Existing fossil fuel extraction would warm the world beyond 1.5 °C

To estimate developed oil and gas reserves, we use Rystad Energy's UCube (Rystad 2020), a commercial bottom-up model. The UCube models the investment and operating decisions of 65 000 fields globally (of which around 25 000 are producing or under construction) based on their geology, costs and projected rates of return. Oil and gas prices are exogenous assumptions in the model; our base case uses an oil price projection of USD 60 per barrel (bbl) (supplementary methods 2).

It is commonly observed that bottom-up reserves data provided commercially to the oil and gas industry is more reliable than freely available and widely used sources (Speirs and Sorrell 2009, McGlade 2014), which have been characterised as 'inaccurate, inconsistent, uncertain and/or contested' (Thompson et al 2009) (supplementary text). Rystad's commercial model incorporates a consistent, independent, expert assessment of economic viability (vital to the concept of reserves) and is updated regularly (supplementary methods 2). Other peer-reviewed studies of oil and gas (Erickson et al 2017, Mercure et al 2021) and the IEA (2020b) use Rystad's data and modelling.

We include under-construction conventional oil and gas projects that have received an FID. In contrast, unconventional shale oil and gas production has two main 'commitment' points: to develop an area (installing infrastructure) and to proceed with wells within it. While either approach could be justified, we conservatively count only those shale reserves accessed by already-drilled wells.

Since coal reserves data are often of relatively poor quality (Zittel and Schindler 2007), we compile a new dataset for the largest coal-producing countries (Trout et al 2022). Although coal deposits are geologically simpler to estimate than oil and gas, reserves reporting is less regulated.

For China, responsible for 45% of global coal production (BGR 2019), we sample close to 300 mines and use a regression analysis to estimate the reserves of the full population of mines (figure 2). For India, the United States, Indonesia, Australia, South Africa and Poland, together representing 37% of global production (BGR 2019), our study collects mine-level data from company financial reporting, government regulatory data and industry data publications (supplementary methods 5). For Russia and Germany (8% of global production), we derive top-down estimates from government and industry sources, respectively (Ministry of Natural Resources and the Environment 2019, DEBRIV 2019). For countries accounting for the remaining 10% of world coal production, we estimate developed reserves as equal to current annual production multiplied by the median ratio of developed reserves to production across our nine studied countries (Mendelevitch 2018).

Zoom In Zoom Out Reset image size

While coal reserves reporting is increasingly being standardised via CRIRSCO, legacy classification systems persist in China, Russia and India, which reflect technical but not necessarily economic extractability (Thomas 2013). For China, we apply an adjustment factor of 0.57 (interquartile range 0.29, 0.83) based on a sample of 30 mines that state reserves under both Chinese and CRIRSCO standards; for Russia we use a defined subset of reserves reported under the Russian system and set wide uncertainty bands; for India, we control reserves estimates against remaining mine life and mine capacity, while setting wide uncertainty bands (supplementary methods 4 and 5).

China: To our knowledge, there is no published dataset nor compiled set of records on all Chinese mines. The National Energy Administration publishes data on the annual production capacity and location of over 3300 operating mines, covering over 90% of national mine operating capacity as of the end of 2017 (supplementary methods 4).

We sample reserves data for 299 mines, covering 42% of China's total operating capacity (Trout et al 2022). For these mines, we directly research recoverable reserves (可采储量), primarily using mine environmental impact assessments, regulatory reports released by the central and provincial governments, and company reports.

We conduct an ordinary least-squares regression of our sample using Long's (2009) generalisation of Taylor's (1977) rule for mine design, relating the scale of the mine (capacity, C) to recoverable reserves (R): $C = b{R^a}$. We rearrange to $R = \alpha {C^\beta }$ and regress on: ${\text{In}}\left( R \right) = \;{\text{In}}\left( \alpha \right) + \beta {\text{In}}\left( C \right)$. The regression (figure 2) gives coefficients with 90% confidence intervals of $\alpha = 40.85$ (36.55, 45.26) and $\beta = 1.22$ (1.15, 1.29). We predict recoverable reserves for unsampled mines using $R = \;40.85{C^{1.22}}$.

