Improving performance of eco-industrial parks

Abstract

Industrial Ecology hypothesises that networks of industries designed to be analogous to the structure and properties of food webs may approach a similarly sustainable and efficient state. Although ecology is the metaphor for designing Eco-Industrial Parks (EIPs), prior research has shown that EIPs are inferior in performance compared to natural ecosystems. One EIP design approach is to enlarge EIPs by combining two or more synergistic networks to create a larger, and hopefully more successful, synergistic mega-network. A quantitative analysis using structural ecosystem metrics is presented in this paper in order to test the potential of this approach. The findings indicate that merely enlarging EIPs by significant amounts may not be the best strategy for improving performance, but that special attention should be placed on the inclusion of key actors like agriculture that act like detritivores and promote more intense internal cycling.

Keywords:

• Industrial ecology
• ecosystems
• design
• manufacturing

1. Introduction

Environmental problems from our production systems are not solvable by a ‘silver bullet’ technology (Allen et al. 2002), rather their inherent complexity needs a systems-based solutions. Achieving truly sustainable manufacturing requires a systems view that goes beyond a single factory or company. This means close collaborations around internal and external value chains and resource networks must be pursued between multiple companies and stakeholders. Only then will production move from the current ‘take-make-waste’ society to a truly cyclical production paradigm.

Nature can give fundamental insight for sustainable manufacturing systems: ecological systems are examples of complex, interconnected, sustainable systems. Interacting species and interacting industries both represent collections of entities that exchange materials and energy forming interconnected networks, ecosystem or food webs and industrial networks or parks (Frosch and Gallopoulos 1989). Thus, despite the fact that the organisation of industrial systems may be produced by concerns unique to human commerce (e.g. policy, supply chain management), human systems can be characterised and examined using the same techniques used by ecologists, and the organisation of ecological system may offer insights.

The organisation of individual actors into ensembles has consequences for material and energy cycling. When multiple co-located facilities interact, industrial networks achieve higher system efficiency through the exchange of ‘waste’ energy and materials industrial symbiosis occurs. The interactions formed through industrial symbiosis are analogous to the close and often long-term interactions between two or more biological species in an ecosystem. Industrial networks characterised by these symbiotic interactions among firms co-located in a bounded geographic area are called eco-industrial parks (EIP). The industries in EIPs share and/or exchange inputs and outputs (for example raw materials, products, process wastes or water) forming an industrial version of an ecosystem, or more specifically a food web (Chertow 2000). Kalundborg, Denmark is a commonly cited example of a successful EIP (Ehrenfeld and Gertler 1997).

A central hypothesis in the field of Industrial Ecology is that such a symbiotic network of industries will approach a similarly sustainable and efficient state to natural ecosystems (Frosch 1992). Ecological principles have in practice been used more as qualitative metaphors for EIP design than as a source for sound design principles (Hess 2010; Isenmann 2003; Jensen, Basson, and Leach 2011). Ecology has the potential to offer much more insight and science-based understanding of symbiotic systems. For example, as detailed in Section 2 of this paper, ecological literature defines metrics that examine properties and interactions within ecosystems. The interspecies interactions within ecosystems are graphically organised into food webs that capture biodiversity, species interactions (particularly feeding relationships) and the structure of links (e.g. between predators and prey). Ecologists measure these aspects in their analysis of food webs using structural measures and metrics that are derived from input–output analyses and matrix theory (Briand 1983; Briand and Cohen 1987; Duchin 1992; Fath 2007; Schoener 1989; Warren 1990a). The metrics developed by ecologists describe and analyse the structures governing food webs, properties of which are highly desirable in industrial systems. The theory that this work builds off of is that these desirable traits may be transferred to industrial networks by mimicking the food web structure, the success of which is measured using the ecological food web metrics. The positive potential of this has already been shown with a carpet recycling network designed using structural food web metrics. The ‘bio-inspired’ carpet network design was found to positively correlate (R² = 0.96) with standard designs which focus solely on cost- and emissions-minimisation (Reap and Bras 2014).

A variety of important structural ecological parameters discussed in detail in Section 2 are used in (Layton, Bras, and Weissburg 2016b) to compare the structures of EIPs to food webs, building upon the quantitative knowledge in ecology. This comparison used a comprehensive data-set of 144 food webs and provided a more ecologically correct understanding of how food webs are organised than previous efforts. The 48 EIPs in (Layton, Bras, and Weissburg 2016b) showed an average and maximum performance well below the average performance characteristic of food webs. The ‘best’ EIP in the collection has a cyclicity of only 3.87 and a linkage density of 3.13 while the average food web has a cyclicity of 6.03 and a link density of 7.69, almost twice as large as the values for the best EIP. This raises an interesting design and research question: ‘How can we improve these values and thus the associated performance of EIPs?’

