbanized and urbanized conditions and their appropriate uses (Davis, 2005; Loperfido et al., 2014; Roy et al., 2008). These LIDs can be controlled effectively throughout an urban landscape and are important for water management of local and regional scale (Davis, 2005; Prez-Pedini et al, 2005; US EPA 2000). In this regard, optimization of bioretention is essential for an effective cost, size, and appropriate combination to improve hydrologic impact. For this study, bioretention was selected from the LID (Damodaram and Zechman., 2012; Hsieh and Davis., 2005) (Figure. 6.4).
157
Figure 6.4. Schematic overview for optimization process.
6.2.4 Calibration and Sensitivity analysis
We conducted auto-calibration and sensitivity analysis by combining the watershed module in K- LIDM with the MATALB software (Table 6.1). The pattern search tool (pattern-search.m) was employed as an algorithm of the auto-calibration. This algorithm is one of the global optimization methods and it can identify an optimal point by a systematic direct search method. It is also a derivative-free evolutionary algorithm and is practical for objective functions by diminishing the error (Cho et al., 2011; Findler et al., 1987; Lewis and Virginia, 2002; Maier and Dandy, 2000). Latin Hypercube-One-factor-At-a-Time method (LH-OAT) was used as the method for executing a
158
sensitivity analysis (Van Griensven et al., 2006). The index for sensitivity is the sum of the squared error (SSE), which considered the comparison between the observed and simulation values (Cho et al., 2012). Each parameter is then ranked according to the results of the sensitivity analysis for a convenient and more effective model calibration. The NashβSutcliffe Efficiency (NSE), the coefficient of determination (R2), and the RMSE-observation standard deviation ratio (RSR) were also employed to evaluate the model performance graphically and statistically (Moriasi et al., 2007;
Nash and Sutcliffe, 1970; Oeurng et al., 2011) (Figure. 6.4(A)).
In addition to sensitivity analysis for the watershed module, we tried additional tests to figure out the characteristics of bioretention in the LID module based on the Global Sensitivity Analysis (GSA).
GSA is one of the mathematical techniques that can identify how the output of model varies according to the changing set of inputs (Pianosi et al., 2015). Among the GSA methods, Morris sensitivity analysis (Morris, 1991) was applied to the bioretention modules in this study. This method uses the elementary effect that is attributable to each input (Nguyen and De Kok, 2007). A detailed explanation of this method can be found in Morris (1991). A visualized sensitivity result was also produced to evaluate the behavior of the model (Morris, 1991). We used the MATLAB toolbox to provide the Morris sensitivity analysis (Pianosi et al., 2015).
6.2.5 Optimizing method using flow duration curve
In this study, we utilized flow duration curves (FDCs) for optimizing the bioretention level and placement. This index is applicable for watershed management and recognizing the variation of streamflow regime behavior in the field of water resources engineering. (Vogel and Fennessey, 1995;
Yokoo and Sivapalan, 2011). The critical exceedance percentiles of streamflow (e.g. Q5, Q10, Q25, Q50, Q75, Q90, and Q95) can be extracted from the FDC analysis (Mandal and Cunnane, 2009) and this index is widely used for low flow and high flow regimes (Mandal and Cunnane, 2009; Mu et al., 2007). Albeit the effect of LID yields a good performance, the effectiveness is significantly unsuitable if the LID size is considerably huge. Therefore, it is evident that LID size is an important factor for cost-effectiveness (Lee et al., 2012b; Jia et al., 2012). The bioretention parameters indicated by Cho et al. (2013) and Minnesota Pollution Control Agency (2016) (Table 4.1) was used for optimizing bioretention. The soil texture type was also considered in the process (Rawls et al., 1982; Leij et al., 1996; Schaap et al., 1998) (Table 6.2).
159
Table 6.1. Parameters for bioretention.
Layer Parameter Value Note
Surface layer Storage depth (in) Below 5.98 Prince Georgeβs County Maryland (1999), ACCWP (2011)
Soil layer Thickness (in) 17.71-35.43 EPA (2010) Storage layer
Riser height (in) 5.90-17.71
EPA (2010)
Minnesota Pollution Control Agency (2016) Riser diameter (in) 4
Underdrain Diameter (in) 4 NCDENR (2007)
160
Table 6.2. Soil properties classified by soil texture.
