• Fri. Sep 25th, 2020

Dimancherouge

Technology

The imperative to reduce carbon emissions in astronomy

In this section, we provide an overview of the emissions that Australian astronomers are responsible for, from the sources of expected greatest significance, in no specific order.

Flights

Relative to the general public, astronomers travel a lot. Reasons include, but are not limited to: conferences, workshops, collaboration, seminars, observing runs, committee meetings, job interviews, and relocation. This is not specific to astronomers though; academics in general are responsible for considerable greenhouse gas emissions due to flying. One case study suggests that business-related flights from university employees contribute approximately two thirds of the emissions of campus operations18. Flights are often the greatest single source of university emissions, with conference attendance accounting for approximately half of those flight emissions19.

Not only does all international travel require flying thousands of kilometres from Australia, but due to the size and low population density of the country, domestic travel often does too. As a point of reference, we collate the approximate greenhouse gas emissions per passenger from direct flights between Australian capital cities in Table 1, according to Qantas. Based on the same carbon calculator, return trips from Australia to Europe or the Americas can comfortably exceed 3 tCO2e per passenger.

Table 1 Typical emissions for (the most) direct return flights between Australian capital cities, according to Qantas

In Australia, aviation was responsible for 22.02 MtCO2e of emissions in 2016 alone (which includes 12.02 MtCO2e from international flights)20. This suggests that aviation is responsible for 4% of the country’s total emissions (or close to 1 tCO2e yr–1 per person on average). While this may sound like a small fraction, it is important to recognize that about half the population will not fly at all in a given year, that most of them will only fly once in that year, and that the vast majority will do so for leisure, not business. For the relatively few people who fly regularly, their personal fraction of emissions from air travel presumably must be much higher than the nominal 4%. As we demonstrate in this section, astronomers are among those people (certainly, at least, in Australia).

CAS budget example

As an example of astronomy’s disproportionately high flight emissions, consider Swinburne University of Technology’s Centre for Astrophysics and Supercomputing (CAS). In 2017, approximately 80% of CAS’s travel budget was spent on flights: AU$301,000 in total (including external funding contributions), with AU$54,000 spent on 134 domestic round-trip flights, and the remaining AU$247,000 spent on 133 international round trips (often including more than two flights). These flights covered the 80 full-time-equivalent (FTE) staff and students in CAS during 2017, meaning each person was responsible for 1.7 domestic and 1.7 international flights on average. A typical domestic return flight from Melbourne produces 230 kgCO2e per passenger (taking a naïve average of the values for MEL in Table 1). Considering Los Angeles as a typical international destination, a return international flight produces 3 tCO2e per passenger (from Qantas’s calculator). Therefore, the average astronomer in CAS was responsible for 5.4 tCO2e in 2017 from flying alone, with 0.4 and 5.0 tCO2e coming from domestic and international flights, respectively. As a rough guide to the average monetary carbon cost of flying, these figures imply 0.57 kgCO2e per AU$ for domestic flights and 1.6 kgCO2e per AU$ for international flights. These figures are comparable to the case study of Stohl21 at a different institute (and research field).

ICRAR-UWA travel records

A further, more detailed example is available from the International Centre for Radio Astronomy Research–University of Western Australia node (ICRAR-UWA). Here, the complete travel records for the 2018 and 2019 calendar years were analysed. Over this time, ICRAR-UWA used three different travel agencies. All work-related travel captured by these agencies was accounted for, regardless of the funding source. Two of those agencies gave direct emissions values for all bookings captured by their systems. For the third agency, we still had access to all flights travelled, but had to calculate the emissions for each flight; for this, we used Qantas’s calculator.

All emissions initially quoted did not differentiate between economy and business class flights. Business class seats occupy roughly triple the area of economy seats (this varies plane to plane, and is often lower for domestic trips and higher for international trips, with one article suggesting a factor of 3.5 is more common for the latter). For the relatively few business class flights listed in the travel records, we multiplied their emissions by three. We emphasize that economy class is the norm for astronomers, and the vast majority of bookings in these records were indeed economy.