Identifying the fundamental physical relationships responsible for the correlations that are beginning to be recognised between bio-inspired network patterns and environmentally superior industrial network designs is crucial to creating concrete design guidelines, which is our long term goal. The approach investigated here looks at increasing the success of EIPs and thereby bringing them closer to food webs by enlarging the EIPs with more interacting actors (i.e. companies and industries). This is a typical approach taken in practice when one company at a time to joins an EIP. In this paper, we go beyond additions of single companies and investigate the effects of combining two or more synergistic EIP networks (thus doubling or even quadrupling the EIP size) to create a larger, and hopefully more successful, synergistic mega-network. In this paper, an analysis of this approach of combining multiple EIPs into a ‘super’ EIP is presented to test its potential for achieving more natural food web like behaviour. First some background on ecosystem metrics is provided.

2. Structural ecosystem metrics

Ecological literature defines metrics to examine ecosystem properties, including species interactions, topology and patterns of mass and energy flow (see, e.g. Odum 1969; Pimm 1982; Warren 1990b; Schoener 1989; Briand and Cohen 1987; Cohen et al. 1993; Ulanowicz 1997). The flows of materials and energy in an ecosystem, highlighted in its food web representation can be represented in a matrix form (food web matrix [F]). The interactions are organised from prey (rows) to predators (columns). Figure 1 shows a hypothetical ecosystem represented as food web with a directional digraph (left) and converted into a food web matrix (right). Since a species (N) can be both predator and prey the result is a square matrix. A value of 1 in the matrix indicates the existence of a directional flow from row to column and a zero indicates no connection. In other words, if predator-j feeds on prey-i, then f_ij = 1; the interaction (or link, L) is accounted for exactly once in the food web matrix. The maximum number of links, L scales as (N)*(N − 1) if cannibalism is not allowed and N² if it is (noted as a 1 on the diagonal).

Figure 1. Left – A food web of a hypothetical ecosystem with species numbered. Right – A food web matrix; f_ij = 1 represents a unidirectional link between prey (i) and predator (j) and a zero represents no link.

A wide variety of metrics have been developed to try to understand the link between structure and behaviour of ecological systems (Bascompte and Jordano 2007; Fath and Halnes 2007), and which can be used to characterise human systems that also exchange matter and energy among individual actors within an ensemble. The structural measures and metrics used most frequently by ecologists can be calculated using the N × N structural food web matrix shown in Figure 1. The benefit to using these metrics for industrial networks is that they can be calculated knowing only structural information, that is, all calculations per Equations (3(3) )–(13(13) ) are simply based on the binary information of whether a link exists between two actors in the matrix, or not, and its direction. The entry f_ij represents the connection between actor i to actor j and is the ith row and jth column entry in the matrix. It has a value of either 1 (link exists) or 0 (no link). From there basic network descriptors can be calculated such as the density of linkages in the system (Equation (1(1) )), the number predators and prey (Equations (5(5) ) and (3(3) ), respectively) and the average number of predator–prey exchanges. Linkage density (L_D, Equation (1(1) )) is the number of links in the system normalised by the number of actors in the system (Dunne, Williams, and Martinez 2002). The number of prey and predators (n_prey and n_predator, Equations (3(3) ) and (5(5) )) are the number of actors that provide and consume resources, respectively, (Schoener 1989), or producers and consumers in industry terms, respectively. The ratio of these two is the prey to predator ratio (P_R, Equation (6(6) )) a numerical representation of the balance of consumers to producers in the system. The number of specialised predators (n_S-predator, Equation (8(8) )) is a subset of n_predator and only counts those consumers who interact with only one actor, highlighting actors who will be particularly sensitive to a change in the makeup of the system. The specialised predator ratio (P_S, Equation (9(9) )) is the fraction of consumers that are feed on, or interact with, only one type of producer. Generalisation and vulnerability (G and V, Equations (10(10) ) and (11(11) )) are subsets of L_D and represent, respectively, the number of producers that a consuming-actor can consume and the number of consumers a producing-actor provides for (Pimm 1982; Schoener 1989). Connectance (C, Equation (12(12) )) is the number of realised direct interactions in a web divided by the total number of possible interactions. It should be noted that if one forbids cannibalism then the denominator is the fraction of nonzero off diagonal elements in the food web matrix [F]. Cyclicity (λ_max, Equation (13(13) )) is the maximum real eigenvalue of the transpose of the food web matrix (flow is then columns to rows in this transpose). Cyclicity represents the presence and complexity of internal cycling in the system (Borrett, Fath, and Patten 2007; Fath 1998; Fath and Halnes 2007; Layton et al. 2012), and is taken as an indication of how cyclic pathways proliferate as the number of steps in the cycle grows. More detailed descriptions of these metrics may be found in Layton, Bras, and Weissburg (2016b) in addition to the references already listed.(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)

