Drivers of global tourism carbon emissions
Input-output analysis
We employ the input-output technique40 to estimate tourism carbon footprints \(\widetilde{Q}\). The input-output framework allows for the estimation of direct and indirect emissions generated through the interdependencies of the tourism sector. Tourism expenditure \(\widetilde{y}\) serves to represent the economic contribution of the tourism sectors to the national economy. It is integrated into Leontief’s model \({L=(I-A)}^{-1}\) as the final demand19, where \(I\) is the identity matrix and \(A\) the interdependency coefficient. The calculation of carbon footprints is facilitated by employing country-specific carbon emission intensities \(q=Q{x}^{-1}\), where \(Q\) represents CO2−e emissions by country and industry sector, and \(x\) denotes industrial output. The formula to calculate the carbon footprints is therefore \(\widetilde{Q}={qL}\widetilde{y}\).
To conduct a comprehensive assessment of the tourism carbon footprints from 2009 to 2020, we utilize a global multi-region input-output (MRIO) model, or the so-called GLORIA environmentally-extended database20. The GLORIA database encompasses 160 countries and 4 rest-of-the-world regions, with each country’s economy represented by 120 economic sectors (Supplementary Method 4 and 5). This extensive coverage enables the estimation of carbon footprints up to 19,680 sectors, providing a detailed understanding of the carbon footprints associated with various tourism-related activities.
We gain deeper insights into the dynamics of global tourism markets by analysing carbon footprints from both the visitors’ residence perspective known as residence-based accounting (RBA) and the tourist destination perspective known as destination-based accounting (DBA)8. Using both RBA and DBA (Supplementary Method 1), we are able to discern to whom the carbon footprints in the tourism sector belong. The RBA approach attributes consumption-based emissions to the country of the visitors’ residence (encompassing emissions from both domestic and outbound tourism), while the DBA approach assigns these emissions to the host country (covering emissions from both domestic and inbound tourism).
Structural Decomposition Analysis
Building upon the input-output framework, our study further explores the factors driving the increase in global tourism carbon footprints from 2009 to 2019, using structural decomposition analysis (SDA)41,42. In this analysis, we first focus on four key determinants: technology improvement \({D}_{{tech}}\), the supply chain \({D}_{{scha}}\), tourism consumption \({D}_{{cons}}\), and population growth \({D}_{{pop}}\). To integrate the population effect \(P\) into SDA, tourism expenditure has to be transformed into per capita values \({\widetilde{y}}_{{pc}}=\widetilde{y}{P}^{-1}\).
Additionally, we incorporate the carbon footprints from the use of private vehicle \(({{{\rm{f}}}})\) as an extra factor due to its computation being conducted outside the input-output model. Let \(\triangle\) represent the change values from 2009 to 2019, then the increase of the carbon footprints from the use of private vehicle is denoted as \(\triangle {{{\rm{f}}}}={{{{\rm{f}}}}}_{2019}-{{{{\rm{f}}}}}_{2009}\). The drivers of tourism carbon footprints are \(\triangle \widetilde{Q}={D}_{{tech}}+{D}_{{scha}}+{D}_{{cons}}+{D}_{{pop}}+\triangle {{{\rm{f}}}}\), with each factor’s SDA formula as follow:
Technology improvement:
$${D}_{{tech}}= \triangle {qL}{\widetilde{y}}_{{pc}}P+\,\frac{1}{2}\left(\triangle q\triangle L{\widetilde{y}}_{{pc}}P+\triangle {qL}\triangle {\widetilde{y}}_{{pc}}P+\triangle {qL}{\widetilde{y}}_{{pc}}\triangle P\right) \\ +\,\frac{1}{3}\left(\triangle q\triangle L\triangle {\widetilde{y}}_{{pc}}P+\triangle q\triangle L{\widetilde{y}}_{{pc}}\triangle P+\triangle {qL}{\triangle \widetilde{y}}_{{pc}}\triangle P\right) \\ +\,\frac{1}{4}\left(\triangle q\triangle L\triangle {\widetilde{y}}_{{pc}}\triangle P\right)$$
