Abstract
There is a growing demand for sustainable products and systems. Sustainability encompasses environmental, social, and economic aspects, often referred to as the three pillars of sustainability. To make more sustainable design decisions, engineers need tools to predict the environmental, social, and economic impacts of products and characterize potential sustainability tradeoffs. To predict the total impact of a product, the quantity of functional units of the product in society and the impact of each product needs to be estimated. This article uses agent-based modeling (ABM), combined with tools such as life cycle assessment (LCA), to predict impacts across all three pillars of sustainability. By using the product impact results, the multidimensional sustainability tradespace can be characterized. The approach described in this article is based on three main components for the predictive modeling of product impacts and the characterization of the sustainability trade space: (i) ABM of product adoption, (ii) the assessment of product impacts, and (iii) an approach for the characterization of product sustainability tradeoffs at the population level. The tradespace characterization uses a Pareto-based method presented visually to find the nondominated solutions in the product impact space. To illustrate and describe how to use the method, a case study is presented that predicts the impact of residential solar panels in a region of the United States under various scenarios. The findings of the case study can help policy makers understand suitable implementation strategies for residential solar panels while considering the impact tradeoffs involved.
1 Introduction
The United Nations Sustainable Development Goals (SDGs) contain aspects of social, environmental, and economic impacts [1]. Social, environmental, and economic impacts make up what are often called the three pillars of sustainability [2] or the triple bottom line [3]. As practitioners work to improve sustainability practices in all three pillars of sustainability, there are often tradeoffs between the different pillars [4,5]. Consumers may not even be aware of the tradeoffs that are made in the products they adopt, such as the potential use of child labor in the mining of metals needed for electric vehicles that reduce greenhouse gas emissions [6]. To make more sustainable design decisions, engineers need tools to predict the impacts of products and navigate potential impact tradeoffs. As society addresses complex, large-scale issues such as climate change, behavior, and technology need to become more sustainable. The move toward electrification, for example, necessitates the installation of technologies and infrastructure that will remain in place for years to come. The long-term horizon, economic costs, and pressing urgency of climate change all highlight the necessity for better design decisions that prioritize sustainability. Pressing societal challenges and long-term investments stand to benefit from predictive impact modeling to accelerate the path to positive impact.
It has been acknowledged that engineers need improved tools to estimate the impact of products and that there are few tools to estimate impacts, especially the estimation of social impacts [7]. Life cycle assessment (LCA) with its accompanying ISO standard is often used to assess environmental impacts [8]. Methods to assess the social impact of products are, compared to LCA, less mature [9]. The business literature has many ways to evaluate economic success, but in the literature related to sustainability, life cycle costing (LCC) and return on investment are frequently used to measure the economic pillar [10]. One common problem with sustainability analysis tools such as LCA is how to scale the results to the population level [11]. In response to this limitation, Liechty et al. developed a method to scale the environmental impacts obtained from the functional unit level to the population level using agent-based modeling (ABM) to understand the adoption of the product in the population [12].
The contribution of this article is to demonstrate an approach for using predictive modeling to scale the social, environmental, and economic impacts of a product from the individual product level to the total impact at the population level and a method for characterizing potential sustainability tradeoffs in all three pillars of sustainability. This article builds on the work of Liechty et al. [12], by examining impacts across all three pillars of sustainability and how to characterize the multidimensional sustainability trade space. Additionally, this work illustrates how to overcome the challenge of scaling impacts per functional unit to the population level. A case study using the impacts of residential solar panels in a region in the United States is used to illustrate the method. This case study makes use of models of residential solar panel adoption from the National Renewable Energy Laboratory (NREL) [13]. The remainder of this article is organized as follows: Sec. 1.1 reviews the previous literature on ABM and product impact assessment, Sec. 2 describes the methodology of this article for triple bottom line impact prediction and trade space characterization, Sec. 3 presents the results of a residential solar panel case study, Sec. 4 provides a discussion of the implications of the results and method, and finally Sec. 5 provides concluding remarks.
1.1 Background.
The approach described in this article is based on three main components for the predictive modeling of product impacts and the characterization of the sustainability tradespace, (i) ABM of product adoption, (ii) the assessment of product impacts, and (iii) an approach for the characterization of sustainability tradeoffs.
