Evaluation of characterization models for the photochemical smog impact category focused on the Brazilian reality

The Life Cycle Impact Assessment (LCIA) is composed of characterization models, and in Brazil, the methodological and scientific LCIA framework is still under development. The research’s aim was to evaluate the literature available characterization models to photochemical smog category. Thus, the contribution of work is recommending one of these models to be used in Brazilian LCA studies, standardizing the studies in Brazil. The methodology consisted of searching the literature and selecting, describing and analyzing the characterization models as well as elaborating a table of criteria for better comparison. Aiming to visualize the differences in the results of each selected model, a case study was applied to analyze the photochemical smog formation potential to the transport of one ton of sugar using two transportation modes (road and railroad). Five characterization models related to smog category were selected, described and compared. Herewith, it was observed that the models present significant differences, that is, each model presents Characterization Factors (CF) for different categories within the environmental impact chain of the photochemical smog (midpoint and/or endpoint), differences in modeling, scale of the model (regional, continental or global), quantity and quality of elementary flows, etc. Those factors have influence in the CF’s calculation and, consequently, the LCA’s results, in the same case study. The criteria table’s results suggested that the model of Van Zelm et al. (2016) – World (midpoint and endpoint), is the best interim option to be used in studies of LCA in Brazil, because it was the model that resulted in the highest grade referring to the established criteria and it presents results on a Global scale. However, the results do not rule out the need for regionalization studies, which would develop a model that presents results and studies directed to the Brazilian reality or adjust the model of Van Zelm et al. (2016) - Brazil.


INTRODUCTION
Lately, the research focusing at sustainable development is increasing, aiming to reduce the environmental and social impacts. One way to improve these studies is through tools that quantify the life cycle impacts of products and services, such as Life Cycle Assessment (LCA), which provides a comprehensive view of potential impacts. The LCA is composed of four stages, the third one -Life Cycle Impact Assessment (LCIA) -allows the calculation of environmental impacts by means of LCIA methods, which are composed by mathematical models (also called characterization models) and facilitate the understanding of the impacts (Hauschild et al. 2011).
The scientific community developed regionalized LCIA characterization models at continental and regional scale to different regions around the world, as well as to a global scale. Therefore, LCA practitioners can choose different LCIA characterization models, for different impact categories. However, there are large differences between the models found in the literature, which lead to two problems: (1) different models yield different results to the same case study and; (2) the uncertainties in the results are high (Renou et al. 2008).
One of the scientific challenges in the LCA community is the regionalization of LCIA methods, once "the location of the source and its surrounding conditions influence the fate of the emitted pollutant and the subsequent exposure it causes" (Potting et al. 2006). This way, LCA studies based on LCIA methods with regionalized characterization models bring more reliable and accurate results, resulting in more credible studies (Bare 2010;Finnveden et al. 2009). LCA studies in Brazil use the methodologies and database of other countries, generating subjectivity in this phase (Associação Brasileira de Normas Técnicas 2014). Therefore, it is necessary that the LCIA methods and their characterization models represent the Brazilian reality (Ministério do  Photochemical smog is a phenomenon that depends on meteorological conditions and certain concentrations of secondary pollutants emitted by the life cycle of products and services (Baird, Cann 2011). The negative impacts are due to their reactive nature that allows them to oxidize organic molecules (Santos 2006). Brazil has large urban centers, which have great potential to emit precursor gases of photochemical smog, such as the Metropolitan Region of São Paulo (Companhia Ambiental do Estado de São Paulo 2015).
With that in mind, this research aimed to compare and provide information about the characterization models for the photochemical smog impact category. In order to analyze the possibility of recommending one of these models to be used in Brazilian LCA studies, standardizing studies in the country. It also aims to apply the selected characterization models in a case study, where it is evaluated the photochemical smog formation potential from transporting one ton of sugar, considering two routes (road and railroad).

METHODOLY
The methodology adopted to compare the characterization models is presented in detail. Then, the information about the considered case study was presented.

Selection and description of characterization models
In order to carry out the study of the LCIA characterization models, the initial procedure was a bibliographical review of the different LCIA methods that present characterization models for the photochemical smog impact category, as well as the models that are not part of the LCIA methods. The selected models were Derwent et al. (1998) takes the name of this one, and when global simply by "Global"; and in some cases also includes the number of sub-regions.

