Unit-level assessment of mitigation pathways for mercury emissions from the global coal power fleet Authors: Guo Yaqin, Yan Liu, Dan Tong, Xinying Qin, Qingru Wu, Steven J. Davis, Xizhe Yan, Dongsheng Zheng, Ruochong Xu, Fei Liu, Ji Gaojian, Guannan Geng, Kebin He, Qiang Zhang Published: 2026 ABSTRACT Atmospheric mercury pollution remains a serious threat for human health and ecosystems, as recognized by the Minamata Convention on Mercury adopted in 2013. Coal-fired power units are major emitters of various environmental pollutants globally, including mercury. Here, we show the historical co-benefits of air pollution control measures and coal power retirement, identify coal-fired power units with disproportionately high mercury emissions, which we define as super emitters, and propose country-specific mitigation strategies. We find that super emitters were primarily concentrated in emerging countries by 2020. Our unit-level analysis highlights that substantial, near-term and cost-effective reductions can be achieved by prioritizing the retirement of super emitters and upgrading pollution control equipment, especially in countries with large coal fleets. In the long term, mercury-specific control technologies will be needed for deeper reductions. Our findings provide insights into customized and stage-based mitigation strategies for phasing down key sources of mercury emissions from coal-fired power units. Introduction Atmospheric mercury poses risks to ecosystems1 and human health2–4, due to its bioaccumulation, long-range transport, toxicity and persistence. These risks are well recognized by the international community, based on which the Minamata Convention on Mercury was established to mitigate mercury emissions and protect both environmental and public health. The coal-fired power sector alone contributed to 5–25% of global anthropogenic atmospheric mercury emissions during 2005–20155–8, and it was also a major emitter of other environmental air pollutants, including ~14% of global PM2.5 emissions in 20108,9. Currently, emission abatement techniques widely employed in the coal-fired power sector are designed primarily for conventional air pollutants, including SO2, NOx, and PM2.5, and they can also synergistically reduce mercury from the flue gas10. For example, fabric filters (FF) designed for trapping particulate matter can remove ~84–93% of mercury from flue gas in bituminous-coal-fired power plants (CFPPs) in the United States (the U.S.), primarily through the capture of particulate-bound mercury (Hgp) and, to a lesser extent, oxidized mercury (Hg2+)11,12. Thus, substantial mercury emission mitigation can be obtained in the conventional air pollutant control campaign targeting SO2, NOx, and PM2.5, as well as carbon neutrality aiming at carbon reduction, without technical barriers. Such co-reduction effects associated with air pollution control devices (APCDs) upgrades and CFPP retirements have already contributed considerably to mercury emission reductions10. The partial employment of mercury-specific technologies (e.g., activated carbon injection, ACI) in the U.S. further ensures ultra-low mercury emissions in coal-fired plants13,14. In contrast, rapid growth in coal power generation and relatively lax environmental regulations in many emerging countries have led to increasing or only slowly declining mercury emissions15–18. These countries now account for the majority of global mercury emissions from the coal-fired plants19. The entry into force of the Minamata Convention on Mercury has further highlighted the need to move beyond the co-benefits of conventional APCDs toward deeper mercury mitigation through dedicated technologies, such as ACI, particularly in high-emission regions10. Meanwhile, it is challenging to determine cost-effective mitigation priorities for these countries, because coal-fired plants or even units differ substantially in installed capacity, power-generation efficiency, and installed control technologies. Therefore, there is an urgent need for customized mitigation strategies based on both the successful experiences of the U.S. and the historical development patterns of regions with large coal-fired power fleets. Previous studies have explored the major factors driving mercury emission changes10,12,20–24, and revealed possible measures targeted at mercury emission mitigation10,25–31, including retirement policy, control technology upgrades, the “promoting the big and quashing the small” policy32, and power generation efficiency improvement11, based on global, regional, city- or plant-level aggregated estimates. These studies have enhanced our understanding of the magnitudes of mercury emission changes and their mitigation potential among some regions. However, most of these studies primarily focus on the drivers of mercury emission changes at the sector-wide level, as well as the measures designed for decarbonization or conventional pollution control from a regional or global perspective33–35. Limited attention has been given to the heterogeneity of contributions to historical mercury reduction across regional units, and the formulation of customized emission reduction strategies, including mercury-specific technologies, among units within the coal-fired power sector. Although some high-resolution analyses and simulations have been conducted for specific regions10,11,29,36, they might neglect both the importance of emissions from some small countries and the valuable experience gained from countries that have achieved substantial reductions in mercury emissions through advanced emission control technologies and regulatory policies, such as the U.S. and Western Europe37. These experiences could inform mercury mitigation in high-emitting regions by supporting the development of unit-specific strategies tailored to local conditions. Moreover, previous cost-efficiency analyses of mercury emission reductions have mostly been limited to the country or sector level38,39. Unit-level cost-effectiveness evaluations that incorporate mercury-specific technologies remain limited. Thus, a detailed and holistic analysis of underlying drivers behind historical success experiences for obtaining easily accessible emission reduction opportunities, subsequent identification of remaining co-reduction opportunities and further mitigation opportunities from mercury-specific technologies from the regional level to the unit level is crucial for formulating strategic planning. In this work, we develop a unit-level mercury emission inventory for global coal-fired power units, and analyze emission characteristics across units in terms of capacity size, pollution control technology and emission intensity. We decompose the unit-level emission changes into four driving factors: activity rate, pollution control technology, newly built capacity, and retired capacity. Based on these identified drivers, we further evaluate future mitigation pathways under different combinations of climate mitigation scenarios, retirement strategies, conventional pollution control policies and mercury-specific technologies. Our results reveal substantial regional heterogeneity in historical mitigation drivers and future mitigation potential, highlighting the importance of customized and stage-based mercury mitigation strategies for different coal-fired power fleets. Results Characteristics of global coal-fired power units Overall, the penetration of control technologies with higher mercury removal efficiency generally increased with unit size and this phenomenon became more pronounced over time (blue-colored bars in Fig. 1). This was mainly caused by the rapid deployment of large units ( ≥ 300 MW) with advanced APCDs (purple lines in Fig. 1), especially in late-industrializing countries (e.g., China, India and countries in rest of Asia), although regional differences still exist (see Supplementary Figs. 4–5 for details on rest of Asia and rest of world). For example, in 1990, electrostatic precipitators (ESP) were the dominant APCDs among different-sized units in the rest of Asia, while by 2020, the penetration of both desulfurization and dust-removing technologies reached nearly 100% in extra-large units ( ≥ 600 MW), and even 41.6% in the smallest units (0–50 MW). In India, due to lax environmental regulations, coal-fired power units were still dominated by ESPs in 2020, although units larger than 300 MW