Life cycle assessment of in‐person, virtual, and hybrid academic conferences: New evidence and perspectives

This study contributes to the debate on the environmental impacts of academic conferences by comparing the life cycle impacts of a sample of real‐world in‐person, virtual, and hybrid conferences with different features and organizers. Results show that virtual formats reduce impacts by two to three orders of magnitude across all impact categories (for global warming, averagely from 941.9 to 1.0 kg CO2eq per person). The hybrid case study, with a share of 69% virtual attendees, displays an average 60% reduction in indicator results, less than ideal cases where the farthest attendees join online. The cross‐conference comparison allowed identifying several drivers of impact variation. For in‐person conferences, some never addressed drivers were uncovered, including the energy sources and systems used to supply the venue or the number of non‐local staff members and exhibitors. For virtual conferences, the main impact driver is the average time spent online by delegates, surprisingly more related to virtual experience design (e.g., synchronous vs. asynchronous presentations) than conference duration. The study further summarizes mitigation options from the literature and proposes new ones, such as selecting a venue supplied by a biomass‐fueled district heating system or with a green electricity contract (around −41 and −1.9 kg CO2eq per person, respectively). Lastly, our work highlights some inconsistencies that affect current conference assessments and proposes new research avenues, advocating the need to shift the focus from optimizing single conferences to considering the optimal portfolio of conferences and other activities for academic societies to meet their members’ needs while minimizing environmental impacts.


INTRODUCTION
Conferences represent a well-established practice in academia to facilitate the creation and dissemination of knowledge (Rowe, 2018).They serve multiple important roles in the career of academics, enabling them-among others-to promote their work, attract feedback, and meet like-minded peers to build new research collaborations (Donlon, 2021).These benefits, however, come with a significant drawback: conferences can be a high resource-demanding and emission-intensive process (Hischier & Hilty, 2002).In the past two decades, scholars in different fields have started to disclose the greenhouse gas (GHG) emissions associated with conferences, especially conference travel, showing alarming figures (Jäckle, 2019;Kuper, 2019).For instance, Klöwer et al. (2020) found that the per-capita footprint of scientists traveling to the 2019 Fall Meeting of the American Geophysical Union was about 3 tons of CO 2eq , greater than the amount many citizens around the world emit over an entire year.
Against this backdrop, a growing movement has started to question the established model of in-person conferencing and advocate the need, well before the Covid-19 pandemic, to leverage the significant improvements in videoconferencing technologies to switch to more sustainable formats, such as virtual or hybrid conferences (Fraser et al., 2017;Reay, 2003).A few pioneers led the way-see Dolci et al. (2011)-but these were isolated cases.The pandemic, however, upended the conference landscape and forced the switch to virtual formats to happen suddenly across all fields.Many regarded it as a great opportunity for academia to reinvent its conferencing model (Jordan & Palmer, 2020).More recently, some conferences have also experimented with hybrid formats, where some attendees join in person and others virtually (Langin, 2021).With the world progressively reopening, organizers are confronted with the task of designing the conferences of the future and are looking for evidence to make informed decisions.
From an environmental perspective, however, Tao et al. (2021, p. 2) highlight that "there is a minimal quantitative understanding of the environmental impacts from different modes of conferences.To understand the sustainability implications of future conferences and inform the policies, it is essential to quantify the environmental footprints of virtual, in-person, and hybrid conferences." Previous studies have extensively analyzed the GHG emissions associated with travel to in-person conferences and ways to reduce them (Burtscher et al., 2020;Coroama et al., 2012;Desiere, 2016;Fois et al., 2016;Jäckle, 2019;Klöwer et al., 2020;Kuonen, 2015;Kuper, 2019;Orsi, 2012;Ponette-González & Byrnes, 2011;Spinellis & Louridas, 2013;Stroud & Feeley, 2015;van Ewijk & Hoekman, 2021).Among these, a few studies also considered virtual and hybrid conferences, either treating them as carbon-neutral scenarios (Jäckle, 2019;van Ewijk & Hoekman, 2021) or assessing their footprint in a simplified way (Burtscher et al., 2020;Klöwer et al., 2020).Some studies adopted a broader scope and conducted comprehensive life cycle assessments (LCA) of in-person conferences' impacts (Astudillo & AzariJafari, 2018;Hischier & Hilty, 2002;Neugebauer et al., 2020), considering activities other than delegate travel, such as accommodation, and impact categories other than climate change, such as human toxicity, thus providing richer insights to organizers.Among these, Hischier and Hilty (2002) also considered the virtual format, even if modeled through a conjectural scenario, whereas none assessed comprehensively the impacts of a hybrid conference.Two recent articles addressed this gap, both leveraging the data from a virtual conference of the Covid era and building in-person and hybrid counterfactuals through scenario analysis (Jäckle, 2021;Tao et al., 2021).Table S1.1 in Supporting Information S1 provides a summary of previous studies regarding their scope and main methodological choices.
Notably, none of the comprehensive LCA studies considered multiple conferences with different features (e.g., size, location, and audience), thus hampering the transferability of the results and making it difficult to highlight potential drivers of impact variation across different conferences.Some non-LCA travel-focused studies considered multiple events (Jäckle, 2019;van Ewijk & Hoekman, 2021)-typically of the same academic society-but their limited scope prevented the identification of drivers of impact variations within processes other than travel and tradeoffs between different impact categories.Regarding format comparison, then, the exclusive reliance on scenario analyses means that the actual choices of stakeholders in real-world settings remain unknown and the projected results need validation.Lastly, while some studies have identified a few issues affecting the validity of format comparisons, such as functional equivalence (Hischier & Hilty, 2002) and ripple/rebound effects (Coroama et al., 2012;Takahashi et al., 2006), to the best of our knowledge, none has comprehensively and critically investigated the inconsistencies related to current conference assessments/comparisons, which may have led to an overemphasis of some issues and the neglection of others.
To address these gaps and move the debate forward, the goals of this study are to (1) quantify and compare the overall environmental impacts of real-world in-person, virtual, and hybrid conferences; (2) identify potential drivers of impact variation across conferences with a different size, location, duration, organizer, and audience; and (3) investigate the inconsistencies that can affect the validity of conference assessment results.
Additionally, a summary of the main mitigation options proposed in the previous literature and an investigation of some never-considered ones are also included in the scope of the study.

