The Morris screening sensitivity analysis (SA) has been used to assess how the uncertainty of input parameters influences the output of the CLARA Simplified Planning Tool (CLARA-SPT). To assess the sensitivity of the tool, four hypothetical waste collection and treatment alternatives, which planned to serve 10,000 people, have been proposed and analysed. These alternatives are (A1) dry sanitation with urine diversion dry toilets (UDDTs), (A2) water-aided sanitation with decentralised treatment units, (A3) water-aided sanitation with central technical treatment and (A4) water-aided sanitation with cesspits. The SA was used to identify the influence of two global and 29 technological input parameters on lifetime costs and residual values of sanitation alternatives. The top two important parameters identified for each alternative are: ‘type of urine transport’ and ‘persons using one UDDT’ for alternative A1, ‘persons served per septic tank’ and ‘required surface area for vertical flow constructed wetland’ for alternative A2, ‘daily diesel generator working hours’ and expected annual growth’ for alternative A3 and ‘cesspit volume’ and ‘expected annual growth’ for alternative A4. Additionally, the Morris SA identified non-linearity and/or parameter interaction response. The SA of the specified alternatives shows that from the 29 technological parameters investigated, a subset of 14 important parameters need estimates that are more accurate, whereas a subset of 15 non-influential parameters can be fixed to a certain value. In particular, two parameters (i.e. cesspit volume and persons using one UDDT) that have been internally fixed in the SPT were found to be important and thus should be made available as input parameters to the user. Overall, the study provides guidance for further modification and simplification of the CLARA-SPT.

INTRODUCTION

Computer models/tools are routinely used to understand, design and manage water resource systems. This is due to their ability to simulate the complexity of water systems and compute high levels of data (Jain & Singh 2003). These models often contain a series of complicated mathematical functions that correlate certain inputs with outputs, which makes it difficult to evaluate the quantitative performance of models (Song et al. 2012). One of the reasons for applying sensitivity analysis (SA) is to examine how the model output is influenced by the effect of different sources of uncertainty in the model inputs (Saltelli et al. 2008). Through the process, the parameters' relative importance can be assessed. The SA can be classified as local, screening and global methods, based on how the input parameters are perturbed over the parameter space (Saltelli 2000; Saltelli et al. 2009). The local SA method simply changes one parameter at a time (OAT) while holding all others at their original value, and measures the derivatives of a change at a single location in parameter space. The screening technique originally proposed by Morris (1991) and further extended by Campolongo et al. (2007) also applies the principle of OAT, but assessment relies on a discrete partitioning of the parameter space. The screening method derives sensitivity indices from a number of incremental ratios of each input parameter with corresponding output changes, called elementary effects. Global SA focuses on how the model output is influenced by individual or paired input parameters over the entire parameter space (Homma & Saltelli 1996; Saltelli et al. 2004). In the sector of water supply and sanitation (WS&S), there is often undue emphasis on selecting appropriate systems for long-term system sustainability. This conventional thinking usually leads to much attention put on the construction phase of the system, while the operating life receives too little attention. The approach makes operation and maintenance of the selected system more difficult, or impossible (Howe & Dixon 1993). Especially developing countries, lacking life cycle-based WS&S planning tools, are facing serious problems in identifying optimal and sustainable solutions toward the improvement of WS&S accessibility (McConville & Mihelcic 2007). Recently, the CLARA Simplified Planning Tool (CLARA-SPT) has been developed to bridge the need for life cycle cost (LCC)-based economic comparison among WS&S alternatives (Lechner et al. 2014). The tool was developed for five African countries: Burkina Faso, Ethiopia, Kenya, Morocco and South Africa. The calculation of costs is based on technologies’ bill of quantities and country-specific unit prices. The tool does not consider environmental, social and health aspects explicitly, since those issues are addressed through the framework of countries’ regulations and standards. On such regard, all comparable systems are expected to achieve the country's legal and client's requirements and environment and health benefits and to be socially acceptable (Lechner et al. 2014). The tool consists of a series of mathematical functions that correlate global and technological input parameters with output parameters. Some of the input parameters are net interest rate, served population, and cesspit volume, whereas output parameters include investment cost, annual operation and maintenance cost, reinvestment cost and residual value of alternatives. Since the recent launch of the tool, the sensitivity of the tool results for various sources of input parameter uncertainty has never been tested. In this paper, the Morris screening method was employed to assess the tool sensitivity, since the method is suitable for the type of model having considerably high numbers of uncertain parameters or consuming considerable time and money to compute (Campolongo et al. 2007). The SA plays an important role in identifying the possible improvements of the tool (for instance more simplification by fixing non-influential parameters) at this early stage of the tool development (Beck 1983; Farr 2011). The main objective of this study is then to assess sensitivity of the CLARA-SPT outputs for the uncertainty of sanitation input parameters. This leads to the identification of important and non-influential input parameters according to the defined hypothetical case study.

