Three sensitivity analysis methods based on derivative, screening and regression are applied to the CLARA simplified planning tool (CLARA-SPT) estimating life cycle cost (LCC) of water supply systems. Derivative-based and standard regression coefficient methods are used to address parameter prioritization and Morris screen method for parameter fixing. The study investigated sensitivity of 26 parameters through three feasible water supply alternatives. The results revealed the importance of all global parameters: period of consideration, net interest rate and expected annual growth rate, for the LCC of all alternatives. The analysis identified ten technological parameters as non-influential, which can be fixed to a certain value. In contrast, two internally fixed parameters (i.e., chlorine dosage and alum dosage) are determined to be important; hence, these parameters need to be available for planners for reasonable cost estimation. Except for period of consideration, all parameters have low standard deviation (σ) value that indicates linear correlation of alternatives' LCC with input parameters and/or insignificant interaction among parameters. This study is valuable for further development of the CLARA-SPT in order to simplify it and to provide guidance for the tool user on the importance of specific input parameters.

ACRONYMS

  • CLARA-SPT

    CLARA simplified planning tool

  • LCC

    life cycle cost

  • OAT

    one parameter at a time

  • RV

    residual value

  • SA

    sensitivity analysis

  • SRC

    standardized regression coefficient

  • TC

    total cost

INTRODUCTION

Environmental decision support tools usually face a problem of conceptualization, uncertainty effect of input parameters on the intended output, over-parameterized and poor result predation, which possibly mislead the decision-making process (Huang & Liu 2008; Sun et al. 2012). One way to achieve performance assessment of a model is by applying sensitivity analysis (SA) to examine how the uncertainty of model input parameters influences the efficiency of the model estimation (Saltelli et al. 2008). The selection of a specific SA method is highly influenced by the main objective of the study, the complexity of the model and the computational time and cost. Often, simple and computationally less expensive SA methods are preferred for models with many input parameters.

According to Saltelli et al. (2000), SA is classified as local, screening and global methods. Relatively local and screening methods that apply the principle of changing one parameter at a time (OAT) are computationally cheaper than global SA methods; whereas, global SA is the most efficient and reliable method that explores the entire parameter space (Homma & Saltelli 1996; Saltelli et al. 2004).

This study addresses the following two objectives of SA (Saltelli et al. 2008):

  1. Parameter prioritization (PP): to identify the most important input parameters that have the greatest potential for uncertainty of the model output, thereby more certainty is demanded.

  2. Parameter fixing (PF): to determine non-influential input parameters that do not cause significant variability on the model outputs. These parameters can be fixed to a certain average value in their acceptable boundaries.

Three complementary SA methods are selectively applied to assess input parameters sensitivity for the uncertainty of proposed water supply alternatives' life cycle cost (LCC). The engaged techniques are derivative-based, standard regression coefficient (SRC) and Morris screening SA methods. Derivative-based and SRC are used for PP and Morris screening for PF (Neumann 2012).

The CLARA simplified planning tool (CLARA-SPT) was launched in February 2014 and is used to estimate LCC of water supply and/or sanitation alternatives. The tool was developed for five African countries: Burkina Faso, Ethiopia, Kenya, Morocco and South Africa (Lechner et al. 2014). Because the tool was developed recently, sensitivity of water supply systems' LCC has not yet been assessed. Conducting SA should contribute to determining conceptualization gaps and potential improvements of the tool (Farr 2011). In this regard, the main focus of this study relies on assessing the sensitivity of CLARA-SPT to identify important and non-influential water supply input parameters for prioritizing and fixing them, respectively.

METHODS

Study area description

The SA of water supply system's LCC computation through CLARA-SPT was tested for one of the nine urban districts of Bahir Dar city named Hidar-11. Bahir Dar city is located at the south of Lake Tana, 11 ° 38′ north latitude and 37 ° 15′ east longitude (MUDHCo 2014). The mean annual precipitation, temperature and aquifer hydraulic conductivity of the area ranges from 1,103 to 1,336 mm, 22 to 29 °C and 0.06 to 19 m/day (W/Yohannes 2010; Wondmagegne et al. 2012).

The estimated population size of Hidar-11 district is about 24,600 in 2014 based on a population of 18,890 in 2008 (TCE 2009) and 4.5% annual population growth rate of the city (CSA 2008). Hidar-11 district is geographically bounded by the river Abay (i.e., a cross boundary river) on the west and Lake Tana on the north-west side (Figure 1).

Figure 1

Bahir Dar city: location of nine districts, selected well fields (WF), spring (SP) and Lake Tana; and reservoirs (Res) for Hidar-11 district.

Figure 1

Bahir Dar city: location of nine districts, selected well fields (WF), spring (SP) and Lake Tana; and reservoirs (Res) for Hidar-11 district.

Bahir Dar's existing water supply system that comes from springs and boreholes is evaluated to be insufficient and unreliable. This is because of water shortage, an aged distribution network, system management problems and frequent electricity power interruption. Currently, the city water supply authority, which is responsible for planning, designing, constructing and operating the water supply system, is exploring additional water supply sources to solve the water shortage problem. As a result, Lake Tana is identified as a potential source to satisfy the ever growing water demand of the city. Users are responsible for their house connection fee, monthly water bill and to maintain the quality of the water at the household level.