METHODS
The LCA method in compliance with ISO 14044 standard was implemented, including its four phases: goal and scope definition, life cycle inventory analysis (LCI), life cycle impact assessment (LCIA), and interpretation of the results (ISO, 2017).

Goal and scope definition
To achieve the goals stated above, we selected a convenience sample of conferences held in different formats (in-person, virtual, and hybrid) and with different features (organizer/audience, size, location, duration, period).Particularly, given the high data and resource requirements of LCA, we targeted the most recent annual meetings organized in each available format by the supporting societies, which guaranteed both the availability of accurate data and the desired heterogeneity (Table 1).

Functional unit
According to LCA standards, "comparisons between systems shall be made on the basis of the same function(s), quantified by the same functional unit(s)" (ISO, 2017, p. 8).When considering different conference formats, though, it has long been evident that videoconferencing solutions do not perform the same functions as in-person meetings (Takahashi et al., 2006).A survey we ran with the members of supporting societies-object of another article we are writing-found that virtual conferences are not able to provide the same networking and socialization opportunities as in-person, even if they provide other functions, such as greater flexibility and accessibility for underrepresented groups.That perfectly aligns with the discussion provided by Hischier and Hilty (2002) and Coroama et al. (2012), who claim that it is difficult to assume functional equivalence when alternatives involving the use of electronic versus conventional media are compared (e.g., telecommunication as a substitute for in-person interaction).New media will always bring advantages and disadvantages with them-that is, different functions-which makes the LCA requirement of functional equivalence less and less adequate in the area of information technology.In Section 3.4, we propose some solutions to tackle functional (in)equivalence in conference format comparisons.For the purpose of the analysis hereby presented, we followed the same pragmatic approach as Tao et al. (2021) and defined the functional unit (FU) as "one average conference participant," which, despite not addressing functional equivalence concerns, provides a sound basis for a fair comparison across different conferences and formats.
As an alternative to our FU, Hischier and Hilty (2002) and Neugebauer et al. (2020) refer to "holding a 3-days conference" in their single case studies, but such an FU would not be suitable for comparing multiple conferences with different sizes and durations.A per participant-day FU may instead sound reasonable having conferences with different durations, but we did not adopt it for two reasons: (i) all in-person conferences in our sample had the same duration when considering pre-conference events, which is quite standard in the OM&OR field; (ii) for virtual conferences, as we show in Section 3, duration is not clearly related to how much people attend the event, which makes it a not relevant functional characteristic-based on our experience, this is the case for in-person conferences too, as attendance is largely constrained by other professional/private commitments.