In Table 2 and Fig. 1, we summarize the findings from ICRAR-UWA’s travel records. While these data have been anonymized, we present statistics for different levels of staff. Where we refer to ‘senior scientists’, we mean all research staff employed at Level C and above in the Australian university employment system, which are effectively all tenured or tenure-track positions, including senior fellows, associate professors and full professors. We broadly label all nominal research staff employed at Level A or B as a ‘postdoc’, all of whom are on fixed-term contracts, which includes research associates and early-career fellows. All remaining staff who are not students fall under the ‘professional’ category. This covers a diverse range of staff, including outreach, administration, computer scientists and engineers. Masters and PhD students are considered separately. Other students are not explicitly accounted for (for example, honours students, of which ICRAR-UWA has none).

Table 2 Summary of greenhouse gas emissions from ICRAR-UWA employees’ work-related flights from 2018 and 2019
Fig. 1: Distributions of ICRAR-UWA staff’s air travel CO2-equivalent emissions.
figure1

Bottom panel: normalized histograms of individuals’ annual emissions for 2018 and 2019 (two entries per person) from each staff type in bins of width 2 t yr–1; bar thicknesses (except for the ‘all’ category) have been artificially reduced to visually separate each distribution. Top panel: mean of each distribution (closed triangles, where open triangles add the contribution from visitors), along with their medians (vertical dashes), and 16th and 84th percentiles (dots connected by horizontal bars). These values are provided in Table 2. Figure produced using matplotlib45.

Unsurprisingly, flight frequency—and thus flight emissions—scales with seniority (as has been found in other studies18). The average senior staff member emits close to 12 equivalent tonnes of CO2 from flying each year (or roughly four return international trips, or three international plus four domestic). Granted, this mean (but not the percentiles) is pulled up by two outlier points in the distribution; removing the factor-of-three assumption regarding emissions of business versus economy seats would reduce this mean to 9.5 tonnes. The average postdoc emits around a third that of an average senior staff member (roughly one international and one domestic trip each year). The average PhD student emits less than half that of an average postdoc (two to three domestic trips each year, or one international trip every two years).

In total, the flight emissions from ICRAR-UWA staff members over the two-year period was 768 tCO2e. A further 86 tCO2e came from guest bookings, that is, travel booked by ICRAR-UWA staff for external visitors and collaborators. It is important that these bookings are not ignored, because if the same study were conducted at those guests’ home institutes, those flights would probably not be captured by their systems. Likewise, there could well be other work-related flights that ICRAR-UWA staff members took over this period that were booked externally and are thus not considered here. Incorporating captured guest flights into our figures compensates for this. In all instances, a senior member of staff was the host for the guests, so this reasonably should only contribute towards the figures for senior staff and totals. We include a second column for means in Table 2 that appropriately takes guest flights into account.

Remarkably, the average per-person emissions from flights of PhD students, postdocs and senior scientists combined at ICRAR-UWA is exactly the same as the estimate for CAS (see the ‘CAS budget example’ section), that is, 5.4 tCO2e yr–1 (excluding guests’ flights). After adding guest bookings, this average increases to just over 6 tCO2e yr–1.

Despite Perth’s relative isolation (it is the second-most isolated major city globally, based on nearest-neighbour distance of cities with populations above one million), the travel budgets of research institutes in Perth do not necessarily exceed that of equivalent institutes elsewhere in Australia. While a domestic trip for those living on the continent’s east coast might mean lower emissions (see Table 1), this is probably counterbalanced by an increase in the number of domestic trips. International travel comes at a heavy carbon cost regardless of the Australian city of origin.