Case study model set-up

A reconnaissance survey of the intervention area confirms the presence of alternative water supply sources: spring, borehole and lake at an approximate distance of 20.7, 8.0 and 6.3 km from the district, respectively. The three feasible alternatives are proposed to satisfy the water demand of the projected population size of about 60,000 in 2034 (20 year planning horizon). Alternatives were labelled as Alt-1, Alt-2 and Alt-3 for a system sourced from the borehole, lake and spring, respectively. Each alternative comprises the appropriate set of technologies under water source, water purification and water distribution functional groups. This example deliberately excluded the costs of individual house connections, since the service provider is not accountable for such expense, rather users are. Involved technologies in the three alternatives are shown in Figure 2, while detailed information of each technology can be seen in the Supporting material (available online at http://www.iwaponline.com/jws/064/126.pdf).

Figure 2

Proposed water supply system alternatives.

Figure 2

Proposed water supply system alternatives.

CLARA-SPT working principle

The CLARA-SPT is designed to estimate and compare LCC of water supply and/or sanitation systems. The water supply part of the tool comprises: water source, water purification and water distribution functional groups (Casielles Restoy et al. 2014). The very first step of LCC computation using CLARA-SPT is to wisely insert all input parameters for each alternative. Consequently, the tool calculates LCC components (i.e., initial investment cost, operation and maintenance cost, reinvestment cost and residual value (RV)) of each and every technology embedded in the alternatives. This is followed by linearly aggregating the net present value of technologies’ cost component to alternatives’ respective LCC component. Since a 20 year planning horizon is considered, the remaining asset value of each infrastructure after 20 years of service is taken as the RV of the systems.

Global parameters of CLARA-SPT (Table 1) are designed to influence the cost of all technologies. However, technological parameters mainly influence the cost of affiliated technology; whereas, internal-technological parameters are design assumptions of technologies, which are internally fixed into certain values.

Table 1

Input parameters label, description, unit and variation range

Label Description Unit Range of value 
Global parameters 
X1 Period of consideration Year 5–30 
X2 Net interest rate 1.5–6.5 
X3 Expected annual growth 1.5–7.5 
User-technological parameters 
X4 Ground water depth 10–80 
X5 Number of boreholes Number 9–30 
X6 Number of disinfection plants Number 1–26 
X7 Number of pumping stations Number 4–40 
X8 Pressure head 20–320 
X9 Trench depth 0.6–2 
X10 Pipe length 1,000–45,300 
X11 Number of surface water tanks Number 2–24 
X12 Number of lake water extractions Number 3–25 
X13 Pump head at intake 12–60 
X14 Number of flocculation sedimentation plants Number 1–25 
X15 Number of surface water treatment plants Number 1–25 
X16 Number of spring water extraction chambers Number 7–5 
X17 Hydraulic conductivity m/s 1.2E-5–2.5E-4 
X18 Water tank elevation 2–20 
X19 Borehole diameter inch 8 and 20 
X20 Kind of disinfectant – Ca(OCl)2, Na(OCl) and Cl2 gas 
X21 Pump installation – Surface and sub-surface 
X22 Kind of flocculant – Alum, ACH and ferric sulphate 
X23 Number of water lines Number 1–2 
Internal-technological parameters 
X24 Diameter distribution 30–70% small; 10–50% middle; 0–40% large 
X25 Chlorine dosage mg/l 1.2–3 
X26 Alum dosage mg/l 15–60 
Label Description Unit Range of value 
Global parameters 
X1 Period of consideration Year 5–30 
X2 Net interest rate 1.5–6.5 
X3 Expected annual growth 1.5–7.5 
User-technological parameters 
X4 Ground water depth 10–80 
X5 Number of boreholes Number 9–30 
X6 Number of disinfection plants Number 1–26 
X7 Number of pumping stations Number 4–40 
X8 Pressure head 20–320 
X9 Trench depth 0.6–2 
X10 Pipe length 1,000–45,300 
X11 Number of surface water tanks Number 2–24 
X12 Number of lake water extractions Number 3–25 
X13 Pump head at intake 12–60 
X14 Number of flocculation sedimentation plants Number 1–25 
X15 Number of surface water treatment plants Number 1–25 
X16 Number of spring water extraction chambers Number 7–5 
X17 Hydraulic conductivity m/s 1.2E-5–2.5E-4 
X18 Water tank elevation 2–20 
X19 Borehole diameter inch 8 and 20 
X20 Kind of disinfectant – Ca(OCl)2, Na(OCl) and Cl2 gas 
X21 Pump installation – Surface and sub-surface 
X22 Kind of flocculant – Alum, ACH and ferric sulphate 
X23 Number of water lines Number 1–2 
Internal-technological parameters 
X24 Diameter distribution 30–70% small; 10–50% middle; 0–40% large 
X25 Chlorine dosage mg/l 1.2–3 
X26 Alum dosage mg/l 15–60 

The system capacity is the main design parameter of the water supply system. For example, for a 20 year period, all systems are designed for 12,615 m3/d capacity, while the capacity rises to 19,591 m3/d for a period of 30 years. Therefore, when one changes the value of the period of consideration, it is required to change the system capacity (i.e., daily demand or flow rate) simultaneously. For this case study, the maximum daily demand is taken as the system capacity, which is estimated based on the Ethiopia design guideline (MOWRD 2002). For details of calculation refer to the Supporting material (online at http://www.iwaponline.com/jws/064/126.pdf).