F I G U R E 1
Life cycle assessment model for academic conferences.

System boundary
To define the system boundary, we mostly referred to Neugebauer et al. (2020) and Tao et al. (2021), the only studies that considered the overall life cycle impacts of academic conferences.We further integrated input from Cavallin Toscani et al. ( 2022), who provided a comprehensive life cycle representation for any type of event.Figure 1 shows the resulting LCA model.On the left side, there are the background processes that provide energy, material, and product inputs to conference activities or dispose of their waste outputs.To model them, we relied extensively on internationally recognized LCI databases (see Section 2.1.3).On the right are the foreground processes that cluster the impactful activities associated with conference organization and delivery.To model them, we mostly relied on primary data provided by conference organizers or secondary data retrieved from literature (see Section 2.2).
Regarding the foreground processes that characterize in-person conferences, Conference organization refers to general planning activities, such as conference-related board meetings, venue inspection visits, organizing committee's and secretariat's activities, track and session chairs' activities, pre-conference participants' activities, and conference materials-including their production, shipping, and disposal.Venue refers to the conference-related use of the venue and buildings, mostly including the energy consumption of conference rooms and equipment.Exhibits include the production, transportation, and disposal of exhibition and/or career fair materials, where applicable. 1 Stakeholders' transport refers to the transportation of attendees, exhibitors, and staff to reach the conference site, and, where applicable, for industry visits.Catering includes the production and transportation of food and beverage items consumed at the conference.Lastly, Accommodation refers to overnight stays of conference stakeholders.
For virtual conferences, Conference organization has the same meaning as before, even though in this context most activities are virtualized and dematerialized.Virtual experience refers to the use of electronic and internet devices to connect event participants during and after the event, including the direct energy use and the indirect energy and materials used for device production.Hybrid conferences are, environmentally speaking, simply a concatenation of previous formats.Additional equipment is required to live stream presentations from the venue (Coroama et al., 2012), but in our model, this extra consumption is absorbed by the Venue process.
Table S1.1 shows how our boundary compares to that adopted in previous studies.As an extension, we considered some never assessed impacts, such as those associated with venue inspection visits, participants' registration activities, the life cycle of exhibition materials, the transportation and accommodation of conference stakeholders other than delegates (i.e., sponsors, exhibitors, and staff), and, for virtual conferences, platform visits and downloads after the conference.As we show in Section 3, some of these items represent significant impact drivers.Conversely, we did not consider some activities/flows, excluded mostly due to the lack of accurate primary/secondary data and/or a lack of relevance to the overall results.These encompass: (a) computer usage for abstract and/or full paper drafting, (b) composite gadgets distributed to attendees (e.g., safety kits), (c) venue water and other material consumption (e.g., cleaning products), (d) travel of stakeholders in their country to reach the airport/station of departure and (e) within the conference location for extra-event activities, (f) energy and water use for food preparation, (g) extra-conference consumption of food and other commodities, and, for virtual conferences, (h) home consumption of food, energy, and water during the conference.
Some of these items-(a), (d), (f), (h), and partly (c) and (g)-were modeled by Neugebauer et al. (2020) and/or Tao et al. (2021) through approximate secondary data and, in general, showed a negligible or very low contribution to the overall results, which justifies our exclusion.An exception is item (h), which Tao et al. (2021) showed to make up most of the virtual conferences' impacts.Its exclusion from our assessment depends on another criterion we adopted, that is, the unclear allocation of impact responsibility to the conference-valid also for items (e) and (g).Differently from Tao et al. ( 2021), indeed, we opted here for a "control" approach (Cavallin Toscani et al., 2022), where only activities under the control of conference organizers are considered, as further discussed in Section 3.4.

