The LC-bioethanol supply chain system is assessed through the development of a spatially explicit model that combines production and logistics. This is based on the modelling approaches commonly applied in the optimisation of multi-site supply chain systems design [15
] and operational planning [17
]. The formulation builds on a model first developed and applied in the context of optimising future hydrogen infrastructures [18
]. The model is formulated as a mixed-integer linear programming (MILP) model in GAMS [19
] and solved to determine cost-optimal supply chain configurations. The modelling approach can be summarised as follows.
Given the following input data:
1) Spatial distribution of biomass supply
2) Spatial distribution of energy demand (ethanol, electricity, heat)
3) Material and energetic requirements of processing steps
4) Technology capital and operating costs
5) Distance, capacity and costs of biomass and ethanol logistics
6) Market structure
a. Hydrated or anhydrous ethanol market
b. Commodity market prices
Determine the optimal:
1) Regional purchase and supply strategy
2) Facility location
3) Facility scale and process-unit composition
4) Logistical interconnectivity and material flows
5) Production costs
The model formulation requires a large amount of information (input data) to be captured analytically within model parameters. The methods used to model process economics (including economies of scale), supply-demand distributions, logistics and processing performance are therefore presented. We also present mathematical formulations that are considered to have a significant impact on the model behaviour or that should be of specific interest to the reader. A concise summary of the full mathematical formulation is provided in Additional file 1
scenarios are developed for feedstock and processing technology.
Capital costs for both processing and logistics units are annualised through a periodic payment of total installed capital cost (C
) as an annuity (R
) as shown in Equation 1. A capital lifetime (n
) is assigned specifically to each system component. A moderate discount rate (i
) of 8% is assumed, representing the risk associated with the return on investment relative to an alternative allocation of capital. This figure is lower than that used by Kaylen et al [10
] (15%) in order to represent the reduced risk in bioethanol investment anticipated under increased oil prices, and in line with increasing fiscal policy support for alternative energy technologies.
Operating costs are assigned on an annual throughput basis (for example, in dollars per odt per year for raw materials). They account for required treatment-specific feedstocks (that is, enzymes, acids, denaturant), utilities (water, electricity, heat), labour, maintenance and overheads. Labour, maintenance and overheads are allocated between process components relative to fraction of total capital cost.
Economies of scale
The economies of scale available in process unit capital and operating costs represent a key cost driver of the spatial system configuration, resulting in a preference (in the absence of other factors) for large, centralised facilities. Process plant economies of scale are typically captured through a continuous power law relating plant scale (P1) and capital cost (C1) through a scaling factor (α) relative to a base case with plant scale P0 and capital costs C0:
Econometric studies are required to determine the scale factor (α). Hamelinck et al [9
] and Wooley et al [14
] identified the scaling factor for individual components for both current and future LC-bioethanol technologies. This facilitates disaggregation of the plant into specific processing steps in order to allow plant scale and process-unit composition at each location to be assessed. Capital cost scale factors are identified as 0.8 for alternative hydrolysis and fermentation technologies [9
]. Economies of scale in operating costs were identified by Kaylen et al [10
]. They identified the significant economies of scale available in administrative plant overheads (α = 0.25) compared with those operating costs linear with production (that is, α = 1 in the case of stoichiometric pretreatment treatment reagents).
The hypothetical geographical area of study is discretised into a grid of homogeneous regions. We always use a 5 × 5 grid, that is, 25 regions. However, the size of each region is varied between 25, 50 and 100 km2. This allows the impact of local supply and demand distributions, system boundaries and the optimal configuration of multiple-plant infrastructures to be assessed at a range of length scales. For example, an optimal plant configuration identified at the 25 km2 scale may not have access to sufficient easily accessible ('endogenous') resources in order to reach a truly 'optimal' plant scale. Furthermore, the optimum at the 25 km2 scale may not remain optimal when the boundary is expanded to encompass a larger region containing additional plants. The optimal scale for assessment, balancing local spatial detail with global plant interactions, can therefore be identified as that scale which first approaches the minimum unit ethanol production cost (dollars per litre). This has important implications in spatially explicit infrastructure modelling wherein spatial resolution represents the dominant computational cost.
Hypothetical demand and supply scenarios are assigned through the specification of rural, semi-rural and urban land-cover types for each region. These are characterised by their agricultural land and population densities (Table ). Values were derived from an assessment of the UK land-cover database [20
] and regressed against population density derived from UK census data [21
]. These values are therefore representative of UK and, more generally, European agricultural conditions. This does not affect the generality of the framework proposed here, as other regions can be considered by using different parameters. The discussion of the results, however, will necessarily be focussed on the UK and EU.