SA techniques

SA of the case study involves 26 parameters: three global parameters, 20 user-technological and three internal-technological parameters (Table 1). Five of the technological parameters (i.e., borehole diameter, kind of disinfectant, pump installation, type of flocculent and number of water lines) are categorized as qualitative/subjective parameters. Their influence was assessed using Equation (1), by which the relative cost change of the alternative from its base value is computed 
formula
1
where is the respective alternative base output at the reference value of X, and is the alternative output obtained after changing the input parameter from the existing qualitative options in the tool.

Sensitivity of the remaining 21 quantitative parameters was evaluated using derivative-based, SRC and Morris screening methods.

Derivative-based method

The derivative-based SA is a commonly applied method using first-order partial differentiation of the dependent variable Y, with the function of input parameter . In this method, the sensitivity of Y (i.e., alternative's total cost (TC) and RV) was assessed by changing OAT, while keeping all other parameters constant at the base value. Parameter perturbation at ±5, ±10, ±15 and ±20% from the base value of Xi was tested. According to Saltelli et al. (2000), the derivative-based sensitivity coefficient is estimated by Equation (2) 
formula
2
The quotient is used to normalize the unit effect of the derivative. For a small change of the ith input parameter, the partial derivative can be approximated as a fraction of finite difference of output value to the change of input parameter. In this regard, Equation (2) can be approximated by the ratio of percentage change of Y to percentage change of Xi (Equation (3)), assuming that Y is linear (Hamby 1994) 
formula
3

To compute for K number of input parameters, minimum simulation number N = K + 1 is required.

SRC method

The working principle of regression-based SA method is random selection of input parameters to execute a Monte Carlo simulation so as to find out a multivariate linear function with the output parameter (Equation (4)). Here, the magnitude of statistically significant regression coefficients are used to estimate SA indices of corresponding input parameters. To remove the unit difference of input and output parameters, standardization of regression coefficients is needed (Kutner et al. 2005). The SRC of a parameter (βi) calculated by Equation (5), is taken as the sensitivity index of the associated parameter, if the coefficient of determination R2 is greater than 0.70 (Saltelli et al. 2004). The greater value of R2 shows linearity and absence of input parameters’ interaction (Haahtela 2010). 
formula
4
 
formula
5
where Y is the dependent output parameter obtained after regression analysis of K number of input parameters and N number of simulations, b0 is the aggregated error factor (intercept) of the regression equation, bi is the coefficients or slope of Xi, and are the standard deviations of Xi and Y, respectively.

Literature recommended ranges of N from 500 to 1,000 irrespective of input parameter numbers (Saltelli et al. 2008). In this paper, sensitivity of the 21 quantitative parameters was assessed by conducting 987 simulations.

Morris screening method

The fundamental principle of the Morris screening method relies on the concept of elementary effect of input parameters on the model result (Morris 1991). It measures sensitivity of large numbers of parameters through partitioning of the parameters space at low computational cost (Saltelli et al. 2004). The method is mainly used to identify non-influential parameters to address parameter fixing objectivity of SA. The elementary effect of the ith parameter represents the relative difference between the tool output obtained after parameter perturbation by Δ, , and the base output Y(X) for the given reference input parameter value (Equation (6)). 
formula
6
where are parameters sampled from K dimensional, P-level grid parameter space. As depends on the location of the random sample, R numbers of replicate runs for each parameter are summarized by mean and standard deviation to obtain sensitivity indices of each parameter (Morris 1991). measures the overall influence of the ith parameter on the output uncertainty, while assesses non-linearity and/or interaction of the parameter with others. When the model is non-monotonic the value of is affected by negative and positive values of, which leads the values to cancel each other out. To avoid this effect, Campolongo et al. (2007) suggested referring to the mean of the absolute value instead of to calculate the sensitivity index of a specific parameter (Equation (7)). 
formula
7
According to Morris (1991), the required number of simulations is estimated by , where R ranges from 4 to 10 and K is the number of input parameters. In this study, eight replicates of each quantitative input parameter varying OAT by ±10, ±20, ±30 and ±40% from their base values were tested. Therefore, for the 21 quantitative parameters a total of model simulations were performed.

Cut-off value setting for parameter categorization

In this study, a 5% of significant level (i.e., 95% confidence level) was chosen to classify non-influential parameters (Frey & Patil 2002). The assumption is that economic value difference of alternatives that fall within a ±5% margin can be considered approximately identical economic options. Hence, the parameter with is categorized as a non-influential parameter that can be fixed to a certain value. For parameter prioritization, a 10% significant level was selected, i.e., alternatives having more than 10% economic value deviation are considered as distinguished options. Therefore, the parameter with is classified as important. Sensitivity indices that range between 0.1 and 0.05 represent still sensitive parameters, which are neither so important to prioritize nor non-influential to fix them (Hamby 1994).