3. Combining eco-industrial parks

The 48 EIPs analysed in Layton, Bras, and Weissburg (2016b) showed performances well below the average performance characteristics of food webs. This is of concern since these EIPs are meant to produce sustainable and efficient systems, and directs us to the questions (1) Why? and (2) How do we improve them? A quick response to increase the ecological performance of these networks would be to increase the size of the EIPs (following the ‘more is better’ approach (see Kemerling 2007, 358). Several EIPs were chosen and grouped together based on shared materials and energy exchanges. These groupings are theoretical since the EIPs are not co-located and therefore the material exchanges would have to occur over great distances. This method follows other theoretical EIP creations (Chertow 1999; Chertow and Lombardi 2005; Bennett et al. 1998; Bodini and Bondavalli 2002; Bodini, Bondavalli, and Allesina 2012; Brown, Gross, and Wiggs 1997; Johnson et al. 1999; Kellogg et al. 1999) for the sake of understanding how to design successful EIPs. The investigation is still valuable, however, by helping to identify factors that limit EIP performance; achieving greater performance by coupling EIPs suggests that individual EIPs each contain only a fraction of available linkages necessary for high levels of material and energy cycling. This would imply that existing barriers relate more to historical contingency in EIP construction than technological or other forces.

EIPs that had common material flows were connected with links across the EIP boundaries; no linkages inside the original EIPs were modified. For example, in Combo 1, shown in Figure 2, the original EIPs Kymi and Lubei both had material flows of water exchanged within their boundaries. New linkages crossing the boundaries of EIPs Kymi and Lubei were added to reflect the common material flow of water: water from the Aquaculture plant in Lubei was connected to the both the chloride dioxide and the calcium carbonate plants in Kymi. Water from the Pulp and Paper Plant in Kymi was likewise newly connected to the Turbo Generator in Lubei. All of the new connections between the original EIPs form the new and larger combination network Combo 1. All of these original and new connects making up Combo 1 can be seen in Figure 2. This results in a maximally connected ‘combo-EIP’ based on existing resource flows. Ecological metrics were calculated for each of the combo-EIPs following the Equations (1(1) )–(13(13) ) and compared to the values of the individual EIPs in each group to look for an improvement in matching ecosystem performance. The metrics were also calculated for the grouped EIPs before the new connections were added to highlight the effect of the new connections apart from the increase in actors, since some metrics are not independent of the network size.

Figure 2. Combination of four individual EIPs into Single Combined EIP (Combo 1 – Table 1).

A sample representation of four individual EIPs (Lubei Industrial Park, Mongstad, Wallingford, and Kymi EIP) and their material exchanges combined into a larger EIP is shown in Figure 2 while diagrams of each EIP individually can be found in the supplemental material of (Layton, Bras, and Weissburg 2016b). The new connections made between EIPs are shown by the arrows crossing the boxed boundaries surrounding each individual EIP. The coloured arrows represent the common material and energy flows while the grey arrows are those flows unique to a single EIP.

The results of the new combination networks (Combo 1–5 above) are made clear by comparing them to the original EIPs used to make each combo. The changes are made more clear by the differences before and after in the metrics linkage density and cyclicity. Linkage density captures the number of species in the system and any changes in the number of linkages. Cyclicity on the other hand captures any changes in the network structure due to how the new or lost linkages interact with the rest of the system. Connectance could be used interchangeably with linkage density as they both capture changes in actors and links. Linkage density is preferably to connectance, however, in that it does not require systems of similar size if used for comparisons. Natural ecosystems have far more actors than the EIPs. The prey to predator ratio, generalisation and vulnerability are of interest only if additional information about the system is available so that changes in the behaviour of the actors may be made.