accounted for 61.6% and 50% of installed capacity and mercury emissions by 2020 (Fig. 1e–h). In contrast, the U.S. and most countries in Europe saw significant technological upgrades in their coal fleet, even though the distribution of installed capacity has changed only slightly over the past three decades.Fig. 1Shares of coal-fired generating capacity by pollution-control technologies across capacity categories, the proportion of capacity and mercury emissions in each capacity category and regional numbers of units.a–d Show China, e–h India, i–l Europe, m–p the U.S., q–t rest of Asia and u–x rest of world, with columns representing 1990, 2000, 2010, and 2020 from left to right. Bars indicate the shares of total coal-fired capacity equipped with different pollution-control technologies. The red, blue and purple lines with markers indicate the estimated percentages of total mercury emissions and total capacity for each capacity category, and the number of coal-fired units, respectively. Capacity categories are listed as smallest (below 50 MW), small (50–100 MW), mid-sized (100–300 MW), large-sized (300–600 MW), and extra-large units ( ≥ 600 MW). Shares of coal-fired generating capacity by pollution-control technologies across capacity categories, the proportion of capacity and mercury emissions in each capacity category and regional numbers of units. a–d Show China, e–h India, i–l Europe, m–p the U.S., q–t rest of Asia and u–x rest of world, with columns representing 1990, 2000, 2010, and 2020 from left to right. Bars indicate the shares of total coal-fired capacity equipped with different pollution-control technologies. The red, blue and purple lines with markers indicate the estimated percentages of total mercury emissions and total capacity for each capacity category, and the number of coal-fired units, respectively. Capacity categories are listed as smallest (below 50 MW), small (50–100 MW), mid-sized (100–300 MW), large-sized (300–600 MW), and extra-large units ( ≥ 600 MW). The proportion of emissions generally varied with the proportion of the installed capacity, but increasing disproportionalities emerged due to the penetration of advanced APCDs among larger units (blue and red lines in Fig. 1). For example, in Europe, the smallest-sized units emitted 11.2% of mercury emissions in 1990, approximately twice their share of power-generating capacity (5%). By 2020, their contributions to emissions have reached 15%, representing a striking 34% increase, due to outdated control technologies in small units (50–100 MW) in Eastern Europe and technological advancements in Western Europe (Supplementary Fig. 6). In India, the widespread application of ESPs across all capacity category between 1990 and 2010 led to an emissions profile closely aligned with installed capacity, but by 2020, the application of Wet Flue Gas Desulfurization (WFGD) with FF (WFGD + FF) in large units made significant disproportionalities, with ~37% of the capacity accounting for only 23.2% of the national emissions. This indicates that these small units could serve as strategic targets for mercury mitigation, either through technological upgrades or early retirement. Unit numbers also indicate a shift in the global coal fleet from predominantly small units to larger and ultra-large units. Meanwhile, the geographical focus on coal power has shifted from Europe and the U.S. toward Asia and other regions. In Europe and the U.S., the number of coal-fired units declined rapidly, particularly small ones. For example, the number of the smallest units in the U.S. fell from 488 in 1990 to 201 in 2020. In contrast, most capacity categories increased in other regions. In India, the number of the smallest units rose from 177 to more than 550, representing an increase of about 211%. Furthermore, China experienced the most pronounced growth in extra-large units, increasing from only 2 in 1990 to 749 in 2020. Identification of super-mercury-polluting coal-fired power units Figure 2 illustrates annual mercury emissions during 1990–2020 (colored by capacity category), as well as the distributions of unit-level power-generating capacity and mercury emissions in 1990, 2000, 2010, and 2020, with units ranked according to mercury emission intensity (defined as mercury emissions per unit of electricity generation, and units with the highest emission intensity on the left). We identify the units accounting for the top 10% of country-level mercury emissions within each region as super-polluting units. The distributions of different mercury species, including Hg2+, Hgp, and gaseous elemental mercury (Hg0), are shown in Supplementary Fig. 7. Regional mercury emissions during 1990–2020 are summarized in Supplementary Table 2. Large disproportionalities in mercury emissions and generating capacity can be observed among regions, especially for units with high mercury emission intensities. Across all the regions, super-polluting units accounted for only a small fraction of total power-generating capacity, ranging from as little as 0.4% in India in 2010 to at most 6.6% in China in 1990, and the majority of super-polluting units belonged to the smallest and small (50–100 MW) capacity categories.Fig. 2Cumulative and annual mercury emissions of coal-fired power units across capacity categories.a, c, e, g, i, k show the cumulative distributions of unit-level generating capacity and mercury emissions for China, India, Europe, the U.S., rest of Asia and rest of world, respectively. Within each row, the four left-hand panels represent 1990, 2000, 2010 and 2020 from left to right. Units are sorted by mercury emission intensity, with the highest-emitting units placed on the left. The dashed red lines indicate the share of generating capacity contributing to 10% of cumulative mercury emissions. b, d, f, h, j, l show annual regional mercury emissions by capacity category from 1990 to 2020 for the corresponding regions. Colors denote capacity categories. Cumulative and annual mercury emissions of coal-fired power units across capacity categories. a, c, e, g, i, k show the cumulative distributions of unit-level generating capacity and mercury emissions for China, India, Europe, the U.S., rest of Asia and rest of world, respectively. Within each row, the four left-hand panels represent 1990, 2000, 2010 and 2020 from left to right. Units are sorted by mercury emission intensity, with the highest-emitting units placed on the left. The dashed red lines indicate the share of generating capacity contributing to 10% of cumulative mercury emissions. b, d, f, h, j, l show annual regional mercury emissions by capacity category from 1990 to 2020 for the corresponding regions. Colors denote capacity categories. These super-polluting units have become more concentrated in a smaller proportion of capacity from 1990 to 2020, and this shift was largely attributable to the commissioning of newly built units, which were generally larger and cleaner, as well as the prioritization of pollution control upgrades for relatively larger facilities. For example, in the rest of Asia, the super-polluting units accounted for 4.8% of total capacity in 1990, but only 2.2% in 2020 (Fig. 2i). Similar patterns were observed in other regions, with decreases in capacity shares ranging from 0.4% to 5.5%. More details for the rest of Asia, the rest of the world and Europe are included in Supplementary Figs. 8–10. Such disproportionalities highlight the importance of a small subset of super-polluting units in reducing mercury emissions. Emission patterns from coal-fired units of different capacity categories varied significantly across regions. Despite fluctuations observed in specific periods across different regions, mercury emissions from units with capacity less than 100 MW generally decreased from 1990 to 2020 globally, with the notable exception of India and the rest of Asia. In contrast, emissions from units larger than 300 MW in most regions increased from 1990 to 2010, followed by divergent trends between 2010 and 2020. Emissions declined sharply in China and the rest of the world, while continuing to rise in India and the rest of Asia. These emission increases in emerging regions indicate that newly built large units, not just smaller or older units, should also be considered for future deep emission mitigation. In comparison, mercury emissions from different-sized units in the U.S. and Europe