Criteria for comparative analysis
In order to compare and evaluate the conformity of the characterization models studied, three main criteria and their sub-criteria were established. These criteria were adapted from the general criteria elaborated by the Research Network in LCIA -Brazil, present in Ugaya et al. (2016) and Ugaya et al. (2019), taking into consideration the study of the selected characterization models. Thus, the three main criteria (Scope, Scientific Robustness and National CF) classifies as the scoring procedure in Table 1.

Application of the models: a case study
The LCIA methods, composed by characterization models, optimize and facilitate LCA studies.
To complete the comparison and analysis, a case study applied the selected characterization models in the previous step, which are: Derwent et al. (1998)

Path and database selection
In order to compare the LCIA characterization models, they were applied to the same case study.
It simulated one ton of sugar's transportation, leaving the center of Campinas/São Paulo/Brazil to the Seaport of Santos, through two routes: road and railroad. The railroad route (186 km  To analyze the LCIA models and the different routes, the SimaPro ® software version 8.3.0.0 was used (PRé 2016). Due to the lack of a Brazilian database, the information of the transport model's types comes from the database ecoinvent 3 -allocation, recycled content -unit present in the SimaPro and which represents the European types of transport. For railroad transport, it was used the database of a train powered by diesel, with particle filter. To road transport, it was used the database of a lorry powered by diesel, with a load capacity greater than 32 tons, which syntax is present in Table 3. The reference region (continent, country, and specific region), the impact category (evaluation level) of the models, the relative percentage of the routes and the final reference unit was detailed, in order to facilitate the evaluation and comparison between the characterization models.
The characterization models present different final reference units, which makes them not directly comparable to each other. However, through a relative percentage of the scenarios studied, we can compare the models and the paths (a procedure also known as internal normalization of results). In this way, the path with greater impact capacity received 100% of impact power and the path with the lowest capacity received a percentage relative to the one with the greatest impact.

RESULTS AND DISCUSSION
The description of the characterization models and their comparison were shown. Also, the results of the case study were presented.

Selection and description of characterization models
With the bibliographic review, we selected, described and embedded in the cause-effect chain five LCIA characterization models. The cause-effect chain adapted from ILCD Handbook (Hauschild et al. 2011), represents that the impacts at the endpoint level are the end-of-chain damage (damage on human health, crops and ecosystems), and the damage at the midpoint level are all the impacts that precede them ( Figure 2).
It is also observed in Figure 2, that there are three models that present CF at midpoint level: (1) Van

Description of the LCIA characterization models
The Table 4 presents a description of the main characteristics of each characterization model evaluated.  (2013) do not yet make part of any LCIA method.

Criteria for comparative analysis
The results of the criteria table (Table 2) and the models' studies supported its comparative analysis. This aimed to compare and analyze the characterization models and allowed to detect the pros and cons of each model studied. Therefore, three groups of comparison divided the models: (1) those with CF for midpoint level; (2) those with CF for damage to ecosystems; and (3) those with CF for damage on human health.

1) Models that have midpoint CF (POCP or OFP):
The model of Van Zelm et al. (2008, has midpoint CF (OFP) to 137 elementary flows and its resolution (0.25° × 0.25°) represents the formation and dispersion of smog in Europe.
It is part of the LCIA method ReCiPe 2008, which has recognition by the scientific community.
The model description presents transparency and accessibility of the equations, variables and chemical transport model (Lotos-Euro). According to these reasons, high compliance was considered to scope and scientific robustness criteria, and not applicable to the national CF criterion.
The model of Derwent et al. (1998), Jenkin and Hayman (1999) (Bartholomé, Belward 2005). Therefore, the model shows moderate compliance to the scope criterion, high to scientific robustness, and moderate/low to national CF criterion. long as there is a chemical transport model that represents the formation and dispersion of the photochemical smog in the region of interest, as well as regional secondary parameters.
3) Models that have CF for damage on human health: The Brazil, but only for two elementary flows: NOx and NMVOC. It considers a low resolution (1° × 1°) and a single CF for Brazil as a whole. Due to this information, the scope criterion has moderate compliance, the scientific robustness criterion has high compliance, and the national CF criterion has moderate/low.
The model of Krewitt et al. (2001) has endpoint CF (damage on human health and damage to crops) for two elementary flows: NOx and NMVOC, in Europe. It makes use of chemical transport models to set up the formation of the photochemical smog, which have an adequate resolution (0.5° × 0.5°). The model, used by the LCIA method EcoSense, has a reliable scientific basis, but with limitations, i.e., has no transparency and accessibility of equations and variables. With this, the model shows high compliance to scope criterion, moderate/low to scientific robustness and it is not applicable to national CF.