The purpose of committed CO₂ emissions accounting (Davis et al 2010, Davis and Socolow 2014) is to quantify future cumulative emissions implied by existing capital investments at one part of the energy system—in our case, extraction facilities. For fossil fuel-burning infrastructure, key variables are expected operational lifetime and, in the case of power plants, capacity factor (Davis and Socolow 2014, Tong et al 2019). For fossil fuel extraction facilities, the key constraining variable is the volume of reserves that can be economically extracted. If all committed reserves are extracted, the resulting CO₂ emissions would be largely unaffected by where or how they are burned (IPCC 2006). The following identity describes our approach to CO₂ accounting (specific applications are in supplementary methods 6):

$$\begin{align} & \text{CO}_2 \,\,\text{from developed reserves by fuel} \nonumber\\ &\quad = \;R \times h \times f \times b\end{align}$$

where:

• R = quantity of reserves in physical units (barrels of oil, condensate and natural gas liquids (NGLs), cubic feet of gas, or tonnes of coal).
• h = net calorific value (NCV) of the fuel (terajoules (TJ) per unit of fuel).
• f = default CO₂ emissions factor (tCO₂ per TJ), derived from carbon content (kgC TJ⁻¹) and assuming 100% oxidation of carbon.
• b = proportion of CO₂ emitted to the atmosphere, rather than stored in non-energy uses (%).

A small proportion of extracted fuels is converted into non-energy uses (e.g. plastics and lubricants, fertilisers and carbon fibres). Some uses lead to long-term storage of carbon, whereas others delay the release of carbon into the atmosphere (Marland and Rotty 1984). We use net carbon storage factors calculated by Heede (2014) (8.02% for liquids; 1.86% for gas; 0.016% for coal) to account for this in our results.

For all data listed above, uncertainties are assessed by expert judgment, according to the quality of the sources, and as described in supplementary methods 2, 4 and 5. To calculate the reserves quantities and emissions associated with developed reserves, we perform a Monte Carlo analysis over all the parameters to which we allocate an uncertainty (supplementary methods 6). This includes reserves estimates, oil density, coal rank, and NCV and emissions factors per fuel (figure S3 available online at stacks.iop.org/ERL/17/064010/mmedia), as well as uncertainties specific to our Chinese coal reserves methodology (supplementary methods 4). Each parameter is defined by a lognormal probability distribution function based on a best estimate, a low and high estimate, and the probability range that the high-low range spans.

Our central estimate of global developed reserves by fuel, at the start of 2018, is: 870 billion barrels (Gbbl) of oil, 3027 trillion cubic feet (Tcf) of gas, and 220 billion tonnes (Gt) of coal. Using the UCube model, which allows us to apply consistent definitions, our estimate of developed reserves amounts to 66% and 51% of the total reserves of oil and gas, respectively (tables S2 and S3). Well-known sources such as BGR (2019) and BP (2019) give larger estimates of total proven reserves, but also rely on government data that are often politicised and/or use inconsistent definitions for various countries (supplementary text). Welsby et al (2021) provide more technically robust estimates of proven oil and gas reserves, shown in tables S2 and S3, that are close to our estimates of total reserves. Coal reserves are more uncertain, being less regulated. BGR estimates total (developed and undeveloped) coal reserves at five times our developed reserves estimate (table S4).

We find that developed reserves are concentrated in China (coal), Russia (oil, gas and coal), Saudi Arabia (oil) and the United States (oil, gas and coal) (figure 3(a); table 1).

Zoom In Zoom Out Reset image size

We find that burning developed reserves of coal, oil and gas would produce 936 Gt CO₂, with a 90% confidence interval of 880–992 Gt CO₂ (figure 3(b)). This comprises 47% from coal (446 Gt CO₂ [399–494]), 35% from oil (323 Gt CO₂ [300–347]) and 18% from gas (165 Gt CO₂ [152–187]). Ninety percent of these emissions come from just 20 countries (table S1).

We compare these committed CO₂ emissions to estimates of the remaining carbon budgets for limiting warming to 1.5 °C (50% probability) and 2 °C (83% probability) above pre-industrial levels, as reported by the Intergovernmental Panel on Climate Change (IPCC) (Canadell et al 2021) and adjusted to their values as at the start of 2018 (using Friedlingstein et al 2019, 2020). We interpret the respective 50% and 83% probabilities as best reflecting the Paris Agreement goals (UNFCCC 2015): to hold global average temperature rise to 'well below' 2 °C while 'pursuing efforts' to limit it to 1.5 °C. Committed emissions from fossil fuel extraction facilities are 60% larger than the remaining 1.5 °C carbon budget (580 Gt CO₂) and exhaust 95% of the 2 °C budget (980 Gt CO₂).

Developed oil and gas reserves together exhaust more than four fifths of the 1.5 °C budget. The remaining one fifth is equivalent to just 6 years of coal production at present rates. Assuming such rapid closure of coal mines is not feasible, some developed oil and gas reserves, alongside coal, would need to be kept in the ground.