METHODS

Sanitary systems case study

To test the sensitivity of CLARA-SPT, four hypothetical sanitation alternatives were created. Each alternative in this case study serves 10,000 people. The economic lifetime for this example is 25 years. The four sanitation alternatives labelled as A1, A2, A3 and A4 are:

  • (A1) dry sanitation with urine diversion dry toilets (UDDTs);

  • (A2) water-aided sanitation with decentralised treatment units;

  • (A3) water-aided sanitation with central technical treatment; and

  • (A4) water-aided sanitation with cesspits.

Table 1 shows each alternative technology used to collect, transport and treat municipal waste up to WHO standards either to reuse the effluent for agriculture or for safe disposal to the environment (World Health Organization 2006). A detailed description of the alternatives including the technologies used and the input parameters for each alternative is provided in the supplementary material (available online at http://www.iwaponline.com/wst/071/497.pdf).

Table 1

Technologies used in the sanitation chain of each alternative

 
 

CLARA-SPT computational setting

Sanitation alternatives that were purpose-built for SA of the CLARA-SPT comprise two functional groups of waste collection and waste treatment. Principally, the CLARA-SPT is developed to calculate LCC of alternatives, but, as a prerequisite, initial and recurrent costs of all embedded technologies of each alternative are internally computed. This is followed by net present value (NPV) computation of future costs and residual value (i.e. operation and maintenance cost (O&M cost), reinvestment cost (RIC) and salvage/residual value) of included technologies. Finally, the tool linearly aggregates initial investment cost, NPV of O&M cost and NPV of RIC of all embedded technologies in order to get total costs of the alternative. The NPV of the residual value after 25 years of service is presented as the remaining asset value of the system (Casielles Restoy et al. 2014). The SA of the CLARA-SPT involves two global parameters (i.e. net interest rate and expected annual growth) and 29 technological parameters (whereby 24 are user input parameters and five are internally fixed). The description of the input parameters is shown in Table 2. Global parameters are general variables, which are able to influence the cost of all technologies included in the alternatives. Whereas technological parameters are variables that only influence hosting technology cost, irrespective of others. Internal parameters are built-in design assumptions of various technologies during the tool development process (for example, the cesspit volume was set to a fixed value). Those internal parameters are neither visible nor changeable by the tool user.