National extrapolation

Official figures submitted as part of the 2019/2020 mid-term review of the Australian astronomy decadal plan suggest there are currently 365.2 FTE research staff nationwide, covering academic levels A–E, that is, junior postdocs through to full professors. These figures will be made public as part of the mid-term review process. Consistent with our earlier definition, if we consider postdocs to hold temporary contracts and be employed at either academic level A or B, then postdocs account for 166.2 of those FTEs. That leaves 199 ‘senior scientists’, which we again consider as those at academic level C and above, and/or those with permanent employment. Five additional FTEs fall outside the standard university employment levels, which we do not categorize here. 326.5 FTE astronomy PhD students are enrolled nationwide, as are 72 FTE masters students. An earlier figure from 2014 suggested 242 support staff were also employed across the country22, which we equate to our ‘professional’ category.

Combining these numbers with the means in Table 2 (including the guest contribution to senior scientists) gives an estimate of the total national emissions from flights as 4,190 tCO2e yr–1.

Supercomputer usage

As described in a recent white paper23, the estimated computing requirements of Australian astronomers is 400 million CPU core-hours (MCPUh) per annum, expected to rise to 500 MCPUh yr–1 by 2025. This is split across many computing facilities, including both domestic and international supercomputers. Each has its own energy efficiency and is powered by different sources. It is therefore non-trivial to translate this level of computer processing into a rate of CO2-equivalent emissions.

The three most notable supercomputing centres for Australian astronomers are the National Computing Infrastructure (NCI) in the Australian Capital Territory (ACT), the Pawsey Supercomputing Centre in Western Australia (WA), and the OzSTAR supercomputer in Victoria. We contacted each of these to request official figures on the energy/emission requirements that would allow us to estimate astronomers’ computing carbon footprint as accurately as possible. Unfortunately, Pawsey was the only centre that responded with data. We therefore extrapolate from these data to estimate the national computing emissions of Australian astronomers.

Figures provided to us privately by Pawsey show that the centre consumed 10.94 GWh of electricity in the 2018/2019 financial year, <100 MWh of which came from their own solar panels. While one of Pawsey’s two solar inverters was down for much of this period, we can reasonably estimate that 99% of the electricity powering Pawsey comes from the grid. 25% of Pawsey’s computing resources are allocated to astronomy through the dedicated Galaxy supercomputer24. We can therefore estimate that Australian astronomers require 2.7 GWh yr–1 of electricity for their Pawsey usage alone (this is probably a lower limit, as other machines at Pawsey—for example, Magnus—are used by astronomers too). In southwest WA, electricity currently carries a carbon cost of 0.75 kgCO2e kWh–1 (we account for both ‘scope 2’ and ‘scope 3’ emissions when considering mains power consumption throughout this paper; in principle, this includes the emissions associated with extraction and burning of the fuel used to produce the electricity, as well as losses in transmission)25. 2.7 GWh yr–1 at Pawsey therefore translates to 2.0 ktCO2e yr–1. 51.1 MCPUh were consumed on Galaxy for radio astronomy during the 2018/2019 financial year24, implying a carbon cost of 40 tCO2e MCPUh–1.

Given the above, we estimate the net power required to run code on a supercomputer that includes all overheads and cooling to be 53 W per core. In theory, this value could actually be higher for many facilities, as Pawsey uses a groundwater cooling system that should reduce the energy requirements of cooling. Nevertheless, if we assume that 53 W per core is typical for most supercomputers, then we need only consider where other commonly used facilities are, and the emissions per kWh there. In Victoria and the ACT, electricity emissions are 1.17 and 0.92 kgCO2e kWh–1, respectively25. Despite these being official numbers from the Australian Government, we highlight an important caveat regarding the emissions from electricity use in the ACT later on. For now, we take those numbers at face value. Assuming a ratio of 3:2:1 for NCI:Pawsey:OzSTAR (ACT:WA:Victoria) usage in astronomy (a difficult ratio to gauge with publicly available information), this gives an average of 0.905 kgCO2e kWh–1 or 48 tCO2e MCPUh–1.

It is important to note that only 60% of Australian astronomers’ supercomputer usage is from domestic facilities23. The average emissions per kWh for countries in the Organisation for Economic Co-operation and Development (OECD) is roughly half that of Australia’s. Accounting for this—assuming it reflects where the offshore supercomputers that Australian astronomers use are—reduces the average emissions for Australian astronomers’ supercomputing time to 38 tCO2e MCPUh–1.