The internal cycling, as measured by cyclicity – Equation (13(13) ), in food webs is very strongly influenced by the presence of detritivores. Although this influence is known, the amount of cycling in natural ecosystems dwarfs any man-made system. Over half of all the material in a natural food web is connected to a decomposer-type species such as fungi, which recycles unused material or dead matter (detritus) and returns it back to the system. Decomposers ensure the presence of food web pathways that include all other species in the system because the connections due to this consumption pattern contribute-to many other existing cycles. Even limited connections to an actor that functions similarly in an EIP would dramatically increase connectivity, and thereby efficiency.

Typical detritivores for an EIP are actors that function in waste treatment (i.e. composting), recovery and recycling (i.e. repair, remanufacture, reuse, resale), but agriculture (i.e. farm, zoo, landscaping, green house, golf course) is also a detritus-type actor for an EIP. Additionally, to qualify as a detritus-type actor there must be at least one link entering and leaving said actor. This last criterion is based on the fundamental functional description of a detritus/decomposer in a food web and ensures that the detritus-type actor is an active participant of the EIP. Agricultural actors typically have these characteristics and thus increase the cyclicity of an EIP when present (Layton, Bras, and Weissburg 2016a). The combination EIPs also seek to increase the number of these detritivore-type facilities available to any actor in the network.

Agriculture facilities are an obvious way of incorporating detritivore behaviour in the industrial network and therefore the question regarding the impact of agriculture on an EIP success in mimicking ecosystem structure and function is a potentially important question for designers of EIPs. Of the 48 EIPs investigated in Layton, Bras, and Weissburg (2016b, 34) had some type of agricultural component and 14 did not. Some of the benefit of an agriculture component in an EIP has to do with the ability to use a mixture of diverse byproducts, such as organic wastes such as food or paper wastes, animal effluent, compost and fertiliser for a variety of purposes. Integrated bio-systems (IBS) are a subset of EIPs that are heavily based on agriculture (Abuyuan et al. 1999; Hardy, Hedges, and Simonds 2000). The benefit of an agriculture component in terms of its ability to make an industrial network look more ecological has not yet been investigated using the ecosystem metrics presented in this paper.

3.1. EIP Combo 1: Lubei Industrial Park, Mongstad, Wallingford, and Kymi EIP

The four EIPs in Combo 1 were paired due to a common use of water, steam, fly ash, wastewater, electricity, hydrogen, carbon dioxide, chlorine and sodium hydroxide. Lubei Industrial Park, designed to be located in China, is outlined in Mathews and Tan (2011). The Kymi EIP located in Kymenlaakso, Finland is outlined in Sokka, Pakarinen, and Melanen (2011). Both the Wallingford EIP in Wallingford, Connecticut and the EIP Mongstad in Mongstad, Norway are outlined in Reap (2009). The Lubei and Mongstad EIPs both have aquaculture as their active agriculture (AA) actor. Nine different materials and energy streams can be exchanged between the four EIPs if they were co-located, but considering distances between the EIPs only five flows can realistically be exchanged.

3.2. EIP Combo 2: GERIPA, Gladstone and Montfort EIP

The three EIPs in Combo 2 were paired due to a common use of soil and other organic wastes, fly ash and biogas. GERIPA, which stands for Geração de Energia Renovável Integrada á Produção de Alimentos, is an IBS (integrated bio-system) designed for Brazil and is outlined in Ometto, Ramos, and Lombardi (2007) and Reap (2009). Gladstone is a proposed addition to an existing EIP in Gladstone, Australia and is outlined in Corder (2005, 2008) and Reap (2009). The Montfort Boys Town is also an integrated bio-system located in Suva, Fiji and can be found in Reap (2009). The active agriculture actors in the three EIPs GERIPA, Gladstone and Monfort include, respectively, farming and a biodigester, biomass and fertiliser production, and farming, aquaculture, and fertiliser production. Six different materials and energy streams were able to be exchanged between the three EIPs, four of which realistically could be exchanged taking into account the distances between the EIPs.