RESULTS AND DISCUSSION

SA results of all 26 input parameters for alternatives’ TC and RV are presented in this section. Positive values of the derivative-based index indicate that an increase of the parameter leads to a rise of the output value, while negative values indicate the inverse effect.

Global parameters SA

The relative significance level of the global parameters X1 (consideration period), X2 (net interest rate) and X3 (expected annual growth rate) for alternatives’ TC and RV are presented in Table 2. Both derivative-based and SRC methods confirmed the importance of all global parameters except X3 for alternatives’ RV. The negative sensitivity index of X2 shows the decrease of the alternative's present economic value with the rise of net interest rate. Similarly, longer consideration period (X1) clearly has the consequence of further devaluing the system's RV (Jovanovic 1999; Pearce et al. 2006). Highest standard deviation (σ) of parameter X1 for both TC and RV indicates the presence of significant non-linearity and/or interaction with other parameters.

Table 2

Global parameters’ sensitivity indices for alternatives’ TC and RV

Label Alt-1
 
Alt-2
 
Alt-3
 
Sd β μσ Sd β μ* σ Sd Β μ* σ 
TC 
X1 1.722 0.827 1.585 0.306 1.711 0.877 1.767 0.393 1.531 0.829 1.553 0.292 
X2 − 0.328 − 0.262 0.329 0.027 − 0.353 − 0.201 0.353 0.029 − 0.316 − 0.246 0.317 0.026 
X3 0.366 0.328 0.367 0.029 0.394 0.252 0.394 0.031 0.352 0.274 0.352 0.028 
RV 
X1 − 1.453 − 0.835 1.500 1.041 − 1.114 − 0.200 1.456 1.812 − 1.017 − 0.575 0.811 0.254 
X2 − 0.773 − 0.435 0.776 0.092 − 0.773 − 0.460 0.776 0.092 − 0.773 − 0.660 0.776 0.092 
X3 0.084 0.053 0.461 0.037 0.093 0.061 0.461 0.037 0.091 0.096 0.461 0.037 
Label Alt-1
 
Alt-2
 
Alt-3
 
Sd β μσ Sd β μ* σ Sd Β μ* σ 
TC 
X1 1.722 0.827 1.585 0.306 1.711 0.877 1.767 0.393 1.531 0.829 1.553 0.292 
X2 − 0.328 − 0.262 0.329 0.027 − 0.353 − 0.201 0.353 0.029 − 0.316 − 0.246 0.317 0.026 
X3 0.366 0.328 0.367 0.029 0.394 0.252 0.394 0.031 0.352 0.274 0.352 0.028 
RV 
X1 − 1.453 − 0.835 1.500 1.041 − 1.114 − 0.200 1.456 1.812 − 1.017 − 0.575 0.811 0.254 
X2 − 0.773 − 0.435 0.776 0.092 − 0.773 − 0.460 0.776 0.092 − 0.773 − 0.660 0.776 0.092 
X3 0.084 0.053 0.461 0.037 0.093 0.061 0.461 0.037 0.091 0.096 0.461 0.037 

Grey highlighted and bold: important; bold italic: non-linearity and/or parameter interaction.

Technological parameters SA

Qualitative parameters SA

Results in Table 3 show non-influential effect of X21 (pump installation type) for both TC and RV of all alternatives . Similarly, X20 (kind of disinfectant) and X22 (kind of flocculent) did not have any effect on the RV of alternatives. This is because these parameters are hardly associated with alternatives’ infrastructure cost contribution on the system's RV. However, their main influence on the alternative's operation and maintenance cost result in an un-neglected effect of X20 and significant effect of X22 for respective alternative's TC. Parameter X19 (borehole diameter) was found to be important for Alt-1 TC and sensitive for its RV . This can be explained by the fact that a larger diameter for drilling needs is more costly than a smaller diameter. Parameter X23 (number of water lines), either one or two water lines, was also determined to be important for Alt-2 TC. The amount of discharge flow per water line is affected and influences the pipe diameter and thus the TCs.

Table 3

Qualitative parameters relative sensitivity indices for alternatives' TC and RV

Label TC
 
RV
 
  Δrr -Alt-1 Δrr-Alt-2 Δrr-Alt-3 Δrr -Alt-1 Δrr-Alt-2 Δrr-Alt-3 
X19 0.100   0.051   
X20 0.075 0.072 0.086 0.000 0.000 0.000 
X21 0.002 0.002 0.004 0.012 0.010 0.012 
X22   0.108   0.000  
X23   0.158   0.010  
Label TC
 
RV
 
  Δrr -Alt-1 Δrr-Alt-2 Δrr-Alt-3 Δrr -Alt-1 Δrr-Alt-2 Δrr-Alt-3 
X19 0.100   0.051   
X20 0.075 0.072 0.086 0.000 0.000 0.000 
X21 0.002 0.002 0.004 0.012 0.010 0.012 
X22   0.108   0.000  
X23   0.158   0.010  

Light grey highlighted and bold: important; dark grey highlighted and italic: non-influential parameter.