(1)
The supply chain:
$${D}_{{scha}}= q\triangle L{\widetilde{y}}_{{pc}}P+\,\frac{1}{2}\left(\triangle q\triangle L{\widetilde{y}}_{{pc}}P+q\triangle L\triangle {\widetilde{y}}_{{pc}}P+q\triangle L{\widetilde{y}}_{{pc}}\triangle P\right)\, \\ +\,\frac{1}{3}\left(\triangle q\triangle L\triangle {\widetilde{y}}_{{pc}}P+\triangle q\triangle L{\widetilde{y}}_{{pc}}\triangle P+q\triangle L{\triangle \widetilde{y}}_{{pc}}\triangle P\right) \\ +\,\frac{1}{4}\left(\triangle q\triangle L\triangle {\widetilde{y}}_{{pc}}\triangle P\right)$$
(2)
Tourism consumption:
$${D}_{{cons}}= {qL}\triangle {\widetilde{y}}_{{pc}}P+\,\frac{1}{2}\left(\triangle {qL}\triangle {\widetilde{y}}_{{pc}}P+q\triangle L\triangle {\widetilde{y}}_{{pc}}P+{qL}\triangle {\widetilde{y}}_{{pc}}\triangle P\right) \\ +\,\frac{1}{3}\left(\triangle q\triangle L\triangle {\widetilde{y}}_{{pc}}P+\triangle {qL}\triangle {\widetilde{y}}_{{pc}}\triangle P+q\triangle L{\triangle \widetilde{y}}_{{pc}}\triangle P\right) \\ +\,\frac{1}{4}\left(\triangle q\triangle L\triangle {\widetilde{y}}_{{pc}}\triangle P\right)$$
(3)
Population growth:
$${D}_{{pop}}= {qL}{\widetilde{y}}_{{pc}}\triangle P+\,\frac{1}{2}\left(\triangle {qL}{\widetilde{y}}_{{pc}}\triangle P+q\triangle L{\widetilde{y}}_{{pc}}\triangle P+{qL}{\triangle \widetilde{y}}_{{pc}}\triangle P\right) \\ +\,\frac{1}{3}\left(\triangle q\triangle L{\widetilde{y}}_{{pc}}\triangle P+\triangle {qL}\triangle {\widetilde{y}}_{{pc}}\triangle P+q\triangle L{\triangle \widetilde{y}}_{{pc}}\triangle P\right) \\ +\,\frac{1}{4}\left(\triangle q\triangle L\triangle {\widetilde{y}}_{{pc}}\triangle P\right)$$
(4)
Tourism expenditure
Tourism Satellite Account (TSA) is the primary data source to capture tourism-related expenditure, which specifies all expenditure relating to domestic and inbound trips spent on businesses registered within the territory43. TSAs are only available for 62 countries, and the data is re-categorized based on the Tourism Satellite Account 39 sectors classification system (Supplementary Method 6).
For the 113 countries that do not report TSA, we resort to alternative sources. Inbound tourism expenditure data is obtained from the UN Tourism (World Tourism Organization), and domestic tourism expenditure data is sourced from the World Travel and Tourism Council (WTTC). These sources are used to compile the total tourism expenditure, referred to as internal tourism expenditure, for each country.
The aggregated internal tourism expenditure data must be allocated to different tourism sub-sectors. To disaggregate tourism expenditure into sectors for each countries, the following procedure is used. We first choose 31 goods and services in the GLORIA that tourists may purchase. Domestic household and government consumption data across 31 goods and services is then used to disaggregate domestic tourism expenditure into sectors. Total household and government consumption (domestic + exports) is chosen to disaggregate inbound tourism expenditure—this proxy considers export values consumed by overseas households, in this case, inbound tourists. This approach provides valuable insights into the interaction between tourism spending and various sectors of domestic economies, as well as international consumption patterns. The procedure also captures the dynamics of a country’s economic conditions, including large-scale downturns such as those experienced during the COVID-19 pandemic. For more information on the estimation process and equation, please see Supplementary Method 6.
