Carbon intensity of global crude oil trading and market policy implications


Scope and resolution of the life cycle assessment (LCA)

This analysis uses life cycle assessment (LCA) to account for the well-to-refinery-entrance greenhouse gas (GHG) emissions from the petroleum supply chain. The foundation of the LCA is a network representing the global oil supply chain where oil fields, shipping terminals, pipeline stations and refineries are nodes; pipelines and shipping routes are edges. At the level of each producing country, the network is jointly used with a multi-objective optimization algorithm to estimate crude blending. In conjunction with data on crude demand at refineries, this enables a resolution at the level of individual supply chain pathways, as illustrated in Fig. 1.

Fig. 1: Case study of crudes from Saudi Arabia to India: the supply chain network, blend formation and tracking of crude barrels from sources to destinations (volume shown in kilo-barrels/day or kbbl/d).
figure 1

Our method captures the global supply with a well-to-refinery-entrance scope. To detail its different aspects, the infrastructure details for Saudi Arabia and its connectivity to India are shown here for illustrative purposes. As shown by the network architecture (a), the supply chain consists of nodes (oil fields, shipping terminals, pipeline stations and refineries) and edges (pipelines and shipping routes). The sizes of field nodes and refinery nodes are scaled in proportion of production volume and total intake volume, respectively. To estimate blend formation, we use the network with a multi-objective optimization algorithm that estimates how crude from oil fields combines to form blends (b). Tying these elements with information about crude demand at refineries, we track crude at the level of individual supply chain pathways from fields to refineries (c).

The “well-to-refinery-entrance” scope points toward two emission categories—crude extraction (upstream) and crude transportation (midstream). The crude blending algorithm, in conjunction with field-level crude extraction CI calculated by OPGEE16, constitutes the former, while the latter is estimated using mode-specific emission models (refer to “Methods”). The resolution of the LCA enables the accounting of life cycle emissions at different levels of aggregation. As shown in Fig. 2, this not only generates estimates of carbon intensity for global crude blends but also destination-specific CI to inform policymaking addressing refineries and/or petroleum products.

Fig. 2: Case study of crudes from Saudi Arabia to India: carbon intensity (CI) estimates (in kg-CO2eq/barrel) at different levels of policy-relevant aggregation (sample refineries in India selected based on refining volume >20 kilo-barrels/day or kbbl/d).
figure 2

After estimating the upstream and midstream carbon intensities for every individual pathway, we aggregate them at the refineries (a); the bars shown at the chosen refinery nodes represent the blend-level well-to-refinery-entrance CI in kg-CO2eq/bbl. The weighted carbon intensities of crude blends from Saudi Arabia to India (b) are not only variable across the relevant four crude blends but also exhibit a wider variability when accounting for different refinery destinations. This demonstrates that the heterogeneity in upstream and midstream emissions leads to each refinery in a country having a unique profile of crude blend carbon intensities.

Carbon intensity of marketed global crude oil blends

Figure 3 shows the volume and upstream CI of marketable global crude blends, where the latter is computed by coupling the blend estimation algorithm with the field-level crude production CI. Uncertainties are quantified by varying parameters of the algorithm weighting different factors, such as proximity, pipeline connectivity, etc., in the optimization approach (refer to Supplementary Note 4). We examine blend-level variability within and between countries in addition to the aggregated country-level variability.

Fig. 3: Carbon Intensity (CI) associated with crude extraction aggregated at the level of crude blends (top 3 blends by volume with >0.1 million barrels/day or Mbbl/d) in the top 15 oil-producing countries.
figure 3

The key blends (based on the aforementioned criterion), their CI, the aggregate country-level volumes and CI are shown for the chosen set of countries. This illustrates how producing countries compare against the global average and how blends compare against the respective country averages. Note that some countries (Russian Federation and Angola) exhibit low uncertainty in carbon intensity for each crude blend and low variability across crude blends; conversely, other countries (Canada, Venezuela, and Iran) have wide uncertainty within and wide variability across crude blends. Source data are provided as a Source Data file.