Table 2

Parameters' label, description, unit, references value and variation range

Label Description Unit Reference value Variation range Label Description Unit Reference value Variation range 
Global parameters Technological parameters (user input) 
X1 Net interest rate −3–6 X17 Urine emptying interval day 14 7,14,30 
X2 Expected annual growth 0–8 X18 Type of faeces transport – Small truck Donkey cart, small truck, big truck 
Technological parameters (user input) X19 Type of urine transport – Vacuum truck Vacutug, vacuum truck 
X3 Distance to treatment site km 0.5–12 X20 Precipitation agent type – Magnesium oxide Bittern, magnesium oxide, magnesium sulphite 
X4 Number of HFCWs and VFCWs Number 50 2–100 X21 Type of faecal sludge transport – Vacuum truck Vacutug, vacuum truck 
X5 Average trench depth 1.4 0.5–2.8 X22 Type of intermittent loading (VFCW) – Siphon Siphon, pump 
X6 Required area of HFCW per person m2 1.5 1–3 X23 Screen type – Fine Fine, coarse 
X7 House connections with manhole 30 0–60 X24 Jet aerator – No Yes, no 
X8 House connection pipe length 2.5–10 X25 Treatment level – DN DN, nitrification, DN+ ASS 
X9 Persons served per septic tank Number 20 5–2,000 X26 Daily diesel generator working hours 0–4 
X10 TS content (primary sludge) 5–10 Technological parameters (internal assumption) 
X11 Required area of VFCW per person m2 1–4 X27 Cesspit volume m3 18 4.5–36 
X12 Length of sewera 300, 400, 17,760, 300 100–27,000 X28 Ratio of UDDT material to bulk material Number 0.5–2 
X13 Number of pumping stations Number 1–6 X29 Ratio of dewatered sludge to bulk material Number 0.67 0.3–1.4 
X14 Pumping station pressure head 10 5–15 X30 Persons using one UDDT Number 15 8–25 
X15 Price of bulk material €/kg 0.2 0.1–0.4 X31 Persons per sewer connection Number 15 5–25 
X16 Number of sludge drying beds Number 1–4      
Label Description Unit Reference value Variation range Label Description Unit Reference value Variation range 
Global parameters Technological parameters (user input) 
X1 Net interest rate −3–6 X17 Urine emptying interval day 14 7,14,30 
X2 Expected annual growth 0–8 X18 Type of faeces transport – Small truck Donkey cart, small truck, big truck 
Technological parameters (user input) X19 Type of urine transport – Vacuum truck Vacutug, vacuum truck 
X3 Distance to treatment site km 0.5–12 X20 Precipitation agent type – Magnesium oxide Bittern, magnesium oxide, magnesium sulphite 
X4 Number of HFCWs and VFCWs Number 50 2–100 X21 Type of faecal sludge transport – Vacuum truck Vacutug, vacuum truck 
X5 Average trench depth 1.4 0.5–2.8 X22 Type of intermittent loading (VFCW) – Siphon Siphon, pump 
X6 Required area of HFCW per person m2 1.5 1–3 X23 Screen type – Fine Fine, coarse 
X7 House connections with manhole 30 0–60 X24 Jet aerator – No Yes, no 
X8 House connection pipe length 2.5–10 X25 Treatment level – DN DN, nitrification, DN+ ASS 
X9 Persons served per septic tank Number 20 5–2,000 X26 Daily diesel generator working hours 0–4 
X10 TS content (primary sludge) 5–10 Technological parameters (internal assumption) 
X11 Required area of VFCW per person m2 1–4 X27 Cesspit volume m3 18 4.5–36 
X12 Length of sewera 300, 400, 17,760, 300 100–27,000 X28 Ratio of UDDT material to bulk material Number 0.5–2 
X13 Number of pumping stations Number 1–6 X29 Ratio of dewatered sludge to bulk material Number 0.67 0.3–1.4 
X14 Pumping station pressure head 10 5–15 X30 Persons using one UDDT Number 15 8–25 
X15 Price of bulk material €/kg 0.2 0.1–0.4 X31 Persons per sewer connection Number 15 5–25 
X16 Number of sludge drying beds Number 1–4      

a‘Length of sewer’ is used in all four alternatives; its reference values represent the value for A1, A2, A3 and A4, respectively.

DN: denitrification, DN + ASS: denitrification and aerobic sludge stabilization, HFCW: horizontal flow constructed wetland, TS: total solids, UDDT: urine diversion dry toilet; VFCW: vertical flow constructed wetland.

Reference values in Table 2 for the global parameters were arbitrarily defined and technological input parameters were estimated using the descriptions provided in the CLARA-SPT user manual (Casielles Restoy et al. 2014). Starting from these reference results, Morris screening SA was employed to assess sensitivity of the CLARA-SPT results. In this study, the individual input parameter was perturbed OAT at a proportion of ±10, ±20, ±50 and ±100% from its reference value. This is a systematic procedure to deal with the conceptual uncertainty of parameters, scientific assumptions embedded into the tool and input–output correlations (Sun et al. 2012).

Morris screening SA method

The Morris screening SA method is an economical and qualitative method that relies on sensitivity measure through partitioning of the parameter space. Morris SA provides easily interpreted sensitivity indices and is used to estimate sensitivity of a large number of parameters at low computational cost (Saltelli et al. 2004). But, the method merely provides relative importance of parameters by ranking the input parameters based on importance and non-linearity and/or interaction (Alam & McNaught 2004). Moreover, this method is applied to screen out the non-influential parameters, which helps to reduce the parameter space and complexity of the SA. The Morris method presents a measure of sensitivity by computing elementary effect di. The elementary effect of the ith parameter (di(xi)) represents the relative difference between the tool output obtained after parameter perturbation by Δ, y(x1,…, xi−1, xi + Δ, xi+1,…,xk), and the output for the given reference parameter value y(x) (Equation (1)). Each input parameter of the tool xi, i = 1,…, k is assumed to vary across p number of levels in the parameter space. 
formula
1
where x1,…, xi−1, xi, xi+1,…xk are parameters sampled from K dimensional, p-level grid parameter space. As the elementary effect di depends on the location of the random sample, r replicates are computed for each parameter, and the cumulative distribution of the r elementary effects is analysed. Each parameter's is summarised by mean (μi) and standard deviation (σi) to obtain the sensitivity indices (Morris 1991; Alam & McNaught 2004). The μi measures the level of importance of ith parameter in determining the output uncertainty, while σi indicates higher-order effect of the parameter that is non-linearity and/or interaction of a parameter with others. When the model is non-monotonic, the value of μi is affected by negative and positive values of di, which might even each other out. Hence to avoid the problem of opposite sign effect, it is suggested by Campolongo et al. (2007) to use the mean of the absolute value of the elementary effects (μi*).

