Bottom-up emission inventories provide primary understanding of
sources of air pollution and essential input of chemical transport
models. Focusing on
Bottom-up emissions are estimated through comprehensive parameterization of fuel consumption, industrial production, emission factors and mitigation measures and spatially allocated to satisfy the chemical and climate model requirements. Uncertainties of emissions have been qualitatively illustrated (e.g., Granier et al., 2011; Saikawa et al., 2017) or quantitatively analyzed (Streets et al., 2003; Zhao et al., 2011; Guan et al., 2012; Hong et al., 2017), inferring significant gaps in activity statistics and control measures' assumptions in emission inventories developed for different spatial scales (i.e., global, regional, or city scale; Zhao et al., 2015).
Extensive comparisons of emission inventories have been conducted to
illustrate the impacts of variable emissions on the model simulation
results (e.g., Saikawa et al., 2017; Zhou et al., 2017). Although they
provide important indications on the extent of discrepancies, there
are still gaps for applying the comparison results to improve the
inventory accuracy:
Comparisons have been conducted for the total anthropogenic
sources, instead of by sectors, subsectors, and sources. Inconsistency of
source categories included in inventory models were not overviewed
or analyzed. Few studies go into the comparisons on a specific parameter level
because the technology-based framework for each inventory was not
publicly available. Top-down and bottom-up comparisons have not been comprehensively
combined to infer the potential uncertain parameters for all key
sectors.
To further improve the accuracy of emission estimation, we compared and
evaluated the global ECLIPSE (Evaluating the Climate and Air Quality Impacts
of Short-Lived Pollutants; Klimont et al., 2017) and MIX Asian (Li et al.,
2017) inventories due to the following reasons: (a) Up to the time of paper
preparation, ECLIPSE and MIX were the only publicly accessible gridded
emission datasets that include both
Another motivation of this work is to discuss the “fitness” of current
developed inventories (specifically ECLIPSE and MIX) and modeling work
performed with them for policy-relevant discussion. The inventories and the
relevant modeling work are playing an increasingly important role for policy
discussion in Europe and most recently more and more in Asia on different
scales. However, there are no systematic and officially approved methods and
inventories but a variety of scientific products. While a lot of effort has
been made to validate emission estimates with measurements, higher source and
spatial resolution of inventories and projections will also serve discussion
about how to shape future policies to reduce impact of air pollution. In this
work, we compared the ECLIPSE and MIX emissions over China at a detailed
activity-source level. What we focused on in this paper is the bottom-up
comparison detailed to a specific parameter contributing to the differences
between the two widely used gridded emission inventories (ECLIPSE and MIX),
combined with top-down validations from the satellite observations. We
compared the activity rates and emission factors derived from several key
parameters for the largest sources for each sector and subsector.
Discrepancies in the methodologies used, data sources, technology penetration
assumptions and spatial emission patterns are discussed and illustrated.
Furthermore, we combined the bottom-up comparisons with top-down evaluations
based on observations of the OMI (Ozone Monitoring Instrument) aboard the
Aura satellite
(Levelt et al., 2006). To our knowledge, this is the first emission inventory
assessment work in which parameter-level comparison and remote sensing
evaluations are combined. OMI data provide essential constraints on emission
estimates, spatial distributions and trends (Wang et al., 2012; Liu et al.,
2016). In recent work described by Geng et al. (2017), OMI
Methodology and data used are summarized in Sect. 2. Bottom-up comparisons of emissions are illustrated by decomposing the elements of inventory development in Sect. 3.1. Section 3.2 presents the evaluations and constraints from a satellite perspective. A summary of key reasons leading to emission discrepancies is provided in Sect. 3.3. Finally, Sect. 4 gives the concluding remarks.
Spatially specific emission inventories of air pollutants and greenhouse gases are among key inputs for chemical transport models (CTMs) and climate models. ECLIPSE (Klimont et al., 2017) and MIX (Li et al., 2017) emission inventories have been applied in numerous modeling activities at global (Stohl et al., 2015) and regional levels, within the ECLIPSE and MICS-Asia (Model Intercomparison Study for Asia) Phase III projects, respectively. In general, both inventories use a dynamic technology-based methodology to estimate anthropogenic emissions by multiplying activity rates with technology-specific emission factors for each source by administrative unit (province or county) (Klimont et al., 2017; Li et al., 2017). Then, spatial proxies are used to distribute emission estimates by province or county to grids to satisfy the needs of model simulation. The key features of both inventories are listed in Table 1.