To account for inflation over time, constant-price tourism expenditure (base year 2009) for the period 2009–2020 was calculated by first converting local currencies to US dollars and then adjusting for inflation using the USA Consumer Price Index across 44 categories, obtained from the US Bureau of Labor Statistics44.
This “convert-first, then deflate” approach is preferred by most researchers45. The main condition for adopting this procedure is that the sectoral output composition between countries should be similar46. Tourism satisfies this condition as its production of tourism services mainly consists of transportation, accommodation, food, recreational activities, and retailing. Every destination uses these components and similar production structure to construct “tourism”.
From a feasibility perspective, this approach is further supported by the availability of high-quality currency converters and deflators. Exchange rates against a common currency (the U.S. dollar) are available for all countries from 2009 to 2020. The availability of the U.S. Consumer Price Index across 44 categories, including those specific to food away from home, motor fuel, and transport services, allows for sector-specific adjustments. This helps minimize the noise that would occur from applying a single consumer price index across all expenditures. We did not employ the ‘deflate-first, then convert’ procedure because several countries lack consumer price index data from the World Bank database. The literature suggests that similar results are achieved whether the ‘convert-first, then deflate’ or ‘deflate-first, then convert’ approach is used47.
Tourist flow
The Tourism Satellite Account and international tourism expenditure provide detailed spending patterns for domestic and inbound tourism in a given country. However, they do not account for outbound tourism spending. To estimate how much residents spend abroad per country, we rely on international bilateral travel flows. These flows facilitate the allocation of international tourism spending to each destination from their respective origins, and vice versa (inbound spending from country A to country B = outbound spending from country B to country A). By mapping out bilateral travel patterns and expenditures for each origin-destination pair, we can estimate outbound tourism spending per country.
The UNWTO data provides bilateral travel volume between countries. In instances where tourist flow information is absent from the UNWTO dataset (around 30% of the bilateral travel data), we resort to a modelling approach, namely the gravity model48. This model considers factors that influence tourist movements, including the population size of both the visitor and the destination countries, and the physical distance between countries. The complete picture of tourist flows subsequently determines how each country is responsible for consuming inbound and outbound tourism.
Emissions
We include 29 types of emissions, ranging from carbon dioxide (CO2) to sulfur hexafluoride (SF6) obtained from the GLORIA database to capture detailed information about greenhouse gases (GHG) emissions (Supplementary Method 7). This inclusion allows us to capture all potential air pollutant emissions, including those from fuel combustion, industrial processes, residential activities, as well as emissions from animals and land use. We then utilize the 100-year time horizons Global Warming Potentials (GWP) provided by the Intergovernmental Panel on Climate Change49 (IPCC) to standardize the impact of different gases relative to CO2, forming carbon dioxide equivalent (CO2-e).
Both aviation and fuel-based private vehicle use are major contributors to the overall tourism carbon footprints. To ensure robustness in our analysis, we conduct calculations of these two components using additional data. We utilize the database from ForwardKeys to estimate air transport carbon emissions for individual countries. ForwardKeys monthly CO2 calculations draw inspiration from the ICAO methodology, analysing CO2 emissions on a per-route and per-aircraft basis. It first considers aircraft fuel consumption and flight distance while excluding the CO2 component that are related to freight. Passenger service aviation emissions are then allocated to countries where airlines register. This allocation process aligns with the United Nations’ ‘System of Economic and Environmental Accounting (SEEA)’ and the World Tourism Organization’s ‘Measuring the Sustainability of Tourism (MST)’ framework50,51,52. We then calculate aviation emissions related to tourism based on the proportion of air transport used for tourism purposes and fed this information to GLORIA to calculate indirect emissions in the supply chain (e.g., truffle in France to be air transported to restaurants in Australia). For estimating emissions from private vehicle use, we use specific fuel prices for 144 countries and the emission factor (CO2 per litre) to convert visitor expenditures on gasoline into carbon emissions.
