Global patterns of daily CO₂ emissions reductions in the first year of COVID-19

Emissions of CO₂ from fossil-fuel combustion and cement production—the dominant sources of anthropogenic greenhouse gases—became highly variable during the COVID-19 pandemic, offering an opportunity to investigate their drivers. Fossil CO₂ emissions are commonly estimated from activity data (for example, energy consumption) and emissions factors, and their global uncertainties (±6% to ±10%) are relatively low compared to other air pollutants. Satellite observations provide consistency checks for large point sources and cities. Pandemic-related lockdowns had marked impacts on human activity and the Earth system, reducing air pollutants and CO₂ emissions. Because direct, low-latency activity data for global CO₂ emissions are scarce, the pandemic’s effect can be inferred by combining available activity data with high-frequency proxies (for example, government policy stringency as confinement levels or mobility data). The availability of daily or sub-daily proxy data across sectors enables near-real-time, high temporal resolution estimates. To address this gap, the authors developed and updated the Carbon Monitor daily CO₂ emissions dataset for the full year 2020, combining inventories with near-real-time activity data across power generation, industry, ground transportation, aviation, maritime transportation and residential fuel use, covering hundreds of cities and many countries. This allows evaluation of national energy-use patterns and disentangling drivers including seasonality, working days, weather, the economy and the pandemic, despite greater uncertainties than official inventories.

The study builds on decades of emissions estimation methods using activity data and emissions factors and on prior assessments of global carbon budgets and trends. It references satellite-based validations of inventories for cities and point sources, and multiple early COVID-19 studies that estimated emissions changes using confinement levels or mobility data. It situates the work relative to international inventories and datasets (EDGAR, UNFCCC, IEA, BP), which have reporting lags of 6–16 months, motivating near-real-time approaches.

The authors computed daily CO₂ emissions since January 2019 by sector and country using inventories and near-real-time activity data, assuming emissions factors were constant between 2019 and 2020 so that daily emissions scale with activity. Scope covered six sectors: power generation, industry (including cement), ground transportation, domestic and international aviation, international shipping, and residential consumption. Key steps: 1) Disaggregate 2019 annual emissions to daily using activity patterns. 2) Scale to 2020 using day-specific activity ratios. Data and sector-specific approaches: - Power: Daily national thermal electricity production was used as activity; for China daily coal consumption (Zhedian Company) was used to disaggregate monthly thermal generation; Russia used hourly thermal production from SO-UPS. - Industry: Monthly emissions in 2019 were derived from production data or industrial production indices (IPI). For 2020, monthly emissions were estimated from year-on-year changes in industrial output; where recent data lagged (one to two months), Trading Economics forecasts were used. Daily disaggregation employed daily thermal electricity generation. - Ground transportation: Urban traffic congestion indices (TomTom) in 416 cities were used as proxies. A sigmoid model mapped congestion level to car counts and thereby emissions, calibrated using Paris data. - Aviation: Domestic and international aviation emissions were estimated from real-time flight distances (Flightradar24). - International shipping: Daily changes proxied using the Baltic Dry Index. - Residential: Proxies for residential and commercial building fuel use were applied. US-specific workflow: State-level sectoral emissions (2017 baseline updated to 2018 with EIA energy consumption) were disaggregated to months via monthly consumption, then to days with state-level daily indicators; for the last two months of 2020, daily indicators and scale factors were used due to missing monthly data. ROW (rest of world) countries: Emissions reductions associated with workplace closures were parameterized using reductions observed in India, the United States, Europe, Brazil, Russia and Japan for four closure levels (Oxford Workplace Closing Index). Each ROW country’s daily 2020–2021 emissions were computed by weighting 2019 emissions by the country’s daily closure index. Baseline simulation for 2020: To isolate COVID-19 impacts, a counterfactual daily baseline was simulated by fitting linear regressions to monthly emissions for 2015–2019 per sector and country to project 2020 monthly baselines, then scaling 2019 daily profiles to those monthly totals. For sectors without monthly data (ground transport, residential, domestic aviation), yearly functions were applied; no baseline was simulated for international aviation and shipping. Uncertainty: Sector-specific 2-sigma uncertainties were estimated assuming normal errors and uncorrelated errors across sectors and countries. Power: ±10% total (activity ±5%, emissions factor ±13%). Industry: ±30% (output ±20%, emissions factors ±14% to ±28% in available countries). Ground transport: ±9.3% from regression prediction intervals. Residential: ±40% based on comparisons with French daily gas data. Aviation: ±10.2% comparing flight distance vs counts. International shipping: ±13% per IMO. Combined with EDGAR 2019 baseline uncertainty (±7.1%), the overall uncertainty in the 2020 vs 2019 annual change is ±13.6%. Correlation analysis: Pearson correlations among five indicators were computed for two periods (Mar–May 2020 and Oct–Dec 2020) across ten countries (US, India, UK, France, Germany, Italy, Spain, Russia, Brazil, Japan): daily COVID-19 deaths, government stringency index, time spent at residences, power demand change, and daily CO₂ changes relative to baseline.