1.1.1 Agent-Based Modeling.
ABM is a bottom-up modeling approach in which the behavior of individual entities is modeled based on rules rather than on a top-down governing equation [14]. One of the main benefits of ABM is the ability to capture emergent phenomena from agent interactions [15]. There have been many applications of ABM, including segregation [16], water management [17], bullying [18], combat simulation [19], team problem solving [20], among many other applications [21]. The main application of interest in this article is the use of ABM to predict product adoption. ABM has been used in many cases for the adoption of products and the diffusion of technology [22–24]. One of the advantages of ABM when modeling product adoption is the ability to examine product adoption patterns across different segments of a population. Although ABM has been used to model product adoption in many cases, ABM has only recently been used to predict product impacts [12,25]. In the cases where it was used to predict the impacts of the product, it was not used to predict the impacts in the three areas of sustainability. The work in this article will use the results of the ABM to predict the social, environmental, and economic product impacts simultaneously.
1.1.2 Product Impact Assessment.
Interest in the assessment of the impacts of products has increased in recent years [24]. Movements such as the Environmental, Social, and Corporate Governance framework (ESG) [26] and B Corp Certification [27] have spread in the business sector due to the consumer demand for companies to have a more positive social and environmental impact. With increasing consumer demand for improved sustainability, methods for measuring sustainability and product impacts have also grown [28]. This subsection will describe the advances and current approaches to assess environmental, economic, and social impacts.
The evaluation of environmental impact using LCA is the most established and mature of the three pillars of sustainability [8]. An LCA measures the impact of a product or system based on the raw materials needed and the processes used at different stages of the product life cycle [29]. Although LCA was first used around 1970, the practice began to be standardized in the early 1990s by the Society of Environmental Toxicology and Chemistry (SETAC) [30]. In 1996, the basic standard for LCA was published as ISO 14040 [31]. Although the ISO standard outlines the basic principles of LCA, there are different LCA approaches such as the IMPACT World+ method [32] and the ReCiPe 2016 method [33]. LCA output will vary depending on the method used, but, in general, the output is impact metrics per functional unit in various categories of environmental impact. LCA is not well equipped to incorporate the dynamic nature of sociotechnical systems [34], but when coupled with a method such as ABM, the dynamic sociotechnical system can be taken into account [35].
In the sustainability literature, LCC is one of the most common methods of assessing economic impact and is often used as the only metric for economic impact in sustainability assessments [36]. LCC is the sum of the costs associated with the manufacture, distribution, use, and disposal of a product or system. The use of only a single metric gives a narrow view compared to the more holistic approach of the various LCA indicators. Wood and Hertwich propose the need for a wider view of economic impacts than that provided by the LCC [37].
Assessing the social impact of products is a relatively new field compared to LCA [24]. The social impact of a product can be defined as the impact of the product on the day-to-day lives of persons [38]. Different methods have emerged to assess the social impact of products, such as social impact assessment [39], social life cycle assessment (SLCA) [40], and the product impact metric [41]. Most methods of assessing social impact have focused on assessing current impacts rather than predicting future impacts, although we are beginning to see cases of their use to predict impacts [42].
Conducting a LCSA is a complex process. Different approaches have been taken to combine the different portions of LCSA. Some studies keep the three analyses separate [43,44], while others apply a summation of the analyses with or without weightings to aggregate the results using multicritera decision analysis methods [28]. Aggregation methods using a sum will either have to be normalized or will be inappropriate to combine due to differences in units. Generally these methods present impacts in terms of the impact per functional unit, and do not calculate a total impact at the population level. Although impact per functional unit is an important measure, it does not take into account the way that changes to a product design or system design, that will change the impact at the functional unit level may also change patterns of product adoption that will ultimately change the total impact at the population level.
1.1.3 Sustainability Tradeoff Characterization.
There are cases where improvements in the different pillars of sustainability may conflict with each other [45]. In the presence of conflicts, a design team will need methods to characterize the trade space. There are multiple methods to improve decision making in the presence of tradeoffs, including multicriteria decision-making methods such as the analytic hierarchy process and TOPSIS [46], and Pareto-based methods to find and visualize the set of nondominated solutions [47,48]. Using a Pareto-based method presented visually allows for the nondominated solution set to be easily understood by stakeholders and is relatively simple to incorporate into discussions surrounding design decisions [12].
