The Carbon Footprint of Conference Papers

The action required to stem the environmental and social implications of climate change depends crucially on how humankind shapes technology, economy, lifestyle and policy. With transport CO2 emissions accounting for about a quarter of the total, we examine the contribution of CO2 output by scientific travel. Thankfully for the reputation of the scientific community, CO2 emissions associated with the trips required to present a paper at a scientific conference account for just 0.003% of the yearly total. However, with CO2 emissions for a single conference trip amounting to 7% of an average individual’s total CO2 emissions, scientists should lead by example by demonstrating leadership in addressing the issue.


Introduction
The environmental and social implications of climate change depend not only on Earth's systemic responses, but also on how humankind shapes technology, economy, lifestyle and policy [1]. Action should not be postponed, as it is argued that we have already surpassed a safe threshold in atmospheric carbon dioxide concentration (from a 280 ppm pre-industrial value to 387 ppm today, with a proposed boundary threshold of 350 ppm) [2]. Changes in economy, lifestyle, and policy, entail changes in human behaviour, which will ultimately require decisions involving moral questions. Decisions should not be put off, considering that decisions that delay mitigation may have the greatest effect on the cost-risk distribution for returning global temperature increase to sustainable levels [3]. Science has an important role in framing the discussion and informing policy makers and the public [4]. This work adds to the discussion by highlighting the contribution of science itself to global carbon dioxide output [5]; in particular, to investigate the annual contribution of CO 2 output by travelling to scientific conferences to present a paper. These emissions could directly affect the environment, but also reflect badly on science, as demonstrated by the derisive press coverage of the 2009 Copenhagen summit's CO 2 footprint [6,7].
In terms of policy, choices to mitigate climate change may focus on market mechanisms (e.g., subsidies, trading schemes, or taxes), information disclosure (e.g., energy efficiency labeling schemes), and behavioral science [8]. Our focus is on information disclosure. We examine emissions associated with scientists travelling to present their work at conferences that publish their proceedings through indexed imprints. This is a subset of their total travel, as a part of their travel miles involve non-conference travel. However, conference travel is integral to scientists' work and, in contrast to other kinds of travel, its purpose is tied to science's core function. Conference trips are also, at least in theory, discretionary in the sense that they can be substituted through the use of various communication technologies. Furthermore, the emissions we study are also a subset of the total travel associated with conferences, because some conferences do not publish indexed proceedings, and many scientists attend conferences without presenting a published paper. Extrapolating total conference travel from our data through the use of conference attendance figures is difficult, because, according to our experience, attendance at conferences by scientists who do not have a paper to present tends to be biased toward those living relatively near the conference's location.
We show that CO 2 emissions associated with the trips required to present papers at scientific conferences account for 0.003% of the yearly total travel emissions. This is a bit more than the total transportation emissions for Geneva in a recent year, at about 800 kt CO 2 (1 kt is 10 6 kg), or less than the total transportation emissions for Barcelona, at about 1236 kt CO 2 [9]. Thankfully for the reputation of the scientific community, the environmental impact of the scientific conference trips we examine seems to be overblown. However, with CO 2 emissions for a single conference trip amounting to 7% of an average individual's total CO 2 emissions, scientists should lead by example in addressing the issue.