remained on a downward trend throughout the 1990–2020 period. Specifically, in the U.S., emissions from different-sized units have decreased by 85.9–96.7%, with the mid-sized units (100–300 MW) experiencing the largest reduction of 96.7%, dropping from 8.9 tonnes in 1990 to 0.3 tonnes in 2020. Regional disparities in plant-level emission intensity The geographical distribution of global coal-fired plants is illustrated in Fig. 3 and Supplementary Fig. 11. Plant-level mercury emission intensity was calculated by aggregating emissions and power generation across all units within each plant, and therefore differs from the unit-level metric. Overall, plant-level mercury emission intensity exhibited a consistent decline from 1990 to 2020, primarily driven by the deployment of cleaner, large-scale plants and the implementation of advanced pollution control technologies. These developments were particularly impactful in regions such as China, the U.S., and the rest of Asia. Over time, the number of plants with high emission intensities decreased, and their spatial distribution became increasingly concentrated. For example, in 1990, 36.4% of the plants (839 out of 2306) exceeded the global average emission intensity of 0.067 kg GWh−1, with the majority located in China (343 plants). By 2020, this proportion had dropped to 32.8% (1089 out of 3323), with the geographical hotspot shifting toward India, which alone hosted half of such plants. Moreover, relatively high-emission-intensity plants were predominantly located in India in 2020, which hosted over 90% of all identified super-polluting units (Supplementary Fig. 12). The rapid expansion of coal-fired power capacity, often in the absence of advanced APCDs, contributed to persistently high emission intensities in India and Central Asia (Fig. 3b). Meanwhile, Eastern Europe, where technological upgrades lagged behind during 1990–2020, continued to exhibit relatively high plant-level emission intensity, making it another focal region in 2020. Notably, despite the low European average emission intensity among units with capacity above 100 MW in 2020, <0.035 kg GWh−1, small-scale plants with high emission intensity persisted in Europe, especially Eastern Europe, warranting continued attention.Fig. 3Plant-level and regional mercury emission intensity across generating capacity categories.a, b show emission intensity of individual coal-fired plants in 1990 and 2020, respectively. The size of each point represents the installed capacity of the corresponding plant, and the color indicates its emission intensity. The number of operating coal-fired plants in each year is also shown. c shows regional mercury emission intensity for each capacity category. Maps of mercury emission intensity in 2000 and 2010 are shown in Supplementary Fig. 11. Map boundaries are based on standard maps provided by the Standard Map Service System of the Ministry of Natural Resources of the People’s Republic of China, and the official boundaries were not altered. Plant-level and regional mercury emission intensity across generating capacity categories. a, b show emission intensity of individual coal-fired plants in 1990 and 2020, respectively. The size of each point represents the installed capacity of the corresponding plant, and the color indicates its emission intensity. The number of operating coal-fired plants in each year is also shown. c shows regional mercury emission intensity for each capacity category. Maps of mercury emission intensity in 2000 and 2010 are shown in Supplementary Fig. 11. Map boundaries are based on standard maps provided by the Standard Map Service System of the Ministry of Natural Resources of the People’s Republic of China, and the official boundaries were not altered. Overall, mercury emission intensities generally declined as unit capacity increased across nearly all regions, as newer and larger units were generally equipped with more efficient end-of-pipe control technologies (Fig. 3c). For example, the emission intensities of units larger than 600 MW in Europe, 0.015 kg GWh−1 in1990 and 0.007 kg GWh−1 in 2020, accounted for <20% of those of units smaller than 50 MW (0.087 kg GWh−1 in 1990 and 0.071 kg GWh−1 in 2020). In particular, the mean emission intensities of units larger than 100 MW in the U.S. approached zero in 2020, whereas units smaller than 50 MW still exhibited an emission intensity of 0.027 kg GWh−1. Most other regions exhibited similar patterns. From a cross-regional perspective, the progress of mercury emission reductions during 1990–2020 differed substantially among regions, reflecting varying levels of pollution-control technology upgrades. China experienced the most significant decline in emission intensity, decreasing from 0.064–0.126 kg GWh−1 in 1990 to 0.003–0.054 kg GWh−1 in 2020 across different capacity categories, reaching or even surpassing the levels of European units within the same capacity categories. This suggests that rapid pollution-control technology upgrades enabled a substantial technology catch-up effect. In contrast, emission intensities in India showed relatively limited improvement across all capacity categories and remained comparatively high in 2020 (0.043–0.106 kg GWh−1). In addition, units smaller than 100 MW in Europe and the rest of the world showed limited declines in emission intensity and still maintained relatively high levels ( ≥ 0.067 k g GWh−1), suggesting that older small-scale units may remain a key challenge for future global mercury mitigation efforts. Unit-level drivers of mercury emission changes We further estimated the key factors driving unit-level mercury emission changes, including activity rate, pollution control technology, retired capacity, and newly-built capacity, as shown in Fig. 4, and the regional contributions from each driving factor are summarized in Supplementary Fig. 13. For surviving units, the influence of activity rate varied by regions and capacity categories, but the median unit-level contributions to emission changes were mostly close to zero, with limited volatility. Extreme emission changes from activity rate shifts were rare, except in specific capacity categories and regions, such as India. Regarding pollution control technologies, the major unit-level contributors to emission reductions shifted from Europe and the U.S. to other regions, with the exception of India (Fig. 4d–f). For example, more than 80% of the total emission reductions from APCD upgrades during 1990–2000 occurred in Europe and the U.S., amounting to emission reductions of 22.2 tonnes (Supplementary Fig. 13). In contrast, during 2010–2020, coal-fired power units in China, rest of world and rest of Asia emerged as the primary contributors. In particular, most units in China individually reduced emissions by only ~50 kg, contributing to a cumulative reduction of 37.1 tonnes.Fig. 4Decomposition of changes in unit-level mercury emissions across regions.Contributions are shown for changes in activity rate (a–c), upgrades of pollution control technologies (d–f), retirement of coal-fired units (g–i), and newly built coal-fired units (j–l). Columns represent capacity categories of 0–100 MW, 100–300 MW, and ≥300 MW, respectively. Each point represents the mercury emission change of an individual coal-fired unit resulting from a given driving factor during a corresponding study period, and colors indicate the six regions. Within each region, boxplots summarize the distributions for 1990–2000, 2000–2010 and 2010–2020 from left to right. Global and regional summaries of the unit-level decomposition results are provided in Supplementary Fig. 13. Decomposition of changes in unit-level mercury emissions across regions. Contributions are shown for changes in activity rate (a–c), upgrades of pollution control technologies (d–f), retirement of coal-fired units (g–i), and newly built coal-fired units (j–l). Columns represent capacity categories of 0–100 MW, 100–300 MW, and ≥300 MW, respectively. Each point represents the mercury emission change of an individual coal-fired unit resulting from a given driving factor during a corresponding study period, and colors indicate the six regions. Within each region, boxplots summarize the distributions for 1990–2000, 2000–2010 and 2010–2020 from left to right. Global and regional summaries of the unit-level decomposition results are provided in Supplementary Fig. 13. The