Setting of SA cut-off values

It is important to specify a cut-off (threshold) value to be able to classify parameters as being important or non-influential. In this study, the cut-off value of 0.10 (10%) was arbitrarily selected for assessing the alternative's total cost and benefit sensitivity. Hence, the parameter with μ* < 0.10 was classified as being non-influential and parameter with σ > 0.10 indicated the presence of non-linearity and/or interaction of parameters with others. The rationale is that a parameter change result beyond ±10% deviation from the reference output values has been considered as an important effect during economic comparison of alternatives at the preplanning stage. Likewise, when the alternatives' output variation is within ±10% margin, they are not economically distinguished as different options. In general, a higher value of μi* and σi respectively, indicates relative importance of the parameter and responsibility of the parameter for introducing non-linearity and/or interaction on the output uncertainty.

RESULT AND DISCUSSION

Results obtained from the Morris SA are presented in the following sections. Morris sensitivity indices (μ* and σ) of the alternative's costs and residual values are presented in the tables. Important parameters are highlighted in light grey, non-influential parameters are written in italic and parameters that show non-linearity are highlighted in dark grey.

Alternative's total cost SA for global parameters

Table 3 shows the Morris SA results for the global parameters X1 (net interest rate) and X2 (expected annual growth rate). Both parameters are important (μ* > 0.10) but have linear behaviour (σ < 0.10). X1 and X2 merely influenced recurrent costs of alternatives through technologies on CLARA-SPT computational set-up. The recurrent cost share of alternative A1, A2, A3 and A4 is about 64%, 28%, 66% and 32%, respectively, from their respective total costs on the reference/base computation. Due to these facts, sensitivity of X1 and X2 is greater on alternative A1 and A3, where the recurrent cost share is significantly high.

Table 3

SA result of alternatives’ total cost for global parameters (light grey highlighted: important; non-highlighted: linear interaction)

Label Alternatives
 
A1
 
 A2
 
A3
 
A4
 
μσ μσ μσ μσ 
X1 0.275 0.051 0.130 0.025 0.283 0.057 0.130 0.028 
X2 0.357 0.077 0.173 0.043 0.370 0.084 0.175 0.040 
Label Alternatives
 
A1
 
 A2
 
A3
 
A4
 
μσ μσ μσ μσ 
X1 0.275 0.051 0.130 0.025 0.283 0.057 0.130 0.028 
X2 0.357 0.077 0.173 0.043 0.370 0.084 0.175 0.040 

Alternative's total cost SA for technological parameters

It is well understood that not all 29 technological parameters are involved in all alternatives; rather only parameters that are linked with embedded technologies of each alternative (Table 1) are included. For instance, parameter X6 is only used in alternative A1 and A4, where horizontal flow constructed wetlands (HFCWs) are part of the solution, while parameter X12 is used in all alternatives since sewers are used in all four alternatives. In order to display relative sensitivity of parameters for each alternative, the absolute mean value μ* of each parameter elementary effects (abscissa) is plotted against the standard deviation σ (ordinate) of parameter elementary effects in Figure 1. Table 4 shows the Morris SA results for the 29 technological parameters for alternatives’ total costs.

Table 4

Morris SA result of alternative's total cost (light grey highlighted: important, italic: non-influential, dark grey highlighted: non-linearity and/or interactions)

  Alternatives
 
Type Label A1
 
A2
 
A3
 
A4
 
μσ μσ μσ μσ 
Technological parameter X3 0.019 0.024 0.010 0.006   0.003 0.001 
X4 0.038 0.010 0.055 0.003   0.038 0.011 
X5 0.024 0.001 0.050 0.002 0.075 0.002 0.024 0.001 
X6 0.171 0.001     0.169 0.001 
X7   0.011 0.001 0.013 0.001   
X8   0.007 0.007 0.024 0.001   
X9   0.371 0.222     
X10   0.008 0.002     
X11   0.299 0.001     
X12 0.053 0.003 0.063 0.005 0.151 0.003 0.053 0.003 
X13     0.013 0.002   
X14     0.015 0.001   
X15 0.000 0.000   0.001 0.000   
X16   0.007 0.000   0.021 0.004 
X17 0.271 0.383       
X18 0.009 0.013       
X19 0.569 na       
X20 0.017 0.000       
X21   0.074 na   0.121 na 
X22   0.128 na     
X23     0.009 na   
X24     0.036 na   
X25     0.139 0.196   
X26     0.374 0.096   
Internal parameter X27       0.542 0.521 
X28 0.009 0.001       
X29     0.027 0.002   
X30 0.552 0.287       
X31 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 
  Alternatives
 