3.3. EIP Combo 3: Brownsville, Burnside, Clark Special Economic Zone and Kawasaki EIP

The four EIPs in Combo 3 were paired due to a common use of soil and other organic wastes, waste plastic, used oil and tires, steam, water and waste water. The Brownsville EIP was located in Brownsville, TX and is outlined in (Martin et al. 1996). The Burnside EIP is in Nova Scotia, Canada and can be found in Cote (2009). The Clark Special Economic Zone was proposed for the Philippines (Reap 2009). The active agriculture actors in the Clark EIP are the result of landscaping, a golf course, a greenhouse and composting. Seven different materials and energy streams were able to be exchanged between the four EIPs, four of which realistically could be exchanged taking into account the distances between the EIPs (locations range from Texas to Canada to the Philippines to Japan). Info on the Kawasaki EIP can be found in Hashimoto et al. (2010) and Mathews and Tan (2011).

3.4. EIP Combo 4: Kymi and Wallingford EIP

The two EIPs in Combo 4 were taken from Combo 1 to test the lack of presence of an active agricultural component. Three different materials and energy streams were able to be exchanged between the two EIPs, only one of which realistically could be exchanged taking into account the distance between Finland and Connecticut where the two EIPs are located.

3.5. EIP Combo 5: Brownsville, Burnside and Kawasaki EIP

The three EIPs in Combo 5 were taken from Combo 3 to test the lack of presence of an active agricultural component. Four different materials and energy streams were able to be exchanged between the three EIPs, three of which realistically could be exchanged taking into account the distances between the EIPs (locations range from Texas to Canada to the Japan).

4. Discussion

4.1. General findings

In this paper, we focus on the effect of combining EIPs in order to test the hypothesis whether this creates a larger industrial system that more closely mimics networks found in nature. Combining multiple EIPs into a large ‘mega’ network like presented in Section 3 might be considered a giant accomplishment in practice by policy makers, but as shown does not necessarily increase ecological performance significantly. A mere opportunistic and arguably random approach to enlarging and combining EIPs may not result in EIP networks with structures that performance like ecological food webs.

The additional linkages provided by the combinations did not effect, or had very limited effect on the number of predator and prey type actors in the system. Only the existing material flows were used to create the new connections between companies, therefore the number of prey and predators, or consumers and producers, remained the same. Only in very few cases new material flows could be added based on additional information found in the descriptions of the individual industrial groupings. The added connections also had no effect on the overall number of actors in the system. Therefore, we are not concerned with the five metrics relating to these values in this discussion (N, L, n_predator, n_prey, P_R). An exception is with Combo 1 and Combo 2, which show small changes in predators and prey before and after the new linkages were added. These changes are due to the addition of new consumers based on the available detailed descriptions of the original EIPs. These descriptions contained additional information about what participating companies were exchanging and receiving and therefore connections within the bounds of the individual EIPs could be added. Without this detailed information about the EIPs’ input and output flows such new predator and prey-type actors were not created.

The effect of the additional linkages between EIPs for the rest of the metrics was consistently strongest for linkage density (L_D), connectance (C), generalisation (G) and vulnerability (V). These metrics showed an increase between the individual EIPs and the combined EIP networks. All four of these metrics are directly influenced by the number of linkages in the network and thus the basic effect of the addition of linkages is reflected. Tables 1–5 show an equivalent percentage increase across all of these metrics for all five combination networks.

Table 1. Combo 1 EIP made up of Kymi EIP, Lubei Industrial Park, Mongstad EIP and Wallingford EIP.

EIP Combo 1	AA component?	N	L	npredator	nprey	C = L/N^2	LD	PR	PS	V	G	λmax
Kymi EIP	N	8	14	7	7	0.22	1.75	1	0.714	2	2.13	1.82
Lubei Industrial Park	Y	9	17	7	8	0.21	1.89	1.14	0.429	2.13	2.22	0
Mongstad EIP	Y	11	20	10	8	0.17	1.82	0.8	0.3	2.5	1.57	1.55
Wallingford EIP	N	12	18	11	9	0.125	1.5	0.82	0.545	2	2.8	1
Combo 1 pre links added		40	72	35	32	0.045	1.8	0.914	0.486	2.25	2.06	1.82
Combo 1 post links added		40	169	36	33	0.106	4.23	0.917	0.111	5.12	4.69	4.48
% change		0	135	3	3	135	135	0	−77	128	128	146

Table 2. Combo 2 EIP is made up of GERIPA, Gladstone and Montfort IBS.

