2.1.1 MEIC

The integrated inventory INTAC is built upon MEIC, a comprehensive database with extensive coverage across time periods, space, species, and sectors. Developed by Tsinghua University since 2010 (http://meicmodel.org.cn, last access: 15 December 2022) (Li et al., 2017a; Zheng et al., 2018), MEIC provides monthly emissions for air pollutants and CO₂ in China from 1990 to the present at a resolution of 0.25° × 0.25°. It caters to the demand for timely and accurate estimates of atmospheric emissions and has gained widespread adoption by both domestic and international research institutions. We use 2017 emissions from MEICv1.3 in this study.

MEIC employs several strategies to improve emission estimation parameters. This includes categorizing emission sources across ∼ 800 sectors, utilizing a technology- and big-data-driven approach for dynamic emission characterization, and employing a localized emission factor database (Li et al., 2017a; Zheng et al., 2018). Emission estimates for power, on-road, and residential sources are enhanced through the use of unit-level data (F. Liu et al., 2015), county-level emission estimates (Zheng et al., 2014), and integration of extensive household surveys (Peng et al., 2019), respectively. MEIC builds a database encompassing temporal allocation profiles (ranging from yearly to monthly, daily, and hourly) (Li et al., 2017b), spatial allocation proxies (from province to county and further to grids) (Geng et al., 2017; Li et al., 2017b; Zheng et al., 2017), and a speciation framework for NMVOCs involving five mechanisms (CB-IV, CB05, SAPRC-07, SAPRC-99, and RADM2) (Li et al., 2014) to support the development of model-ready gridded emissions.

Among the seven inventories, MEIC has the lowest priority and is only considered when the other six cannot provide the necessary emissions for a specific city and source.

2.1.2 The industrial point source emission inventory for China

The proxy-based method used for spatial allocation in MEIC introduces biases in emission mapping, especially at the kilometer scale (Zheng et al., 2021, 2017). To significantly reduce the uncertainty, we merged an industrial emission inventory with detailed information on ∼ 100 000 facilities into INTAC.

Compiled by the MEIC team at Tsinghua University for the year 2013 (Zheng et al., 2021) and updated by them for 2017, this point-based inventory combines three databases investigated under the guidance of the Chinese government, offering a comprehensive overview of industrial facilities. It includes details on the locations, activity rates, production technology, end-of-pipe pollution control devices, and other parameters. It is worth noting that the facility-level activity data were corrected using provincial activity data from MEIC as a total constraint to ensure consistency with national totals from statistics (Zheng et al., 2021). The facility-level, technology-based approach allows for dynamic tracking of emission fluctuations resulting from technological advancements and tightening emission regulations. Crucially, the use of facility geolocations, rather than relying on spatial proxies like urban population, enables the derivation of gridded industrial data at a resolution of ∼ 1 km. This approach significantly avoids misallocating emissions from rural to urban areas at fine grids, as supported by previous studies demonstrating its effectiveness in mitigating simulated biases in air pollutant concentrations within densely populated regions (e.g., Zheng et al., 2021). For temporal variations it employs the same monthly profiles as MEIC, including the production of various industrial goods or gross domestic product (GDP), as outlined in Li et al. (2017b). The NMVOC speciation also aligns with the MEIC model. This inventory takes priority over MEIC, indicating that only few industrial sources not covered in this inventory are substituted with MEIC.

2.1.3 The YRD air pollutant emission inventory

Regional emission inventories within YRD provide a more accurate representation of emissions compared with the national-scale MEIC, as proven by ground and satellite observations (Yang and Zhao, 2019; Zhang et al., 2021; Zhao et al., 2017a, 2018, 2020; Zhou et al., 2017b). This improvement is attributed to the avoidance of outdated or non-localized emission calculation parameters commonly present in large-scale inventories like MEIC. Here, we merge the 2017 YRD air pollutant emission inventory into INTAC to achieve state-of-the-art estimates for rapidly changing emissions over this core area (An et al., 2021; Gu et al., 2023; Zhou et al., 2017b).

Localized field surveys and measurements greatly enhance the reliability of calculation parameters within the YRD inventory. Highly resolved emissions for the power sector are acquired through on-site monitoring with high temporal resolution (Y. Zhang et al., 2019), rather than relying on static and outdated average emission factors. Facility-level information (e.g., the removal efficiencies) obtained from local investigation and a segment-based industrial process method enhances the understanding of both the quantity and spatial patterns of industrial emissions. Considering meteorological factors and land use conditions during agricultural processes results in more accurate seasonal and spatial distributions of NH₃ emissions. (Zhao et al., 2020). An investigation of in-use machinery is conducted to capture the seasonal emission patterns from off-road machines (Zhang et al., 2020). Real-world surveys are performed to determine grain straw ratios and household burning proportions, facilitating the quantification of emissions from biomass-fueled stoves. The PM_2.5 and NMVOC speciation profiles are updated based on multi-instrument sampling and analysis in both current and previous studies (Huang et al., 2018; Zhao et al., 2017a), satisfying the need for simulating PM_2.5 chemical components and O₃. The YRD inventory is collected with a spatial resolution of 0.1° and an annually temporal resolution in this study. Only CB05 volatile organic compound (VOC) species are collected.