Results
We examine emissions associated with scientists travelling to present their work at conferences. We base our study on author and conference location data obtained from conference papers. We obtained our primary set of conference paper bibliographic details from the Scopus digital library by retrieving details of randomly sampled conference proceedings papers published over the period 1998-2008. This selection yielded a sample of 2.8% of the population's papers.
In general, total air passengers per year increased dramatically from 2001 to 2008, with a negligible decline in 2002 and 2009 [10,11] (Table 1). Over the same period, although average CO 2 emissions of scientific conference travel fell from 2001 to 2005, they increased again to the year 2000 levels in 2008 ( Figure 1 and Table 2 ). Over the year, although average emissions per paper are fluctuating, CO 2 total emissions per month are considerably higher during the spring and autumn months, which are popular for holding conferences (Table 3).
Author countries in the southern hemisphere fare quite badly in terms of the associated CO 2 emissions, while author countries with low emissions are those near conference locations (us, Canada, Mexico); see Figure 2, Table 4 and Table 5. However, we found no correlation between country wealth [12] and average emissions per paper-a country may lack financial resources, but when its scientists travel they do not necessarily fly less miles. At the same time, the papers published by authors in the country are correlated with country wealth in logarithmic transformation (r~0:4053508, p-value~0:00167, a~0:05, n~46, Pearson correlation test, permutation test used for hypothesis testing, 100,000 sampled permutations) and are therefore also correlated with the total emissions due to papers published by authors in the country (r~0:4030809, p-value~0:00250, a~0:05, n~46).
Two factors seem to increase the CO 2 emissions associated with a conference location: distance and popularity (Table 6); at the country level (Table 6) the southern hemisphere again fares particularly badly. On the other hand, conference countries and locations (Table 7) associated with low CO 2 emissions are those located off the beaten track.
A location's popularity as a conference location (Table 8) doesn't seem to be associated with travel distance and the consequent CO 2 emissions (r~0:08643165, p-value~0:17803, a~0:05, n~109, test between the average CO 2 emissions of a location and number of papers presented there, Pearson correlation test, permutation test used for hypothesis testing, 100,000 sampled permutations). None of the low CO 2 locations appear in the list of the ten most popular locations, while Honolulu, which is the eighth worst destination from a CO 2 emission perspective, is also famously popular.
Although the CO 2 emissions associated with a us-based author travelling to a conference are relatively low, the us as a conference hosting country contributes a lot to CO 2 emissions, both through the number of presented papers and the emissions associated with them. As we can see in Figure 3, the West Coast and Hawaii are leading in these two aspects.
Most CO 2 emissions in our study are attributed to travel to usbased conferences (Table 9), with travel within the us being the highest source of emissions. Within the us, (Table 10) travel to sunny California is the source of all but one of the top ten locations with the highest CO 2 emissions; the other is travel from California to Florida.
Looking at the most common trips at the country level (Table 9) we find most of the worst offenders in terms of total CO 2 emissions. However, the list also includes a lot of travel within Switzerland and Japan, which generates an order of magnitude fewer total emissions, and probably even fewer if one takes into account that these trips are often made by train. A similar pattern  is not apparent when we look at common trips within the us (Table 10). Travel within California generates a full quarter of the CO 2 emissions of the worst offender, namely travel from New York to California, indicating the need for improving the state's rail links. Finally, we tried to estimate the total carbon footprint of science travel associated with presenting papers at conferences. We calculated the average amount of CO 2 emissions per conference paper to be 801 kg; this figure comes from data from the 32,264 papers for which we were able to calculate their emissions. To establish the total number of conference papers published in a (recent) year, we undertook an overlap analysis [13] of two bibliographic databases, Scopus and ISI Web of Science. We estimated a total of 1.17 million conference papers in 2008 with a 95% confidence interval of ½1,030,754{1,313,585.
For this number of conference papers per year the emissions amount to 939 kt CO 2 in 2008. Total CO 2 emissions were at 28.962 Gt in 2007, with international aviation emissions totalling 411.6 Mt CO 2 [14]. Assuming that the increase from 2007 to 2008 followed a 3% annual trend [15], science travel emissions accounted for about 0.003% of all emissions or 0.228% of international aviation emissions in 2008.
This may not seem much. On a per capita basis, however, the total per capita emissions were 4328 kg CO 2 (2754 kg CO 2 for non-oecd countries and 10,969 kg CO 2 for oecd countries) [14]. Since a conference trip corresponds on average to 801 kg CO 2 , the share of conference travel in the mean CO 2 footprint of an average person is far from negligible. One may counter that scientists are probably a very biased sub-group within the populations of the world, with a higher than average CO 2 footprint, and therefore the CO 2 emissions associated with their conference travel form a relatively smaller percentage of their total CO 2 footprint. However, this argument as an excuse for a scientist's higher CO 2 emissions does not hold much water under any of the four prominent proposals for allocating them in the future, namely, equal per capita entitlements, rights to subsistence emissions, priority of the least well-off, or equalizing marginal costs [16].
Science has the duty to understand and explain climate change, to inform policy discussions, and to work out alternatives. This is an important responsibility. Scientists should therefore lead by example in the efforts to solve the problem.