2 Methods
The general process used in this article is outlined in Fig. 1. To illustrate and describe how to use the modeling approach, a case study is presented that predicts the impact of residential solar panels in the state of Utah in the United States under various scenarios. A scenario in this article is defined as a particular set of modeling inputs. For example, a particular solar panel type and price projection are the different scenarios investigated. As with all product impact models and assessments, the case study does not predict all product impacts. It predicts a subset of indicators for certain aspects of environmental, social, and economic impacts. Prediction or assessment of a subset of product impacts remains valuable for making design decisions that result in improved product impacts.
2.1 Model Planning.
The first stage of the process is model planning. Model planning requires defining the objective of the model and the types of scenarios that will be used to explore the impact space of the product. For the case study, the objective is to predict the environmental, social, and economic impacts of residential solar panels in the state of Utah between the years 2022 and 2050. Although the model can be used for the entire United States or any state in the country, the results presented in this article are limited to the state of Utah because the computational resources required to model the entire country are high. The results for Utah are meant to serve as an illustration of using the method. To understand the impact space, scenarios with different product designs and pricing structures were investigated. This involved using three types of solar panels, each under three different pricing scenarios. The parameters of the solar panels can be seen in Table 1, with the values for the solar panel performance derived from estimates published by NREL [49,50]. After initial model planning, the ABM and the impact assessment per functional unit can be carried out in parallel.
Solar panel performance parameters
| Efficiency | Degradation | Lifetime | |
|---|---|---|---|
| Panel type | (%) | (%/yr) | (yrs) |
| Monocrystalline silicon | 21 | 0.5 | 25 |
| Polycrystalline silicon | 19 | 0.5 | 25 |
| Copper–indium–diselenide (CIS) | 12 | 1.0 | 25 |
| Efficiency | Degradation | Lifetime | |
|---|---|---|---|
| Panel type | (%) | (%/yr) | (yrs) |
| Monocrystalline silicon | 21 | 0.5 | 25 |
| Polycrystalline silicon | 19 | 0.5 | 25 |
| Copper–indium–diselenide (CIS) | 12 | 1.0 | 25 |
2.2 Distributed Generation Market Demand Agent-Based Modeling.
The ABM model used in the analysis is the Distributed Generation Market Demand Model (dGen) developed by NREL [13]. This model predicts the adoption of renewable energy in a region using a Bass diffusion model method [51]. The agents in the model represent potential customers with their respective housing developments. These agents represent a weighted group of individuals living in the marked county relative to the real population. The agents are then randomly assigned locations relative to the chosen region. Each location has different attributes that are based on data sampled from the region. This includes energy consumption, energy costs, economics, and geopolitical influences. These attributes are then randomly assigned to the agents. The agent is also given a solar adoption parameter based on market diffusion models. These models are based on past technological diffusion models [52], as well as payback period and percentage of monthly bill savings of the agents. The dGen model was calibrated and validated using historical solar panel adoption data [53]. dGen uses a solar database from the National Solar Radiation Database. This database is used to estimate the solar irradiation incident on a region. This database is used to approximate yearly weather in the model.
To model residential housing, dGen uses a stochastic system to choose the roof area, azimuth angle, and slope of each agent based on information collected from the region. These parameters are then used with the solar model and the input of the solar panel to calculate the power per square area developed by the solar panels. To calculate the economic impact of solar panels, dGen uses regional data on energy costs, state incentives, and assumes one of two business models. The agent can own or rent solar equipment. One of the business models are randomly assigned to the agents, with each business model having an equal probability of being assigned.
The design variables changed in the model scenarios included different types of solar panels. These files included performance increase estimates per year, as well as degradation estimates. The photovoltaic wholesale price was varied as well. The model had built-in default wholesale price scenario files that were used. These files included a high, medium, and low range.