2.1.4 The PRD emission inventory

The regional emission inventories within the PRD region have demonstrated enhanced reliability compared with previous studies (Huang et al., 2021; Sha et al., 2021; Zheng et al., 2012). The PRD emission inventory developed by Jinan University captures spatial and temporal variations within the PRD region under emission control policies, serving as a foundation for supporting air-quality modeling (Huang et al., 2021; Sha et al., 2021).

The PRD inventory exhibits notable accuracy improvements achieved by means of big-data-driven estimation methods, updated spatial–temporal allocations, and localized NMVOC speciation profiles. Gridded hourly open biomass burning emissions are quantified by fusing the fire radiative power data from three satellites, and hourly shipping emissions are estimated using high-frequency automatic identification system (AIS) records. A total of 31 monthly profiles and 10 spatial proxies are updated to reflect spatial–temporal patterns of emissions influenced by economic growth and energy consumption structural adjustment. Approximately 90 % of industrial emissions are disaggregated using exact locations, and novel proxies (e.g., farmland production potential) have been developed for several sectors. The NMVOC speciation is carried out through extensive localized measurements and literature reviews, manifested as a collection of 480 NMVOC source profiles across eight sectors and 380 species. The species relevant to the SAPRC-07 chemical mechanism are collected in this work. Additionally, the inventory encompasses 800 source categories, placing particular emphasis on incorporating new sectors relevant to VOC emissions. Activity rates are improved by utilizing extensive field surveys and data mining efforts involving investigations of production data for 10 000 industrial plants and the gathering of activity-relevant information for 50 million vehicles. Emission factors that reflect the local context are obtained or revised based on source measurements and the latest research findings. These updates help mitigate uncertainties in emission estimates for the PRD region. The PRD inventory is initially collected at a monthly resolution and a spatial resolution of 0.05°, with detailed spatial–temporal allocation proxies as outlined in Huang et al. (2021).

2.1.5 The open biomass burning emission inventory in China

As a significant source of CO₂, BC, OC, and other pollutants, open biomass burning profoundly influences air quality, climate change, and human health (Reisen et al., 2013). A case study in summer 2011 for the YRD region revealed that during a severe haze episode, open biomass burning contributed to 37 %, 70 %, and 61 % of PM_2.5, OC, and elemental carbon (EC) emissions, respectively (Cheng et al., 2014). To address the absence of this source in MEIC, we integrate a high-resolution open biomass burning emission inventory by Peking University into INTAC (Huang et al., 2012a; M. Liu et al., 2015; Song et al., 2009; Yin et al., 2019).

The inventory applies satellite observations to tackle considerable uncertainties associated with provincial statistical data and overcome the coarse resolution found in previous studies (Ni et al., 2015). The estimation of biomass consumption in the inventory is based on the fire radiative energy (FRE) approach, which depends on the energy emitted by fires. This approach helps reduce the biases introduced by burned-area algorithms, especially for small-scale fires. The inventory utilizes the high-spatial-resolution land cover dataset GlobeLand30 derived from multi-spectral images to classify biomass fuel types. Eventually, daily emissions from forest, grassland, cropland, and shrubland are calculated at a 1 km resolution. The reasonableness is validated by comparing with other datasets such as the fourth version of the Global Fire Emissions Database (GFED). The initially collected inventory lacks model-ready VOC species.

2.1.6 The shipping emission inventory in East Asia

In recent years, maritime trade in the East Asian region has significantly increased (United Nations Conference on Trade and Development (UNCTAD), 2014), resulting in a surge in shipping emissions with substantial impacts on air quality and climate. Previous studies have indicated that East Asian shipping emissions accounted for 16 % of the global total in 2013. Shipping emissions made a growing contribution to the rise in annual mean PM_2.5 concentrations, reaching levels as high as 5.2 µg m⁻³ in 2015 (Lv et al., 2018). To address the omission of this emission source in MEIC, we integrate the shipping emission inventory in East Asia for 2017 into INTAC (H. Liu et al., 2016; Liu et al., 2019).

The inventory introduces an innovative approach based on comprehensive and dynamic ship activity data. A static dataset of approximately 66 000 vessels is compiled as a foundation, using information from Lloyd's Register and the China Classification Society. This dataset encompasses various ship properties including ship category, hull shape, engine rotational speed, engine capacity, maximum speed capability, and build year. High-quality AIS data are used to capture ship activities, incorporating the Maritime Mobile Service Identification identifier, geographical location, real-time speed, and time-related information. The AIS data are also employed to generate gridded emissions from shipping at a spatial resolution of 0.1°. The inventory enhances our comprehension of regional-level shipping emissions and significantly alleviates biases arising from the misallocation of marine fuels, as observed in global studies (Endresen et al., 2007). The collected shipping inventory provides emissions at an annual resolution for seven species including SO₂, NO_x, CO, NMVOC, PM_2.5, BC, and OC.