Type Label A1
 
A2
 
A3
 
A4
 
μσ μσ μσ μσ 
Technological parameter X3 0.019 0.024 0.010 0.006   0.003 0.001 
X4 0.038 0.010 0.055 0.003   0.038 0.011 
X5 0.024 0.001 0.050 0.002 0.075 0.002 0.024 0.001 
X6 0.171 0.001     0.169 0.001 
X7   0.011 0.001 0.013 0.001   
X8   0.007 0.007 0.024 0.001   
X9   0.371 0.222     
X10   0.008 0.002     
X11   0.299 0.001     
X12 0.053 0.003 0.063 0.005 0.151 0.003 0.053 0.003 
X13     0.013 0.002   
X14     0.015 0.001   
X15 0.000 0.000   0.001 0.000   
X16   0.007 0.000   0.021 0.004 
X17 0.271 0.383       
X18 0.009 0.013       
X19 0.569 na       
X20 0.017 0.000       
X21   0.074 na   0.121 na 
X22   0.128 na     
X23     0.009 na   
X24     0.036 na   
X25     0.139 0.196   
X26     0.374 0.096   
Internal parameter X27       0.542 0.521 
X28 0.009 0.001       
X29     0.027 0.002   
X30 0.552 0.287       
X31 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 

na: refers to σ estimation that is not applicable since the parameter perturbation choice is less than three.

Figure 1

Parameter screening and relative sensitivity using Morris SA for A1, A2, A3 and A4 total cost: μ* indicates the importance and σ the degree of non-linearity and/or interaction caused by parameter Xi.

Figure 1

Parameter screening and relative sensitivity using Morris SA for A1, A2, A3 and A4 total cost: μ* indicates the importance and σ the degree of non-linearity and/or interaction caused by parameter Xi.

Parameters X6, X17, X19 and X30 were determined to be important parameters for A1 total cost but only X17 and X30 show non-linear behaviour. For A2 total cost, X9, X11 and X22 were identified to be important parameters, among them only X9 with non-linear behaviour. For the total cost of A3, only X25 shows non-linearity, while X12, X25 and X24 were found to be important parameters. Finally, for A4 total cost, the three parameters X6, X21 and X27 were important whereby only parameter X27 shows non-linearity. The results confirmed that some parameters are important for one alternative's total cost but not for others. This is because of the different cost share and planning set-up of the technologies to which the parameters are linked in the associated alternatives. For example, parameter X12 (sewer length), presented at all four alternatives through sanitary sewer technology, was identified to be an important parameter only for A3, but not for A1, A2 and A4. This is because of:

  1. greater cost share of sanitary sewer in A3 (15% of A3 total costs) compared to A1, A2 and A4 (5, 10 and 5% of respective alternatives total cost);

  2. small-sized and shorter sanitary sewer network layout was planned for A1, A2 and A4 to transport grey and septic effluent to 50 decentralized treatment sites, unlike A3, where the sewer network is planned to transport the entire wastewater to a central treatment plant.

Therefore, from this result it is possible to conclude that X12 became an important parameter for alternatives that contain longer and bigger size sewer networks than the one having a compacted sewer network. Moreover, parameter X21 (type of faecal sludge transport) is found to be important for A4 but not for A2. This can be explained by the fact that the volume of collected sludge from the cesspit in A4 is much more than the primary sludge collected from septic tanks in A2. In all alternatives, parameter X31 (persons per sewer connection), an internal design assumption used for calculating the distribution of pipe diameters for the sewer network, was found to be a non-influential parameter irrespective of the type of sewer layout. This indicates that settlement density variation at a connection point did not cause sewer diameter change for a small sanitary sewer system, such as the one in our example for 10,000 persons.

Alternatives residual value SA

The results of the Morris SA (Table 5) show that X12 (length of sewer) was found to be an important parameter for all alternatives’ residual values, because the predefined lifetime of the sanitary sewer (50 years) is long enough to retain considerable asset value at 25 years. The residual value share of sanitary sewer was identified to be 20%, 30%, 19% and 37% of the total residual value of alternative A1, A2, A3 and A4, respectively. Accordingly, the importance of parameter X12 on the residual value of alternative A2 and A4 (μ* = 0.238 and 0.302) is relatively greater than A1 and A3.