EIP Combo 2	AA Component?	N	L	npredator	nprey	C = L/N^2	LD	PR	PS	V	G	λmax
GERIPA IBS	Y	8	15	6	8	0.23	1.88	1.33	0.167	1.88	1.8	1.93
Gladstone (2008)	Y	23	25	13	14	0.05	1.09	1.08	0.692	1.79	1.17	0
Monfort Boys Town	Y	9	11	7	7	0.14	1.22	1	0.571	1.57	2.43	1
Combo 2 pre links added		40	49	28	26	0.031	1.23	0.929	0.607	1.88	1.75	1.93
Combo 2 post links added		40	124	32	29	0.078	3.1	0.906	0.344	4.28	3.88	4.01
% change		0	153	14	12	153	153	−2	−43	127	121	108

Table 3. The Combo 3 EIP is made up of the Brownsville EIP, Burnside EIP, Clark Special Economic Zone and Kawasaki.

EIP Combo 3	AA Component?	N	L	npredator	nprey	C = L/N^2	LD	PR	PS	V	G	λmax
Brownsville EIP	N	16	22	10	12	0.086	1.38	1.20	0.400	1.83	1.14	1.41
Burnside EIP	N	11	20	10	9	0.165	1.82	0.90	0.300	2.22	2.20	2.05
Clark Special Zone	Y	20	51	19	17	0.128	2.55	0.90	0.263	3.00	2.00	3.34
Kawasaki EIP	N	8	16	8	8	0.250	2.00	1.00	0.500	2.00	1.62	1.88
Combo 3 pre links added		55	109	47	46	0.036	1.98	0.979	0.340	2.37	2.32	3.34
Combo 3 post links added		55	235	47	46	0.078	4.27	0.979	0.149	5.11	5.00	3.94
% change		0	116	0	0	116	116	0	−56	116	116	18

Table 4. The Combo 4 EIP is made up of the Kymi EIP and Wallingford EIP.

EIP Combo 4	AA Component?	N	L	npredator	nprey	C = L/N^2	LD	PR	PS	V	G	λmax
Kymi EIP	N	8	14	7	7	0.22	1.75	1.00	0.714	2.00	2.13	1.82
Wallingford EIP	N	12	18	11	9	0.125	1.50	0.820	0.545	2.00	2.80	1
Combo 4 pre links added		20	34	18	16	0.085	1.70	0.889	0.611	2.13	1.89	1.82
Combo 4 post links added		20	53	18	16	0.133	2.65	0.889	0.333	3.31	2.94	2.81
% change		0	135	3	3	135	56	0	−45	56	56	55

Table 5. The Combo 5 EIP is made up of the Brownsville, Burnside and Kawasaki EIP.

EIP Combo 5	AA Component?	N	L	npredator	nprey	C = L/N^2	LD	PR	PS	V	G	λmax
Brownsville EIP	N	16	22	10	12	0.086	1.38	1.20	0.400	1.83	1.14	1.41
Burnside EIP	N	11	20	10	9	0.165	1.82	0.90	0.300	2.22	2.20	2.05
Kawasaki EIP	N	8	16	8	8	0.250	2.00	1.00	0.500	2.00	1.62	1.88
Combo 5 pre links added		35	58	28	29	0.047	1.66	1.04	0.393	2.00	2.07	2.05
Combo 5 post links added		35	93	28	29	0.076	2.66	1.04	0.321	3.21	3.32	2.34
% change		0	116	0	0	62	60	0	−18	60	60	14

The greatest changes occurred in specialised predator fraction (P_S) and cyclicity (λ_max), making these two metrics most interesting. The number of consumers that interact with only one producer (P_S) decreased in all five combinations. This is due to the new connections added between EIPs. These new connections allow actors from other EIPs to participate in a larger set of existing material exchanges (Tables 1–5). This is desirable as it is a sign of a more robust network, having fewer actors who depend on only one type of interaction. Robust actors can recover if a resource supplier were to be removed by moving to another supplier with the same or equivalent resource.