Agricultural land cover and population densities for rural, semi-rural and urban region types
Regional typologies are mapped onto the grid to generate the two 'generic' spatial distributions considered here. The Centralised distribution represents a central urban region with a peripheral semi-rural and rural boundary region. The Corner-Point distribution has the urban region located at the corner of the system, again with a peripheral semi-rural and rural boundary. This imposes a hard boundary on the urban demand epicentre, representative of a coastal or national border. These distributions are presented schematically in Figure .
Spatial distribution scenarios. (a) Centralised. (b) Corner-Point. The size of the circles indicates the magnitude of the demand/supply.
A 10% fractional availability of agricultural land for biomass sourcing is assumed, approximating the current EU set-aside quota. Feedstocks are characterised in terms of their lignin, hemicellulose and cellulose fraction and the higher heating value derived from the component fractions. The Current scenario process feedstock represents a generic crop residue such as wheat straw or corn stover. Harvested yield is assumed at 5 odt ha-1 year-1. The Future scenario feedstock is assumed to represent a high-yielding hybrid poplar. Harvested yield is assumed at 25 odt ha-1 yr-1 for a three-year coppice cycle. A summary of feedstock properties is provided in Table . A farm gate commodity cost of $53.9 odt-1 is assumed for both feedstocks (converted from UK cost data) in order to allow economies of scale and logistics cost drivers to be isolated.
Demand is assumed continuous at 2000 W per capita for electricity and heat and 980 W per capita for gasoline road-transport fuel demand [22
]. Ethanol is assumed as a direct substitute for gasoline energy demand. The potential for heat provision from the ethanol refinery is limited to 10% of the regional heat demand, reflecting network installation and heat loss constraints in radial heat distribution.
The calculation of absolute regional demand requires subsequent allocation of per capita demand to total regional population, itself a function of population density and absolute spatial length scale. Population density was allocated in defining each of the three regional typologies (Table ). Despite projected US and UK population increases of approximately 45% and 7% respectively by 2050, population growth is not considered when developing the Future scenario. The spatial distribution of demand, rather than its absolute magnitude, remains the dominant driver for optimal trade-offs in the system.
A generic process flowsheet for the LC-bioethanol production process is presented in Figure . This represents the network of feedstock, product and intermediate commodities, process technologies and their respective material and energetic interconnectivity. Incorporated process technologies and relative energetic flows are specific to the Current technology scenario.
A process flowsheet for the Current technology scenario. The thickness of each arrow is representative of the relative energy content of that stream.
The process flowsheet is composed of the pretreatment and fermentation process (Process
), which generates a low-density ethanol titre (5.0wt%EtOH
, L) from a biomass feedstock (B). The ethanol titre is concentrated through a purification train consisting of a stripping column (Stripping
) to generate a medium-density intermediate (35.0wt%EtOH
, M), a rectification column (Rectification
) to generate the ethanol-water azeotrope (94.0wt%EtOH
, H) and a membrane purification process (Purification
) to generate the pure, anhydrous ethanol product (P). The Future
scenario eliminates the need for the stripping step as fermentation titres are assumed to approach 35.0wt%EtOH
through developments in microbial resistance to ethanol concentration. In addition, high titres via process intensification of fermentation (for example, fermentation with simultaneous ethanol stripping [23
]) have already been demonstrated.
Stripping (or Rectification in the Future scenario) also generates a silage residue stream (S), which contains the unconverted cellulose and lignin fractions and the process water removed in the stripping column. This is passed to a solids separation unit (Solids Separation) which generates wet fuel (WF) and waste water (W) streams. The wet fuel is subsequently dried (Drying) to generate a dry fuel (DF) which is converted into hot-utility (HU) and electricity (E) in a combined heat and power unit (CHP).
process technology is assumed to represent a single-stage saccharification and fermentation (SSF) process with pretreatment and fermentation conversion efficiencies of 75% and 95%, respectively. Overall conversion was assumed equal for both cellulose and hemicellulose fractions. The Future
processing technology is envisaged to embody the principles of CBP with pre-processing and fermentation conversion efficiencies of 98% and 95%, respectively [9
A transition to CBP technologies will significantly reduce the capital and operating costs of processing (that is, pre-processing, hydrolysis and fermentation) by an estimated 63% (see [9
]). In optimal single-plant systems, increasing unit logistics costs balance against decreasing unit process and capital costs as the scale of the system increases (see [13
] for a more detailed discussion). Thus, a reduction in capital and operating cost intensity, as embodied in the transition to CBP technologies, results in a downsizing of optimal single plants. This effect is countered in this work through the assumption of increased biomass yields per unit area in the Future
scenario; this serves to reduce unit logistics costs.