With all of this in mind, we estimate that the total emissions from Australian astronomers’ supercomputer usage is 15 ktCO2e yr–1. This is nearly quadruple the value from flights (see the ‘National extrapolation’ section). Dividing across all senior staff, postdocs and PhD students gives a mean supercomputing carbon footprint of 22 tCO2e yr–1 per researcher. Note that we have implicitly assumed that cores on local clusters in Australia carry the same power and carbon requirements as cores on supercomputers; the 400 MCPUh yr–1 figure should include the use of local clusters. Similar to flights, we expect that many of the people being averaged over will require relatively negligible computing time, and thus the mean emissions per researcher will be much less than the actual emissions of the researchers who have a heavy reliance on high-performance computing.

While it is difficult for us to quote an uncertainty on this number within a specified confidence interval, we can take 28 tCO2e yr–1 per astronomer as a fair upper limit (the figure we would have derived had we not accounted for the lower overseas emissions for electricity). Because of the considerable production of renewable energy that the ACT is responsible for26, one can argue that emissions from NCI should be treated as zero. Taking that argument, while maintaining the assumption that 30% of Australian astronomers’ computing is done in the ACT, would lead to a value of 14 tCO2e yr–1 per researcher. This provides a reasonable estimate of a lower bound.

Observatories and telescopes

Another important source of emissions is the operation of observatories and telescopes. We sought information from several observatories regularly used by Australian astronomers regarding their emissions from operations (for example, power consumption). While the information provided to us is not a complete accounting of all relevant domestic and international observatories (not all places we contacted supplied data), we can place a meaningful lower limit on the total electricity and emissions requirements for Australian astronomers to conduct observations.

In a private communication, the Australia Telescope National Facility (ATNF, part of CSIRO) provided us with the electricity consumption of all observatories they operate over a one-year time frame. The sites considered include the Australia Telescope Compact Array (ATCA), the Parkes Observatory, the Mopra Radio Telescope, and the Murchison Radio-astronomy Observatory (MRO). ATCA, Parkes and Mopra all use mains power (with backup diesel generators) and are all situated in New South Wales (NSW). Those three sites consumed a combined total of 3,760 MWh of electricity over the year ending 29 February 2020, including all the telescopes, buildings and integral facilities on-site. ATCA accounts for 1,920 MWh, with 70% of its observing time allocated to Australia-based principal investigators (PIs) in 201827. Parkes accounts for 1,550 MWh, with 55% allocated to Australian PIs in 201827. The remaining 290 MWh covers Mopra, although it is harder to obtain a fraction of time spent by Australian PIs on this telescope. An earlier report from 201528 shows eight programmes were run on Mopra in the year prior, with three-eighths of the first-name observers identified as belonging to Australian institutions. Given the carbon cost of 0.92 tCO2e MWh–1 for mains power in NSW25, the combined operation of ATCA, Parkes and Mopra produces 3.5 ktCO2e of emissions per year, with a contribution based on Australian astronomers’ usage of 2.2 ± 0.1 ktCO2e yr–1.