Combining EIPs increased cyclicity and reduced the disparity between the performance of human vs. natural systems. The average cyclicity for food webs is 6.03 and the best EIP out of the original group of 48 in Layton, Bras, and Weissburg (2016b) had a cyclicity of only 3.87. Combo 1 and Combo 2 both have a drastically higher cyclicity, 4.48 and 4.01, respectively, than any of the individual EIPs (Table 6). These two combination networks saw the biggest increases in cyclicity due to the additional connections made; a 146 and 108% change, respectively. Cyclicity is a crucial structural aspect contributing to the working of ecosystems. Without cyclicity the structural connectivity of the network is measured via the metric connectance (C) and the extra feedback connections caused by flows to detritus and back to the system through the detritus are neglected (Fath and Halnes 2007). Detritus actors process low quality resources and convert them into a form that is easily processed by the rest of the ecosystems actors. Flows to detritus are made up of pathways such as death, excrement or exfoliation. This low quality energy processing is similar to standard methods of increasing the thermal efficiency of thermodynamic power cycles by rerouting material streams with low quality energy back into the system (Layton et al. 2012). The cycling measured by cyclicity is fundamentally different from other cyclic structures in the ecosystem. This type of cycling is able to work independent of any structural hierarchy inherent in the system (Fath and Halnes 2007). The cyclicity of the individual EIPs making up each of the five combinations ranged from zero, meaning no cycles are present, to less than 2, meaning some complex cycling is present. A cyclicity much greater than one is representative of complex and abundant internal cycles between the actors in the network, so by combining EIPs together more connections were able to be made, resulting in a network with a more complex and ecosystem-like structure.

Table 6. Combination EIPs compared against averages for food webs data-set.

	λmax	LD	PR	PS	V	G	N	L	npredator	nprey	C = L/N^2
Food web averages (Layton, Bras, and Weissburg 2016b)	6.03	7.69	1.13	0.100	8.82	9.69	57	523	47	51	0.155
Combo 1	4.48	4.23	0.917	0.111	5.12	4.69	40	169	36	33	0.106
Combo 2	4.01	3.10	0.906	0.344	4.28	3.88	40	124	32	29	0.078
Combo 3	3.94	4.27	0.979	0.149	5.11	5.00	55	235	47	46	0.078
Combo 4	2.81	2.65	0.889	0.333	3.31	2.94	20	53	18	16	0.133
Combo 5	2.34	2.66	1.04	0.321	3.21	3.32	35	93	28	29	0.076

4.2. Effects of agriculture in EIPs

The two groupings, Combo 1 and 2, that had the largest increases in cyclicity (146 and 108% change, respectively) were also made up of the most EIPs with an agricultural component. The question regarding the impact of agriculture on an EIP success in mimicking ecosystem structure and function is a potentially important question for designers of EIPs. Of the 48 EIPs investigated in Layton, Bras, and Weissburg (2016b, 34) had some type of agricultural component. The five combination EIPs made in this paper investigate possible effects of agriculture in an industry network. Combo 4 and Combo 5 in particular were created to test the added value of having an agriculture component, neither of the two groups have an EIP with an agriculture component. Improvements are still seen from the individual EIPs to the larger combined EIP, however, not as significant as changes seen for Combo 1, Combo 2 and Combo 3, which all had an agricultural component in one of more of the EIP building blocks. Perhaps the best way to look at the added value of agriculture is between Combo 1 and Combo 4, and Combo 3 and Combo 5. The two agriculture EIPs in Combo 1 bring cyclicity up to 4.48, without these two components cyclicity only reaches 2.81, below that of the best single EIP. The singular EIP with an agriculture component (landscaping, a golf course, a greenhouse and composting) in Combo 3 brings the cyclicity for the entire group up to 3.94, without it the cyclicity is 2.34. These findings indicate that adding agricultural components, which can act as detritivores, does have a beneficial effect. But, the addition of actors that exhibit detritivore behaviour is more important than merely adding agriculture in itself.

4.3. Effects of physical proximity between EIPs

Some literature points out that the physical proximity of ecosystems is something that industrial networks cannot recreate and therefore any hope for a successful analogy is lost (Husar 1994). Ecosystems often have a physical proximity that is becoming more uncommon in today’s global economy, an increasingly larger cross section of business interactions are done between companies that may not even speak the same language (Milner 2014). Species proximity results in low energy expenditures for transportation of materials and energy in addition to relatively short reaction times in the face of perturbations. The energy expenditures of transportation in an industrial setting may not be that distinct from the energy which an animal, especially a migratory animal, must expend to feed for example storks have a system boundary that extends over 12,000 km (van den Bossche 2005). Industrial networks need not be collocated to reap some of the benefits from cyclical interactions that result from mimicking the structure of food webs. In addition to these location independent benefits, constant improvements in infrastructure and transportation are creating more cost effective solutions that once in place can result in minimal energy transfer requirements. A greater distance between networks, however, does make the exchange of things such as wastewater and steam unrealistic, two materials which are very commonly and successful exchanged between collocated industries. Thus, distance should not be a deterrent to the implementation of food web structure and creation of new industrial networks, only recognised such that the best choices as to what is exchanged may be made. Benefits such as longer paths that better use the entirety of a material and the robustness and stability that results from a diverse exchange system in today’s global economy do not depend on proximity.