2.1.7 PKU-NH₃

As a prominent alkaline component in the atmosphere, ammonia plays a crucial role in atmospheric chemistry and also terrestrial and aquatic ecosystems through its participation in atmospheric reactions and deposition processes. This study integrates PKU-NH₃, a high-resolution ammonia emission inventory for China developed by Peking University. PKU-NH₃ is designed to track the evolution of NH₃ emissions amid the rapid increase in grain and meat production in China over the past few decades (Huang et al., 2012b; Kang et al., 2016). This inventory offers a better grasp on NH₃ emissions in China through the application of a process-based method and more reliable emission factors, in contrast to previous studies (Kurokawa et al., 2013; Li et al., 2017b). Top-down NH₃ inversion through satellite observations provides additional validation for the accuracy of PKU-NH₃ (Paulot et al., 2014).

Earlier studies of NH₃ emissions commonly used fixed EFs, overlooked some ammonia emission sources, and had coarse resolutions (Ohara et al., 2007; Streets et al., 2003). Unlike previous approaches, PKU-NH₃ incorporates dynamic and multifactorial EFs and more comprehensive emissions sources. The determination of emission factors takes into account various parameters related to local conditions and agricultural practices. When estimating NH₃ emissions of synthetic fertilizer application, the model considers five types of fertilizers and factors such as soil acidity, ambient temperature, fertilizer application technique and dosage, wind speed, and in situ measurements of NH₃ flux. For livestock waste, NH₃ emissions are calculated using a mass-flow approach across four phases of manure management, considering variables such as animal rearing types, temperature, and wind speed. In addition, NH₃ emissions from other small sources are quantified, including agricultural soil, nitrogen-fixing crops, crop residue compost, excretion of rural populations, open biomass burning, waste disposal, gasoline vehicles, diesel vehicles, and industrial processes. The NH₃ emissions are allocated from provinces into 0.1° grids based on spatial proxies such as land cover, rural population, and other relevant indicators. Monthly emission factors shaped by meteorological conditions are used to calculate NH₃ emissions from fertilizer application and livestock source at a monthly level.

2.2.1 Source mapping

To merge inventories under a unified emission source classification system, the emission sources in the MEIC model are categorized into 88 standard sectors for mapping (Table S1 in the Supplement). The first-level category comprises 10 subcategories, namely, stationary combustion, industrial process, mobile source, solvent use, agriculture, dust, biomass burning, storage and transportation, waste treatment, and other sources. These subcategories are then further subdivided into 88 second-level sources, which take industrial classification for national economic activities for reference. For example, the industrial process sector encompasses emission sources such as the manufacturing of non-metallic mineral products, manufacturing of chemical fibers, manufacturing of foods, and smelting and pressing of ferrous metals. In the initial step of integration, the sectors in each emission inventory are mapped to the standardized two-level sources.

2.2.2 Species mapping

In the next step, non-methane volatile organic compounds (NMVOCs), particulate matter (PM), and NO_x in each inventory are converted into model-ready species to support CTMs. The species mapping process is grounded in the chemical species mapping methods in the MEIC model (Li et al., 2017b, 2014). The model supports aerosol chemical schemes such as AER05 and AER06. NO_x emissions are allocated to NO and NO₂ emissions based on ground observations. The step-by-step NMVOC speciation framework developed in Li et al. (2014) is employed to generate emissions for various gas-phase chemical mechanisms commonly used in CTMs, including CB-IV, CB05, SAPRC-07, SAPRC-99, and RADM2. The framework incorporates an explicit-assignment approach and updated profiles based on both local measurements and the SPECIATE database v.4.5. The sources abundant with oxygenated volatile organic compounds (OVOCs) are identified, and the incomplete profiles with missing OVOC fractions are corrected. The accurate speciation mapping helps reduce uncertainties in model-ready emissions. For inventories providing speciated VOC emissions for certain mechanisms (e.g., the YRD inventory for CB05 and the PRD inventory for SAPRC-07), we use their emissions directly, or alternatively, we utilize MEIC's speciation framework to generate model species for the five chemical mechanisms.

