Scaling up gas and electric cooking in low- and middle-income countries: climate threat or mitigation strategy with co-benefits?

The analysis included LMIC countries with a population of at least 1 million households using polluting solid fuels and/or kerosene in 2018 (the baseline year) based on the WHO global database of primary fuels/ energy used for cooking [10]. The analysis includes over 2.6 billion polluting fuel users from 77 countries. The scenarios are described below and summarized in table 1.

The business as usual (BAU) scenario assumes that past rates of change in primary household fuel choice are extended into the future. Past household cooking fuel choices were calculated for rural and urban areas of each country using WHO’s household cooking fuel choice database, which includes six fuel categories: Biomass (consists mainly of unprocessed firewood, but also includes crop residues and dung), Charcoal, Coal, Kerosene, Gas (includes LPG, natural gas, and biogas), and Electricity [10]. Average rates of change for each fuel and other key assumptions are described in SI section 3.

The Full Transition-LPG (FT-LPG) scenario makes two key assumptions: (a) households using polluting fuels (Kerosene, Coal, Charcoal and Biomass) transition steadily to LPG such that use of polluting fuels falls to zero by 2040 and b) any electricity use at baseline evolves as in the BAU scenario. Therefore, by 2040, all cooking is done with either LPG or a mix of LPG and electricity.

The Full Transition to LPG and Electricity (FT-LE) scenario follows the same pattern as the FT-LPG scenario, but uses electricity as the main transition, while allowing BAU rates of LPG to continue. The Full Transition to Electricity examines a hypothetical transition to 100% cooking with electricity (including LPG-using households transitioning to exclusive use of electricity by 2040). The Intermediate (IT) scenario assumes BAU rates for each clean option increase by 50%.

Household fuels emit a wide variety of pollutants including well-mixed GHGs like carbon dioxide (CO₂), methane (CH₄), and nitrous oxide (N₂O) as well as SLCFs like black and organic carbon (BC and OC), non-methane volatile organic carbon (NMVOC), sulfur and nitrogen oxides (SO_x and NO_x ) and carbon monoxide (CO) [24]. BC and OC are constituents of fine particulate matter (PM_2.5), exposure to which is associated with numerous health risks [25]. Exposure to SO_x , NO_x , and CO can also damage health.

Emissions from ‘upstream’ processes like extraction, refining, and transportation of fossil-based cooking fuels (kerosene, LPG, and coal) were derived from Greenhouse Gases, Regulated Emissions, and Energy Use in Transportation (GREET), a life cycle model developed by Argonne National Labs primarily for analyzing transportation fuels [26]. GREET was also used to estimate transportation emissions associated with charcoal. GREET’s default parameters reflect current US conditions, but the model allows users to adjust inputs to reflect other regions. For the global analysis, we determined the dominant consumers of each fuel from our sample of 77 countries, as well as the major exporters supplying these countries and adjusted GREET accordingly. We describe specific approaches for each fuel in SI section 4.

For upstream emissions from charcoal production, the average emission factors for CO, NO_x , PM_2.5, SO_x , BC, OC, CH₄, N₂O, CO₂, and NMVOC were compiled from previous collections of field emission factors [19, 27, 28] and converted to a delivered energy basis using an assumed stove efficiency [19] (table S9). For biomass, we assumed that wood is locally collected and has no upstream emissions. The end use emission factors for biomass, charcoal, kerosene, coal and LPG were taken from previous compilations of lab and field data [6, 19, 27, 28].

Shares of electricity production and total production for different years were taken from the World Bank Development indicators [29] with missing data supplemented by International Energy Agency (IEA) statistics [30]. Future grid mixes were simulated using projections from the IEA’s World Energy Outlook ‘Stated Policies’ scenario through 2040 [31]. Lifecycle grid emissions per kWh of electricity produced from each type of power generation were estimated using GREET [26]. Plant-specific emissions are included in GREET’s database with some tunable parameters. GREET’s default values assume a US-based grid and power plant feedstock. We changed some parameters for coal, as mentioned above. Parameters for other feedstocks were left at their default values. We describe grid projections, emission factors, and additional assumptions in SI section 5.

When woody biomass is harvested and burned as fuelwood or charcoal, much of the carbon in the wood is converted to CO₂, contributing to climate change. However, depending on the rate of harvest and land management practices, some, or all the woody biomass regenerates, reducing the impact of the CO₂ emitted during combustion. When there are imbalances between woody biomass harvest and regrowth, CO₂ emitted during combustion is not fully sequestered by regenerating trees. This is called ‘non-renewable biomass’ (NRB). The ratio of NRB to woody biomass consumption is the ‘fraction of non-renewable biomass’ (fNRB). We use country-specific pan-tropical estimates of fNRB from a previous study [7] to account for net CO₂ emissions from wood and charcoal (see SI section 6 for more detail).