Derivative-based SA indices of quantitative parameters

The results obtained from derivative-based SA for alternatives' TC are shown in Figure 3. The vertical distance from the horizontal axis refers to the importance of the parameter. Parameter X8 (pressure head) was detected as the most important parameter for all alternatives' TC and RV. Because the relative share of pumping station costs is highest for Alt-3 (54%) its influence on Alt-3 TC is higher than on Alt-1 and Alt-2 TCs (36 and 37% of TC are pumping station costs for Alt-1 and Alt-2, respectively). Moreover, X10 (pipe length) was found to be important for all alternatives' TC and RV. The result also shows the importance of X25 (chlorine dosage) only for alternatives' TC but not for RV, since infrastructure cost of alternatives has never been influenced by X25. Parameter X4 (ground water depth) was classified as an important parameter, since the LCC of borehole development largely depends on drilling depth of the well (Jagals & Rietveld 2011). On the other hand, X5 (number of boreholes) and X9 (trench depth) were found to be important for RV of related alternatives, but not for their respective TC. The reason might be either their influence is more governed by capital costs, or the hosting technologies of these parameters (i.e., borehole, transport main and/or distribution network) have longer lifetimes than 20 years of consideration. For TC of Alt-2, X13 (pump head at intake) and X26 (alum dosage) were identified to be important parameters but not for RV. This is due to the substantial effect of X13 on Alt-2 operation and maintenance costs and overshadows its effect on capital cost, whereas X26 is totally an operation-related parameter.

Figure 3

Derivative-based SA of quantitative user-technological parameters for Alt-1, Alt-2 and Alt-3 TC.

Figure 3

Derivative-based SA of quantitative user-technological parameters for Alt-1, Alt-2 and Alt-3 TC.

SRC-based SA indices of quantitative parameters

The coefficients of determination R2 of Alt-1, Alt-2 and Alt-3 TC are 0.98, 0.93 and 0.97 and thus >0.70, respectively. This indicates the validity of SRC sensitivity measure and high degree of linearity for all alternatives' TC. The method was also valid for RV of Alt-1 and Alt-3, where R2 is 1.00 and 0.99, respectively. However, it was found to be invalid for RV of Alt-2, since estimated R2 = 0.34. The result obtained from SRC confirmed the importance of X8, X4 and X26 for the TC of related alternatives (Figure 4 and Table 4). Parameter X10 (pipe length) was determined to be important only for TC of Alt-3 despite its involvement in all alternatives. The reason is that the presence of longest transport main (20.7 km) in Alt-3 resulted in a relatively higher cost share of 9% than 2% for 8.0 km and 3% for a 7.5 km long transport main of Alt-1 and Alt-2, respectively. However, X10 was classified as an important parameter for RV of all alternatives, since 50 year lifetimes of pipes leaves considerable RV in the system at the end of 20 years.

Table 4

SA results of alternatives' TC

Label Alternatives' TC
 
Alt-1
 
Alt-2
 
Alt-3
 
Sd β μ* σ Sd β μ* σ Sd β μ* σ 
X4 0.207 0.201 0.209 0.005         
X5 −0.030 −0.023 0.038 0.019         
X6 0.002 0.015 0.002 0.000 0.002 0.015 0.002 0.000 0.002 0.009 0.002 0.000 
X7 0.007 0.034 0.008 0.002 0.004 0.013 0.004 0.001 0.008 0.018 0.009 0.003 
X8 0.339 0.267 0.339 0.001 0.325 0.182 0.323 0.002 0.513 0.352 0.513 0.002 
X9 0.021 0.013 0.021 0.000 0.022 0.010 0.022 0.000 0.039 0.020 0.039 0.000 
X10 0.105 0.073 0.105 0.000 0.107 0.050 0.109 0.002 0.192 0.115 0.192 0.000 
X11 0.004 0.035 0.004 0.000 0.004 0.064 0.004 0.000 0.009 0.072 0.010 0.000 
X12      0.008 0.011 0.009 0.002     
X13      0.123 0.082 0.123 0.002     
X14      0.007 0.062 0.006 0.001     
X15       0.001 0.004 0.001 0.000     
X16          0.012 0.007 0.012 0.000 
X17          −0.004 −0.005 0.004 0.001 
X18          0.002 −0.002 0.002 0.000 
X24 −0.011 −0.005 0.011 0.000 −0.011 −0.004 0.011 0.000 −0.013 −0.005 0.013 0.000 
X25 0.100 0.041 0.093 0.002 0.100 0.028 0.090 0.002 0.108 0.041 0.108 0.002 
X26      0.150 0.104 0.150 0.002     
  PP PF PP PF PP PF 
Label Alternatives' TC
 