2.2.3 Temporal disaggregation

The seven emission inventories are collected at different temporal resolutions (Table 1) and need to be temporally allocated to a unified monthly scale for integration. Monthly emissions from PKU-NH₃, the PRD inventory, the industrial point source inventory, and MEIC can be directly used for data merge. Daily-level emissions from the open biomass burning inventory are aggregated to monthly scales through summation. For annual inventories (e.g., the YRD inventory), sector-specific monthly profiles derived from the MEIC model are used for disaggregation (Li et al., 2017b). For instance, monthly power generation data from the National Bureau of Statistics describe variations in monthly power emissions. Industrial production or the GDP from the National Bureau of Statistics is employed to account for monthly emission fluctuations related to industrial heating, boilers, cement, iron and steel, and other industrial processes. Monthly emission factors calculated by the International Vehicle Emissions model are applied to on-road vehicles. Considering the insignificant monthly variations in automatic identification system data for marine shipping, the annual shipping emissions are uniformly disaggregated across the months.

2.2.4 Spatial allocation

The seven inventories are in different data formats, including point source and gridded formats at varying resolutions, necessitating spatial harmonization for integration. Although the industrial point source inventory and the open biomass burning inventory can accurately pinpoint the specific geographical locations of emission sources, the other five inventories rely on numerous spatial proxies to disaggregate emissions into grids, which inevitably introduce uncertainties at very fine resolutions. Therefore, we re-grid the final product to 0.1° to ensure high-level spatial accuracy. Gridded emissions finer than 0.1° resolution are aggregated to 0.1°, which is performed in the open biomass burning inventory and the PRD inventory. For the industrial point source inventory, latitude and longitude coordinates are employed to directly position them within grid locations. Area sources in MEIC are allocated to grids using spatial proxies within the MEIC model (Li et al., 2017b). For instance, industrial sources are assigned to grids based on urban population (Schneider et al., 2009). The road network (Zheng et al., 2014) serves as a proxy for disaggregating emissions of on-road vehicles, while rural population (Schneider et al., 2009) is used as the proxy for fertilizer and livestock sources. It is important to mention that uncertainties may arise at city borders if emissions from adjacent cities come from different inventories during the integration process. To mitigate biases introduced by border issues, all emissions at 0.1° resolution are first uniformly downscaled to 1 km in preparation for the spatial–temporal coupling process, and then they are re-gridded back to 0.1° for the final product.

2.2.5 Spatial–temporal coupling

Finally, following the procedures outlined in Sects. 2.2.1–2.2.4, all inventories are preprocessed to a standardized format encompassing 88 sectors, various species, a spatial resolution of 1 km, and a monthly temporal resolution. This preprocessing prepares the inventories for merging, ultimately resulting in the generation of a standardized data cube.

The integration is carried out at source-by-source, species-by-species, and grid-by-grid levels, with the process guided by the priority order of each inventory (Table 1). MEIC serves as the default inventory in our integration, offering extensive spatial and species coverage along with spatial proxies, temporal profiles, and NMVOC speciation methods within the model. The remaining six emission inventories are assigned a predefined priority order. The industrial point source emission inventory for China takes precedence over industrial emissions in MEIC, substituting proxy-based spatial allocation with precise geographical coordinates. This extends the applicability of MEIC from a resolution greater than 0.25° to finer scale (Zheng et al., 2021, 2017). To achieve fine-grained emission characterization in critical areas, the YRD and PRD emission inventories enriched with localized data and advanced methods are incorporated to update emissions in these areas. While MEIC comprehensively estimates emissions for ∼ 800 source categories in China, there may still be omissions for certain emission sources. The inclusion of inventories for open biomass burning and East Asian shipping helps partially fill this gap. PKU-NH₃, generated by a process-based model to provide a comprehensive understanding of China's NH₃ sources, is utilized to replace all NH₃ emissions in other inventories. The prioritization is performed city by city. For emissions of a particular species from a specific emission sector, when multiple inventories overlap in city grids, the estimates from the highest-priority inventory are selected as the final emissions. Through this step the integrated inventories are developed based on the configured output settings such as map projection and spatial–temporal attributes.

3.1.1 By sector

Table 2 displays emissions specific to power, industry, residential, transportation, agriculture, solvent use, shipping, and open biomass burning sectors in INTAC. For pollutants primarily originating from fuel combustion and industrial processes (e.g., SO₂, NO_x, CO, PM₁₀, and PM_2.5), the power, industry, and transportation sources collectively contribute significantly to their emissions, ranging from 56 % to 83 %. Industrial sources take a leading role in various atmospheric pollutants, contributing more than 30 % for SO₂, NO_x, CO, NMVOC, PM₁₀, and PM_2.5 emissions. Due to low combustion efficiency and a lack of emission control measures, residential sources exhibit a high emission factor for products of incomplete combustion, leading to 40 % of CO, 48 % of BC, and 73 % of OC emissions. Solvent sources exclusively produce NMVOC emissions, contributing 33 % to the overall emissions. The complexity of VOC emission origins is evident in the diverse range of contributing sources. Agricultural sources dominate NH₃ emissions, comprising an 83 % share of total emissions. As described in Sect. 2.1.7, PKU-NH₃ incorporates a wide variety of NH₃ sources, providing a more comprehensive understanding of the sectors contributing to NH₃ emissions. Insignificant sources may exert a large influence in specific regions or periods, such as during large wildfires or in cities with heavy traffic. Additionally, the contribution of the supplemented open biomass burning source, especially for OC (7 %) and NMVOCs (6 %), cannot be overlooked.