LCI databases and LCIA methods
The SimaPro software and LCI databases therein implemented were employed to perform the analyses.Particularly, we made extensive use of the ecoinvent 3.8 database (Wernet et al., 2016) as our primary source of LCI data for background processes.When available, datasets with geographical coverage related to the country/region where the conference took place were selected, otherwise, globally averaged datasets were used.In rare cases in which the needed process datasets were not available-especially for food production-we further relied on Agri-footprint 5.0 (Paassen et al., 2019) database.The list of employed datasets for each analyzed conference is available in Supporting Information S2-S8.
For the LCIA, in line with our goal to run a comprehensive assessment and comparison, we followed Tao et al. (2021) and selected the ReCiPe method, at the midpoint level with the hierarchist perspective, one of the most established and comprehensive methods worldwide (Huijbregts et al., 2017).

Life cycle inventory
For each conference, primary and secondary data were collected for all foreground processes in Figure 1.Sources of primary data were the conference organizers and/or society leaders, who gave us access to conference-related documents and answered our questions in ad hoc interviews.
Examples of primary data include transportation information for venue inspection visits by organizing staff, quantities of purchased conference materials, the actual amount of time delegates were connected during virtual conferences, and so on.Despite this data availability, many assumptions and secondary data had to be used to make provided information usable in an LCA setting or fill data gaps.We strived to remain consistent and when there were gaps for a conference, missing data were either extrapolated from other conferences in the sample or derived from other LCA studies or online resources (e.g., e-commerce websites for the material composition of purchased products, when not specified).As prescribed by ISO 14044, all assumptions, data sources, and collection/calculation procedures were documented in ad hoc data collection sheets to increase transparency and replicability.Particularly, for all conferences, we created a spreadsheet that details the list of modeled activities and related sub-activities/materials, and, for all sub-activities, it specifies: the name of the linked ecoinvent dataset containing relevant background data, the measured/calculated value of the flow, 2 the unit of measure, and all relevant information and assumptions regarding data collection/calculationincluding the indication of secondary data sources, if used.These spreadsheets are available in Supporting Information S2-S8, to which we refer for inquiries regarding specific flows or activities.In Supporting Information S1.2, instead, the general logic and main assumptions we used to calculate the inventory of each unit process are described.

Life cycle impact assessment results
Regarding the impact quantification and comparison across formats, Table 2 reports the LCIA results for the analyzed conferences.The first row related to the GW category shows the per-capita carbon footprints, with an average of 941.9 kg CO 2eq for in-person conferences and 1.0 kg for virtual-almost three orders of magnitude of difference.The last instead related to the WC category displays the water footprints, with an average of 2602.7 L for in-person and 9.0 L for virtual-more than two orders of difference.For other impact categories as well, the difference between in-person and virtual is mainly between two and three orders of magnitude.To understand the scale of such difference, the carbon footprint of an average in-person attendee alone is about twice the total footprint of EurOMA 2021 (488.3 kg CO 2eq ).
TA B L E 2 LCIA results.2020), Stakeholders' transport dominates most impact categories, with particularly large shares in those driven by fossil fuel consumption, such as GW (∼93% averagely), OF-HH/TE (∼97.6%), and FRS (∼93%).Accommodation instead is the primary contributor to MEu (∼63.9%) and WC (∼56.8%), with significant shares also for FEc and MEc, driven by the material and energy consumption of hotel operations.The Venue process has an average share of 2.5% across all categories and conferences, with a maximum of 9.5% for IR, due to the consumption of electricity produced from nuclear power in some conference locations (e.g., Finland for EurOMA 2019).Food production for Catering is responsible for a large share in the LU category (37%), and the shares of 18.8% and 12.1% in the WC and MEu categories.The latter two are also significantly driven by Conference organization, with contributions of 8.1% and 9%, respectively, mostly due to the production of conference materials.Lastly, Exhibits have a neglectable impact across all categories (average of 0.05%).