Alt-1
 
Alt-2
 
Alt-3
 
Sd β μ* σ Sd β μ* σ Sd β μ* σ 
X4 0.207 0.201 0.209 0.005         
X5 −0.030 −0.023 0.038 0.019         
X6 0.002 0.015 0.002 0.000 0.002 0.015 0.002 0.000 0.002 0.009 0.002 0.000 
X7 0.007 0.034 0.008 0.002 0.004 0.013 0.004 0.001 0.008 0.018 0.009 0.003 
X8 0.339 0.267 0.339 0.001 0.325 0.182 0.323 0.002 0.513 0.352 0.513 0.002 
X9 0.021 0.013 0.021 0.000 0.022 0.010 0.022 0.000 0.039 0.020 0.039 0.000 
X10 0.105 0.073 0.105 0.000 0.107 0.050 0.109 0.002 0.192 0.115 0.192 0.000 
X11 0.004 0.035 0.004 0.000 0.004 0.064 0.004 0.000 0.009 0.072 0.010 0.000 
X12      0.008 0.011 0.009 0.002     
X13      0.123 0.082 0.123 0.002     
X14      0.007 0.062 0.006 0.001     
X15       0.001 0.004 0.001 0.000     
X16          0.012 0.007 0.012 0.000 
X17          −0.004 −0.005 0.004 0.001 
X18          0.002 −0.002 0.002 0.000 
X24 −0.011 −0.005 0.011 0.000 −0.011 −0.004 0.011 0.000 −0.013 −0.005 0.013 0.000 
X25 0.100 0.041 0.093 0.002 0.100 0.028 0.090 0.002 0.108 0.041 0.108 0.002 
X26      0.150 0.104 0.150 0.002     
  PP PF PP PF PP PF 

Light grey highlighted and bold: important; dark grey highlighted and italic: non-influential.

Figure 4

SRC-based SA of quantitative user-technological parameters for Alt-1, Alt-2 and Alt-3 TCs.

Figure 4

SRC-based SA of quantitative user-technological parameters for Alt-1, Alt-2 and Alt-3 TCs.

Unlike the derivative-based method, SRC result showed the importance of X11 (number of surface water tanks) and X7 (number of pumping stations) for RV of Alt-1, and X11 for RV of Alt-3.

Morris screening SA indices of quantitative parameters

The calculated mean value of parameter absolute elementary effects μ* (abscissa) was plotted against the standard deviation σ (ordinate) of the parameter elementary effect for Alt-2 TC (Figure 5). From these figures we can clearly observe the relative sensitivity (μ*) of qualitative user-technological parameters and their linearity effect (σ). The value of μ* was used to determine non-influential parameters. The results in Tables 4 and 5 reveal the non-influential effect of X6 (number of disinfectant plants), X7 (number of pumping stations) and X11 (number of surface water tanks) for all alternatives' TC and RV. Whereas X9 (trench depth) and X24 (diameter distribution), which are capital cost items, were identified to be non-influential for all alternatives' TC but not for RV. This can be explained by the fact that the 50 years' lifetime of water distribution networks and transport mains leaves substantial RV in the system at the end of 20 years of consideration. In line with this, a previous study finding of small water supply systems agreed the independence of distribution system's TC on the pipe diameter, since the capital cost of pipe sizes used in the small system are relatively constant (Jagals & Rietveld 2011).

Table 5

SA results of alternatives’ RV

Label Alternatives’ residual value
 
Alt-1       Alt-2       Alt-3       
Sd β μ* σ Sd β μ* σ Sd β μ* σ 
X4 0.137 0.100 0.137 0.002         
X5 0.128 0.178 0.131 0.005         
X6 0.002 0.002 0.002 0.000 0.002 −0.067 0.002 0.000 0.001 0.011 0.001 0.000 
X7 0.029 0.136 0.032 0.013 0.012 −0.003 0.017 0.012 0.010 0.093 0.017 0.014 
X8 0.140 0.077 0.140 0.000 0.156 0.091 0.157 0.002 0.121 0.102 0.121 0.001 
X9 0.140 0.060 0.140 0.000 0.174 0.078 0.174 0.000 0.150 0.097 0.150 0.000 
X10 0.306 0.148 0.306 0.000 0.375 0.186 0.385 0.009 0.357 0.265 0.357 0.000 
X11 0.022 0.152 0.025 0.002 0.025 −0.149 0.028 0.002 0.029 0.336 0.033 0.003 
X12     0.018 0.017 0.020 0.004     
X13     0.014 0.030 0.014 0.000     
X14     0.010 0.079 0.010 0.000     
X15     0.001 −0.011 0.001 0.000     
X16         0.030 0.053 0.030 0.000 
X17         −0.009 −0.013 0.010 0.002 
X18         0.008 0.012 0.008 0.000 
X24 −0.074 −0.025 0.074 0.000 −0.085 −0.030 0.085 0.000 −0.050 −0.026 0.050 0.000 
X25 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 
X26     0.000 −0.009 0.000 0.000     
  PP PF PP PF PP PF 
Label Alternatives’ residual value
 