5. Conclusions and future work

This paper demonstrates that significantly increasing the size of EIPs helps but is potentially not enough by itself to generate positive food web changes. Doubling or even quadrupling the size of the EIPs by combining them did not necessarily guarantee expected changes in terms of ecological behaviour. The most positive changes result not from added actors being added, but from additional linkages. Interestingly, this implies that reducing the size of a network while increasing links would be more positive than merely adding actors to the system without regard to potential opportunities for exchanges. For example, it is unlikely that high cyclicity values can be achieved in EIPs that lack actors fulfilling the role of detritus/decomposers. This suggests that EIP designers must incorporate analogous interactions in their industrial networks to achieve the strong cycling characteristic of food webs.

Furthermore, it has been noted in ecological literature that there may in fact exist a point where a more streamlined network, essentially a network with less diversity, has negative repercussions in the form of overdependence and reduced robustness to random perturbations. A hypothesis within ecology is that diversity may be a strong contributor to the stability of a system: when one actor is removed the system may adapt or recover by another actor(s) stepping in to fulfil the supporting role (Korhonen and Snäkin 2005). Principles and metrics for assessing ecosystem robustness and stability along these lines seem also relevant and applicable to improving sustainability of industrial systems (Layton, Bras, and Weissburg 2015). The natural tendencies for ecosystems as they mature is for the interactions to become more selective, shifting the focus from production towards efficiency (Odum 1969). Mature ecosystems obtain efficiency by way of an increase in use of existing actors, essentially using what is available as completely as possible. This results in the desirable property of a complex structure with an abundance of connections between species (Fath and Halnes 2007). These lessons from ecology could support design of EIPs that function truly like ecosystems.

Future work, therefore, should focus on how to establish guidelines that can support a priori ‘design’ of EIPs that have superior performance and truly act like natural ecosystems, rather than ad hoc, opportunistic expansion and combination efforts. Furthermore, given the importance of detritivores, future work should focus on analysing the effect of (a) identifying appropriate industries that can act as detritivores and (b) assessing the effect of including them in EIPs in order to identify the most appropriate detrivore. Additionally, rather than merely focusing on man-made, engineered systems, the potential of including and connecting EIPs to natural systems should also be explored. For example, rather than a conventional waste water treatment facility, one could also use a wetland system with specific plants to filter out pollutants. The design of such bio-EIPs is difficult but could lead to some truly innovative symbiotic and sustainable systems.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Astrid Layton, PhD, is an assistant professor in the Department of Mechanical Engineering at Texas A&M University. Her research interests include using bio-inspired analysis techniques to design sustainable human networks and systems, including but not limited to industrial resource networks and complex energy systems. After receiving her PhD degree from Georgia Tech in 2014, she was a visiting lecturer at Georgia Tech Lorraine in Metz, France, after which she joined Texas A&M in 2017.

Bert Bras, PhD, is a professor at the George W. Woodruff School of Mechanical Engineering at the Georgia Institute of Technology since September 1992. His research focus is on sustainable design and manufacturing, including design for recycling and remanufacture, bio-inspired design, and life-cycle analysis with applications in automotive and energy systems. He has authored and co-authored over a 150 publications. From 2001 to 2004, he served as the director of Georgia Tech’s Institute for Sustainable Development and he received the 2007 Georgia Tech Outstanding Interdisciplinary Activities Award.  In 2014, he was named a Brook Byers Professor of Sustainability. In 2016, he was appointed as the Associate Chair for Administration in the G.W. Woodruff School of Mechanical Engineering.

Marc Weissburg, PhD, is a professor in the School of Biology at the Georgia Institute of Technology. His research interests include chemical ecology, sensory ecology and biologically inspired design. He obtained his PhD in Ecology and Evolutionary Biology from SUNY Stony Brook, and has an active research program in sensory ecology. He is a co-founder of Georgia Tech’s Center for Bio-Inspired Design (CBID) and has taught biologically inspired design (BID) for undergraduate students, practicing professionals, and at workshops for 10 years. His interdisciplinary efforts in BID include infrastructure and industrial ecology, the pedagogy of bio-inspired design and informal science education using bio-inspired design.

Acknowledgements

This material is based upon work supported by the National Science Foundation [grant number CMMI-0600243], [grant number CBET-0967536] and [grant number CBET-1510531]. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.