Blends in Russia show less inter-blend variation in carbon intensities—standard deviation of 3.32 kg-CO2-equivalent/barrel (kg-CO2eq/bbl) versus the global standard deviation of 32.08 kg-CO2eq/bbl. In addition, the volume-weighted country upstream CI of 48.38 kg-CO2eq/bbl is close to the global volume-weighted average of 45.03 kg-CO2eq/bbl. This is due to the low standard deviation in the field-level CI (volume-weighted country standard deviation of 12.76 kg-CO2eq/bbl versus the global standard deviation of 34.49 kg-CO2eq/bbl) and the presence of proximate blend clusters connected by common large-scale infrastructure such as the ESPO pipeline network. More generally, the former is the key driver behind inter-blend variability and uncertainties. For example, the low inter-blend variability in Angola can be contrasted to the high inter-blend variability in Canada based on the respective volume-weighted field-level distributions (The mean and standard deviation of the weighted CI distribution of Angolan fields are 50.34 and 6.88 kg-CO2eq/bbl, respectively, whereas for Canadian fields are 71.73 and 46.62 kg-CO2eq/bbl, respectively).

With a range of 3.4–181.6 kg-CO2eq/bbl, the Middle East region shows significant variability, primarily down to differences in field-level CI as described by Masnadi et al. 16 The uncertainties in the CI of Iranian blends are, in general, higher than some of the other major producing countries (major countries defined as the top 15 oil-producing countries as indicated in Fig. 3) due to the greater number of blends (~2.5 million-barrels/day spread over 11 blends) in the country (and less degree of differentiation between crude properties, which makes the blending algorithm sensitive to the weighting parameters. On the other hand, the presence of a predominant blend in Saudi Arabia (~9.8 million-barrels/day spread over 6 blends) and Iraq (~3.1 million-barrels/day spread over 5 blends), namely Arab Light and Basrah Light, respectively, results in low uncertainties. More generally, the presence of fewer blends and one predominant blend indicates lower uncertainties due to the resulting stability of optimal solutions found through the gradient-descent approach (refer to “Methods”).

A similar degree of inter-blend variability is seen across Latin America—blends from Mexico, Brazil and Argentina are found to be near the global volume-weighted average, whereas Venezuelan blends have significantly higher CI due to the heavy oil type of reservoirs and the use of carbon-intensive operational practices (e.g., steam flooding)16.

In North America, the energy-intensive Oil Sands Synthetic blend from Canada has the highest carbon intensity among the major global blends (144.5 kg-CO2eq/bbl). This closely tracks the fields with similar API density from the oil sands region, which shows a carbon intensity range of 82.5–160.2 kg-CO2eq/bbl and a volume-weighted mean of 139.3 kg-CO2eq/bbl, thus attesting to the efficacy of the blending algorithm.

Figure 4 shows the cumulative well-to-refinery-entrance CI at the blend level by combining the upstream and midstream CI for the 20 highest volume global crude blends, in addition to showing the variability within midstream emissions. The global volume-weighted midstream CI of 5.37 kg-CO2eq/bbl, as shown in the sub-figure, contributes ~10% to the well-to-refinery-entrance emissions (refer to Supplementary Note 8 for comparisons with relevant literature). Although in magnitude, the average upstream CI is 9 times the midstream CI, the variation in midstream CI, for a given blend, across all supply chain pathways is significant, as shown in the right sub-figure. All the distributions in the chosen set are asymmetrical and have long tails indicative of the complexity of crude transportation networks; these skewed, irregular patterns emphasize the need to identify specific opportunities for policy intervention instead of applying a blanket approach.

Fig. 4: Well-to-refinery-entrance carbon intensity (CI) with the variability in crude transportation CI (in kg-CO2eq/barrel) for the top 20 global crude blends by volume.
figure 4

Segmenting well-to-refinery-entrance carbon intensity into upstream and midstream (a) demonstrates the wide variability in CI, with Arab Heavy and Western Canadian Select representing the low and high bounds, respectively. This sub-figure also illustrates how the proportion of upstream and midstream CI varies across blends. Violin plots (b) show the distribution of midstream CI. Specifically, they illustrate the volume-weighted distribution of midstream CI values (thicker parts of the violin indicate higher probabilities) that the listed crude blends exhibit across different supply chain pathways in the global network. The dashed black line shows the global volume-weighted average, the white dots show the blend-specific volume-weighted averages, and the dashed white lines show the volume-weighted quartiles for each blend. Source data are provided as a Source Data file.

Notable examples showing high variability include West Texas Intermediate (WTI) from the U.S. and Maya from Mexico. Given that WTI is a benchmark blend centralized in Cushing, Oklahoma, and that it is consumed in 49 North American refineries, the corresponding midstream entails high variability in pipeline miles traversed through an extensive and well-connected pipeline transport system. While for Maya, the variation is explained by a large spread of destinations ranging from domestic refineries to shipped exports to Southeast Asia. Like WTI, the Bakken blend shows high transportation CI due to the long distances between the source fields (Bakken region in Central North America) and destination refineries, which are as far out as the Gulf Coast and the East Coast of the U.S.