Figure 2 consolidates 88 standardized emission sources into 25 categories, allowing for a more detailed analysis of sectoral emission patterns compared with Table 2. Owing to substantial coal use in industrial and power sectors, along with sulfur-rich ship fuels, prominent contributors to SO₂ emissions include power, shipping, stationary combustion, and manufacture of non-metallic mineral product sources, accounting for 15 %, 13 %, 12 %, and 12 %, respectively, to total SO₂ emissions. This indicates that achieving further reductions in SO₂ emissions requires the implementation of more energy-efficient end-of-pipe control measures and adoption of low-sulfur fuels. The dominant origins of NO_x emissions are from the freight truck, power generation, and shipping sectors, representing 21 %, 15 %, and 13 %, respectively, of the total emissions. Both freight trucks and vessels extensively use compression ignition engines, which are prone to generating NO_x emissions under high-temperature and oxygen-rich conditions. Implementing strict vehicle standards is crucial to effectively reduce NO_x emissions from exhaust gases. Coatings, other industrial processes, and passenger vehicle sources together account for 51 % of anthropogenic NMVOC emissions. The major contributors to primary PM_2.5 emissions include biomass fuel, the manufacture of non-metallic mineral products, and the smelting and pressing of ferrous metal sources, making up 22 %, 17 %, and 10 %, respectively, of the total emissions. It is noteworthy that the use of biomass fuels (e.g., rice straw and firewood) for cooking or heating in rural areas results in considerable PM_2.5 emissions especially in provinces like Sichuan, Anhui, Shandong, and Heilongjiang.

3.1.2 By fuel type

Figure 3 illustrates the proportions of major air pollutant emissions in 2017 for each fuel type. Fossil fuel combustion significantly dominates the emissions of PM₁₀, PM_2.5, CO, BC, SO₂, and NO_x, with proportions ranging from 38 % to 80 %. Coal combustion accounts for 56 % of SO₂ emissions, with power, residential activities, and industrial production as the primary emitters. Furthermore, petroleum combustion, mainly from marine vessels, constitutes 20 % of SO₂ emissions. For NO_x emissions, petroleum combustion contributes 48 % of the total, predominantly arising from freight trucks (5.2 Tg), marine vessels (3.1 Tg), and passenger vehicles (1.0 Tg). Coal combustion processes, such as power and industrial boilers, also result in substantial NO_x emissions (31 %). The biomass fuel source causes 53 % of OC emissions. Emissions of NMVOCs and NH₃ are primarily associated with non-combustion processes.

Figure 3Fuel-specific distributions of major air pollutant emissions in the 2017 INTAC for China.

Download

Figure 4Spatial distributions of major air pollutant emissions in the 2017 INTAC for China.

Table 3Anthropogenic emissions (in Gg) of air pollutants by province in the 2017 INTAC for China. Emissions from the shipping emission inventory in East Asia are not included.

Download Print Version | Download XLSX

3.1.3 Spatial distribution

We present the gridded emission maps of major air pollutants in Fig. 4. Emissions from anthropogenic sources in China exhibit significant spatial heterogeneity. Due to economic growth and industrial activities, air pollutant emissions are primarily concentrated in the central and eastern regions of China, especially in economically developed urban clusters such as the Beijing–Tianjin–Hebei (BTH) region, the YRD, and the PRD and in regions like Sichuan and Chongqing. These four key areas, as depicted in Fig. S2, collectively account for 25 %, 33 %, 35 %, 37 %, 30 %, 35 %, 33 %, 27 %, and 29 % of the national emissions of SO₂, NO_x, CO, NMVOCs, NH₃, PM₁₀, PM_2.5, BC, and OC, respectively. Moreover, the emission maps at a fine spatial resolution of 0.1° × 0.1° present the local variations in emission patterns, identifying numerous hotspots in small areas and showcasing distinct gradients in emissions. Table 3 shows the provincial-level emissions (except Hong Kong SAR, Macao SAR, and Taiwan), and a map depicting provincial boundaries is displayed in Fig. S2. The emission levels in specific provinces are determined by factors such as resource endowments, industrial structure, energy consumption, and emission control measures. Taking SO₂ as an example, the top five provinces are Shanxi, Shandong, Hebei, Guizhou, and Inner Mongolia, collectively accounting for 36 % of the national total SO₂ emissions. Guizhou Province, located in the southwest of China, is characterized by high-sulfur coal and a relatively gradual implementation of pollution control measures, which result in elevated SO₂ emissions. In the other four provinces, large-scale heavy industries have led to substantial coal consumption and correspondingly higher SO₂ emissions. Provinces with a less industry-focused economic structure and lower energy consumption, including Tianjin, Hainan, Qinghai, Beijing, and Tibet, exhibit the lowest SO₂ emissions accounting for approximately 2 % of the national total.