Texture class Residual saturation (ΞΈr)
Porosity (ΞΈ)
Saturated Hydraulic
Conductivity, Ks (cm/hr) n Bubbling pressure (cm)
Sand 0.02 0.417 21 3.19 7.26
Loamy sand 0.035 0.401 6.11 2.39 8.69
Sandy loam 0.041 0.412 2.59 1.61 14.66
loam 0.027 0.434 1.32 1.31 11.15
Silt loam 0.015 0.486 1.68 1.53 20.76
Sandy clay loam 0.068 0.33 0.43 1.39 28.08
Clay loam 0.075 0.39 0.23 1.49 25.89
Silty clay loam 0.04 0.432 0.15 1.37 32.56
Sandy clay 0.109 0.321 0.12 1.41 29.17
Silty clay 0.056 0.423 0.09 1.39 34.19
Clay loam 0.09 0.385 0.06 1.2 37.3
161
A multi-objective optimization is ideally suited for modeling by using multiple conflicting objectives in an almost real-world scenario (Deb, 2014). In this study, we used a controlled elitist genetic algorithm (gamultiobj.m; a variant of NSGA- II) from the MATLAB software. This method is one of the most efficient, multi-objective, evolutionary algorithms that use the elitist approach (Deb et al., 2002). The solutions in this method are sorted by the degree of dominance within a given population and the algorithm finds the solution that will preserve the population diversity along the first non-dominated front to find the entire Pareto-optimal region (Dorn and Ranjithan, 2003; Lee et al., 2012). It is more imperative to determine the appropriate the size of bioretention in order to consider its cost and hydrologic impact of it. Soil type is one of the major elements producing the hydrologic effect of bioretention. We used not only the level of biorientation but also soil type in terms of variables to multi-objective optimization. To do so, the optimal range of distributed bioretention and centralized bioretention was 0β100% of each sub-basin and whole basin, respectively. We also considered 11 different soil types for optimization (Table 4.3). Each soil type is assigned according to a nominal scale and is then added as a variable in the multi-objective optimization.
The aims of the bioretention optimization in this study were: 1) to lessen the effect of flood (Q5/Q50, Q25/50); 2) to restore the base flow (Q95/Q50, Q75/Q50); and, 3) to improve the hydrological properties that can also amend the behavior of flood and base flow (Q25/Q50, Q75/Q50) using a controlled elitist genetic algorithm and considering the size of bioretention (Table 3) (Figure. 6.4(B)).
Table 6.3. Index for the optimization process.
Index Equation
High flow (Q5/Q50+Q25/50)/2 Low flow (1/Q95/Q50+1/Q75/Q50)/2 Combined (Q25/Q50+1/Q75/Q50)/2
162
6.3.1 Case study
6.3.2 Site description
We selected the Pungyeoungjeoncheon (PY) basin (35Β°10β²13.5β³N, 126Β°48β²54. 3β³E) in Gwangju and Jangseong City, Republic of Korea for applying optimization of Bioretention (Figure. 6.5). The basin has a total area of 56.25 km2 and the upper regions of the watershed mostly constitute agricultural areas that are used for rice production. Industrial and residential areas are located in the middle and lower regions of the watershed. ArcGIS (Version 10.2) was used to delineate 14 sub- basins of the watershed using the digital elevation model (DEM) from the National Geographic Information Institute. The meteorological and geographical data were acquired from a nearby Gwangju weather station (Gwangju, Republic of Korea) and National Geographic Information in Korea. The average annual rainfall is 1391mm and the average annual maximum and minimum temperature are 29.3Β°C and -1.9 Β°C. The daily flow rate was obtained from the Water Resource Management System to calibrate and validate the model. The period from 2010 to 2012 was used for the model calibration period for flow rate and the period from 2013 to 2014 was used for the model validation. The average and standard deviation of flow rates for the calibration were 2.35 and 5.77 (CMS), respectively, with maximum and minimum figures of 100.4 and 0.11 (CMS), respectively.
The average and standard deviation of flow rates for the validation were 1.93 and 5.56 (CMS), respectively, with maximum and minimum figures of 119.5 and 0.28 (CMS), respectively.
163
Figure 6.5. The study area in Gwangju and Jangseong city (35Β°10β²13.5β³N, 126Β°48β²54. 3β³E).