Comparing the midstream CI distributions, we find that blends with a large export footprint, e.g., Arab Medium (100% exported), Merey (>93% exported), and Basrah Light (>94% exported), have multi-modal distributions. This is due to the prevalence of shipped exports and specific features of trade lanes connected to the key import hubs across different continents.

Crude transportation CI from producer to consumer countries

The variability in transportation CI aggregated at the country level shows noticeable patterns in the supply chain (Figs. 5 and 6). Specifically, Fig. 5 illustrates the volume-weighted average CI associated with crude transportation from a given producer country to a given consumer country, while Fig. 6 illustrates the CI trends at the region level.

Fig. 5: Midstream carbon intensity (CI) and trade volumes between producer and consumer countries.
figure 5

This figure illustrates how supply chain traceability allows us to see the pairings between producer and consumer countries and the associated trade volumes and carbon intensities for each of these pairings. This is a level of detail aggregated from the individual source blend and destination refinery pairs. Blank values in the visualization matrix correspond to producer, consumer country pairs that do not have a crude trading relationship. Source data are provided as a Source Data file.

Fig. 6: Volume-weighted midstream carbon intensity (CI) in from selected oil-producing regions, segmented by consumer regions and crude transport modes.
figure 6

Volume-weighted midstream CI in kg-CO2-equivalent/barrel (kg-CO2eq/bbl) from the Middle East (a), Latin America (b), Africa (c) and Russia (d) segmented by consumer regions and crude transport modes. Midstream characteristics are highly variable, making the life cycle carbon intensities attributed to crude transportation highly dependent on the consumer region. The main drivers guiding this heterogeneity are total shipping distances, the proportions of pipeline and ocean transport and the overall transport efficiency. (C.A.—Central Asia, SE Asia—Southeast Asia) Source data are provided as a Source Data file.

The extensive pipeline systems in the U.S. and Canada together account for ~40% of the total pipeline miles in the world while representing ~23% of the total refining volume and ~17% of the total crude production volume22. These pipeline miles span across a distributed, decentralized network of refineries (~34% of the global number of refineries). In addition, the global volume-weighted average per mile CI of pipeline transport is 2.5 times that of shipping transport. These factors together increase the CI of crude transport in the region to 8.7–12.1 kg-CO2eq/bbl against the global average of 5.37 kg-CO2eq/bbl. In comparison, the CI of the pipeline system in Russia (with extensions into Western/Eastern Europe and China) exhibits a range of 1.5–5.1 kg-CO2eq/bbl, with the differences due to the fact that overall pipeline miles are comparable to crude production (~12% of total global pipeline miles and ~12% of total crude production) as opposed to North America and higher centralization (~62 refineries compared to ~100 in the U.S.). Additionally, given the regions of Eastern Europe, Western Europe and China represent ~88% of Russia’s net export volume, the corresponding midstream CI is skewed toward pipeline transport, unlike other exporting regions.

Among shipped exports, as seen in Fig. 6, the volume-weighted shipping CI from Latin America to Asia is 10.7 kg-CO2eq/bbl in contrast to that from the Middle East, which is 5.2 kg-CO2eq/bbl. This difference is attributable to inefficiencies in shipping, the usage of smaller tankers (all things equal, tankers with larger capacities result in lower per-barrel emissions), and longer distances (route carbon intensity has a correlation coefficient of ~0.74 with route distance). The differences can also be seen in the country-level breakdown as shown in Fig. 5—for example, CI values from Venezuela to India, Colombia to China, Mexico to Japan are 15.29, 16.05 and 14.10 kg-CO2eq/bbl, respectively; those from Iraq to India, Iran to China, Saudi Arabia to Japan are 5.18, 2.07 and 4.20 kg-CO2eq/bbl, respectively. This is consistent with the patterns in crude tanker activity that indicate high traffic of ultra and very-large crude carriers (ULCCs, VLCCs) with capacities >2 million barrels from the Middle East. Segmenting shipped exports from the Middle East based on destinations, we observe that the CI of trade with North America is two times more than that with South and Southeast Asia. The primary driver behind this difference is the shipping distance—the volume-weighted average shipping distance from the Middle East to North America is 2.5 times the shipping distance to South and Southeast Asia.