Materials and Methods
We obtained our primary set of conference paper bibliographic details from the Scopus digital library by retrieving details of conference proceedings papers published over the period 1998-2008. We sampled the papers in a random fashion by selecting those whose author identifier-a system-assigned ten digit integerlast three digits ended in one of the following twenty combinations: 001, 111, 222, …, 999, and 120, 121, …, 129. The sample's coverage decreases over the years, varying from a high of 5.1% in 1999 to a low of 1.8% in 2008.
We ensured the reproducibility of our Scopus queries by limiting each query's results to papers entered into the system before July 1st 2009, capturing in effect the state of the database on that particular day. For this we used Scopus's (undocumented) ORIG-LOAD-DATE predicate, and specified as its argument the date measured in elapsed seconds (1,246,406,400) from January 1st, 1970 (the so-called Unix epoch). Because the results of each query were larger than the number we could download from Scopus, we divided each query into halves, based on the paper's publication year. Thus a typical query pair would be pubyear bef 2004 and pubyear aft 1997 and srctype(p) and AU-ID(*120) and ORIG-LOAD-DATE BEF 1246406400. and pubyear bef 2009 and pubyear aft 2003 and srctype(p) and AU-ID(*120) and ORIG-LOAD-DATE BEF 1246406400.
To calculate the CO 2 emissions per conference we assumed that a traveling author requires a single flight to get to the conference venue, and the flight would connect the departure and arrival points of latitude w and longitude l by the shortest possible arc, whose length d we calculate by using the Haversine formula [17]: Our assumptions underestimate the actual carbon footprint per travel, as trips seldom use the ideal path, and flight connections add take-offs and landings that increase CO 2 output. When applying the method described in the Act on CO 2 Calculator Version 2.0 [18], we distinguished only between short-haul and long-haul flights at 3700 km and assumed that scientists travel only economy class.
To determine the geographical coordinates of the author's and the conference's location we used two gazetteers (geographical dictionaries): the us National Geospatial-Intelligence Agency's (nga) database of foreign geographic feature names and the us Geological Survey (usgs) topical gazetteer files. We also used tables of large us cities from the us Census Bureau, and expanded country and administrative division codes according to the us Federal Information Processing Standard 10-4. In total, out of 63,034 papers in our database, of which 59,522 had data on both the author and the conference location, we fully geolocated 32,264 papers, pinning down 83% of the available conference locations and 61% of the available correspondence addresses. The travel emissions associated with presenting papers at a conference, the corresponding percentage over the total emissions, the average CO 2 emissions for each paper, and the corresponding number of papers appear in Table 11 in terms of the author's country and in Table 12 in terms of the conference's country. Although we include only geolocated papers in our results, the ratio of   Table 2 ). We matched conference locations in the gazetteers among many locations with the same name using a series of increasingly rough heuristics looking for: a unique name and state (e.g. Anaheim, ca), a unique name (e.g. Kuala Lumpur), a country's capital (Paris), or for a unique or major city and a country (Beijing, China). Author addresses in our data set were always tagged with a country, and we therefore matched them looking either for a city in a specified state and country, or for a unique or major city and a country (Beijing, China). In addition, we cleaned up postcodes located adjacent to city names by matching them according to country or region-specific standards, and we created various aliases for countries and administrative regions, which would cause violent convulsions to many diplomats.
To establish the total number of conference papers published in a (recent) year, we undertook an overlap analysis of two bibliographic databases, Scopus and isi Web of Science. We proceeded as follows.
If p a is the fraction of all papers in the world indexed by the first database and s a is the number of papers in the first database (its   Assuming that each database indexes independently, then if n a is the number of papers returned for a query by the first database, n b is the number of papers returned for the same query by the second database, and n o is the number of papers returned for the same query by both databases, we have n o n~p a n b n so that p a~n o n b .
Substituting we get n~s a n b n o .
We executed queries in both Scopus and isi Web of Science in February and March of 2010. Since both databases limit the number of results that can be downloaded for each query, we took into account only queries returning no more than 500 papers. We also took out of the calculations queries returning less than 50 papers, as in this case the overlap (n o ) could be very small creating outliers. Papers were matched if they were published in the same year and they had the same start and end page.
The queries were single words that we required to be matched exactly, for material published in proceedings in 2008. The words were selected by trawling the titles of paper titles that were published in 2008 in the journals Science and Nature. In Scopus, the queries were of the form:

TITLE({science}) AND SRCTYPE(p) AND PUBYEAR IS 2008
while in ISI the queries were of the form: TI = science AND PY = 2008 having selected the Conference Proceedings Citation Index-Science (cpci-s)-1990-present and Conference Proceedings Citation Index-Social & Humanities (cpci-ssh)-1990-present.
In the end, we had 80 result sets that met our criteria. From these we estimated a total of 1,172,169 conference papers in 2008 with a 95% confidence interval of ½1,030,754{1,313,585.