Figure 6.6 is the schematic diagram of the K-LIDM that was applied in this study. The diagram shows the integration of the distributed and centralized bioretention across the watershed. Distributed bioretention takes charge of the runoff from each subbasin before entering the runoff of each subbasin in reach while centralized bioretention is in charge of the whole basin through the main reach. After applying bioretention, we generated the optimal size and the soil type of bioretention by considering FDC
164
Figure 6.6. A schematic diagram of the K-LIDM in application with distributed and centralized bioretention.
165
6.3.2 Calibration, validation and sensitivity analysis
We executed the sensitivity analysis before the calibration process for a more effective calibration in the K-LIDM. Table 4.4 shows the sensitivity rank, determined by LHOAT, and the calibrated values of the parameters in the watershed module, yielded by the pattern search tool of MATLAB (Lewis and Virginia, 2002). The parameters involved in the interflow (INTFW and INFILT), base flow (BASETP), and ground water functions (DEEPER) are ranked as the most influenced parameters in K-LIDM (Table 4). Their results were similar to those of other studies in which the HSPF model took on a central role (Jairo et al., 2013; Xie and Lian, 2013). DEEPFR and INFILT can affect normal flow, hence; these parameters show that normal flow is the most important for calibration (Jairo et al., 2013) and the results also indicate that it is important to consider ground water for watershed modeling.
166
Table 6.4. Parameters for Korea Low Impact Development Model K-LIDM.
Parameter Description Unit Value Rank
LZSN Lower zone nominal soil moisture storage in 9.98 7
INFILT Index to the infiltration capacity of the soil in/hr 0.04 4
KVARY Variable groundwater recession parameter 0 10
AGWRC Basic groundwater recession parameter 0.98 5
DEEPFR Fraction of groundwater inflow which will enter deep ground water 0.98 1
BASETP Fraction of remaining potential ET which can be satisfied from base flow 0.19 3 AGWTP Fraction of remaining potential ET which can be satisfied from active groundwater storage 0 10
CEPSC Interception storage capacity in 0.1 10
USZN Upper zone nominal soil moisture storage in 1.252 6
INTFW Interflow inflow parameter 1.08 2
IRC Interflow recession parameter 0.31 8
LZETP Lower zone ET parameter 0.1 10
NSUR Manning's N for the assumed overland flow plane 0.2 9
167
Figure 6.7(a) showed the plot of standard deviation vs. mean of the Elementary Effects (EEs) of each parameter in the bioretention module. The figure also includes for parameter in bioretention module with the confidence bound of each value. These results were calculated by using Morrisβs sensitivity analysis among the GSA. The saturated hydraulic conductivity (Ks) and constant n of the Van Genuchten equation have considerable mean values of EEs. This indicates that Ks and n can significantly affect the bioretention mechanism. However, the standard deviations of EEs for these parameters are high compared with those for the other parameters. Both parameters significantly influenced the effects of the ensemble of factors (Saltelli et al., 2008). By contrast, as indicated by their low standard deviations, porosity and residual saturation weakly interacted with other parameters. Ks, bubbling pressure, and n have broad confidence bounds for their mean and standard deviation of EEs, indicating that these parameters possess significant uncertainties (Figure 6.7(a)) (Saltelli et al., 2008). The convergence plot for analyzing the sensitivity variations with the model evaluations is shown in Figure 6.7(b). Each dashed line of the parameter indicates the confidence bound of the mean of EEs as the model iteration increased. The estimated Morrisβs sensitivity is plotted against the gradually increasing number of model evaluations (Yang., 2011). The EEs mean values of Ks and n have considerable variations with the model evaluation. Initially, the difference between Ks and n is relatively large, however; this difference decreases as the number of model evaluation increases (Figure 6.7(b)). This result showed that these parameters are sensitive to sample size, indicating that sample size significantly affects the sensitive analysis. However, porosity, residual saturation, and bubbling pressure are weakly affected by model evaluation.
Our model yielded an acceptable performance during calibration (2010-2012) and validation (2013-2014) as shown in Figure. 6.8. The value of NSE and R2 were greater than 0.7 for both periods.
This means that the model performance can be evaluated as βvery goodβ and it has a good representation of the watershed hydrology (Moriasi et al., 2007). Although K-LIDM showed acceptable prediction accuracy, it was found that the model underestimated the peak flow. Many researchers have the same problem; the model tends to underestimate the simulated values due to the time step of simulating the high peak flow (Cho et al., 2012; Kirsch et al., 2002).