Comparing the different sources of transportation emissions (pipeline, shipping and other), we observe that while the global volume-weighted averages for pipeline and shipping are similar (2.55 and 2.61 kg-CO2eq/bbl), there exist significant inter-regional differences as discussed above and demonstrated by Fig. 6. On a global basis, the shipping emissions estimated here are within the bounds of previous studies, where these emissions have been estimated to be up to 14% higher or 20% lower than the present results (refer to Supplementary Note 8 for the detailed comparison of these CI values with relevant literature). Note that the “other” category represents the heuristic edges in the network, which can be conceptually interpreted as the intra-field pipeline connections and approximations to substitute for missing pipeline data. The global volume-weighted average for the “other” edges in the network is 0.21 kg-CO2eq/bbl, i.e., ~8% of the core pipeline emissions (refer to “Methods” and Supplementary Note 3 for more details about the network creation process).

These averages can be further contextualized through the scale of the different modes of transportation in terms of barrel-miles per day: ~33,000 for pipeline, ~145,000 for shipping (and 3000 for other) and CI per unit distance: 5.56 kg-CO2eq/bbl per thousand miles for pipeline (hence also for the “other” category, refer to Supplementary Note 1) and 2.28 kg-CO2eq/bbl per thousand miles for shipping.

Net carbon footprint attributed to consumer countries

Upon aggregating the well-to-refinery-entrance CI at all global refineries, Fig. 7 illustrates the attribution of CO2eq emissions to countries based on the crude blends they consume. The map shows all countries with refining capacity included in the assessment. We further examine countries that process >1 million barrels of crude per day (which cumulatively represent ~78% of the total refining volume included in the study) in the accompanying bar chart and sort them in order of the corresponding net annual kg-CO2eq emissions. Compared to the global volume-weighted average of 50.46 kg-CO2eq/bbl, among these countries, the well-to-refinery-gate carbon intensity varies from 8.84 to 86.39 kg-CO2eq/bbl.

Fig. 7: Well-to-refinery-entrance carbon intensity (CI) with the variability in CI and absolute emissions for countries with major refining volume.
figure 7

The map (a) shows the well-to-refinery entrance CI for countries with refinery volumes greater than 20,000 barrel/day. The results in this figure are aggregated at the level of consuming countries and demonstrate wide variability across the world’s top crude oil refiners (b), ranging from 8.84 to 86.4 kg CO2eq/barrel, among countries with refinery volumes greater than 1 million barrels/day. This variability leads to countries with the lowest CIs having less total emissions attributable to well-to-refinery-gate emissions (c) than some other countries with half the refining volume. Source data are provided as a Source Data file.

We observe that among producers, the net well-to-refinery-entrance carbon footprint aggregated at the country level based on crude consumption is a strong function of the domestic upstream CI. The main reasons for this are that, on average, upstream CI makes up for 90% of the net CI as discussed earlier, most major producers consume crude produced domestically, and the presence of established and efficient pipeline routes to key domestic refineries—e.g., Saudi Arabia, Iraq, UAE and Russia all consume crude produced 100% domestically (the U.S. is an exception with a significant import footprint, which thus adds transportation emissions due to the resulting shipping activity). Consequently, countries such as Venezuela, Canada and Algeria, with above-average upstream CI, rank accordingly upon aggregation of the well-to-refinery-gate CI. Furthermore, within this subset, countries that are more expansive (i.e., with relatively greater pipeline miles) have a relatively higher transportation CI, most notably Canada and the U.S. This is based on the proportionality between length and emissions observed in the fluid mechanics-based approach used for estimating pipeline emissions (refer to “Methods” and Supplementary Note 7 for more details).

Among countries that are predominantly importers, the variability in the blend upstream CI and the midstream CI associated with specific source-destination routes drives the net CI aggregated at the country level based on the crudes they consume. Additionally, importers relying on shipping have a marginally higher refining-attributed CI, as seen by comparing regions of Western Europe (access to pipeline systems from Scandinavia, Russia) and Asia (sparse pipelines; heavy reliance on shipping for imports).

These findings are relevant to regulators that seek to encourage low-carbon sourcing and supply chain pathway prioritization by differentiating crude blends at the point of refinery intake. While Fig. 7 provides a snapshot of the net annual impact of source crudes, the methodology, resolution of the LCA and the modular nature of the analyses act as enablers for climate-aware trade decisions in the near future.