In-person
Virtual conferences, instead, have a more stable pattern across categories, with Virtual experience responsible for greater impact shares than Conference organization-59.2%versus 40.8% on average-both driven by the energy consumption of electronic devices and the material consumption for their production.
Focusing on INFORMS 2021-the first hybrid conference ever assessed-the impact breakdown is reported for both in-person (Figure 2) and virtual (Figure 3) attendees.For the former, the impact profile looks equivalent to that of an in-person event.Impacts are larger than other conferences for reasons that are explained in the next section (e.g., greater traveled distances).For virtual attendees, instead, the impact profile looks quite different from other virtual conferences.That occurs because the Conference organization stage includes the planning activities for the whole event, including those for the in-person component that are more energy and material demanding.The Virtual experience stage is instead similar to INFORMS 2020.What is not captured in previous graphs is the overall benefit of a hybrid event (see Table 2).Compared to INFORMS 2019, an average 60% reduction is obtained in the overall per-capita indicator results.This is around 20% less than what was projected by Tao et al.' (2021) simulation with the same share of virtual attendees (68.6%).Indeed, they considered an ideal scenario in which the farthest attendees join online.In real cases, that can hardly be enforced by organizers without incurring discriminatory behavior.However, they might design ad hoc incentives and promotion campaigns to attract a larger share of virtual attendees and prompt faraway delegates to attend online.a Share of travelers and Average traveled distance do not include the airport-venue connection for air travelers, while Share of transport GHG emissions does.

Drivers of impact variation
Regarding potential drivers of impact variation, our identification strategy was primarily based on spotting significant differences across the impact profiles of conferences within the same format and working backward to find the reasons behind them.
For in-person conferences, a large variation can be seen in the impacts of Stakeholders' transport, with INFORMS and POMS attendees displaying greater travel impacts than EurOMA ones.As expected, this is mainly related to the average traveled distances, with POMS and INFORMS attendees traveling greater distances than EurOMA ones (see Table 3 for a summary of travel metrics).EurOMA has indeed a more concentrated audience with a prevalence of medium-haul travelers from Europe-typical of European conferences (Desiere, 2016).The American conferences, instead, have a larger share of long-haul travelers, mostly coast-to-coast travelers in the US and inter-continental travelers.POMS 2019, particularly, has a larger share of non-US attendees, which explains its greater average distance than INFORMS 2019.This does apply to INFORMS 2021, probably due to its location in Los Angeles which allowed for better flight connections.
The greater distances traveled by INFORMS 2021 attendees, however, seem not to explain alone their much larger transportation footprint.
We discovered an extra impact associated with the transportation of staff and exhibitors.EurOMA and POMS represent typical small/middle-size societies with contained organizational structures.In their conferences, mostly local student volunteers are hired as staff and exhibitors encompass only a few publishers and companies.INFORMS is instead a more structured organization that combines local staff with full-time staff traveling from event to event.Its conferences further involve a large exhibition space with dedicated personnel from many companies and recruiting institutions.
The allocation of staff and exhibitors' travel to attendees explains the impact surplus: +11.2% Hybrid (in addition to the above)