- Global: 2020 fossil CO₂ emissions totaled 33.1 GtCO₂, a decrease of 2,232 MtCO₂ (±304 MtCO₂), or −6.3% (range −7.2% to −5.5%) relative to 2019. Mean daily emissions were 90.5 MtCO₂/day in 2020 versus 96.8 MtCO₂/day in 2019 (−6.6%). Global GDP fell 3.6% in 2020. - Temporal pattern: Largest weekly decline occurred in week 15 (6–12 April 2020), −17% versus same week 2019. April 2020 emissions fell 16.3% versus April 2019. Emissions recovered from late April with reopenings; by December 2020, emissions were only 0.5% below December 2019. Baseline counterfactual analysis indicates a larger annual COVID-19-related reduction of 6.5% after accounting for historical trends. - Country/region changes (2020 vs 2019): Brazil −9.7%; United States −9.5% (≈−480 MtCO₂); Russia −8.0%; India −7.9% (≈−195 MtCO₂); EU27 & UK −7.3% (≈−226 MtCO₂; UK −8.8%, France −9.0%, Germany −7.2%, Italy −7.1%, Spain −12.7%); Japan −4.7%; China +0.9% (≈+89 MtCO₂). China’s share of global emissions rose from 29.6% (2019) to 31.9% (2020), while international bunkers’ share fell from 3.9% to 2.6%. - Sectoral contributions to the 2020 global decrease: Ground transportation −709 MtCO₂ (−10.9%), contributing 32% of total decrease; Power −554 MtCO₂ (−4.1%), 25%; International bunkers (aviation + shipping) −503 MtCO₂ (−36.9%), 23% total, with international aviation ≈16% and international maritime ≈7%; Industry −265 MtCO₂ (−2.6%), 12%; Domestic aviation −112 MtCO₂ (−30.8%), 5%; Residential −89 MtCO₂ (−2.5%), 4%. Sectoral shares shifted: power 39.2% and industry 29.4% of 2020 totals (up from 38.3% and 28.3% in 2019); ground transport 17.5% (down from 18.4%); aviation 1.6% (down from 2.8%); international shipping 1.8% (down from 2.1%); residential 10.5% (slightly up from 10.1%). - Persistence: Ground transport emissions were 10.9% lower in 2020, with largest monthly drops in April (−33.7%) and May (−26.2%), and smaller drops in November (−9.7%) and December (−6.8%). Power and industry bottomed in April (−10.0% and −9.9%), rebounding to slightly above 2019 from August (Aug–Dec average growth +1.0% and +2.5%), yet still down cumulatively for 2020 (−2.5% and −1.4%). - Correlations and waves: In Mar–May 2020, daily CO₂ reductions strongly correlated in magnitude with COVID-19 deaths, government stringency, mobility at residences, and power demand (reported r≈0.9, 0.9, 0.8, 0.8 in absolute value). In Oct–Dec 2020, relationships between CO₂ changes and deaths or stringency largely weakened or disappeared; CO₂ changes remained strongly related to power demand (r≈0.9) while the reduction in demand was far smaller in the second wave (−11.3 GWh/day) than in the first wave (−1,877.6 GWh/day).

The study demonstrates that COVID-19 containment measures and associated reductions in human activity produced the largest observed annual drop in fossil CO₂ emissions to date, concentrated in mobility-related sectors. The temporal dynamics—steep declines in early 2020 followed by rapid rebounds from late April—highlight how emissions closely track policy-driven and behavioral changes. Subsequent waves with similar or higher health impacts produced much smaller emissions responses, indicating adaptation, less stringent or more targeted measures, and partial resumption of activities, particularly outside the most mobility-dependent sectors. Country contrasts underscore differing timelines and recovery patterns: China’s swift rebound yielded a net annual increase, while the United States and EU/UK made the largest absolute contributions to the global decrease. Sectoral analysis clarifies leverage points for mitigation, with transport—especially international aviation and shipping—being highly sensitive to mobility restrictions, while power and industry rebounded as demand recovered. The magnitude of the 2020 decline is comparable to the sustained annual reductions needed to keep warming to 1.5°C, underscoring the scale and speed required for structural energy transitions as opposed to temporary, disruptive reductions. Near-real-time monitoring of daily emissions can inform timely policy responses and track the climate implications of economic recovery strategies (green vs brown).

The authors provide a near-real-time, daily global CO₂ emissions dataset for 2020, quantifying a 6.3% (2,232 MtCO₂) reduction relative to 2019, driven primarily by mobility-related sectors during the first pandemic wave, with rebounds thereafter. The work advances methods to integrate inventories with high-frequency activity proxies and constructs a 2020 counterfactual baseline to isolate COVID-19 impacts. Findings show that temporary restrictions can produce deep but short-lived emissions cuts; achieving climate targets requires structural changes in energy and transport systems that deliver sustained annual reductions comparable in magnitude to 2020. Future work should expand monitoring coverage, improve sector- and country-specific activity data and emissions factors at high frequency, refine proxy-to-emissions relationships (especially in transport and residential sectors), better represent regions with limited data (ROW), and integrate near-real-time emissions with policy and economic indicators to guide mitigation and recovery planning.

- Emissions factors were assumed constant between 2019 and 2020; real-world variability (fuel mix, plant efficiency) may alter inferred changes. - Proxy data limitations: traffic congestion indices were mapped to emissions via a model calibrated in Paris, with uncertain transferability to 416 cities; residential proxies may not capture all fuel-use dynamics. - Data lags and gaps required use of forecasts for industrial output (Trading Economics) and proxies for several sectors and regions; international aviation and shipping lacked baseline simulations. - ROW countries’ emissions changes were inferred from relationships derived in a subset of countries using Oxford closure indices, which may not reflect local conditions. - Baseline (counterfactual) emissions for 2020 were simulated using linear trends from 2015–2019; nonlinearity or structural changes could bias the baseline. - Uncertainties are relatively large for some sectors (residential ±40%, industry ±30%); overall annual change uncertainty is ±13.6%. - Correlation analyses are observational and do not establish causation; sign conventions (reductions vs levels) complicate direct interpretation of correlation signs. - Satellite validation is limited to some large sources; city- and plant-level checks are not comprehensive.