Implications for decarbonization policy

At the highest resolution of the LCA, i.e., individual source field to destination refinery pathways, carbon intensities vary from 0.74 to 39.41 gCO2/MJ with a volume-weighted average of 9.01 gCO2/MJ or 50.46 kg-CO2eq/bbl (this average compares to the IEA’s estimate of 57.23 kg-CO2eq/bbl5). This heterogeneity in life-cycle emissions represents an untapped decarbonization opportunity that can be realized through policy action.

Specifically, through our approach, we fill the gaps related to supply chain traceability which have limited policy efforts such as the LCFS by CARB11 and the Fuel Quality Directive by European regulators13 that sought to differentiate between sources of crude. The future success of such policies built on high-fidelity life-cycle assessment could impact crude oil pricing and cause a non-linear shift in the global supply curve. These estimates, along with the high resolution of the LCA, lay the foundation for effective decarbonization policies that can prioritize low-carbon trades. The different levels of aggregation (pathway, blend trade, country) facilitate the flexibility needed to implement these policies. This flexibility could be utilized through multiple policy channels with regional (e.g., CARB in California) and sectoral (e.g., CORSIA by the ICAO for the aviation sector) scopes, thus creating a multi-pronged structure to incentivize CI-based crude differentiation. Furthermore, the significance of these emissions over the 30-year horizon (as shown in Fig. 8) can motivate real-time, granular carbon emissions estimation and reporting from industry. Reporting systems can use appropriate technology (e.g., blockchain)23, thereby leading to better and more data. This can enable more accurate emission inventories and, in turn, lead to more effective decarbonization through policy action and business strategy. Additionally, with the high granularity and broad coverage of our approach, the analysis creates the framework to fill the gaps that would eventually lead to the convergence of model-based approaches and reporting practices.

Fig. 8: Scenario analysis—oil trade prioritization optimized for life-cycle CO2eq.
figure 8

We estimate the total reduction potential in well-to-refinery-gate CO2eq emissions by analyzing a variety of Shared Socioeconomic Pathway (SSP) scenarios. The estimation is performed by (1) considering the time series of oil supply, (2) ranking crude trade supply chain pathways from highest-to-lowest carbon intensities (CI) and (3) fulfilling supply by prioritizing the low-carbon pathways. The left sub-figure (a) shows this mitigation curve generated by removing marginal barrels of crude at any given supply level based on CI. To plot all scenarios along this curve, we use adjusted volumes, i.e., volumes scaled by the ratio of the scenario-specific 2015 supply value and the net supply from the crude production data to ensure all scenarios have the same starting point. Source data are provided as a Source Data file.

To estimate the potential decarbonization impact of the pathway-level CI heterogeneity, we consider oil supply projections from a diverse array of Shared Socioeconomic Pathways (SSP) scenarios2,24. We then use the pathway-level CI estimates to meet future demand by prioritizing low-carbon pathways, i.e., fulfilling the reduction in the forecasted annual number of crude barrels by eliminating supply from pathways having the highest carbon intensities, as illustrated in Fig. 8. The left sub-figure in Fig. 8 shows the different forecasting models and SSP scenarios under consideration with the projected values of oil supply in 2050; we exclusively look at scenarios with future oil supply less than present levels. Next, for every incremental decrease in supply, we quantify the corresponding CO2eq saved by phasing out the high-carbon supply chain pathways and generate the CO2eq savings curve shown in the left sub-figure.

For a subset of model/scenario combinations, we then examine the forecasted time series to estimate the average annual carbon intensities with the net CO2eq savings according to the aforementioned demand fulfillment minimizing overall CO2eq. Under 1.5 °C scenarios up to 2050, this corresponds to additional CO2eq savings of 1.5–6.1 Gt with an average of ~4.5 Gt across all models with SSPs 1–4. This is comparable in magnitude to removing 100 million new gasoline-powered passenger cars, assuming a typical car is driven for 10 years and emits 4.6t CO2 per year.

It is worth emphasizing that these savings are additional, beyond the savings from reduced supply or CO2eq emissions management, and thus can be potentially realized without other capital-intensive interventions. Thus, these emissions reduction can be realized in addition to, and not instead of, process-oriented decarbonization options like reduced flaring and carbon capture. Consequently, market-oriented decarbonization based on crude CI differentiation is a valuable piece in the overall decarbonization puzzle (refer to Supplementary Note 9 for additional details on the emissions reduction opportunity).


Leave a Reply

Your email address will not be published. Required fields are marked *