2.3 Environmental Impact Assessment.
Although there are many methods for LCA, the ReCipe2016 [33] method is well established in the literature. This study uses the ReCiPe2016 method with a hierarchical perspective to perform a midpoint assessment. A midpoint analysis was selected because it has the most detailed indicators related to environmental impact, with 17 indicators. There are numerous guides for completing an environmental LCA [29,54,55]; therefore, this article will not give a detailed explanation of how to carry out an LCA. Rather, this research will provide insight into how LCA results are integrated with ABM results. Across the various social, environmental, and economic impacts, there is some overlap between the three categories. For example, human health and safety are often considered social impacts [56], but in the ReCiPe 2016 method, there are indicators related to human health, namely, human carcinogenic toxicity and human noncarcinogenic toxicity. These two indicators come from the LCA, but in this article, during the analysis of the results, they will be treated as social impact indicators. The output from the LCA needs to be in terms of a functional unit, where the output of the ABM provides the number of functional units needed in the modeled scenario. In the case of the solar panel adoption model under discussion, the functional unit of the LCA is the environmental impact per kilowatt-hour of energy generated, and the ABM outputs the number of kilowatt-hours generated by the solar panels throughout the simulation. The LCA metrics use data from an LCA performed by Milousi et al. [57] using the Ecoinvent v3.4 database, and the LCA metrics for each of the 17 indicators can be seen in Table 2.
LCA metrics per functional unit
| Imact category | Unit | Polycrystalline | Monocrystalline | CIS |
|---|---|---|---|---|
| Global warming | 4.43 × 10−2 | 5.24 × 10−2 | 3.95 × 10−2 | |
| Stratospheric ozone depletion | 2.06 × 10−8 | 2.45 × 10−8 | 1.75 × 10−8 | |
| Ionizing radiation | 4.08 × 10−3 | 4.45 × 10−3 | 3.96 × 10−3 | |
| Ozone formation, human health | 1.05 × 10−4 | 1.20 × 10−4 | 9.09 × 10−5 | |
| Fine particulate matter formation | 1.04 × 10−4 | 1.23 × 10−4 | 9.39 × 10−5 | |
| Ozone formation, terrestrial ecosystems | 1.10 × 10−4 | 1.25 × 10−4 | 9.26 × 10−5 | |
| Terrestrial acidification | 2.21 × 10−4 | 2.47 × 10−4 | 2.07 × 10−4 | |
| Freshwater eutrophication | 3.78 × 10−5 | 4.07 × 10−5 | 4.62 × 10−5 | |
| Terrestrial ecotoxocity | 1.17 | 1.13 | 4.62 × 10−1 | |
| Freshwater ecotoxicity | 1.16 × 10−2 | 1.17 × 10−2 | 1.30 × 10−2 | |
| Marine ecotoxicity | 1.53 × 10−2 | 1.54 × 10−2 | 1.69 × 10−2 | |
| Human carcinogenic toxicity | 4.17 × 10−3 | 4.33 × 10−3 | 4.19 × 10−3 | |
| Human noncarcinogenic toxicity | 1.63 × 10−1 | 1.64 × 10−1 | 2.00 × 10−1 | |
| Land use | 1.23 × 10−3 | 1.23 × 10−3 | 9.60 × 10−4 | |
| Mineral resource scarcity | 5.54 × 10−4 | 5.42 × 10−4 | 8.21 × 10−4 | |
| Fossil resource scarcity | 1.08 × 10−2 | 1.27 × 10−2 | 9.40 × 10−3 | |
| Water consumption | m3/kWh | 1.35 × 10−3 | 1.17 × 10−3 | 3.22 × 10−4 |
| Imact category | Unit | Polycrystalline | Monocrystalline | CIS |
|---|---|---|---|---|
| Global warming | 4.43 × 10−2 | 5.24 × 10−2 | 3.95 × 10−2 | |
| Stratospheric ozone depletion | 2.06 × 10−8 | 2.45 × 10−8 | 1.75 × 10−8 | |
| Ionizing radiation | 4.08 × 10−3 | 4.45 × 10−3 | 3.96 × 10−3 | |
| Ozone formation, human health | 1.05 × 10−4 | 1.20 × 10−4 | 9.09 × 10−5 | |
| Fine particulate matter formation | 1.04 × 10−4 | 1.23 × 10−4 | 9.39 × 10−5 | |
| Ozone formation, terrestrial ecosystems | 1.10 × 10−4 | 1.25 × 10−4 | 9.26 × 10−5 | |
| Terrestrial acidification | 2.21 × 10−4 | 2.47 × 10−4 | 2.07 × 10−4 | |
| Freshwater eutrophication | 3.78 × 10−5 | 4.07 × 10−5 | 4.62 × 10−5 | |
| Terrestrial ecotoxocity | 1.17 | 1.13 | 4.62 × 10−1 | |