Share of virtual participation All
Extra number of staff members Stakeholders' transport, Accommodation food-or normalizing the figures against the number of offered meals would reverse this result (Neugebauer et al., 2020).Lastly, a minor difference regards, again for EurOMA, the lower impact of Conference organization in those categories driven by resource consumption (e.g., LU and WC), reflecting the effort of its organizers to reduce conference materials as much as possible.
It is worth noting that some previous considerations are valid because the analyzed conferences had almost the same duration-3−4 days, typical in the OM&OR field.Duration could indeed represent a driver of variation, as some stages like Venue, Catering, and Accommodation depend on it.An increase in the conference days would increase both the relative impact shares of these stages and the overall absolute values, even if probably not linearly for Accommodation and Catering, as many delegates may not be able to extend their stay due to other commitments.Another potential driver not captured above is the accommodation type used by attendees.We had granular data only for INFORMS conferences in terms of the percentage of stakeholders staying in 4-stars and 3-stars hotels and we extrapolated them to EurOMA and POMS, thus not capturing their actual patterns and neglecting other less used accommodation types (e.g., luxury hotels, hostels, rented flats, and so on).As to virtual conferences, a large variation is visible going from INFORMS 2020 to POMS 2021 and then EurOMA 2021, with the impact contributions of Virtual experience and, consequently, the overall indicator results increasing.By analyzing platform analytics data, we found the reason for this in the different amounts of online activity by attendees, with EurOMA attendees being connected in total for ∼8.9 h per person and INFORMS ones for ∼2.8 h (no granular data for POMS).Surprisingly, this is unrelated to the conference duration: EurOMA lasted half as long as INFORMS (3 vs. 6 days) but its attendees stayed connected for around three times as long.It has probably more to do with how the virtual experience was designed.EurOMA chose a synchronous format with all live-streamed presentations, while INFORMS, due to its larger size and related planning difficulties, opted for an asynchronous format with mostly pre-recorded presentations.This different engagement level should be considered by organizers when planning future virtual conferences.Another driver suggested by Tao et al. (2021) is the geographical distribution of attendees, which can affect the upstream production of energy used in electronic devices (Virtual experience stage) and which we did not capture because of the use of globally averaged datasets.
Table 4 sums up the main variation drivers identified in the analysis.

Mitigation options and scenario analysis
Apart from shifting to virtual and hybrid formats, previous studies have suggested several mitigation options to make in-person conferences greener (see the last column in Table S1.1).Most have righteously focused on air travel, being the predominant contributor to environmental impacts-van Ewijk and Hoekman (2021) offer a comprehensive overview in this regard.Main options include the shift to landbound transport for closer air travelers (Desiere, 2016;Neugebauer et al., 2020), the optimization of the conference location based on attendees' origins (Jäckle, 2019;Kuonen, 2015), and the implementation of a multi-hub conference with inter-connected hubs in different

CONCLUSION
This study compared the life cycle impacts of seven real-world in-person, virtual, and hybrid conferences organized by different societies and with different features.The organizing societies belong all to the OM&OR field, which together with the reduced number of cases, may limit the generalizability of the results.Still, our sample presents greater size and heterogeneity than all previous comprehensive LCA studies, which allowed us to uncover significant drivers of impact variation (Table 4).In addition to expected ones, such as the average distance traveled by delegates, new ones were spotted.For in-person conferences, the number of non-local staff members and exhibitors-dependent on society's size and value proposition (e.g., provision of a career fair)-plays an important role in transportation and accommodation impacts.The energy sources and systems used to supply conference buildings instead greatly affect venue impacts.For virtual conferences, the main driver is the time spent online by attendees.
Interestingly, instead of a direct consequence of conference duration, this seems to have more to do with how the virtual experience is designed, with synchronous presentations leading to greater engagement than asynchronous ones-∼9 h of participation per person at EurOMA (3 days, synchronous) versus ∼3 h at INFORMS (6 days, asynchronous).
As to format comparison, our results confirm and extend previous literature.For all societies, virtual conferences were shown to reduce impacts by two to three orders of magnitude across all impact categories: for GW, from an average of 941.9 kg CO 2eq per person to 1.0 kg, making them an indispensable option in a serious path toward decarbonization.Put in perspective, the total carbon footprint of the EurOMA 2021 virtual conference was about half the per-capita footprint of an average in-person delegate.The hybrid case study instead, with a composition of 69% virtual attendees, led to a 60% per-capita carbon footprint reduction, less than what was forecasted by previous studies assuming that the farthest attendees shift to virtual attendance (Tao et al., 2021).That cannot be enforced by academic societies without incurring discriminatory behavior, but they might design proper incentives to increase the share of virtual attendees.Beyond format shift, the study reviewed other mitigation options proposed in the literature and provided new ones to tackle venue impacts, namely the selection of a venue supplied by a district heating system (−41 kg CO 2eq per person from the INFORMS 2019 case) or with a green electricity contract (−1.9 kg per person from scenario analysis).We also assessed a mitigation scenario for virtual conferences, showing that if half attendees had a green electricity contract, an average reduction of 80 g CO 2eq per person would be achieved.
Lastly, to move the agenda forward, the study underlined several inconsistencies that affect present conference studies and proposed some avenues for future research.The most promising one, in our opinion, involves a radical change in focus while investigating the environmental impacts of scientific activities.For the environmental optimization of single conferences, studies are converging toward the multi-hub format (Klöwer et al., 2020;Tao et al., 2021;van Ewijk & Hoekman, 2021).However, do we need those conferences in the first place?The core role of an academic society is arguably not to organize a conference but to meet its members' needs, including the need to network or disseminate research.On the one hand, conferences are not the only way to meet these needs (e.g., seminar series), and on the other, different formats of communication (i.e., in-person vs. virtual) meet various needs differently.Therefore, embracing a portfolio perspective, academic societies should first understand the needs of their members and then identify the optimal number and format of conferences or other activities to meet those needs while minimizing the overall environmental impacts.