Figure 5Inter-comparisons of emission estimates between INTAC and MEIC. Panel (a) shows the difference by sector, and (b) presents the ratio of emissions in INTAC to those in MEIC.

Download

3.2.1 Comparison of emission magnitudes in INTAC with MEIC across sectors and regions

INTAC improves the representation of anthropogenic air pollutant emissions by incorporating a large number of industrial point sources, integrating high-resolution regional inventories, and supplementing missing emission sources in MEIC. Remarkable differences between INTAC and MEIC are illustrated in Fig. 5 across regions and sectors. Compared with MEIC, INTAC shows higher levels of 16.7 %, 11.5 %, 10.8 %, 11.0 %, and 9.1 % for SO₂, NO_x, PM₁₀, PM_2.5, and OC emissions, respectively. However, it indicates lower levels of 6.3 % and 10.6 % for NMVOCs and NH₃. CO and BC emissions exhibit good agreement between the two inventories, with differences lower than 3.9 %. In comparison with MEIC, the supplementary emission sources in INTAC – specifically open biomass burning and marine shipping – account for the majority of increased emissions, contributing 95 %, 89 %, and 74 % for SO₂, CO, and PM_2.5, respectively. Additionally, the incorporation of PKU-NH₃ in INTAC leads to a 21 % decrease in NH₃ emissions from agricultural sources, while NH₃ emissions from residential sources and transportation increase by 99 % and 13.1 times, respectively. Such difference in agricultural sources is mainly caused by the estimates of synthetic fertilizer (Kang et al., 2016), particularly concerning the treatment of fertilizer types and corresponding emission factors.

Many discrepancies between MEIC and INTAC arise from the integration of regional emission inventories. As presented in Fig. 5b, notable disparities are observed in the YRD and PRD regions. Estimates for NO_x emissions in the YRD region are approximately 88 % of those derived from the MEIC model. This highlights an enhanced precision attributable to reliable assessments of denitrification efficiency in power plants and the measured NO_x emission factors for both power plants and boilers within the integrated YRD inventory, as supported by previous studies (e.g., Zhao et al., 2018). INTAC's estimates for NMVOC emissions in the YRD region are 26 % lower than estimates in MEIC. The overestimation in MEIC mainly results from the uncertainties in solvent use source, particularly coating, printing, and dyeing processes. The integrated YRD emission inventory employs more accurate calculation parameters for NMVOCs such as statistical data from local city yearbooks, industry association reports, and apparent consumption of solvents. Furthermore, the speciation profiles of NMVOCs are localized and corrected based on the literature research and measurements. In the PRD region, the NO_x emissions from INTAC are 41 % higher than MEIC estimates, with non-road sources and non-metallic mineral products contributing 45 % and 40 %, respectively, to this difference. The PRD inventory employs a detailed calculation approach for shipping emissions based on AIS data, in contrast to the simplified approach for inland waterway sources in MEIC. The NO_x emissions from industrial processes of brick and flat-glass manufacturing are not considered in MEIC, a deficiency which is addressed in the integrated PRD inventory. INTAC's NMVOC emissions are approximately 59 % of those from MEIC. The disparity is particularly notable in industrial and solvent use sources, contributing 49 % and 35 %, respectively, to the observed difference. In INTAC nearly half of the VOC emission factors for industrial solvent sources are based on local measurements, and a preference for raw-material-based calculations over product-based ones reduces uncertainty in the estimation. For significant VOC-emitting sources like cleaning solvents, MEIC employs an emission factor of 1000 g kg⁻¹, whereas the PRD inventory uses 850 g kg⁻¹. In the case of oil refineries, the emission factors are 2.76 g kg⁻¹ for MEIC and 1.82 g kg⁻¹ for the PRD inventory.

Figure 6Inter-comparisons of the percentage of point, on-road, and area source emissions between INTAC and MEIC.

Download

3.2.2 Impact of point source contributions

The most accurate method for obtaining emissions at finer-scale grids relies on spatial allocation based on precise geographical coordinates. In MEIC the majority of emission sources are represented as area sources and distributed onto grids using spatial proxies, such as urban population, except for power plants. In contrast, the increased proportion of industrial point source emissions in INTAC significantly constrains the uncertainties associated with spatial proxies. Figure 6 shows the inter-comparisons of the percentage of point, on-road, and area source emissions between INTAC and MEIC. Air pollutants, especially those dominated by industrial combustion sources like SO₂, NO_x, PM₁₀, and PM_2.5, exhibit a significantly higher proportion of point source emissions within INTAC compared with MEIC. In MEIC the proportion of point source emissions for SO₂, PM₁₀, NO_x, and PM_2.5 is 17 %, 9 %, 19 %, and 7 %, respectively. However, in INTAC these percentages substantially increase to 66 %, 54 %, 52 %, and 48 %, respectively, indicating a more accurate representation of spatial patterns. For other species with emissions mainly from area sources (e.g., residential and transportation), there are limited improvements in the proportion of point source emissions in INTAC.