| Freshwater ecotoxicity | 1.16 × 10−2 | 1.17 × 10−2 | 1.30 × 10−2 | |
| Marine ecotoxicity | 1.53 × 10−2 | 1.54 × 10−2 | 1.69 × 10−2 | |
| Human carcinogenic toxicity | 4.17 × 10−3 | 4.33 × 10−3 | 4.19 × 10−3 | |
| Human noncarcinogenic toxicity | 1.63 × 10−1 | 1.64 × 10−1 | 2.00 × 10−1 | |
| Land use | 1.23 × 10−3 | 1.23 × 10−3 | 9.60 × 10−4 | |
| Mineral resource scarcity | 5.54 × 10−4 | 5.42 × 10−4 | 8.21 × 10−4 | |
| Fossil resource scarcity | 1.08 × 10−2 | 1.27 × 10−2 | 9.40 × 10−3 | |
| Water consumption | m3/kWh | 1.35 × 10−3 | 1.17 × 10−3 | 3.22 × 10−4 |
The system boundaries used for the LCA include the product life cycle phases from production to disposal at the end of the solar panel. Impacts were calculated for a photovoltaic system that included the solar panels, mounting hardware to an existing roof, inverter, and necessary wiring. All components except the solar panels were the same in each analysis. The usable lifespan for the system was assumed to be 30 years. See the study by Milousi et al. [57] for more details on the LCA.
2.4 Social Impact Assessment.
Different methods can be used to assess social impact. SLCA can be used in this method if using an approach that will result in quantitative results at the midpoint or endpoint level. This study uses a method proposed by Stevenson et al. [42], where relevant social impact categories are selected from the 11 social impact categories identified by Rainock et al. [56]. Once the relevant categories are identified, indicators are selected to measure those categories. The creation of jobs and the reduction in utility bills, which relate to the impacts on the paid work category, have previously been used to measure the social impact of solar panels [58,59]. In addition, there are impacts on human health with indicators obtained from the LCA results. Metrics obtained from the LCA are measured by impact per kilowatt-hour, while the creation of new jobs is measured as the number of jobs per megawatt of installed solar capacity. In the United States, there are 5.2 jobs created per megawatt of installed solar capacity [60]. This is likely a conservative estimate because it covers the entire solar industry. Residential solar typically requires more labor than industrial-scale projects. The mean amount that utility bills are reduced per customer that adopts a solar power generation system is output from the ABM. For a summary of social impact metrics, see Table 3.
Social impact indicators summary
| Indicator | Unit | Value |
|---|---|---|
| Jobs created | Jobs/megawatt | 5.2 |
| Utility bill reduction | $ | Output from ABM |
| Human carcinogenic toxicity | Varies by panel type | |
| Human noncarcinogenic toxicity | Varies by panel type |
| Indicator | Unit | Value |
|---|---|---|
| Jobs created | Jobs/megawatt | 5.2 |
| Utility bill reduction | $ | Output from ABM |
| Human carcinogenic toxicity | Varies by panel type | |
| Human noncarcinogenic toxicity | Varies by panel type |
2.5 Economic Impact Assessment.
Economic impact assessment in sustainability is typically performed using LCC [61], where the total cost of obtaining the system, its maintenance and operation cost, and the cost of disposal are summed. In the case of residential solar panels in this model, only the cost of obtaining the system and its maintenance and operation cost were considered due to the high uncertainty associated with solar panel recycling at the end of the 25-year life. Solar panel recycling costs are likely to change dramatically by the time the first panels, that were installed in 2022 within the model, are recycled in 2047. The life cycle costs are output from the ABM. In addition to LCC, this model also considers the payback period for the initial investment in a solar panel system, which is also output from the ABM.