F
Impact breakdown by conference stage for in-person conferences.The underlying data for this figure can be found in Supporting Information S9.F I G U R E 3 Impact breakdown by conference stage for virtual conferences.The underlying data for this figure can be found in Supporting Information S9.TA B L E 3 Travel patterns by type of transport within in-person conferences.
Figure S1.1 in Supporting Information S1 shows the ReCiPE indicator results per guest•night for different accommodation options modeled in ecoinvent, highlighting the large difference that can arise.
locations and attendees traveling to the closest hub (Coroama et al., 2012; Klöwer et al., 2020)-see Parncutt et al. (2021) and Tao et al. (2021) for the issue of hub selection.These options can be

Type of transport Conference Landbound a Air-Very short haul Air-Short haul Air-Medium haul Air-Long haul Total or average
for the per-capita travel footprint of INFORMS 2021, +2.1% for INFORMS 2019, and +0.2/0.5% for POMS and EurOMA.The same applies to Accommodation impacts, which are greater for INFORMS conferences ceteris paribus.The effect is proportional to the ratio between the number of non-local workers and the number of attendees: around 1 every 100/200 at EurOMA and POMS, 1 every 50 at INFORMS 2019, and 1 every 10 at INFORMS 2021.Interestingly, the total number of staff members at INFORMS 2021 was greater than in 2019 (222 vs. 150), against a number of in-person attendees that was less than one third.This may be attributable to several reasons (e.g., greater complexity of running a hybrid event and unexpected reductions in the number of in-person attendees), which need further investigation.Another significant difference regards Venue impacts, which are much lower across all categories for INFORMS 2019: for GW, 2.9 kg CO 2eq per person versus 56 kg of POMS 2019.This is due to the different energy sources and supply systems used to produce the heat consumed at the venue.For INFORMS 2019, indeed, a district heating system fueled by wood waste was used to supply conference buildings, less impactful than the traditional gas boilers employed in other venues.Other potential drivers for Venue impacts are the period and location in which the conference is held, which we did not capture because of the use of yearly and geographically averaged data.Spring and fall conferences are likely less energy demanding than winter and summer ones, depending on the location.Another variation can be noticed in the impacts of Catering, which are larger across all categories for EurOMA 2019, a surprising result since its organizers purposefully offered a vegetarian-only menu for most meals.This can be explained by the consolidated practice in European conferences of covering most meals within registration fees, as opposed to most American conferences where only a few receptions and luncheons are offered-4 meals per registrant at EurOMA versus 1.5 at POMS and 0.5 at INFORMS.Integrating the food consumed outside the conference-often fast TA B L E 4 Main drivers of impact variation for academic conferences.