Table 5

Morris SA result of alternative's residual value (light grey highlighted: important, italic: non-influential, dark grey highlighted: non-linearity and/or interactions)

  Alternatives
 
Type Label A1
 
A2
 
A3
 
A4
 
μσ μσ μσ μσ 
Technological parameter X3 0.014 0.034 0.000 0.000   0.000 0.000 
X4 0.020 0.003 0.127 0.002   0.030 0.018 
X5 0.090 0.003 0.191 0.006 0.098 0.002 0.137 0.005 
X6 0.051 0.001     0.079 0.001 
X7   0.045 0.001 0.016 0.001   
X8   0.029 0.026 0.031 0.001   
X9   0.152 0.102     
X10   0.003 0.002     
X11   0.105 0.001     
X12 0.197 0.010 0.238 0.013 0.196 0.003 0.302 0.016 
X13     0.007 0.002   
X14     0.001 0.001   
X15 0.000 0.000   0.000 0.000   
X16   0.002 na   0.015 0.017 
X17 0.377 0.533       
X18 0.026 0.036       
X19 0.027 na       
X20 0.000 0.000       
X21   0.095 na   0.070 na 
X22   0.308 na     
X23     0.002 na   
X24     0.011 na   
X25     0.186 0.262   
X26     0.086 0.059   
Internal parameter X27       0.412 0.479 
X28 0.012 0.002       
X29     0.014 0.001   
X30 0.491 0.281       
X31 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 
  Alternatives
 
Type Label A1
 
A2
 
A3
 
A4
 
μσ μσ μσ μσ 
Technological parameter X3 0.014 0.034 0.000 0.000   0.000 0.000 
X4 0.020 0.003 0.127 0.002   0.030 0.018 
X5 0.090 0.003 0.191 0.006 0.098 0.002 0.137 0.005 
X6 0.051 0.001     0.079 0.001 
X7   0.045 0.001 0.016 0.001   
X8   0.029 0.026 0.031 0.001   
X9   0.152 0.102     
X10   0.003 0.002     
X11   0.105 0.001     
X12 0.197 0.010 0.238 0.013 0.196 0.003 0.302 0.016 
X13     0.007 0.002   
X14     0.001 0.001   
X15 0.000 0.000   0.000 0.000   
X16   0.002 na   0.015 0.017 
X17 0.377 0.533       
X18 0.026 0.036       
X19 0.027 na       
X20 0.000 0.000       
X21   0.095 na   0.070 na 
X22   0.308 na     
X23     0.002 na   
X24     0.011 na   
X25     0.186 0.262   
X26     0.086 0.059   
Internal parameter X27       0.412 0.479 
X28 0.012 0.002       
X29     0.014 0.001   
X30 0.491 0.281       
X31 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 

na: refers to σ estimation that is not applicable since the parameter perturbation choice is less than three.

Particularly, the three technological parameters X12, X17 and X30 were determined to be important parameters for alternative A1 residual value, where most parameters were linearly correlated except X17 and X30. For alternative A2 residual value, unlike its total cost, an additional three parameters (X4, X5 and X12) were identified to be important on top of X9, X11 and X22. This is because of the longer service life of sanitary sewers for X5 and X12; frequent reinvestment of the vertical flow constructed wetland (VFCW) (every 10 years) for X4, X11 and X22; and the highest capital cost share of septic tanks (50% of A2 total cost) for X9. The highest σ value response of X9 on residual value of A2 indicates the response of non-linearity. For residual value of alternative A3, only X25 showed the presence of non-linearity, while X12 and X25 were found to be important parameters. For alternative A4 residual value, parameters X5, X12 and X27 were identified to be important, but only X27 perturbation results in non-linearity and/or parameter interactions effect.

Technology SA

The results in Table A1 (available online at http://www.iwaponline.com/wst/071/497.pdf) clearly show that impact of input parameters on the associated technology costs was much higher than the impact observed on alternatives. Two parameters, X15 (price of bulking material, for composting) and X31 (persons per sewer connection) were identified to be non-influential parameters (μ* < 0.1) for both associated technologies and alternatives. Therefore, it is better to internally fix parameter X15 to a certain average value as X31 has been fixed in the planning tool. The following illustrative examples provide explanations about the effect of some parameters on the cost of related technologies. For instance parameter X3 (distance to treatment site) is a required input data for ‘urine collection, faeces collection’ and ‘faecal sludge collection’ technologies. Despite its importance on the cost of ‘faeces collection’ (μ* = 0.19) and septic sludge collection (μ* = 0.11), the parameter was found to be non-influential for ‘urine collection (μ* = 0.03), ‘cesspit sludge collection (μ* = 0.02), and respective alternatives’ cost and residual value. On the other hand, parameter X26 (daily diesel generator working hours) was identified to be the most important parameter for pumping station total cost (μ* = 12.01), where the diesel generator operated only for 2 hours. This is due to the fact that every working hour of the diesel generator cost 56% more than the cost of a pumping station operated with electric power (Ketema et al. 2014). This parameter is also an important parameter for alternative A3 (μ* = 0.37); however, the effect on total costs is much lower. The case of parameter X25 (treatment level), shows different influence levels of a parameter on affiliated technologies (Figure 2). Parameter X25 was found to be important only for total costs and residual values of ‘buffer tank’, ‘SBR’ (sludge blanket reactor) and belt filter press, whereas it was found to be not influential on total costs and residual value of ‘sludge thickener’ and ‘composting’. In addition, parameter X25 is important for the total cost and residual value of the whole alternative A3.