2.6 Results Scaling and Integration.
Once the total impact at the population level is identified for each impact metric, comparisons can be made between the categories of social, environmental, and economic impacts.
2.7 Tradeoff Characterization.
This article uses a visually presented Pareto-based method to identify impact tradeoffs and facilitate discussions about the impact tradespace. To visualize the results, three scatterplot matrices are created. The three matrices are the comparisons between the social, environmental, and economic product impacts. Each dimension of the matrix contains an impact category; e.g., each row contains a social impact metric, and each column contains an environmental impact. By creating scatterplot matrices, pairwise comparisons can be made between all impact metrics. For each pairwise comparison, the Pareto front can be highlighted to show the curve of nondominated solutions [45]. Then the scatterplot matrices are ready to be used to help characterize the impact trade space.
3 Results
Using the dGen model, scenarios were explored for the three types of panels, each with the projections of low, medium, and high prices provided by NREL in the model. This gave nine different scenarios to compare the results. The results will be compared for adoption and energy generation and product impacts.
3.1 Solar Panel Adoption.
The first result examined is the adoption and energy generation of the residential solar panels installed during the scenarios. From Fig. 2, we see that the price scenario is the main differentiating factor for the number of kilowatts of solar capacity installed, as well as the kilowatt-hours of energy generated. The difference between the results of the polycrystalline silicon and monocrystalline silicon panels is minimal. The CIS panel lags behind both in solar capacity installed and in energy generated, due to the lower efficiency of this type of panel. In the high-price scenario, the CIS panel has almost the same installed solar capacity, but it is still behind in energy generation.
Kilowatts of solar capacity adopted and kilowatt-hours of power generated at the population level
Kilowatts of solar capacity adopted and kilowatt-hours of power generated at the population level
3.2 Impact Analysis.
The results of the product impacts at the population level reveal tradeoffs in some cases. Figures 3–5 show scatterplot matrices, where each axis represents one of the three pillars of sustainability. Between the three figures, all pairwise impact comparisons are made. As with solar panel adoption, the price scenario is the largest differentiator between the different scenarios. Due to space constraints within the article, only a subset of the environmental impacts is visualized here to ensure that the visualization fits on a single page. Data for all impact metrics are available upon request.
Social versus environmental impacts, with Pareto front line or highlighted optimal solution
Social versus environmental impacts, with Pareto front line or highlighted optimal solution
Social versus economic impacts, with Pareto front line or highlighted optimal solution
Social versus economic impacts, with Pareto front line or highlighted optimal solution
When examining environmental impacts alone, there is not a single panel type that performs better in all cases, see the vertical axis in Fig. 3. When examining each pricing scenario individually, the thin-film CIS panel performs better in all environmental impact metrics shown in the figure, except for the mineral resource scarcity metric due to the more scarce metals required for its construction. High-cost scenarios have the lowest overall environmental impacts, but that is only due to fewer agents that adopt solar panels based on the high price of the system. The high-cost scenarios also have some of the lowest social impacts for utility bill savings and job creation, and see the horizontal axis in Fig. 3.
In social impact metrics, the monocrystalline panel performs best in terms of mean savings in utility bills and job creation, and see the horizontal axis in Fig. 3. The thin-film CIS panel has the lowest human carcinogenic toxicity. In the human noncarcinogenic toxicity metric, the polycrystalline panel is the best performing, but only marginally better than the monocrystalline panel. Monocrystalline and polycrystalline panels have similar materials required for construction.
For each pricing scenario, there is no significant difference between the panel types for the measured economic impacts. There are also no significant differences in price between the panel types, so the low, medium, and high pricing scenarios provided by NREL in the dGen model are identical between the three panel types. Future work could examine panel pricing scenarios in more detail.