contributions of both newly built and retired units to emission changes were substantial throughout the study period and were most pronounced during 2000–2010, with newly built capacity mainly concentrated in China, India and rest of Asia. Meanwhile retired capacity was primarily from Europe and the U.S. during 1990–2000 and 2010–2020, while China led the emission reduction from unit retirements during 2000–2010. For example, China and India were together responsible for 70.8% (1990–2000), 88.8% (2000–2010), and 78.8% (2010–2020) of increasing emissions from newly built units globally, while Europe and the U.S. only contributed to 4.7%, 1.3% and 0.9% of the associated emission increases. Although the dominant newly-built units gradually shifted from small- and medium-sized to large-sized units (Supplementary Fig. 14), their contributions to mercury emission increases were highly disproportionate. For example, during 2010–2020, units smaller than 100 MW contributed 20.9 tonnes of additional mercury emissions from newly-built capacity, whereas units larger than 300 MW contributed only 5.0 tonnes (Supplementary Fig. 13), despite the fact that 189 large units and only 114 small units were commissioned. This indicates that in some countries, some newly built small units may still be highly polluting, and emission intensity rather than unit count dominates their contribution to overall emissions. In addition, in China—the largest contributor—the main driver was the rapid expansion of installed capacity, while the widespread adoption of advanced pollution control technologies helped limit the associated increase in mercury emissions. In contrast, in other emerging regions, the main contributor to increased emissions was the high emission intensity of newly built units, especially in India (Fig. 1). Regarding the impact of retired capacity, retirement activities in the U.S. and Europe were initially dominated by units smaller than 100 MW between 1990 and 2020, although emission reduction from units larger than 300 MW increased, suggesting a gradual shift of retirement efforts toward large units. However, each retired unit typically yielded limited emission reductions, due to their relatively low unit-level emission intensity and limited coal consumption. For example, the average emission reductions from retired units in the two regions were less than 15 kg per year, together contributing to 22.3 tonnes during 1990–2020. In contrast, although fewer units were retired in India, the rest of Asia and the rest of the world, each retired unit often contributed substantially to emission reductions, in some cases reaching up to 200 kg per year of mercury emission reduction. This pronounced effect is primarily attributed to the poor end-of-pipe control performance of those retired units. Notably, some small units in emerging regions made disproportionately significant contributions to emission changes. For example, in India and the rest of Asia, some newly built small- and medium-sized units exhibited emission increases comparable to, and in some cases even greater than, those of large units (Fig. 4j–l). Additionally, in China, the emission reductions achieved through pollution control technology upgrades in a few small units individually exceeded the median contributions of most large units (red points in Fig. 4d–f). Such regional shifts and substantial disparities among individual units underscore the critical role that certain units play in offsetting broader emission reduction efforts. Customized and phased deep emission mitigation options Based on the aggregation of unit-level mercury emissions from 2020 to 2060 under scenario groups representing different emission mitigation pathways, the regional mitigation strategies under SSP1–2.6 and SSP1–1.9 are depicted in Fig. 5. A detailed cost-efficiency analysis is provided in Supplementary Discussion. Globally, different strategy combinations result in varying regional emission trajectories. Specifically, the short-term performance-based retirement policy demonstrates effectiveness across most countries and delivers sustained emission reductions throughout the study period, yielding cumulative emission reductions ranging from 18.5 (from 39.4 tonnes to 20.9 tonnes in the U.S. under SSP1–1.9) to 253.8 tonnes (from 3032.3 tonnes to 2778.5 tonnes in India under SSP1–2.6). For example, shifting the retirement policy from historical to performance-based retirement could cut cumulative U.S. emissions by ~50%, corresponding to 18.5 tonnes (about seven times the 2020 emissions) under SSP1–1.9 and 29.6 tonnes (~12 times the 2020 emissions) under SSP1–2.6. However, the implementation of strategic retirement policies in Europe and the U.S. entails high capital costs (Supplementary Discussion), due to the expensive coal power infrastructure. Despite substantial absolute emission reductions, the strategic retirement policy displays relatively limited mitigation benefits in emerging regions, with cumulative reduction rates generally below 20%0 relative to the no-retirement baseline.Fig. 5Projected evolution of regional mercury emissions during 2020–2060 under various integrated mitigation options.a–f show China, India, Europe, the U.S., rest of Asia and rest of world, respectively. Colored lines show annual mercury emissions under different mitigation scenarios, with blue and red indicating SSP1–1.9 and SSP1–2.6 pathways, respectively. The line shades represent different mitigation options. Bars show cumulative mercury emissions during 2020–2060 for each scenario, with numbers indicating the corresponding cumulative emissions. Projected evolution of regional mercury emissions during 2020–2060 under various integrated mitigation options. a–f show China, India, Europe, the U.S., rest of Asia and rest of world, respectively. Colored lines show annual mercury emissions under different mitigation scenarios, with blue and red indicating SSP1–1.9 and SSP1–2.6 pathways, respectively. The line shades represent different mitigation options. Bars show cumulative mercury emissions during 2020–2060 for each scenario, with numbers indicating the corresponding cumulative emissions. The co-reduction from APCDs upgrading is particularly significant for countries with large coal-fired power fleets. For example, under the Best Available Technology (BAT) scenario, 68.2% (from 2778.5 to 884.5 tonnes) and 58.7% (from 1772.5 to 711 tonnes) of the future cumulative emissions can be mitigated in India under SSP1–2.6 and SSP1–1.9, respectively, because over 80% of India’s units are currently not equipped with regional BAT. In contrast, the BAT policy shows limited effect in the U.S., where most inefficient units with low mercury removal efficiency are expected to be retired within the first five years. Notably, with the implementation of regional BAT, cumulative emissions in Europe decrease by 42.4% and 40.4%, corresponding to total reductions of 36.8 tonnes under SSP1–2.6 and 27.1 tonnes under SSP1–1.9, respectively. While retrofitting APCDs to BAT in Europe typically costs more than strategic retirement, this may potentially restrict the deployment of BAT technologies. The mitigation potential of these measures will be exhausted once all the units are equipped with BAT or the inefficient units are fully retired. Given the sustained high demand for coal-fired power generation, particularly in India and China under SSP1–2.6, mercury-specific technologies, such as ACI, will be crucial for achieving near-zero mercury emissions in the long term. For example, without ACI, the annual mercury emissions under the BAT scenario are projected to remain at 2 tonnes in India and 3.9 tonnes in China even in 2060 under SSP1–2.6—approximately 80% and 160% of the U.S. emissions in 2020, respectively. Therefore, under SSP1–2.6, applying BAT-ACI to all suitable units in India and China would reduce annual mercury emissions to 1.0 and 0.5 tonnes by 2060, respectively, and relative to the BAT scenario, the additional deployment of ACI would further reduce cumulative mercury emissions by 29% in China and 5% in India. However, reduction potentials from further application of ACI in the U.S. and Europe are limited, due to the already high ACI penetration ( ~ 50% in the U.S. by 2020) and the imminent retirement of the remaining units without ACI. Thus, their long-term mercury reduction should rely primarily on accelerating