Figure 2

Parameter X25 relative sensitivity on related technologies and on alternative A3: μ* indicates the importance and σ the degree of non-linearity or interaction.

Figure 2

Parameter X25 relative sensitivity on related technologies and on alternative A3: μ* indicates the importance and σ the degree of non-linearity or interaction.

SUMMARY AND CONCLUSIONS

SA is usually employed to assess the model output sensitivity to individual parameters and to assess the influence of parameter interactions for a wide range of parameter space of a model. In this study, Morris SA was performed for the CLARA-SPT to highlight the relative effect of various sanitation parameters on the lifetime cost and residual value of alternatives. When applied to CLARA-SPT, the method detected important parameters that need more accurate estimates, and non-influential parameters which open up the ground to further simplify the tool by fixing those parameters. In reference to the threshold value of 10%, 15 technological parameters have been identified as non-influential (factor fixing), where as the remaining 14 technological parameters have been identified to be important (factor prioritization) (Table 6). These parameters predominantly influence the lifetime costs and residual values of sanitation alternatives, and hence these parameters should be known in more detail to make reliable economic comparison of alternatives.

Table 6

Important and non-influential technological input parameters

Important
 
Non-influential
 
User input parameters User input parameters 
X4a Number of HFCWs and VFCWs X3 Distance to treatment site 
X5a Average trench depth X7 House connections with manhole 
X6b Required area of HFCW per person X8 House connection pipe length 
X9a,b Persons served per septic tank X10 TS content (primary sludge) 
X11a,b Required area of VFCW per person X13 Number of pumping stations 
X12a,b Length of sewer X14 Pumping station pressure head 
X17a,b Urine emptying interval X15 Price of bulk material 
X19b Type of urine transport X16 Number of sludge drying beds 
X21b Type of faecal sludge transport X18 Type of faeces transport 
X22a,b Type of intermittent loading (VFCW) X20 Precipitation agent type 
X25a,b Treatment level X23 Screen type 
X26b Daily diesel generator working hours X24 Jet aerator 
Internal assumptions
 
Internal assumptions
 
X27a,b Cesspit volume X28 Ratio of UDDT material to bulk material 
X30a,b Persons using one UDDT X29 Ratio of dewatered sludge/bulk material 
  X31 Persons per sewer connection 
Important
 
Non-influential
 
User input parameters User input parameters 
X4a Number of HFCWs and VFCWs X3 Distance to treatment site 
X5a Average trench depth X7 House connections with manhole 
X6b Required area of HFCW per person X8 House connection pipe length 
X9a,b Persons served per septic tank X10 TS content (primary sludge) 
X11a,b Required area of VFCW per person X13 Number of pumping stations 
X12a,b Length of sewer X14 Pumping station pressure head 
X17a,b Urine emptying interval X15 Price of bulk material 
X19b Type of urine transport X16 Number of sludge drying beds 
X21b Type of faecal sludge transport X18 Type of faeces transport 
X22a,b Type of intermittent loading (VFCW) X20 Precipitation agent type 
X25a,b Treatment level X23 Screen type 
X26b Daily diesel generator working hours X24 Jet aerator 
Internal assumptions
 
Internal assumptions
 
X27a,b Cesspit volume X28 Ratio of UDDT material to bulk material 
X30a,b Persons using one UDDT X29 Ratio of dewatered sludge/bulk material 
  X31 Persons per sewer connection 

aImportant for alternatives’ residual values.

bImportant for alternatives’ total cost.