Uncertainties
Two levels of uncertainty arise from modelling global tourism carbon emissions. The first is based on the assumptions used in the MRIO model. This model assumes that emissions coefficients and production structures are stable for a given year, leading to constant multipliers40. In other words, economies of scale, short-run price effects, and supply constraints do not affect production coefficients. Under these assumptions, tourism expenditure and emissions are presumed to have a linear relationship. However, this may not accurately reflect the operation of some tourism subsectors, which often experience strong seasonality, price fluctuations, and varying utilization rates53,54.
The second source of uncertainty arises from the data and parameters adopted in the model. The GLORIA database Release 057 serves as the source of GHG emissions data for this study. This data aligns with the EDGAR 7.0 dataset, which provides GHG emissions data updated only through 2019. Consequently, GHG emissions data beyond 2019 are estimated values55. This may introduce uncertainty into the 2020 tourism emissions results.
In addition, GLORIA’s updated emissions satellite uses two emissions data sources: the EDGAR database (emissions by activities) and the OECD/Eurostat databases (emissions by industries). Proxies and assumptions were employed to derive the activity-by-industry emissions estimates needed by GLORIA. While the adjustment procedures provide reliable estimates for most sectors, road transport remains ambiguous. The current method can only divide road transport emissions into private vehicles (allocated to households) and allocate the remainder to road transport, postal, and courier services, but not to other sectors55.
Data quality is crucial for global tourism emissions modelling, with tourism expenditure and input-output tables being the key parameters. We introduce a three-grade system of “very good,” “good,” and “fair” to represent the reliability of parameters and emissions results in our study.
“Very good” applies to countries that produce their own tourism satellite accounts and national input-output tables. Tourism satellite accounts provide detailed spending breakdowns by sector, allowing for an accurate match of consumption (tourism expenditure) to the MRIO sectors. Additionally, having a national input-output table ensures that local economic information is directly embedded into GLORIA, accurately representing the economic structure of an economy, which is essential for deriving emissions multipliers.
The second grade, “good,” includes countries where either the official tourism satellite account or the national input-output table is missing. In such cases, a disaggregation procedure is required to develop a comprehensive tourism spending profile, or trade data is used to estimate the economic multipliers of a given economy.
“Fair” applies to countries lacking both official tourism satellite accounts and input-output tables. In these instances, a higher level of uncertainty in the results is expected, as these two key parameters are based on estimation.
Our model currently includes 62 countries with TSA and IO tables, classified as ‘very good,’ covering 92% of global tourism expenditure (see Supplementary Method 6). The ‘good’ group comprises 81 countries, while the ‘fair’ group includes 32 countries—most of which are small countries with a limited scale of tourism activities.
Modelling aviation emissions remains a key challenge in mapping out the global tourism carbon footprint. This study relies on ForwardKeys data to assign passenger emissions (based on a weight-per-passenger estimation, including luggage) to individual countries using CO2 emissions on a per-route and per-aircraft basis. Our analysis shows that tourism air transport contributed 1.1 Gt CO2 in 2019, including direct emission (passenger transport, 0.96 Gt) and indirect emissions (freight transport due to supply chain activities, 0.14 Gt). While our results on aviation are slightly larger than existing references56,57, it is unclear what is summarized under “aviation” and whether freight aviation emissions are included in these reports.
Overall, the high coverage rate of 92% global tourism expenditure data based on 62 TSA data in the model enables us to model global tourism emissions with a high level of reliability. A Monte Carlo simulation indicated that tourism carbon emissions in 2019 were between 5.1 and 5.5 Gt CO2-e, with a 99% likelihood range (Supplementary Method 2).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
link