Figure 7Spatial pattern analysis of emissions in INTAC using SO₂ emissions as an example. Panels (a) and (b) display the spatial distributions of SO₂ emissions in MEIC and INTAC, respectively. MEIC emissions have been downscaled from 0.25 to 0.1° for comparison. To compare MEIC and INTAC in detail, the YRD region is zoomed in on. Panels (c)–(e) show spatial distributions of SO₂ emissions in MEIC and INTAC, and also their difference. Circles in (e) represent the center of a city. Panel (f) compares cumulative percentage of SO₂ emissions in INTAC with those in MEIC across different spatial resolutions. The gridded SO₂ emissions, ranging in resolution from 0.05 to 1.0°, are cumulated in descending order of population. The percentage annotations in different colors indicate the level of accumulated SO₂ emissions in INTAC at various spatial resolutions when SO₂ emissions in MEIC reach 50 % accumulation. Panel (g) shows correlation coefficient between SO₂ emissions in INTAC and multiple spatial proxies at different grid sizes.

To further assess the impact of point sources, Fig. 7 takes SO₂ and the YRD region as an example to compare the spatial emission patterns between INTAC and MEIC. Figure 7c–e reveal that MEIC tends to overestimate emissions in urban centers and underestimate emissions in rural areas compared with INTAC. Amid economic growth and rapid urbanization, MEIC's use of urban population as a proxy for spatial allocation becomes impractical as many factories relocate from city centers to rural areas. To elucidate the difference between population-based and point-source-based allocation methods in emission mapping, we present the cumulative percentage of SO₂ emissions in MEIC and INTAC based on descending population orders in Fig. 7f. We use the grid groups where densely populated areas contribute 50 % of SO₂ emissions in MEIC as an example, and we compare them with the cumulative percentage in INTAC across various grid sizes. The results show that at a resolution of 0.05° INTAC only accounts for 17 % of the emissions, while it reaches 48 % as the grid size increases to 1.0°. This suggests that at a fine grid scale MEIC tends to allocate more emissions to densely populated urban areas, while INTAC allocates a larger proportion to suburban and rural areas, aligning better with the real-world emission spatial patterns. This mitigation of bias through INTAC is especially notable at finer resolutions. The close cumulative percentage at 1.0° in the two inventories can be attributed to the fact that urban and suburban areas often fall within the same grid, leading to a decreasing enhancement in the spatial accuracy achieved by INTAC. Figure 7g further presents the correlation between the spatial patterns of SO₂ emissions in INTAC and various spatial proxies. At a resolution of 1.0° the correlation coefficients between emission distributions and factors (i.e., road networks, nighttime lights, total population, urban population, and rural population) fall within the range of 0.55–0.79. Nevertheless, at a resolution of 0.05° the correlation coefficients range from 0.05 to 0.13. This indicates that at higher spatial resolutions INTAC substantially reduces the bias introduced by spatial proxies in MEIC.

Table 4The discrepancies between simulated SO₂, NO₂, and PM_2.5 concentrations, and observed values for 74 major cities at a resolution of 12 km, using MEIC and INTAC as emission inputs. The statistical metrics used for comparison include R, MB, and RMSE. The bold font represents the difference in modeling performance between INTAC and MEIC.

Download Print Version | Download XLSX

3.3.1 Overall performance in key regions

We conducted simulations using the WRF–CMAQ model driven by INTAC and MEIC separately to evaluate the improvements in modeled air pollutant concentrations. Table 4 evaluates the simulated emissions in 74 major cities (locations depicted in Fig. S2) against in situ observations, with corresponding scatterplots shown in Fig. S3. INTAC demonstrates improved agreement between modeled concentrations and ground-level observations, which benefits from the integrated high-resolution inventories. Compared with MEIC, INTAC leads to a decline in the mean bias of simulated major pollutant concentrations by 2–14 µg m⁻³, a reduction in the root mean square error by 4–19 µg m⁻³, and a decrease in the normalized mean error by 4 %–71 %. This finding indicates that INTAC produces a more accurate overall characterization of emissions in China. Furthermore, given that atmospheric pollution monitoring stations are mainly located in urban areas in China, the observed differences suggest that INTAC can mitigate the overestimation of major pollutant concentrations in urban centers. As discussed in Sect. 3.2.2, MEIC overestimates emissions in urban areas and underestimates them in rural and suburban areas, consequently introducing uncertainties into air-quality modeling. The improved accuracy in spatial distributions within INTAC significantly contributes to enhancing the overall accuracy of air pollutant modeling.