There are substantial inherent uncertainties in any reserves estimate in relation to consistency in reporting standards, geological uncertainty and future economic viability (McGlade 2012) (supplementary methods 1). Our estimates of committed emissions are subject to further uncertainty related to reserves' energy and carbon content and net carbon storage in non-energy uses (supplementary methods 2–6).

We assess the combined effect of such uncertainties by performing a 1000-run Monte Carlo simulation. Figure 4 displays the probability distribution of committed emissions results by fuel for the most significant countries. Chinese developed coal reserves are the largest single contributor to global committed emissions (228 Gt CO₂) and account for the largest uncertainty in our results, with a 90% confidence interval of 188–272 Gt CO₂.

Zoom In Zoom Out Reset image size

We do not find significant variation in results beyond 1000 runs, whilst results remain robust when uncertainties in reserves data are doubled: the mean outcome for developed reserves and committed emissions does not change significantly. For countries with more uncertain parameters, reflected in wider probability ranges (e.g. Saudi Arabia, Iran and Iraq), doubling the uncertainty level increases the 90th percentile range for reserves and emissions by more than 25% (supplementary methods 6; figure S4).

Uncertainties may be larger than we capture because we assume they are uncorrelated. Correlations might arise, for example, from systemic biases in project-level reserves estimation by particular companies, or from effects of national regulatory regimes for reporting reserves. Further assessing correlation between errors would enhance this analysis.

There are larger uncertainties in remaining carbon budgets. The IPCC Sixth Assessment Report cites 380–980 Gt CO₂ as the 66% confidence interval for staying within 1.5 °C, accounting for uncertainty in the climate response to cumulative CO₂ emissions (Canadell et al 2021). Our finding that global committed CO₂ from developed reserves exhausts the 1.5 °C budget is robust to that range. The 66% confidence interval for staying within 2 °C is considerably wider, 980–2380 Gt CO₂. Given the Paris Agreement commits to staying 'well below' 2 °C, we interpret the budget that provides the highest certainty (83%) of staying below that limit as most consistent with the Agreement's aims.

We do not separately address the impact of methane or other short-lived climate pollutants associated with fossil fuel extraction, processing and transport. However, remaining carbon budgets are sized to account for the projected contribution of methane and other non-CO₂ gases to future warming. If methane emissions are not reduced at the rate assumed, then remaining carbon budgets could be smaller (or vice versa), adding further uncertainty (Canadell et al 2021).

This study also does not consider non-energy CO₂ emissions from land use and cement manufacturing, which leave less room in remaining carbon budgets for fossil fuels, nor does it factor in carbon dioxide removal (CDR), which could create more room (Rogelj et al 2018). Keeping cumulative fossil fuel-related CO₂ emissions well within maximum carbon budgets would be the most precautionary approach to avoid overshoot of either 1.5 °C or 2 °C. Cement process emissions are particularly hard to abate, while cement is a vital construction material (IEA 2020a). Most integrated assessment models assume land-use emissions will be negative by mid-century (Rogelj et al 2018), yet they are still positive and climate change impacts threaten the long-term ability of trees to sequester carbon (Anderegg et al 2020). Technologies to capture and store carbon could reduce the current CO₂ emission rate of fossil fuel combustion, but not eliminate it, while CDR technologies carry inherent risks, being unproven at scale (Larkin et al 2018). One reason our study reaches a stronger conclusion than the IEA (2021) finding that no additional fields and mines are needed is that the IEA scenario includes some CDR and significant carbon capture and storage (CCS) of fossil fuel emissions. The IEA acknowledges that CCS availability is one of the greatest uncertainties in its scenario; three decades of efforts to deploy CCS have largely failed (Wang et al 2021).

The UCube model enables us to separately test the sensitivity of developed oil and gas reserves to future price assumptions. Whereas around 80% of global coal production is consumed in the country where it is extracted (BP 2019), and thus sensitive primarily to national market and political conditions, oil and (to some extent) gas are traded on a global market and price can be introduced as a global variable. Our central estimate is based on a long-term Brent oil price of USD 60/bbl ($2020). A high long-term oil price of USD 80/bbl ($2020) increases estimates of commercially viable developed oil and gas reserves by 2%, while a low case of USD 30/bbl $2020) decreases estimates by 22% for oil and 16% for gas (table S5). In the low-oil-price scenario, committed CO₂ from oil and gas falls to 390 Gt. But, including coal, committed emissions from developed fields and mines would still exceed the 1.5 °C budget (50%).