3.3 Tradeoff Characterization.
In Figs. 3–5, we have highlighted the Pareto front. For the cases where there is a single optimal scenario, that point has been highlighted, such as in the first row and second column of Fig. 3. The Pareto front represents the set of optimal scenarios on the tradeoff curve. The figures highlight different tradeoffs that exist in the product impact space. One tradeoff that occurs in many cases is between the social impacts of utility bill savings and job creation with the environmental impacts. We observe that the scenarios with the highest utility savings and jobs created also have high environmental impacts because there are more solar panels adopted in the model, and see the first and fourth columns in Fig. 3. There are some scenarios and impact comparisons where there is an optimal solution and no tradeoff exists. In some cases, there may be an easy choice in a tradeoff such as where there is a greater improvement in one metric for a small penalty in the other, such as when examining the carcinogenic toxicity column in Fig. 4. In other cases, tradeoff visualizations will be just one data point in the decision-making process. The data from the modeling and tradeoff visualizations allow one to examine where impact tradeoffs occur across a wide range of environmental, economic, and social impact indicators. Although it is difficult to make explicit rules of when to prioritize different impact categories in the decision-making process, identifying impact tradeoffs and presenting them in a way that can be understood by various stakeholders has the potential to improve design decisions regarding sustainability.
3.4 Comparison to Other Power Generation Sources.
In cases where a new technology is replacing an older technology, such as the transition to renewable energy sources from fossil fuel energy sources, it is useful to examine the net impact of the technology. In the example of residential solar power, when examining the impact of solar power in isolation, the lowest environmental impacts are the cases where the fewest panels are installed. This does not consider the power source that is being replaced by residential solar panels. To examine the total impact of residential solar power scenarios combined with existing power sources, this article will use the scenario with the highest kilowatt-hours of power generated by residential solar power systems as a baseline to compare against. The highest kilowatt-hour generating scenario was the monocrystalline panel with the low-cost case; this scenario will serve as the baseline case. For the other scenarios with less power generated by the solar panels, the difference in power generated between a particular scenario and the baseline will be made by either coal or natural gas power; see Fig. 6 for the proportion of residential solar power to the other power source.
This analysis will examine each power source individually rather than a percentage of numerous types of power generation. Coal and natural gas power will be used for the comparison. Data were not available for each indicator used in Sec. 3.2. Therefore, it is still necessary to make comparisons of the analysis with only solar panel impacts, as presented in Secs. 3.2–3.3, and solar panel impacts combined with the other power generation methods. These two analyses will enable a more holistic understanding of the various impacts of the new technology. The economic impact analysis that combines solar and other energy sources uses levelized cost of energy (LCOE) as a metric for the lifetime cost. This is a method for the lifetime cost of energy per unit of power generation that takes into account system costs, operations, maintenance, end of life, and the price of carbon emissions. Data were obtained from reports by the International Energy Agency (IEA) [62] and NREL [63], and for the LCOE values of the different energy sources, see Table 4. The LCOE uses $ 30/ton for the carbon emission cost. Data for the LCA of coal and natural gas production were obtained from work by Rapa et al. [64].
Levelized cost of energy for different sources
| Energy source | LCOE USD/MWh |
|---|---|
| Residential solar | 81 |
| Coal | 110 |
| Natural gas | 45 |
| Energy source | LCOE USD/MWh |
|---|---|
| Residential solar | 81 |
| Coal | 110 |
| Natural gas | 45 |
After the total impacts are calculated, the visual comparisons can be used as described in Sec. 3.3 to characterize the tradespace. Figures 7–9 show the comparison plots. In these figures, a power scenario is presented using each pair of panel type and other power generation methods, as well as the baseline case where the residential solar comprises all the impact metric. In these comparisons, similar to the previous comparisons with only solar power generation in Sec. 3.3, there is not a single scenario that is best or worst across all comparisons. Scenarios using coal as the other energy source are on the Pareto front less often than other scenarios, but in impact measures such as terrestrial ecotoxicity, the scenarios containing monocrystalline and polycrystalline panels have higher negative environmental impacts. The Pareto front is highlighted in each figure. Of all the comparisons, only three had an optimal scenario; the other comparisons all had tradeoffs to some extent.