For the global health impact assessment, we did not estimate contributions of emissions to specific health outcomes. Instead, we estimated changes in emissions of (and assumed exposure to) the following health-damaging pollutants: PM_2.5, CO, NO_x , and SO_x [32]. This research group carried out a parallel study of national-level assessments in five priority countries using the ABODE model to estimate health impacts associated with changes in exposure [33]. The results of these health models are discussed in a separate paper.

We estimate climate impacts of the different scenarios using the FaIR v1.6.2 climate model, which is a simple emissions-based model that accounts for non-linearities in the carbon cycle and includes simplified processes representing greenhouse gas, aerosol, ozone, and other forcings from precursor emissions [34–36]. We use the multi-species configuration, with the RCP4.5 scenario as a baseline for the BAU scenario [36, 37]. RCP4.5 is a pathway for stabilization of radiative forcing by 2100, projecting somewhere in the region of 2.5 °C–3 °C global mean temperature increase above pre-industrial by the end of this century, and is the closest RCP scenario to current global climate policy [38]. Uncertainty was calculated based on runs using parameters derived from a 2237 member ensemble developed for analysis in IPCC’s Sixth Assessment Report [37]. This ensemble is observationally constrained in order to span the assessed range in climate system uncertainty, including global temperature change (1850–2019), atmospheric CO₂ concentrations (1750–2014), change in ocean heat content (1971–2018), and assessments of equilibrium climate sensitivity, transient climate response and airborne fraction of CO₂ emissions [37, 39]. Differences in emissions between the BAU and other scenarios were calculated for CO₂, CH₄, N₂O, SO_x , CO, NMVOC, NO_x , BC, and OC, and added to the RCP4.5 baseline emissions trajectory. The resulting emissions trajectories were run in the FaIR climate model to obtain annual changes in GHG concentrations, climate forcing, and temperature projected to 2040. To estimate uncertainties, we report the median and 5th, 25th, 75th and 95th percentile values of the ensemble simulations, with emissions held constant. Therefore, uncertainty estimates do not represent uncertainties in scenarios, only in climate system response [40].

Figure 1 compares energy consumption, number of fuel users and energy consumed under BAU and a full transition to LPG (FT-LPG). Under BAU, demand for biomass, coal, and kerosene decrease by 25%, 96%, and 25%, respectively, between 2018 and 2040. However, these patterns are not consistent across all regions. For example, BAU biomass demand decreases steeply in Asia/Pacific regions but increases in Africa. In addition, BAU charcoal consumption increases globally over this period by over 125%, driven mainly by growth in urban consumption in Africa. Clean fuel use also increases under BAU: LPG consumption increases by 50% and electricity demand for cooking doubles.

Zoom In Zoom Out Reset image size

To gauge health impacts, we report changes in emissions of health-damaging pollutants including fine PM_2.5, CO, NO_x and SO_x . Here, and below, we refer to emissions from cooking fuels in the 77 countries included in our analysis. In 2018, approximately 7.6 Mt of PM_2.5 were emitted from cooking fuels (figure 2(A)). Under the BAU scenario, this is projected to decline by 24%, driven by transitions from polluting fuels to LPG and electricity described above. Under the IT scenario, a modest increase in clean fuel adoption results in a 12% reduction of PM_2.5 emissions relative to BAU. Under full transitions to either LPG or electricity, PM_2.5 emissions from residential cooking are nearly eliminated.

Zoom In Zoom Out Reset image size

CO emissions follow a similar pattern as PM_2.5 (figure 2(B)). In 2018, we estimate that 71 Mt of CO was emitted from cooking fuels. Under the BAU scenario, CO emissions decrease 5% to 67 Mt by 2040 (CO declines less than PM_2.5 due to growth in charcoal consumption). In the IT scenario, 2040 CO emissions are ∼10% lower than BAU, and in the full transitions to either LPG or electricity, CO emissions decrease by over 95%.

NO_x emissions result primarily from atmospheric nitrogen and increase with higher combustion temperatures. LPG, coal, and thermal power generation occur at higher temperatures than biomass combustion in household stoves; therefore life-cycle NO_x emissions are higher from LPG, gas and electric cooking [41]. SO_x emissions primarily occur because of fuel-bound sulfur [42], which occurs in trace amounts in biomass but can be several percent by weight in coal [43]. Under the BAU scenario, NO_x emissions increase by ∼20% between 2018 and 2040 and SO_x emissions decrease by a similar magnitude (figures 2(C) and (D)). Under IT and FT-LPG scenarios NO_x emissions increase further because of increased LPG consumption. In contrast, SO_x emissions are lower in all scenarios relative to BAU except for FT-Elec. In that scenario, SO_x emissions are nearly 30% higher than BAU due to increased coal-based electricity consumption (see SI section 10 for a discussion of regional variation in health-relevant emissions).