Key features of ECLIPSE v5a and MIX emission inventories.
The ECLIPSE dataset is a global emission inventory for the
period of 1990 to 2010 extended by projections to 2050 in 5-year
intervals with monthly variations, developed with the GAINS model
(Amann et al., 2011). Primary sources of activity data are the
International Energy Agency (IEA, 2012) for fuel use and the UN Food and
Agriculture Organization for agriculture (FAO,
Emissions are distributed to grids at a specific resolution
(
In this work, we use a gridded ECLIPSE v5a dataset (current legislation,
CLE;
available at
MIX was developed for 2008 and 2010 (including monthly
variation) by combining the up-to-date regional inventories. For
China, the monthly MEIC dataset (available at
Monthly gridded emissions of MEIC are generated by applying
source-based spatial and temporal profiles (Li et al.,
2017). Provincial emissions of MEIC are firstly distributed to counties,
then further distributed to grids. The former process was based on
statistics by county (i.e., GDP (gross domestic product), industrial
GDP, total population, urban population, rural population,
agricultural activity, vehicle population), and the latter was based
on gridded maps as spatial proxies (i.e., population density map, road
network). For power plants, locations were determined using Google
Earth following the unit-based methodology. Gridded emission products
of MEIC v1.1 at a resolution of
GEOS-Chem is an open-access global 3-D CTM widely used by about 100
research groups worldwide. The model is driven by the GEOS (Goddard
Earth Observing System) meteorological dataset and includes complete
In this work, the Asian-nested grid GEOS-Chem model v9-01-03 driven by GEOS-5
was used to simulate
We developed the top-down emission inventories for
Top-down
The top-down emissions were further determined based on
Eq. (
Following Cooper et al. (2017), we limited
Following the framework of gridded emission inventory development, we conducted parameter-level comparisons between ECLIPSE and MIX and quantified the reasons causing the emission differences for each sector. Starting from the emission comparisons for all of China in Sect. 3.1.1, we further compared emissions by province in Sect. 3.1.2 and gridded emissions in Sect. 3.1.3.
As shown in Table 1, the activity rates were assigned independently by two inventories. As a global emission inventory, ECLIPSE mainly relies on international statistics of IEA. Differently, MIX obtains the official statistics of energy consumption and industrial output from NBS (National Bureau of Statistics) or MEP (Ministry of Environmental Protection) of China. We can expect high independency for the determination of emission factors between ECLIPSE (GAINS model) and MIX (MEIC model). As two independently developed inventory models, the source classification, technology and removal efficiencies of control facilities in GAINS and MEIC are expected to be different, although they both refer to up-to-date measurements and peer-reviewed data. Different methods were developed in two inventory models for specific sectors, including power plants, transportation, and agriculture. For power plants, the spatial proxies were essentially consistent between ECLIPSE and MIX. For other sectors, emissions were gridded independently by two emission inventories (see Table S1).
Although a comprehensive dataset on fuel consumption and product yield, surveys on technique penetration and measurements of emission factors are incorporated in the inventories, there are several additional assumptions made to characterize some sources for which information is either incomplete or missing. For ECLIPSE and MIX, assumptions are made independently and data sources are often different. Particularly, MEIC developed high-resolution emissions based on unit-based information for the power sector and county-level emissions for transportation.
Figure 1 shows the comparisons of China's emission estimates in 2005 and 2010
between two inventories for four key sectors: power, industry, residential
and transportation. For 2010, ECLIPSE estimates about 28
Emissions of
As shown in Fig. 1, emission trends from 2005 to 2010 are similar in
two inventories, indicating analogous assumptions of technology
evolution driven by economic growth and implemented air quality
policies in ECLIPSE and MIX. In general, MIX estimates larger changes
by sectors in the analyzed period. Specifically, for power plants,
MIX estimates a decline for
The fuel consumptions of MIX and ECLIPSE among different sectors in 2010 are presented in Table S2. Owing to different source structure in each of the models, there are sometimes significant discrepancies for specific sectors. For example, for coal, the total consumption is relatively consistent, within 10 % on a mass basis, while in the road transport sector MIX has 28 % higher diesel fuel use. More details along with discussion of emissions and implied emission factors are provided below.