Final score of comparative criteria analysis
From the results of Table 2 and the comparative analysis of the models, we obtained the final score considering the evaluation of the criteria for each characterization model (  Table 5. Final score of the evaluated characterization models. The results in Table 5

Factors influencing CF calculations
LCA practitioners can choose different LCIA characterization models, for different impact categories; however, there are large differences between the models found in the literature (Renou et al., 2008). The characterization models to the photochemical smog impact category present in this study show differences between them, which relates to several factors that influence the final CF calculation, such as:

Application of the models: a case study
From the models studied, it was possible to obtain results for 14 categories related to the photochemical smog impact category. Table 6 shows the relative impact of each route, for each characterization model. We observed in Table 6 that the models showed differences in the results, to the same case study. According to Renou et al. (2008), this happens because the models found in the literature are different from each other.  To the model of Derwent et al. (1998), Jenkin and Hayman (1999), the road had a photochemical oxidation potential of 11.6% greater than the railroad. The four elemental flows that contributed most to smog formation in the road route were: fossil carbon monoxide ( Krewitt et al. (2001) suggests in case of negative damage factors, it might be defendable to set characterization factors to zero, thus indicating that the potential for environmental damage is low in a specific context.
This way, the NMVOC elemental flow presented 100% of the photochemical smog formation potential.
The analysis showed that the difference in the phenomenon formation potential, between the modes of transport, is not significant. The models with the highest difference in smog formation potential were: Krewitt et al. (2001) (damage on human health) and Derwent et al. (1998), Jenkin and Hayman (1999), for which the railroad route showed to be 25.9% and 11.6% (consecutively) less impacting than road. Aside from these models, the others present an average difference of 2.4% of smog formation potential between the two routes. Small differences in results can be related to the type of fuel and the elementary flows considered in the transport database present in the ecoinvent, and the unit processes that contribute to the impact, within The unit processes that contributed to the impact of each transport were similar between the transport modes and characterization models, however, they have different weights in each characterization model. For railroad transport, the three unit processes that contributed the most, by the models of Van Zelm et al. (20082016), were: the transport by railroad (85.9%); the construction of the railroad (6.1%); and the diesel used by the train in transportation (4.9%).
By the model of Derwent et al. (1998), Jenkin and Hayman (1999), the three unitary processes that contributed the most were: the transport by railroad (36.6%); the diesel used by the train in transportation (24.6%); and the construction of the railroad (22.4%). For road transport, the three unit processes that contributed the most, by the models of Van Zelm et al. (20082016), were: the transportation by truck (69.4%); the road construction (19.2%); and the diesel used by the truck in transportation (9.4%). By the model of Derwent et al. (1998), Jenkin and Hayman (1999), were: the diesel used by the truck in transportation (41.3%); the transportation by truck (24.9%); and the road construction (24.4%).
Regarding the results of the considered case study, the impact potentials to the photochemical smog category were similar to railroad and road transport. However, to other impact categories this difference can be significant, for example to particulate matter formation, land occupation, etc. This way, in order to assess the potential impact of transport modes (road and railroad), it is necessary to consider other impact categories.

CONCLUSION
This research provides subsidies to Brazilian Program of Life Cycle Assessment (in Portuguese "Programa Brasileiro de Avaliação do Ciclo de Vida -PBACV"), taking into account the objectives of its second strategic theme: to identify existing LCIA methods and characterization models for the photochemical smog impact category as well as to collect and make available information about these models.
Individual LCA practitioners can choose different LCIA characterization models, for photochemical smog category. However, there are large differences between the models found in the literature. These differences influence the CF calculations, and to pick one or other model leads to different results and conclusions to the same LCA study. Therefore, it needs to be careful in choosing these models, taking into account the scope of the study and aiming a better LCA.
Based in the analysis of the characterization models and the case study presented in this research, we may conclude that the elementary flows that most contribute to the smog phenomenon formation are NOx, NMVOC, carbon monoxide and sulfur dioxide. Therefore, the presence of CF of these elementary flows in a characterization model inventory to the smog impact category becomes a relevant factor on choosing the most appropriate model.