Figure 8Comparison of modeling performance across key regions (i.e., BTH, YRD, and PRD) when using MEIC and INTAC, respectively, as emission inputs. The statistical metrics used for comparison include R, MB, and RMSE. The regions under comparison comprise BTH, YRD, and PRD.

Download

Figure 8 further compares the overall simulation performance between INTAC and MEIC in three key regions (BTH, YRD, and PRD), with corresponding scatterplots shown in Figs. S4–S6. Regarding PM_2.5 and its precursors, MEIC shows a considerable mean bias of up to 36 µg m⁻³ and a root mean square error of up to 59 µg m⁻³ in key regions. In contrast, INTAC demonstrates maximum MB values of 15 µg m⁻³ and maximum RMSE values of 40 µg m⁻³. The correlation coefficients between simulated and observed concentrations of the three air pollutants are generally lower in MEIC compared with those in INTAC. The modeling performance driven by INTAC, particularly for short-lived pollutants, experiences significant improvement due to their strong correlation with spatial distributions of emission sources. Nonetheless, discrepancies between modeled and observed surface concentrations still exist because of uncertainties from meteorological, physical, and chemical processes within chemical transport models. Moreover, emission sources such as residential, transportation, and agriculture are treated in INTAC as non-point sources, and their allocation to grids using spatial proxies can introduce biases in air-quality modeling. It is noteworthy that ammonium (NH4+) concentrations simulated by INTAC agree better with ground measurements than MEIC (Table S3). While NH4+ concentrations are influenced by secondary chemical reactions, the improved model performance still reflects the benefits from the integration of PKU-NH₃.

3.3.2 Improvements across different spatial resolutions

To provide a more in-depth assessment of improved spatial patterns in INTAC, Fig. 9 categorizes grid cells into different bins based on their urban population and calculates the ratio of simulated pollutant concentrations to ground observations for both INTAC and MEIC in each category. The results demonstrate that as the urban population increases, the enhanced model performance of INTAC over MEIC for SO₂, NO₂, and PM_2.5 becomes more evident. Specifically, when the urban population is less than 50 000, both INTAC and MEIC exhibit a median range of simulated-to-observed concentration ratios close to 1. However, as the urban population exceeds 550 000, the average range for MEIC widens to 1.4–5.2, whereas it remains within the range of 0.9–1.0 for INTAC. This indicates a significant improvement in mitigating the overestimation of emissions in densely populated areas by INTAC. The incorporation of the industrial point source emission inventory for China along with the YRD and PRD emission inventories significantly increases point source shares in INTAC, thus producing better spatial representations for real-world emission distributions and smaller simulated deviations.

Figure 9Comparisons of modeling performance between INTAC and MEIC in different ranges of urban population. The 12 km grids are categorized to different bins according to the urban population residing within each grid. The ratio of simulated pollutant concentrations (Sim) to observed concentrations (Obs) for major pollutants (SO₂, NO₂, and PM_2.5) are calculated. The boxplot presents the upper quartile, median (red dot), and lower quartile of the ratios.

Download

Figure 10Comparison of modeled air pollutant concentrations and ground observations for 74 cities at 36 and 12 km resolutions using MEIC and INTAC, respectively, as emission inputs. The dashed black line represents the observational mean, and the annotations above the bar charts indicate the mean biases between simulated concentrations and the corresponding observed values.

Download

Model performance differences between MEIC and INTAC are influenced by grid size. Figure 10 presents the comparison between modeled SO₂, NO₂, and PM_2.5 concentrations against ground observations for 74 major cities at resolutions of 36 and 12 km. Increasing spatial resolution does not lead to a reduction in simulation errors, especially for MEIC. As the horizontal resolution increases from 36 to 12 km, the mean biases of simulated SO₂, NO₂, and PM_2.5 concentrations using MEIC as input show an increase from 37 % to 143 %, 11 % to 46 %, and −3 % to 15 %, respectively, when compared with in situ observations. In contrast, the simulation results using INTAC as input exhibit better agreement with ground observations, with mean biases for SO₂, NO₂, and PM_2.5 increasing from 23 % to 64 %, −0 % to 17 %, and 2 % to 11 %, respectively. This is due to the fact that the deviations in finer grid cells, whether overestimated or underestimated, tend to cancel out at a coarse spatial resolution. The decoupling between emission spatial distributions with proxies at finer grids leads to more noticeable biases in air-quality modeling. Therefore, the findings suggest that the INTAC developed in this study can effectively constrain uncertainties in emissions and the modeling bias, especially at fine spatial scales. The improvement will help tackle emerging challenges in high-resolution air-quality modeling in China.

Table 5Uncertainties in the inventory components of INTAC (in %) contingent upon the availability of such information.

Download Print Version | Download XLSX