The results confirmed that parameters X27 (cesspit volume) and X30 (persons per UDDT), which are internal parameters fixed to 18 m3 and 15 persons, respectively, are highly influential on the cost and residual value of alternatives. Therefore, it would be better to allow those parameters to be defined by the tool users for their site-specific accurate values instead of fixing them. On the other hand, fixing non-influential parameters to a certain value (for instance, X15 – price of bulking material), which hardly influences the costs of alternatives and has minimal effect on technologies, could be helpful to reduce the number of input parameters required and thus the complexity of the tool. We also conclude that the focus of parameter sensitivity assessment should be shifted from cost and residual value of technologies to the entire sanitation chain economic value. This provides a broader spectrum of system expenses and revenues to achieve the defined sanitation standards. Such an approach helps planners and decision makers to identify the important parameters, which influence economic performance of systems, to optimize the effect. Generally, the study outputs are found to be valuable to further simplify the CLARA-SPT. The interaction among parameters, which has an effect on the tool output uncertainty, should be taken into account during future rectification of the tool. SA of water supply parameters of CLARA-SPT is recommended for future study, since this study only addresses sensitivity of sanitation technology input parameters.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the CLARA project (Capacity-Linked water supply and sanitation improvement for Africa's peri-urban and Rural Areas; Contract # 265676; duration: 1 March to 28 February 2014), a collaborative project funded within the EU 7th Framework Programme Theme ‘Environment (including Climate Change)’. The CLARA-SPT that was developed within this project is available for download for free from http://clara.boku.ac.at/index.php/planning-tool-2 (accessed 27 May 2014). We thank the APPEAR program (Austrian Partnership Programme in Higher Education and Research for Development) for providing financial support for the PhD study of Mrs Atekelt Abebe Ketema.

REFERENCES

REFERENCES
Alam
F. M.
McNaught
K. R.
2004
Using Morris’ randomized OAT design as a factor screening method for developing simulation metamodels
. In:
Proceedings of the 36th Conference on Winter Simulation Conference
(
Ingalls
R. G.
Rossetti
M. D.
Smith
J. S.
, eds),
Washington, DC, USA
, pp.
949
957
.
Beck
M. B.
1983
Sensitivity analysis, calibration and validation
. In:
Mathematical Modelling of Water Quality, Streams, Lakes and Reservoirs
(
Orlob
G. T.
, ed.).
Wiley
,
Chichester, UK
.
Campolongo
F.
Cariboni
J.
Saltelli
A.
2007
An effective screening design for sensitivity analysis of large models
.
Environmental Modelling & Software
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 20 May 2014)
.
Farr
J. V.
2011
System Life Cycle Costing: Economic Analysis, Estimation, and Management
.
Taylor & Francis
,
Boca Raton, FL, USA
.
Homma
T.
Saltelli
A.
1996
Importance measures in global sensitivity analysis of nonlinear models
.
Reliability Engineering and System Safety
52
,
1
17
.
Jain
S. K.
Singh
V. P.
2003
Water Resources Systems Planning and Management
.
Elsevier
,
Amsterdam, The Netherlands
.
Ketema
A. A.
Dalecha
T.
Assefa
E.
2014
Water supply and sanitation system preplanning and CLARA Simplified Planning Tool Application for Arba Minch, Ethiopia
.
Sustainable Sanitation Practice
19
(April),
44
52
.
http://www.ecosan.at/ssp (accessed 23 May 2014)
.
Lechner
M.
Pressl
A.
Langergraber
G.
2014
The CLARA Simplifying Planning Tool
.
Sustainable Sanitation Practice
19 (April)
,
29
35
.
http://www.ecosan.at/ssp (accessed 23 May 2014)
.
Morris
M.
1991
Factorial sampling plans for preliminary computational experiments
.
Technometrics
33
,
161
174
.
Saltelli
A.
2000
What is sensitivity analysis? In
:
Sensitivity Analysis
(
Chan
K.
Scott
E. M.
, eds).
Wiley
,
Chichester, UK
.
Saltelli
A.
Chan
K.
Scott
E. M.
2009
Sensitivity Analysis
.
Wiley
,
Chichester, UK
.
Saltelli
A.
Tarantola
S.
Campolongo
F.
Ratto
M.
2004
Sensitivity Analysis in Practice, A Guide to Assessing Scientific Models
.
Wiley
,
Chichester, UK
.
Saltelli
A.
Ratto
M.
Andres
T.
Campolongo
F.
Gatelli
J. C. D.
Saisana
M.
Tarantola
S.
2008
Global Sensitivity Analysis: The Primer
.
Wiley
,
Chichester, UK
.
Song
X.
Bryan
B. A.
Paul
K. I.
Zhao
G.
2012
Variance-based sensitivity analysis of a forest growth model
.
Ecological Modelling
247
,
135
143
.
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
.
Environmental Modelling & Software
37
,
19
29
.
World Health Organization
2006
WHO Guidelines for the Safe Use of Wastewater, Excreta and Greywater
. Volume 1:
Policy and Regulatory Aspects
,
World Health Organization
,
Geneva
.