3 Method and case studies
3.3 Modelling approach
For the ceMFA mass, nutrient and energy flows as well as cost estimates are transferred into mathematical models through the formulation of system equations. For the two case studies Hamburg and Arba Minch two different approaches for setting up the system models are followed. For Hamburg, every system is represented in one particular equation system. This is done since the systems differ relatively widely. For Arba Minch, there is only one equation system set up for all systems, which are then specified by system parameters that are introduced to reflect the system characteristics.
The latter approach has the advantage that different systems and transition options can be directly compared, using only one equation system for the modelling.
In general, the selection of the system boundaries is crucial for a system analysis.
System boundaries refer to spatial and temporal boundaries as well as to the processes that are taken into account (Sonesson et al., 1997). The selection of the system boundaries impacts on the results and their interpretations, as well as on the data requirements (Dalemo et al., 1997). Thus, they need to be carefully defined on the one hand so as to not to limit the system analysis and to allow a wide view on the respective topics. But on the other hand, the scope of the analysis still needs to be manageable in terms of data collection and the amount of required work.
The systems are made up of relevant processes and flows. Processes can be defined as any activities that transform, transport or store materials. Flows are defined as the links
between the processes. The processes that are included in the system boundaries must reflect the whole chain of actions for the provision of a certain service, i.e. in this study the urban water and wastewater system with integrated nutrient recovery. The following general processes are therefore included in the system boundaries47:
water supply
rainwater management
domestic sanitation, i.e. wastewater management including the management of residues such as sludge
management of organic solid waste
agriculture (water and nutrient management)
As spatial system boundaries for both cases, Hamburg and Arba Minch, the administrative town borders are taken into account. However, referring to the Process Agriculture the system boundaries are extended, since the agricultural areas within the towns’ borders are relatively small and not sufficient to feed the whole population.
Therefore, an area large enough to feed the population of the respective city is considered, i.e. a concept of hinterland is introduced. This is done in order to minimise exports and imports across the system boundaries, and to highlight the urban population’s food and fertiliser requirements. The hinterland area is calculated according to the specific food demand and corresponding area required for crop production; this area differs for the cases of Germany and Ethiopia. This calculation is explained in detail in Sections 4.1.1 and 5.1.1.
The processes and flows, i.e. the variables identified in the system definitions, are linked in a set of equations. Initially, balance equations of the different processes are set up, defining the input flows, outputs flows and stock rate changes of every process. In general, this follows the law of mass conservation by adhering to the following relationship:
inflows + production = outflows + accumulation (3‐2)
Production and accumulation can be combined as stock rate change, which results in the following general mathematical expression for the balance equations (Equation 3‐3).
A more detailed description of the derivation of system equations can be found in the
47 Industry is not included in order to disregard peculiarities and to increase comparability and
generalisability. In addition, it is argued that industrial water and wastewater management can be considered in a relatively detached manner from domestic water, wastewater and nutrient management.
report by Baccini and Bader (1996). The algorithm, which SIMBOX uses for solving the
M: stock rate change (i.e. accumulation or degradation)48 A: flow
n: substances, i.e. mass, water, N, P, K, S, C i: input
j: balance process k: output
Next, model equations are defined to represent the behaviour of the system in a mathematical way. Processes either modify the material under consideration (i.e.
include stock rate changes such as accumulation or depletion), or they are transfer processes. Parameters are used to describe the key characteristics of the systems. For example, transfer coefficients describe the partitioning of a substance in a process, i.e.
the transfer of inputs to outputs. This is defined for each output of a process and is not
tc: transfer coefficient, with
∑
, =1k n
k
tcj
In addition, a substance flow (e.g. nutrient flow) induced by the flow of a good (e.g.
wastewater) can be calculated by multiplying substance concentrations with the mass flow of the good as per Equation 3‐5.
If substance flows are directly available (e.g. nitrogen load in excreta per person and day), these are also used directly or they are appropriately converted49 into total flows.
Modelling of energy and costs is interlinked with the modelled physical flows, i.e. mass or nutrient flows. This means that energy and cost variables are introduced and equations are defined by which energy and cost flows can either be calculated directly (i.e. in the case of fixed costs or energy demand) or as dependant on their relationship with mass and nutrient variables (i.e. variable costs and energy demand).
The modelling is complemented by an uncertainty analysis based on Gauss’ law of error propagation using Taylor series. For this, parameters are assumed to show normal distribution (Gaussian distribution) and small uncertainties. Gaussian error propagation is then used to determine the error or uncertainty of the variables produced by the interacting parameters. The results of the uncertainty analysis are indicated as error margins of the variables. In addition, SIMBOX allows Monte Carlo simulations to determine uncertainties resulting from parameters with probability distributions other than a Gauss distribution. Therefore, the distribution of every parameter is assigned as being normal, lognormal or uniform; this is done by applying best knowledge for each parameter50. Standard deviations (or minimum/maximum values for a uniform distribution) are selected. Monte Carlo samples including random numbers are generated and applied to the parameters, followed by simulation runs.
This allows a more precise calculation of the probability distribution of the variables, provided that good knowledge about the distribution of the parameters is available. For the purpose of this analysis Monte Carlo simulations are carried out for every equation system and the resulting uncertainties are compared with the uncertainties based on normal distributions.
Sensitivity analyses are carried out to determine the most sensitive parameters. That means parameters are identified that contribute to a high degree of uncertainty in important variables. The data quality of these parameters can then be refined in order to decrease the uncertainty of the variables. In addition, the range of the change in a variable when the parameters affecting it are changed, is evaluated. The sensitivity analysis calculates the first order changes in variables due to changes in parameters.
“Relative sensitivities per 10%” (see Equation (3‐6) which is based on work of Baccini and Bader, 1996) can be considered high if the relative change in the variable is higher than the assumed relative change in the parameter. Therefore, the sensitivity analysis is
49 Conversions include for example, multiplication by the number of persons or the number of days in a
year.
50 Truncated normal and truncated lognormal distributions can also be assumed.
used to identify key parameters and to get a better understanding of the system as a
Xi: variable pj: parameter
Furthermore, parameter variations are included in the analysis to show the effect of parameters varying over a larger range. This allows the evaluation of measures with regard to their impact on specific flows or other variables such as specific costs or energy demand.