coal-fired unit retirements. In summary, widespread adoption of BAT in the short term, coupled with expanded ACI deployment in the long term, is essential for sustained mercury emission reductions—particularly in emerging economies, although the associated costs remain a significant challenge. Comparisons across climate targets further indicate that more stringent mitigation pathways can deliver additional mercury emission reductions. Across these regions, cumulative emissions under SSP1–1.9 are lower than those under SSP1–2.6, with the difference being particularly pronounced in India and emerging countries in the rest of Asia. Specifically, under the BAT-ACI pathway, cumulative emissions in India decrease from 839.9 to 696.1 tonnes, while those in the rest of Asia decline from 255.0 to 220.5 tonnes, representing further reductions of 17.1% and 13.5%, respectively, when moving from SSP1–2.6 to SSP1–1.9. These findings further demonstrate the substantial co-benefits of power-sector transformation for mercury emission reductions in the coal-fired power sector. Discussion Our systematic and dynamic estimate of mercury emissions from global coal-fired power units reveals that mercury emissions are increasingly concentrated in a small fraction of coal-fired power units. Although upgraded APCDs of some power-generating units and retirements of small- and mid-sized units have successfully reduced mercury emissions, this progress was offset by the construction of large newly built units in emerging regions such as China, India and the rest of Asia. Going forward, in regions where even conventional pollution control technologies are currently lagging behind, cost-effective approaches include the strategic retirement of older and high-emitting units, and the widespread deployment of BAT. Achieving near-zero mercury emissions in these areas, including China, will heavily depend on the large-scale implementation of mercury-specific technologies such as ACI, and the eventual phase-out of coal-fired power generation. Conventional desulfurization, denitrification, and dust removal devices remain the primary APCDs in most coal-fired units, reflecting the continued prioritization of traditional air pollutants in environmental management. However, the widespread deployment of mercury-specific technologies is often prohibitively expensive and may cause additional side effects in most countries in the short term, especially in countries where young and high-mercury emission intensity units are located7,40. By exploring the easily accessible mitigation opportunities and their cost-effectiveness, such as carbon mitigation policies, pre-combustion treatments, strategic retirements, and upgrades to existing APCDs, our evaluation might help prioritize the most readily achievable actions for mitigating multiple air pollutants41–43. In the long term, achieving near-zero mercury emissions in response to the Minamata Convention on Mercury might become inevitable in regions with high demand for coal-fired power under ambitious climate scenarios like SSP1–2.6 or SSP1–1.9, such as India and China (Supplementary Fig. 15). Thus, highlighting the mitigation opportunities for the application of mercury-specific technologies is of critical importance. Additionally, with high-resolution geospatial and temporal data on potential mercury emissions, our evaluation can better support model analyses on the potential impacts on human health42,44. However, some limitations and uncertainties remain, even though the estimated mercury emissions in our study show relatively good consistency with those of other studies (Supplementary Fig. 16), and our study provides a more comprehensive and accurate assessment of unit-level historical and future emissions (Supplementary Tables 3–4). First, in addition to the uncertainties from the Global Power Emissions Database (GPED) discussed in a previous study45, the lack of some regional measured data on the real-world mercury removal efficiencies of air pollution control technologies, as well as the mercury contents in consumed coal, may introduce additional uncertainty35. Varying proportions of coal are produced and consumed domestically within the country. This coal may undergo pre-combustion processes such as coal washing, but we cannot evaluate their emission reduction effects due to insufficient data. Unit-level mercury emission measurements remain scarce, especially in smaller regions, hampering our ability to cross-check and validate our estimates. Efforts are underway to obtain more measured data for comparison and quality improvement. Second, the trajectory of future emissions depicted in our scenarios reflects an assumption that existing power-generating units are strategically targeted for retirement and replacement with newer, more efficient, and lower-emitting ones. However, this trajectory is uncertain46, as some countries, such as the U.S. and Germany, have already halted the construction of new coal-fired units, while existing units in these countries often exceed 40 years of age45, serve as a source of flexible power supply. Third, under the scenario of application of mercury-specific technologies, we only considered ACI while omitting other potentially more cost-effective mercury-specific technologies. Potential alternatives include halogen injection (HI), which may be more cost-effective than ACI for some coal types and plant configurations. Although this choice was based on the current maturity of available technologies, future technological advancements may render these options viable. Moreover, our cost–benefit analysis of ACI did not include waste handling and disposal challenges, which may be particularly problematic in poor countries and could significantly hinder the widespread adoption of ACI in those regions. In addition, large-scale deployment of ACI may involve additional financial burdens and technical barriers, particularly in regions with limited pollution-control infrastructure. Furthermore, long-term reliance on ACI could potentially create a technology lock-in effect if not accompanied by broader transitions toward cleaner energy systems. Fourth, the resolution of Integrated Assessment Models (IAMs) is relatively coarse compared to our unit-level emission projection model, and the base years of these IAMs may greatly differ from current development. Only one model, the Global Change Analysis Model (GCAM), was used in this study, as different IAMs have various projections of future coal power demand47. These discrepancies introduce uncertainties in coal consumption projections and, by extension, mercury emission estimates. Fifth, future capital costs of CFPPs are likely to vary with technological advancements and inflation indices. Our cost-effectiveness analysis may be affected by uncertainties in these costs, which differ depending on commissioning time and boiler technologies. For instance, ultra-supercritical units generally entail higher capital costs than conventional pulverized coal units48–50. To assess the broader applicability of our findings, we conducted a case study on Indonesia, a representative smaller Asian country, which demonstrates consistency between our model results and official regional data51 (Supplementary Discussions). Regardless, our unit-level and regional assessment of historical mercury emissions and mitigation opportunities offer critical insights for the effective implementation of the Minamata Convention on Mercury within the coal-fired power sector. In addition, our findings support the co-governance of climate change and conventional air pollutants, including mercury, which is particularly vital for regions that require synergistic reductions in both carbon emissions and air pollution, where decision-makers must navigate resource constraints and economic trade-offs to prioritize mitigation investments effectively. Methods Estimates of unit-level mercury emissions We estimate mercury emissions from global CFPPs on the basis of a production-based emissions accounting methodology. The calculation is expressed as Eq. (1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{i,r}={\sum}_{s}{A}_{i,r} \times {C}_{r,i} \times R \times {S}_{s} \times {\prod}_{e}\left(1-{\theta }_{e,i,r,s}\right)$$\end{document}Ei,r=∑sAi,r×Cr,i×R×Ss×∏e1−θe,i,r,swhere i, r, s and e represent the coal-fired unit, the country where the coal-fired unit is located, mercury species (including Hg0, Hg2+, Hgp) and air pollution control technologies, including de-NOx, de-SO2, and de-PM technologies, and mercury-specific technologies (ACI in this study), respectively. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{i,r}$$\end{document}Ei,r is the mercury emissions of coal-fired unit i located in region r, kg yr−1; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{i,r}$$\end{document}Ai,r is the annual coal consumption of coal-fired unit i located in region r, Mt yr−1, including lignite, sub-bituminous coal, other bituminous coal, and anthracite, while auxiliary fuel (e.g., natural gas or petroleum coke) consumption at these power plants are not considered; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{r,i}$$\end{document}Cr,i is the mercury content in coal used in coal-fired unit i in country r, g tonne−1; R is the release rate of coal-fired boilers, including pulverized coal boiler (PC) and circulating fluidized bed boiler (CFB), %; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${S}_{s}$$\end{document}Ss is the average proportion of mercury species across different APCDs, %, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\theta }_{e,i,r,s}$$\end{document}θe,i,r,s are the removal efficiencies of APCDs, %. The unit-level emission factor is estimated as Eq. (2).2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{EF}}}}}_{i}={\sum}_{s}{C}_{r} \times R\times {S}_{s}\times {\prod }_{e}(1-{\theta }_{e,i,r,s})$$\end{document}EFi=∑sCr×R×Ss×∏e(1−θe,i,r,s) The annual coal consumption and APCDs of each coal-fired power unit are derived from GPED, initially developed in previous work (GPED v1.0), and updated in our study (GPED v2.0)52. GPED is a global thermal power unit database that covers 89,058 units across 229 countries worldwide (see Supplementary Figs. 17 and 18, two classification schemes were used: a geographical classification consisting of China, India, the U.S., Europe, rest of Asia, and rest of world; and an income-based classification distinguishing developed and developing countries), covering a total of 13,857 coal-fired units from 1990 to 2020. This database offers the necessary information, including coal type, installed capacity, activity information (e.g., coal consumption and types, power generation, utilization hours), and position information (e.g., longitude and latitude). The end-of-pipe control technology installation information (including de-NOx, de-SO2 and de-PM technologies) for each coal-fired unit, and the end-of-pipe control technology name correspondence table between the databases and this study can be found in Supplementary Table 5. The plant-level installation information of ACI in the U.S. comes from the U.S. Environmental Protection Agency (EPA), available at: https://www.epa.gov/power-sector-modeling/national-electric-energy-data-system-needs. Mercury content in coal Mercury emissions from coal-fired units largely depend on coal quality, mainly the coal type and the mercury content of coal12. The coal type used by each unit is available from the GPED v2.0. The original data are derived from the global power plant database (available at: https://old-datasets.wri.org/), and then subsequently corrected based on the national consumption of different coal types. For the sources of mercury content in coal, we preferentially used measurements representing coal actually consumed in CFPPs, rather than raw coal or mine-mouth coal, because these values already implicitly reflect coal preparation before combustion, such as coal washing, although such preparation practices vary dramatically in different countries owing to the different geological coal-forming environments and coal mines53. Detailed sub-national data were preferentially used where available. For China, province-level mercury contents of coal consumed in CFPPs were adopted10,21,24. For the U.S., although the sub-regional mercury content in raw coal in the U.S. can be found from the U.S. Geological Survey Coal Quality Database (COALQUAL)54, the U.S. EPA Information Collection Request database (ICR) lists the mercury content of samples collected from coal shipments delivered to US electric utilities. Thus, these data are used for the consumed coal of the U.S. coal power sector54,55. We compared these data with the data we are using and found that the mercury content in our data fits well with them (see Supplementary Fig. 19). For countries where multiple measurements of coal consumed in CFPPs were available, such as India, we used the national average mercury contents derived from these CFPP-specific measurements for each coal type. For countries lacking sufficient CFPP-specific data, national mercury contents for each coal type were estimated using the best available information from the U.S. Geological Survey (USGS), ICR, default emission factors recommended by the United Nations Environment Programme (UNEP)56, the U.S. EPA Compilation of Air Pollutant Emission Factors (AP-42), the European Monitoring and Evaluation Programme (EMEP) air pollutant emission inventory guidebook, and local studies29,57–59. The detailed information and data sources can be found in Supplementary Tables 6–8, and the mercury content used in this study is presented in Supplementary Table 9. Moreover, detailed rationale and selection procedures for the mercury contents in coal consumed by CFPPs are provided in Supplementary Methods. In addition, imported coal and domestically sourced coal were treated separately. For each country and year, the mercury content of imported coal was estimated as a weighted average based on the import shares of coal from different origin countries reported in UN Comtrade during 1990–2020, while domestic coal was assigned the mercury content corresponding to coal consumed in CFPPs within that country. Specifically, imported coal was estimated based on national import statistics under HS code 2701 (Coal), including anthracite (2701.11), bituminous coal (2701.12), and other coal categories (2701.19), from UN Comtrade (https://comtradeplus.un.org/). Because detailed end-use information on how imported coal is allocated between CFPPs and other sectors (e.g., industrial boilers) is unavailable, we assumed that the composition of imported coal used in CFPPs follows the overall national import structure. This assumption may introduce uncertainty in countries where imported coal is preferentially used outside the power sector. Thus, we conducted a sensitivity analysis comparing the estimated emissions obtained using (1) differentiated imported and domestic coal mercury contents and (2) only domestically sourced coal mercury content. Release ratio Researchers have summarized the mercury release ratio from several studies and suggest that 1–2% of Hg will remain in coal bottom ash12,21,53, and it differs among boiler types. Here, we assumed that the mercury release ratio of coal (R) was 99.4% for PC and 98.9% for CFB21, when coal is burned in power generating units and before passing through the APCDs. It is worth noting that mercury associated with fly ash is primarily removed during particulate control processes and is therefore considered within the APCD removal efficiencies60. Mercury species Mercury may be present in flue gas as elemental mercury vapor (Hg0), as a vapor of an oxidized mercury species (Hg2+), or as particulate-bound mercury (Hgp)19,61. The distribution of these species varies depending on flue gas temperature and the technical configuration of APCDs11,21. Detailed mercury speciation profiles under different conditions can be found in Supplementary Table 10. Mercury removal efficiency Large quantities of mercury can be removed by air pollution control technologies for SO2 emissions (de-SO2) and air pollution control technologies for particulate matter (de-PM)62,63. De-PM are classified into combination precipitators (COM, usually ESP-FF, FF/ESP-FF + WESP), FF, ESP and others (e.g., cyclone, wet dust removal or none). De-NOx technologies, such as SCR and selective non-catalytic reduction (SNCR), cannot remove mercury directly, but their application could improve the mercury removal efficiency of WFGD, with Hg0 being oxidized to Hg2+, which is more easily captured by de-SO2 and de-PM technologies62,64, thereby improving the mercury removal efficiency by ~20% (7–54%) in existing studies12,65. The application information of the de-SO2 and