The MRO hosts both the Murchison Widefield Array (MWA) and the Australian Square Kilometre Array Pathfinder (ASKAP). The isolation of the MRO in WA means it is not connected to mains power. Instead it is powered by a combination of on-site solar photovoltaics and diesel. Once operating at maximum capacity, the solar array is expected to cover >40% of the site’s electricity needs. As of yet, it has not reached this capacity. Over the 2018/2019 financial year, the MWA and ASKAP consumed a total of 4,110 MWh of electricity: 3,360 MWh for ASKAP, 520 MWh for the MWA, and 230 MWh from transmission losses. 600 MWh of this came from solar energy, and the rest from diesel. An additional 200 MWh was consumed at the Boolardy accommodation facility, with roughly a third of this estimated to come from solar, and the rest diesel. Based on figures from the Australian Government25, the carbon cost of burning diesel for energy is 266 kgCO2e MWh–1 (this covers ‘scope 1’ and ‘scope 3’ emissions, that is, the on-site emissions from the burning of diesel and an approximate consideration of indirect emissions associated with its production and transport; the latter is probably an underestimate in the case of the MRO). This implies that the MRO currently produces greenhouse gas emissions at a rate of 0.95 ktCO2e yr–1. Based on the facts that 87.5% of MWA observing time was led by Australia-based PIs in 2019 and 100% of current ASKAP operations are Australia-led, we estimate Australian astronomers’ contribution to MRO emissions as 0.93 ktCO2e yr–1. Because the MRO is one of the sites for the Square Kilometre Array (SKA), its power consumption is expected to notably increase with time as SKA operations ramp up. An increased fraction in dedicated solar power will help offset any rise in the site’s emissions though.

The W. M. Keck Observatory in Hawaii provided us with an estimate of their CO2 emissions from on-site electricity and vehicle use. The latter is only a minor contributor. No flights to or from the observatory were included (flights to the observatory made by Australian astronomers have already been accounted for in the ‘Flights’ section). The total CO2 emissions reported to us were reduced pro-rata with Australia’s current official proportion of Keck observing time of ten observing nights. Noting that Keck operates two near-identical telescopes, there are 730 possible observing nights per (non-leap) year. Given this, an initial estimate of Australia’s share of Keck’s CO2 emissions is 35 t yr–1. Evidently, this is very small compared to emissions from Australia’s use of its own domestic facilities. The Australian astronomical community has had access to up to 40 nights per year on Keck in the past, but even the emissions from that would be almost negligible compared to the sum of ATNF observatories.

We note that the European Southern Observatory (ESO) has already commissioned a study of the emissions of its sites, but the results were pending at the time of writing. Should these be made publicly available, this could prove a useful resource from 2020 onwards. Now a strategic partner of ESO, Australia’s emissions contribution to the use of those observatories should be taken into account for completeness.

With the information we have, we can confidently place a lower limit on the observatory-based emissions of Australian astronomers of 3.3 ktCO2e yr–1. Contributions from the Siding Spring Observatory (which hosts the Anglo-Australian Telescope, the ANU (Australian National University) 2.3 m telescope, the SkyMapper Telescope, and the UK Schmidt Telescope) and ESO facilities are the most notable exclusions from this estimate. Any involvement that Australian astronomers have in space telescopes has not been accounted for here either.

Campus operations

Office spaces and their machinery also contribute to work-related carbon emissions. While a specific analysis of all the buildings that house astronomy departments in Australia is left for future investigation, we can again use ICRAR-UWA as an example and extrapolate. ICRAR-UWA lies in the Ken and Julie Michael Building at UWA. Figures provided to us by UWA suggest that powering the entire building produces 618,772 kgCO2e yr–1. ICRAR occupies 48% of the building’s floor area, implying that the centre is responsible for 297 tCO2e yr–1. Given that 100 people have a desk at ICRAR-UWA (see Table 2), this implies an average of 3 tCO2e yr–1 per person for office building requirements. Extrapolating this to the 1,000 astronomers and support staff nationwide (see the ‘National extrapolation’ section) implies total emissions of 3 ktCO2e yr–1.

A caveat to the building power requirements of ICRAR-UWA is that this includes powering the Hyades computing cluster. In principle, the emissions from the use of local clusters have already been accounted for in the ‘Supercomputer usage’ section. However, the entire Hyades system only has 92 cores, meaning it must account for less than 0.8 MCPUh yr–1. Recognizing that ICRAR-UWA makes up 10% of the national community, any potential ‘double counting’ of computing requirements must be less than 2% of the total from the ‘Supercomputer usage’ section (that is, <0.3 ktCO2e yr–1). In reality, local clusters like Hyades almost never operate near their full capacity.