Feedstock composition affects the relative process energy flows (see Figure ) through the respective allocation of feedstock HHV through pretreatment and fermentation efficiency, relative to each of the cellulose, hemicellulose and lignin fractions (assumed inert) to ethanol and residual fuel process streams. The resultant residual fraction is assumed to be combusted to provide a 25 bar steam input to a hypothetical Rankine cycle. This is designed to incorporate three pass-out turbines each generating power and steam utility at a specific pressure (11, 4 and 1 bar saturated steam). Turbine pressure ratios are scaled to match the internal process hot utility ratio requirements derived from [9
]. Surplus electricity and heat represent valuable revenue streams.
As the front-end (pre-processing) of current processes require a large amount of heat, it is hard to decouple this from the back-end (utility generation), in particular because heat cannot be feasibly transported over large distances. Future pre-processing methods, identified by [9
], apply steam explosion and compressed liquid hot water in order to hydrolyse the cellulose and hemicellulose fractions. These technologies continue to be hot utility intensive and therefore incompatible with Process
decentralisation. The potential for the development of ambient processing is therefore explored. Proposed technologies include CO2
], oxidative delignification (H2
-catalysed enzymatic hydrolysis), and biological pretreatments (a concise review is provided by [25
hot utility requirements are therefore assumed negligible, substantially improving net energetic efficiency.
Logistics encompass all flows of mass and energy within the processing network. While this can be facilitated through pipeline or conveyor on site, it must be expanded to incorporate road, rail, pipeline and cable modes of transportation between
sites (that is, located within different regions
). Thus both internal
process flows are characterised through a two-tier logistics network. Solid road transport using a 120 m3
capacity trailer is assumed for feedstock biomass and both wet and dry residual fuels. Liquid road transport using 27 m3
liquid tanker is assumed for dilute ethanol solutions (5%, 35% and 94% ethanol by weight) and pure ethanol. Rail logistics are not considered to be competitive owing to their high costs compared with road logistics over the relevant range of transport distances [26
]. Pipeline transport is considered a feasible transport mode for all ethanol intermediate fractions, pure ethanol and wastewater. Heat is not considered mobile between regions, while electricity is assumed transported by existing electric cable at zero cost.
Logistics costs (CL) are modelled for each commodity in terms of duration (CT) and distance (CD) as
The parameters in Equation 3 capture annualised capital, maintenance, labour and fuel costs and general overheads. Logistics costs specific to each commodity, mode of transport, source and destination are then a function of distance (L, assuming an empty return trip), the tortuosity of each mode (τ), transfer speed (ν) and total time spent loading and unloading (LUT). Here index i represents each specific commodity while g and k represent the source and destination region respectively. Logistics for biomass collection and ethanol distribution within each region are derived from an equivalent study completed for each region type.
Intermediate purity ethanol logistics
In addition to the development of a framework for multi-plant infrastructure design, this work is focussed on assessing the potential for spatial decoupling of processes within the processing chain, resulting in distributed processing and centralised purification systems. The drivers for such behaviour can be characterised by two parameters: (1) the logistics ratio (LR)
which represents the ratio between biomass and ethanol logistics costs (applicable at a range of purities); and (2) the economies of scale ratio (EoSR)
which represents the ratio between the economies of scale factor (α) for front-end Process and downstream purification stages.
A decrease in LR
can be achieved through the availability of pipeline technologies for both pure ethanol distribution (this is already standard practice in Brazil) and intermediate titres (that is, dilute 'crude' ethanol). The feasibility of pipeline distribution for slurries exhibiting solids concentrations of up to 30% on a wet basis was investigated by Kumar et al [27
] for the case of corn stover transportation. Pipeline operating costs were assumed as $3.07× 10-3
for pure ethanol [11
] and $9.29 × 10-2
for a slurry representative of fermentation broths containing the residual lignin [27
An increase in the EoSR can be envisioned to represent some degree of efficient downscaling and modularisation of the pretreatment and fermentation processes relative to downstream purification and utility generation. This would imply a shift in the capital and operating cost structure, in particular regarding labour and administrative overheads, such that costs are less dependent on scale. Such a scenario is also consistent with a supportive scheme of subsidies for small-scale producers, which would shift the balance of capital and operating costs within the processing system downstream.
Commodity purchase, sale, processing and logistics are linked through a mass balance specific to each commodity within each region as illustrated in the following equation
The model is then solved in order to minimise total system logistics (both inter and intra-regional), process capital and operating costs:
A Current technology scenario is characterised as an SSF process utilising an agricultural residue feedstock and embedded within either a centralised or corner-point supply-demand distribution at the 50 km2 grid scale. A Future technology scenario, employing an ambient CBP process utilising a hybrid poplar short rotation coppice (SRC) feedstock, is also considered. A summary of the technological parameters relevant to each scenario is provided in Table . Sensitivity to centralised and corner-point distributions, regional scale (25, 50 and 100 km2 regions), and more detailed technological scenarios regarding logistics (Equation 4) and economies of scale (Equation 5) ratios are also explored.
Current and Future scenario technology parameters