de-PM technologies of each unit is provided in the GPED v2.0. The mercury removal efficiencies of conventional end-of-pipe control technologies derived from previous studies12,53,58,66 are listed in Supplementary Table 11. The mercury-specific technologies, such as ACI, were considered in our estimation with the same removal efficiency as in the U.S.6, generally combined with SCR + WFGD + FF. The removal efficiency of mercury from high to low is usually ACI > S + COM > S + FF > S + ESP > FF > ESP > others, and the detailed mercury removal efficiencies of conventional end-of-pipe control technologies for specific countries used in this study can be found in Supplementary Table 12. Uncertainty analysis We performed uncertainty analysis for the emission estimates at the regional level based on a Monte Carlo approach described in previous studies12,18. The uncertainty range was calculated by the 80% confidence interval (CI) around the central estimate from 10,000 Monte Carlo tests with specific probability distributions of input parameters including activity rates (coal consumption) and emission factors. The probability distributions of the parameters were obtained from previous studies12,24,67. The probability distributions of some parameters also differed, as listed in Supplementary Table 13. The emission uncertainty of regional mercury emissions is shown in Supplementary Fig. 20. Driver decomposition We used the Logarithmic Mean Divisia Index to identify how each driving factor contributed to the mercury emission change, because of its path independence, consistency in aggregation, and ability to handle zero values68. The drivers were classified as activity rate, pollution control technology, and installed capacity changes (including newly built capacity and retired capacity). For units that were newly built or retired during a particular time interval, there was only one contributor driving the emission changes, namely newly-built capacity and retired capacity. For the surviving units during a time interval, the emission changes came from the contributions of pollution control technology and activity rate (mainly from operational hours), as the capacity, mercury content, and mercury release ratio were fixed parameters in our work, and energy efficiency could be improved only by reducing the self-consumption rate, which was generally too small to have a meaningful effect. The components are expressed as Eqs. (3)–(7).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{i,r}={A}_{i,r}\times {{{{\rm{EF}}}}}_{i,r}$$\end{document}Ei,r=Ai,r×EFi,r4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {E}_{i,r}={E}_{i,r}^{t+1}-{E}_{i,r}^{t}=\Delta {E}_{{{{\rm{A}}}},i,r}+\Delta {E}_{{{{\rm{EF}}}},i,r}$$\end{document}ΔEi,r=Ei,rt+1−Ei,rt=ΔEA,i,r+ΔEEF,i,r5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {E}_{{{{\rm{A}}}},i,r}=L({E}_{i,r}^{t+1},{E}_{i,r}^{t})\times {{\mathrm{ln}}}\left(\frac{{A}_{i,r}^{t+1}}{{A}_{i,r}^{t}}\right)$$\end{document}ΔEA,i,r=L(Ei,rt+1,Ei,rt)×lnAi,rt+1Ai,rt6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {E}_{{{{\rm{EF}}}},i,r}=L({E}_{i,r}^{t+1},{E}_{i,r}^{t})\times {{\mathrm{ln}}}\left(\frac{{{{{\rm{EF}}}}}_{i,r}^{t+1}}{{{{{\rm{EF}}}}}_{i,r}^{t}}\right)$$\end{document}ΔEEF,i,r=L(Ei,rt+1,Ei,rt)×lnEFi,rt+1EFi,rt7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L({E}_{i,r}^{t+1},{E}_{i,r}^{t})=\frac{{E}_{i,r}^{t+1}-{E}_{i,r}^{t}}{{{\mathrm{ln}}}{E}_{i,r}^{t+1}-{{\mathrm{ln}}}{E}_{i,r}^{t}}$$\end{document}L(Ei,rt+1,Ei,rt)=Ei,rt+1−Ei,rtlnEi,rt+1−lnEi,rtwhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{i,r}$$\end{document}Ei,r is the mercury emissions of coal-fired unit ii in region \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${r}$$\end{document}r (kg yr-1); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{i,r}$$\end{document}Ai,r is the annual coal consumption of unit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${i}$$\end{document}i in region \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${r}$$\end{document}r (Mt yr-1); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E{F}_{i,{r}}$$\end{document}EFi,r is the mercury emission factor of unit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${i}$$\end{document}i in region \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${r}$$\end{document}r (g tonne-1); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {E}_{i,r}$$\end{document}ΔEi,r is the change in mercury emissions between year tt and year t+1t+1; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {E}_{{{{\rm{A}}}},i,r}$$\end{document}ΔEA,i,r and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {E}_{{{{\rm{EF}}}},i,r}$$\end{document}ΔEEF,i,r are the contributions of changes in activity rate and emission factor, respectively, to the total mercury emission change; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L({E}_{i,r}^{t+1},{E}_{i,r}^{t})$$\end{document}L(Ei,rt+1,Ei,rt) is the logarithmic mean weight; tt and t+1t+1 represent two consecutive years; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${i}$$\end{document}i denotes an individual coal-fired unit; and rr denotes the region in which the unit is located. Proposed scenarios for mercury emission mitigation Based on the parameters driving emission changes and mercury emission inventories, we evaluate the potential evolution of mercury emissions for each unit in each region by assuming that the mercury content of coal, the mercury removal efficiency of pollution control technologies, the energy efficiency, and the number of operational hours of each unit remain equivalent to the levels observed in 2020. Three key driving factors were used to simulate mercury emissions by tracking the evolution of each coal-fired unit: carbon mitigation scenarios, retirement strategies, and upgrading of air pollution control technologies. A brief summary of these scenarios can be found in Supplementary Table 14. Climate change We derived future coal-fired electricity demand under two climate mitigation scenarios from GCAM based on our previous studies45,69. To minimize the influence of socioeconomic uncertainty, we adopted the sustainable development pathway SSP1 and combined it with two radiative forcing targets: SSP1–2.6 and SSP1–1.9, corresponding to radiative forcing levels of 2.6 and 1.9 W m−2 by 2100, respectively. SSP1–2.6 represents a moderate mitigation pathway broadly consistent with limiting warming to ~2 °C, whereas SSP1–1.9 reflects a more stringent 1.5 °C pathway with a faster transition away from coal and more rapid deployment of low-carbon energy. Under these scenarios, GCAM directly provides regional coal-fired electricity demand at 10-year intervals from 2020 to 2060. The projected retirement and construction of CFPPs in this study are therefore determined within an energy-transition pathway that already reflects future changes in national power-generation structures. Most retired coal-fired capacity is therefore effectively replaced by the expansion of non-coal electricity sources, particularly renewable energy, represented in the GCAM scenarios. Only when the remaining coal-fired units are insufficient to meet the GCAM-projected coal power demand do we assume the construction of new coal-fired units. Nevertheless, coal-fired plants are still projected to remain part of the electricity system during the transition period, particularly in emerging countries where many existing units are relatively young and where immediate replacement by renewables may involve substantial costs associated with premature retirement and maintaining power-system reliability and flexibility. This study therefore focuses on mitigation strategies for the coal-fired power units that remain in operation during the transition. To ensure consistency with historical observations, we proportionally adjusted the GCAM projections because the modeled coal-fired electricity generation in 2020 differed from the observed historical values. The future coal-power demand from 2020 to 2060 under SSP1–1.9 and SSP1–2.6 is shown in Supplementary Fig. 21. Retirement policies Various retirement strategies could alter the coal power fleet structure and subsequently influence the emission characteristics of operating units in the future. We defined two options for future retirement strategies: The first option assumed that all units would be retired with a 40-year lifetime. Statistical results indicated