An additional caveat is some of the support staff who are based at observatories might already have their office requirements covered in the ‘Observatories and telescopes’ section. It might be more appropriate to only extrapolate the per-person office power requirements to 800 people. With both caveats, the true office-based emissions of Australian astronomers might be as low as 2.2 ktCO2e yr–1.

Summary of emissions

Our findings are that the largest contributor to Australian astronomers’ emissions is supercomputing. At 15 ktCO2e yr–1 for the national community (see the ‘Supercomputer usage’ section), this is more than all other sources of work-related emissions combined. This figure is primarily an extrapolation from power usage data we received from a single supercomputing facility (Pawsey). There are many sources of uncertainty contributing to this figure that we have not quantified precisely. With differing assumptions about how emissions from the ACT and non-Australian supercomputers are accounted for, this value could actually be as low as 9.5 or as high as 19 ktCO2e yr–1.

Despite sometimes garnering the most attention in conversation, flights rank a distant second (at best), totalling 4.2 ktCO2e yr–1 (see the ‘Flights’ section). This figure is largely based on an extrapolation of one institute (ICRAR-UWA), but it is entirely consistent with totals from a second institute (CAS). The formal uncertainty carried through from the jackknifing uncertainties given in Table 2 is effectively negligible, but it does not sufficiently account for potential variation across institutes in the country. At a precision of one significant figure, we can fairly confidently say the value is near to 4 ktCO2e yr–1, assuming the values for emissions provided by travel agencies and airlines do not carry systematic uncertainties greater than 10% (which we do not know). Based on this, the uncertainty in our figure for flight emissions should be of order a few hundred tCO2e yr–1. However, we have not explicitly accounted for the altitude of aeroplane emissions, which is particularly problematic due to the production of contrails29,30. In essence, the effective radiative forcing from aeroplane emissions at altitude could be several times that of their nominal CO2 emissions. This systematic error is probably our greatest source of uncertainty.

The powering of observatories ranks third in emissions at >3.3 ktCO2e yr–1 (see the ‘Observatories and telescopes’ section). This is based on the total power requirements of ATCA, the MRO, the Parkes Observatory and Mopra—accounting for the fraction of Australian PI time on these instruments—with the additional but small contribution from Australia’s time on Keck. There are many other observatories that Australian astronomers use, and thus we can only provide a lower limit here. In reality, the emissions from observatory operations could well exceed that of astronomers’ flights.

Finally, emissions associated with powering astronomers’ office buildings are approximated to be 2.2–3.0 ktCO2e yr–1 nationwide (see the ‘Campus operations’ section). Again, this is based on an extrapolation from ICRAR-UWA, and thus we may have underestimated the true uncertainty.

A visual summary of these four sources of emissions and their estimated uncertainties is provided in Fig. 2. Summed together, the Australian astronomy industry is responsible for emitting 25 ktCO2e of greenhouse gases per year. Dividing this across the combined 691.7 FTE of senior scientists, postdoctoral researchers and PhD students implies an average of 37 tCO2e yr–1 per astronomer. This means the work-based emissions of the average Australian astronomer exceed the combined work-plus-life emissions of the average non-dependant living in Australia by >40%. Globally, this is around five times the average work-plus-life emissions per non-dependant. Hypothetically, if half of all emissions were associated with people’s work (and the other half with their lifestyles), it would follow that an Australian astronomer’s job is around three times as carbon-intensive as the average job in Australia, and around ten times that of the average job globally. While there are surely plenty of examples of other jobs that are equally or more carbon-intensive, no such comparison absolves anyone of responsibility.

Fig. 2: Breakdown of the four sources of Australian astronomers’ emissions considered in this work.
figure2

Error bars provide an estimate of our uncertainties, but should not be interpreted as formal confidence intervals. The value for observatories is a lower limit. ‘Per astronomer’ refers to the 691.7 FTEs including PhD students, postdocs and senior researchers. Values are summarized in the ‘Summary of emissions’ section. Figure produced using matplotlib45.

Source Article