Alt-1       Alt-2       Alt-3       
Sd β μ* σ Sd β μ* σ Sd β μ* σ 
X4 0.137 0.100 0.137 0.002         
X5 0.128 0.178 0.131 0.005         
X6 0.002 0.002 0.002 0.000 0.002 −0.067 0.002 0.000 0.001 0.011 0.001 0.000 
X7 0.029 0.136 0.032 0.013 0.012 −0.003 0.017 0.012 0.010 0.093 0.017 0.014 
X8 0.140 0.077 0.140 0.000 0.156 0.091 0.157 0.002 0.121 0.102 0.121 0.001 
X9 0.140 0.060 0.140 0.000 0.174 0.078 0.174 0.000 0.150 0.097 0.150 0.000 
X10 0.306 0.148 0.306 0.000 0.375 0.186 0.385 0.009 0.357 0.265 0.357 0.000 
X11 0.022 0.152 0.025 0.002 0.025 −0.149 0.028 0.002 0.029 0.336 0.033 0.003 
X12     0.018 0.017 0.020 0.004     
X13     0.014 0.030 0.014 0.000     
X14     0.010 0.079 0.010 0.000     
X15     0.001 −0.011 0.001 0.000     
X16         0.030 0.053 0.030 0.000 
X17         −0.009 −0.013 0.010 0.002 
X18         0.008 0.012 0.008 0.000 
X24 −0.074 −0.025 0.074 0.000 −0.085 −0.030 0.085 0.000 −0.050 −0.026 0.050 0.000 
X25 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 
X26     0.000 −0.009 0.000 0.000     
  PP PF PP PF PP PF 

Light grey highlighted and bold: important; dark grey highlighted and italic: non-influential.

Figure 5

Morris screening relative sensitivity μ* versus linearity σ for Alt-2 TC.

Figure 5

Morris screening relative sensitivity μ* versus linearity σ for Alt-2 TC.

Parameter X5 (number of boreholes) involved only in Alt-1, was found to be non-influential for TC but not for RV. On the other hand, X12 (number of lake water extractions), X14 (number of flocculation-sedimentation plants) and X15 (number of surface water treatment plants) only involved in Alt-2, are determined to be non-influential for both TC and RV. From Alt-3 specific parameters, X16 (number of spring water extractions), X17 (hydraulic conductivity) and X18 (water tank elevation) were classified as non-influential.

It is well understood that X25 (chlorine dosage) and X26 (alum dosage) are operation cost items, thus both parameters are categorized as non-influential for alternatives' RV, but not for TC, where operation costs are included.

The low σ value of all technological parameters indicates the linear behaviour and no significant interaction of parameters for all alternatives' TC and RV (Tables 4 and 5).

Parameter prioritization and parameter fixing

If the parameter is classified as non-influential by Morris sensitivity indices , it is unlikely to be identified as an important parameter both by derivative-based and SRC sensitivity methods (Saltelli et al. 2008). Sensitivity indices obtained from this study were also able to prove this agreement. For a parameter to be classified as a prioritized parameter, it is enough to be identified as important either by the derivative-based or SRC method (Table 6). A parameter with low importance according to factors of prioritization does not necessarily imply the parameter to be non-influential (factor fixing). Parameter X24 (diameter distribution) was seen to be neither important nor non-influential for all alternatives' RV (Table 5).

Table 6

Summary of parameter prioritization and parameter fixing

Important (parameter) Non-influential (parameter fixing) 
Global parameters 
X1a,b Period of consideration       
X2a,b Net interest rate       
X3a Expected annual growth rate       
User-technological parameters 
X4a Ground water depth X6 Number of disinfectant plant 
X5b Number of borehole X7 Number of pumping station 
X8a,b Pressure head X11 Number of surface water tank 
X9b Trench depth X12 Number of lake water extraction 
X10a,b Pipe length X14 Number of flocculation sedimentation plant 
X13a Pump head at intake X15 Number of surface water treatment plant 
X19a,b Borehole diameter X16 Number of spring water extraction chamber 
X20a Kind of disinfectant X17 Hydraulic conductivity 
X22a Kind of flocculant X18 Water tank elevation 
X23a Number of water line X21 Pump installation 
Internal parameters 
X25a Chlorine dosage X24 Diameter distribution 
X26a Alum dosage       
Important (parameter) Non-influential (parameter fixing) 
Global parameters 
X1a,b Period of consideration       
X2a,b Net interest rate       
X3a Expected annual growth rate       
User-technological parameters 
X4a Ground water depth X6 Number of disinfectant plant 
X5b Number of borehole X7 Number of pumping station 
X8a,b Pressure head X11 Number of surface water tank 
X9b Trench depth X12 Number of lake water extraction 
X10a,b Pipe length X14 Number of flocculation sedimentation plant 
X13a Pump head at intake X15 Number of surface water treatment plant 
X19a,b Borehole diameter X16 Number of spring water extraction chamber 
X20a Kind of disinfectant X17 Hydraulic conductivity 
X22a Kind of flocculant X18 Water tank elevation 
X23a Number of water line X21 Pump installation 
Internal parameters 
X25a Chlorine dosage X24 Diameter distribution 
X26a Alum dosage       

aDenotes alternatives’ TCs.

bDenotes alternatives’ RVs.

In general, by applying the three complementary SA methods, one internal and 10 user-technological parameters were identified as non-influential. While the analysis determined two internally fixed (i.e., chlorine dosage and alum dosage), three global and ten user-technological parameters were classified as important parameters for the LCC of water supply system.