Net social versus environmental impacts, with Pareto front line or highlighted optimal solution, carcinogenic toxicity, and noncarcinogenic toxicity values are presented on a logarithmic scale
Net social versus environmental impacts, with Pareto front line or highlighted optimal solution, carcinogenic toxicity, and noncarcinogenic toxicity values are presented on a logarithmic scale
Net social versus economic impacts, with Pareto front line or highlighted optimal solution, carcinogenic toxicity, and noncarcinogenic toxicity values are presented on a logarithmic scale
Net social versus economic impacts, with Pareto front line or highlighted optimal solution, carcinogenic toxicity, and noncarcinogenic toxicity values are presented on a logarithmic scale
Net environmental versus economic impacts, with Pareto front line or highlighted optimal solution
Net environmental versus economic impacts, with Pareto front line or highlighted optimal solution
4 Discussion
The method shown here illustrates how to predict social, environmental, and economic product impacts at the population, and how to characterize sustainability tradeoffs that may exist in the design of products and systems. This work expanded on the work of Liechty et al. [12] by providing methods for visual tradespace exploration in all three dimensions of sustainability. Using a Pareto-based method allows one to easily integrate impact results into the decision-making process. Quantitative impact modeling during the design process is just one tool that can be combined with other qualitative and quantitative methods to improve design decision making.
There are limitations associated with the case study used to demonstrate this method. A limited number of social impacts were considered, and in future work, more social impacts could be considered that focus on more stakeholders in the product life cycle, such as workers in the manufacturing of the panels or utility companies. Future work can also improve the pricing scenarios to illustrate how small differences in panel pricing ultimately affect impact. As with all predictive models that predict trends over many years, the model uses assumptions about future trends that may vary from the predictions. This model has assumptions about solar panel cost, performance, and impact. Estimates are made based on current trends, but ultimately the actual results in the coming years will vary from the predicted results. Ultimately all models are wrong, but predictive models can enable making better choices in the present while making assumptions about the future. While these are limitations within the case study, the general method remains valuable in the decision-making process.
The results of the case study reveal the complexity of improving the impacts of products in a sociotechnical system. There was no single scenario on the Pareto front in every impact comparison. It also illustrates that there is much more to the impact space than changes to product parameters; aspects of the system design, including business and policy decisions, are also important. For example, if subsidies were increased to lower the prices for each scenario, the adoption of solar panels would increase, which may also influence future taxes to pay for subsidies. Although, the specifics for navigating the sustainability trade space will vary based on contextual factors. One approach in moving forward in the decision-making process is to select two to three objectives that the decision-making team deems most important. Once the objectives are selected, the team may investigate the pairwise comparisons for each objective and identify the nondominated solutions in each comparison. The next step would be to identify if there is a scenario that is part of the nondominated set for each of the objective comparisons. To illustrate this, an example using objectives of minimizing total cost, carbon dioxide emissions, and carcinogenic toxicity is used. When examining the pairwise comparisons, the scenario of the CIS panels replacing natural gas power is a nondominated solution in each comparison. In a case where no scenario is a nondominated solution for each objective comparison, other methods will need to be applied. These methods can include examining the distance to the Pareto front or using a multicriteria decision-making method. Future work will examine the use of multicriteria decision-making methods for sustainability tradeoffs at the population scale.
With increasing calls for improved sustainability and progress toward the SDGs, tools to help make better design decisions for impact can become a valuable part of the decision-making process. These tools become more important when products and infrastructure will be in place for many years or decades because it is too costly to iterate on the design of the product, system, or infrastructure.
5 Conclusion
The contribution of this article is to demonstrate an approach for the predictive modeling of the social, environmental, and economic impacts of a product and an approach to characterize potential sustainability tradeoffs in the three pillars of sustainability. Furthermore, this work demonstrates how the impact of a functional product unit scales to the population level across the three pillars of sustainability. Tools to improve the impacts of products will be especially important as society makes significant shifts to make progress in complex societal challenges such as climate change. Many of these large shifts will require massive investments, and the systems may be in place for many years. With large amounts of resources and the high cost of fixing errors, the ability to use predictive modeling for the quantification of product impacts and characterization sustainability tradeoffs will enable improved design decisions regarding sustainability.
Acknowledgment
The authors would like to recognize the National Science Foundation for providing the Grant Nos CMMI-1662485 and CMMI-1632740 that funded this research. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.
Conflicts of Interest
There are no conflicts of interest.
Data Availability Statement
The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.