Figure 3 shows fuel-specific emissions of well-mixed GHGs and BC in 2018 and in 2040 for each scenario. CO₂-equivalent units (CO₂e) represent the climate impacts of well-mixed GHGs (CO₂, CH₄, and N₂O) weighted by their 100 year Global Warming Potentials. In 2018, total emissions of well-mixed GHGs from residential cooking was 1.02 Gt-CO2e. Under BAU, we project these to increase by 12%, to 1.15 Gt-CO₂e by 2040. Under the IT scenario, emissions decrease 2% relative to BAU (to 1.12 Gt-CO₂e) by 2040. More substantial reductions relative to BAU occur under the full-transition scenarios: 15%, 32%, and 26% for the FT-LPG, FT-LE, and FT-Elec scenarios, respectively.

Zoom In Zoom Out Reset image size

In 2018, household cooking emitted ∼1.2 Mt of BC. Emissions decrease to 1.0 Mt by 2040 in the BAU scenario, decline by ∼12% under the IT scenario and are nearly eliminated under full transitions to LPG and/or electricity. OC and NMVOC emissions follow similar patterns (see SI section 10).

Under BAU, compared to a scenario of zero emissions from cooking, household cooking emissions contribute to an increase in effective radiative forcing of roughly 39 mW m⁻² (−97 to 42 mW m⁻²; 95% CI) and 15 millikelvin (−41 to 26 millikelvin; 95% CI) of warming by 2040. Our alternate intervention scenarios result in both reduced forcing and lower temperature (figure 4). By 2040, median temperature is 0.3 millikelvin lower in the IT scenario and 4.7–5.1 millikelvin lower in the full transition scenarios (figure 4(A)). However, the confidence intervals for each scenario significantly overlap with zero (FT-E: −0.0238 to 0.0245, FT-LE: −0.0247 to 0.0267, FT-LPG: −0.0244 to 0.0271, IT: −0.0027 to 0.0035) so that we cannot state with certainty that the transitions would result in net cooling. Nevertheless, full transitions to LPG and/or electricity cause sustained reductions in annual emissions, resulting in median temperature pathways that will continue to diverge from BAU well into the future. Moreover, relative uncertainty in the temperature response should decrease in the future, due to the longer lifetimes of well-mixed GHGs and higher certainty associated with their effects.

Zoom In Zoom Out Reset image size

Short-term climate forcing is dominated by aerosols, with reductions in OC and BC resulting in net warming and cooling, respectively (figure 5). Reductions in well-mixed climate GHGs like CO₂ have minimal impact on temperature during the timeframe of our analysis. We note that cumulative CO₂e emissions do not necessarily correspond well with their associated climate response for mixtures of long- and short-lived GHGs as in our transition scenarios [44, 45]. However, changes in CO₂e (figure 3) provide some indication of the long-term impacts of each scenario. For example, under the BAU scenario, cumulative CO₂e emissions are 24.6 Gt between 2018 and 2040. The three full transition scenarios reduce cumulative emissions of well-mixed GHGs by 2.6–3.5 Gt CO₂e over the simulation period (see SI section 11 for more detailed discussion of GHG and SLCF emissions).

Zoom In Zoom Out Reset image size

SLCFs dominate the projected climate impacts and associated uncertainty through 2040. Figure 5(A) shows the net temperature change and uncertainty (shown as the 5%–95% confidence interval of an observationally constrained, perturbed parameter ensemble) between FT-LPG and BAU scenarios attributed to each pollutant. Figure 5(C) summarizes this for each scenario. BC and OC have the largest impacts as well as the widest confidence intervals due to uncertainties in climate response to aerosol emissions; CH₄, NMVOC, and CO also have substantial impacts, largely due to their roles in tropospheric ozone formation. The net effect could be positive or negative based on overall uncertainty; however, the median response of the ensemble for all scenarios shows that cooling due to decreases in emissions warming agents (BC, CO, CH₄, and NMVOCs) overcomes warming arising from OC reductions. Figure 5(B) shows how regional emission reductions contribute to overall temperature changes by 2040. The African region has the largest impact, driven by reductions in BC, CH_4, NMVOC, and CO from decreased demand for fuelwood and charcoal. Under BAU, the African region has substantial increases in charcoal and fuelwood consumption (figure 1). Under the FT-LPG and other full transition scenarios, which phase out demand for charcoal and fuelwood, SLCF emissions are nearly eliminated.

Changes in PM_2.5, BC, and temperature are well-correlated because BC is a constituent of PM_2.5 and a key driver of short-term temperature change. Figure 6 shows that the modeled short-term temperature changes, though mostly driven by SLCF reductions, also scale with changes in CO₂e. These are reductions for all counties except Thailand and Myanmar (not shown). CO₂e is a more robust indicator of long-term temperature impacts. The generally linear relationship between ΔCO₂e and ΔT is moderated by several factors. For example, variation in the fNRB, a geographically-specific parameter characterizing the sustainability of woodfuel demand (see SI section 6), has a strong influence on ΔCO₂e. Further, the varying relative contributions of CH₄ and CO₂ to total CO₂e reductions (table S10) affects this relationship, as warming from CH₄ is more strongly associated with annual emission changes than warming from CO₂, which is well-understood to be a function of cumulative emissions.

Zoom In Zoom Out Reset image size