CONCLUSIONS

Even if the result found from the three methods shows some discrepancy, reasonable input parameter sensitivity classification has been achieved by complementing the methods. Relatively, the Morris screening method gave a better overview about parameter non-linearity and/or interactions than derivative-based or SRC methods. Parameter non-linearity/interaction is found to significantly contribute to the uncertainty of the tool output. During water supply system planning the highest attention should be placed on the accuracy of the important parameters' value. Special care is also needed for the selection of period of consideration because it interacted with most of the other parameters.

From the CLARA-SPT SA results, input parameters X17 (hydraulic conductivity), X18 (water tank elevation), X21 (pump installation) and all other non-influential parameters can be fixed to a certain value without causing any significant influence on the LCC of water supply systems (Table 6). By doing this, it is possible to reduce the number of unknown variables which will help to further simplify the CLARA-SPT in the future.

In contrast, X25 (chlorine dosage) and X26 (alum dosage) were identified to be important, while the CLARA-SPT internally fixed them into 2 mg/l and 30 mg/l, respectively. Therefore, these two parameters should be made available as accurate values to make realistic LCC estimation of water supply systems.

In general, this study clearly shows the level of sensitivity of water supply input parameters that contributes to further improvement of CLARA-SPT computational efficiency.

ACKNOWLEDGEMENTS

The authors would like to thank the CLARA project for allowing full access to the CLARA simplified planning tool. We acknowledge the APPEAR programme (Austrian Partnership Programme in Higher Education and Research for Development) for providing financial support for the first author.

REFERENCES

REFERENCES
Campolongo
F.
Cariboni
J.
Saltelli
A.
2007
An effective screening design for sensitivity analysis of large models
.
Environ. Modell. Softw.
22
,
1509
1518
.
Casielles Restoy
R.
Lechner
M.
Langergraber
G.
2014
CLARA Simplified Planning Tool v1.5, User Manual. http://clara.boku.ac.at/index.php/planning-tool-2 (accessed 19 December 2014)
.
CSA
2008
Summary and statistical report of the 2007 population census and population size by age and sex
. In:
Commission FDRoEPC
(ed.),
Central Statistical Authority of Ethiopia
,
Addis Ababa
,
Ethiopia
.
Farr
J. V.
2011
System Life Cycle Costing: Economic Analysis, Estimation, and Management
.
Taylor & Francis
,
Boca Raton, FL
,
USA
.
Jagals
P.
Rietveld
L.
2011
Estimating costs of small scale water supply interventions
. In:
Valuing Water, Valuing Livelihoods: Guidance on Social Cost-benefit Analysis of Drinking-water Interventions, with Special Reference to Small Community Water Supplies
, (
Cameron
J.
Hunter
P
Jagals
P.
Pond
K.
, eds),
IWA Publishing
,
London
,
UK
, pp.
149
166
.
Kutner
M. H.
Nachtsheim
C. J.
Neter
J.
Li
W.
2005
Applied Linear Statistical Models
.
McGraw-Hill
,
New York
,
USA
.
Lechner
M.
Pressl
A.
Langergraber
G.
2014
The CLARA simplified planning tool
.
Sustain. Sanit. Pract.
19
,
29
35
.
http://www.ecosan.at/ssp (accessed 19 December 2014)
.
MOWRD
2002
Ethiopian Guidelines Specification for Drinking Water Quality
.
Ethiopian Ministry of Water Resource Development
,
Addis Ababa
,
Ethiopia
.
MUDHCo
2014
Bahir Dar City Administration. http://www.mwud.gov.et/web/bahirdar/home (accessed 19 December 2014)
.
Pearce
D.
Atkinson
G.
Mourato
S.
2006
Cost-Benefit Analysis and the Environment: Recent Development
.
Organisation for Economic Co-operation and Development (OECD)
,
Paris
,
France
.
Saltelli
A.
Karen
C.
Scott
E. M.
2000
Sensitivity Analysis
.
John Wiley & Sons
,
New York
,
USA
.
Saltelli
A.
Tarantola
S.
Campolongo
F.
Ratto
M.
2004
Sensitivity Analysis in Practice, Guide to Assessing Scientific Models
.
John Wiley & Sons
,
Chichester
,
UK
.
Saltelli
A.
Ratto
M.
Andres
T.
Campolongo
F.
Gatelli
J. C. D.
Saisana
M.
Tarantola
S.
2008
Global Sensitivity Analysis: The Primer
.
John Wiley & Sons
,
Chichester
,
UK
.
Sun
X. Y.
Newham
L. T. H.
Croke
B. F. W.
Norton
J. P.
2012
Three complementary methods for sensitivity analysis of a water quality model
.
Environ. Modell. Softw.
37
,
19
29
.
TCE
2009
Bahir Dar Water Supply and Sanitation System Feasibility Study and Preliminary Engineering Design Report
.
Tropics Consulting Engineers PLC
,
Addis Ababa
,
Ethiopia
.
W/Yohannes
S. M.
2010
Numerical groundwater flow modeling of the lake Tana basin, upper Nile, Ethiopia
.
Msc thesis
,
Department of Earth Science, Addis Ababa University
,
Addis Ababa
,
Ethiopia
.

Supplementary data