that coal-fired units generally operated for ∼40 years globally70, which reflects the decision to retire a unit or power plant with the economic consideration of operating costs, replacement costs, and revenues (referred to as historical retirement)45,71. Overaged units older than 40 years that were built before 1980 would be randomly retired between 2020 and 2030. The second option assumed that the super-polluting units (always small, old and inefficient coal-fired units) would be retired early; in other words, units with higher emission intensity would be retired early (referred to as performance-based retirement). Historical retirement enabled coal-fired units to operate for their historically expected lifetime before being replaced (40 years). Performance-based retirement prioritized the retirement of polluting units and slightly reduced the global average lifetime (being ~33 years), as shown in Supplementary Fig. 22. All existing units are assumed to maintain their current power generation efficiency and annual operating hours, indicating that the emission factors and activity rates for each unit will remain at their 2020 levels in the future. Moreover, the mismatch between coal power demand and supply from existing units is projected to be met by newly built coal-fired plants, which is assumed to be met primarily by units equipped with the most advanced pollution control technologies available in each region. As a result, the emission intensity of new capacity is expected to be the lowest within its respective region. Pollution control technology The efficiency of air pollution control technologies determined the level of mercury emissions. Given that many operating units, especially in emerging regions except for China, lacked efficient pollution control technologies by 2020 and that there are some units with low co-reductions of mercury emissions, we design three pollution control technology scenarios. The first option assumes that pollution control technologies would follow legislation and maintain their 2020 levels; the mercury removal rate of all in-fleet units would remain at their 2020 levels by 2060, and newly built units to fill the coal-fired power supply gap will be built with pollution control technologies whose removal efficiency equals the highest removal efficiency of in-fleet units (Business as Usual, BAU). The second option assumes that the BATs of the region where the unit is located would be applied to coal-fired units before 2025 in the U.S., 2030 in Europe and China, or 2040 in other regions71. The third option assumes that all the operating coal-fired units will gradually retrofit with mercury-specific technologies, namely, ACI. Under this scenario, coal-fired units are required to be equipped with BAT technologies before considering the application of ACI, as ACI is relatively more costly and may have more side effects. This scenario is referred to as BAT–ACI. Because only the universal use of BAT alone cannot completely mitigate mercury emissions, given that mercury is non-degradable, bio-accumulative, and widely dispersed in the environment, combining BAT with ACI to further reduce mercury emissions is necessary72. Moreover, this assumption is also supported by its practical application in the U.S. coal power sector73–75. Although there are other mercury-specific technologies, such as HI, we only focused on ACI in this study because it is commercially mature, widely used, and has demonstrated stable performance in mercury mitigation76. Moreover, ACI and HI are often used together to enhance mercury removal efficiency; ACI is the primary control technology, while HI typically serves as a supplementary method. Therefore, when labeling control technologies in CFPPs, only ACI is usually indicated. The removal efficiency of ACI emphasized in this study refers to the general label for mercury-specific technologies used in the U.S. from UNEP. In this yearly dynamic retrofit process under BAT and ACI scenarios, units with high emission intensity are prioritized for retrofitting; and it is required that the remaining operational lifetime should exceed five years after upgrading, to avoid cases where a unit is retired shortly after being equipped with BAT, which would be economically inefficient. Moreover, after the installation of additional APCDs, such as WFGD, net generation will be reduced. After wet scrubbers are installed, electric power consumption at the plant level will increase. Thus, we assume that the additional electricity consumption will be compensated by increased power generation from newly built units. In other words, as more end-of-pipe control devices are installed, coal power generation from newly built units will increase accordingly. The electricity consumption rates of different air pollution control technologies are collected from existing studies77, summarized in Supplementary Table 15. Therefore, based on two climate mitigation strategies, two retirement policy options, and three pollution control technology configurations, we constructed a range of emission reduction scenarios. Unless otherwise specified in the main text, the default retirement policy assumes historical retirement patterns, and the default pollution control technology configuration follows business-as-usual (BAU). Cost analysis The total costs (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{total}}}}}$$\end{document}Ctotal) embodied in mercury mitigation policies include stranded capital costs of strategic retirement (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{stra}}}}}$$\end{document}Cstra), and upgrading costs to improve end-of-pipe control technology (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{upg}}}}}$$\end{document}Cupg) in air pollution control strategies, as calculated in Eqs. (8)–(11).8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{total}}}}}={C}_{{{{\rm{stra}}}}}+{C}_{{{{\rm{upg}}}}}$$\end{document}Ctotal=Cstra+Cupg9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{stra}}}}}={{{\rm{Cap}}}}\times {U}_{{{{\rm{str}}}}{{{\rm{a}}}}}\times \left(1-\frac{{{{\rm{Age}}}}}{{{{\rm{Lifetime}}}}}\right)$$\end{document}Cstra=Cap×Ustra×1−AgeLifetime10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{upg}}}}}={C}_{{{{\rm{upg}}}}}^{{{{\rm{inv}}}}}+{C}_{{{{\rm{upg}}}}}^{{{{\rm{OM}}}}}$$\end{document}Cupg=Cupginv+CupgOM11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{upg}}}}}^{{{{\rm{OM}}}}}={{{\rm{Cap}}}}\times {U}_{{{{\rm{fix}}}}}^{{{{\rm{OM}}}}}\times \left ({{{\rm{Lifetime}}}}-\,{{{{\rm{Age}}}}}_{{{{\rm{upg}}}}}\right)$$\end{document}CupgOM=Cap×UfixOM×Lifetime−Ageupgwhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{Cap}}}}$$\end{document}Cap is the unit-level installed capacity, MW. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{U}}_{{{{\rm{stra}}}}}$$\end{document}Ustra is the unit cost, USD MW-1, which is compiled from previous literature and reports (see Supplementary Table 16 for details and references)48–50. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{Age}}}}$$\end{document}Age is the operating age of the coal-fired unit at the time of retirement (yr). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{Lifetime}}}}$$\end{document}Lifetime is assumed to be the technical lifetime of the coal-fired unit, here taken as 40 years. The upgrading costs comprise the capital cost \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left(\right.{C}_{{{{\rm{upg}}}}}^{{{{\rm{inv}}}}}$$\end{document}Cupginv) and the fixed operation and maintenance (O&M) cost (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{{{\rm{upg}}}}}^{{{{\rm{OM}}}}}$$\end{document}CupgOM), in which the costs per unit capacity are assumed to be the average cost of different types of units in each country, and due to limited available data, the average upgrading cost in the U.S. is used for high-income countries without data, while the average costs in China are used for other countries without available data78. The detailed data and references are presented in Supplementary Table 17. For all future cost analyses, an annual 5% discount rate is applied. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{Age}}}}}_{{{{\rm{upg}}}}}$$\end{document}Ageupg is the age of the unit at the time the APCDs upgrade is implemented (years). Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file Source data Source Data_Figure 1–5 Source Data_Figure 1–5