For power plants, coal combustion contributes more than 95 % of
Activity rates, emissions and emission factors for
The coal consumption of ECLIPSE is
ECLIPSE's
For
Comparison of industrial emissions is most challenging since this
sector includes a multitude of sources with greatly varying emission
characteristics and different representation in the investigated
inventories. Overall, ECLIPSE calculates lower emissions, i.e., 24 and
13 % for
Residential combustion contributes around
Transportation sector contributes more than 25 % to the total
Comparisons of activity, emissions and emission factors
for the transport sector emission estimates of
Table 3 compares the fuel consumption, emission estimates and net emission factors among various vehicle types in two inventories for 2005 and 2010. Parameters for 2005 show a similar difference ratio to those in 2010. Assumptions for diesel combustion sources (on-road and off-road) are the main contributor to emission discrepancies. In 2010, diesel emissions of ECLIPSE are over 50 % lower than MIX estimates.
Provincial emissions were developed using different methodologies for two inventories (see Sect. 2.1). The provincial emission discrepancies between two inventories are attributed primarily to two factors: (i) the differences in activities, emission factors and policy implementation assumptions at the national level (as discussed in the previous section) and (ii) distribution of activities among the provinces – see Sect. 2.1 for principal data sources for the latter.
Figure 2 compares emissions and the relative difference in fossil fuel
consumption by province in ECLIPSE and MIX in 2010. The differences in
provincial
Comparisons of emission estimates and fuel consumption by
province in China in 2010. Values out of
For
The sectorial distributions of emissions by province are generally
consistent between the two inventories, as presented in Fig. S1 in the Supplement. For
Gridded emissions are direct inputs for atmospheric chemistry models
and climate models. We compared ECLIPSE and MIX gridded emissions by
analyzing three components: emissions by grids at
Comparisons of MIX and ECLIPSE gridded emissions on
Sectorial emissions show distinct spatial characteristics (Fig. 3b).
Comparisons of industrial and residential sectors show clear
administrative boundaries as these are typically distributed from
provincial emissions using population-based proxies. Since power
plants are treated as point sources, emissions differ in grids over
the entire country, higher for
The differences of gridded emissions illustrated in Fig. 3 are attributed to the discrepancies in emission estimates nationwide and by provinces (Sect. 3.1.1, 3.1.2) and also method and data of emission spatial allocations (see Sect. 2). For power plants that were treated as point sources, emissions are gridded based on the locations verified by Google Earth (Liu et al., 2015), consistent between ECLIPSE and MIX. For other sectors, ECLIPSE gridded the provincial emissions according to the source-specific layers, and MEIC used a two-step allocation method (province to county, county to grid). The data sources of spatial proxies also differ between the two inventories (see Table S1). We further compared the spatial proxies by sector in Fig. 4.
Emission distribution ratios within provinces in China in
2010 at
In this work, we calculated the distribution ratios, reflecting the
spatial proxies used, by dividing the emissions for each grid by the
provincial emissions for each sector. The distribution ratios between
ECLIPSE and MIX in 2010 are shown in Fig. 4. Good correlations
(slope
In MIX, decreasing emissions for the Beijing and PRD (Pearl River Delta)
regions are estimated, which are dominated by a decline in the power and
transport sectors, in contrast to the increasing emission trend of
ECLIPSE. The different trends of transportation emissions are
attributed to the different assumptions about the legislation effect on
pollution control in the two inventory systems. For Beijing, the
differences in the transportation emission trend are mainly caused by
diesel vehicles. In ECLIPSE, 47 % increases are estimated for
diesel-fueled vehicles, compared to 28 % emission decreases in
MIX. Fuel consumptions show large discrepancies in the trend from 2005 to
2010, with
For the PRD region, gasoline and heavy-duty diesel vehicles contribute to the
different emission trends. Of the emission changes for light-duty gasoline buses,
a 22 % increase is estimated in ECLIPSE,
compared to 12 % emission reduction in MIX. For heavy-duty diesel
vehicles, the trends in fuel consumption (
In this section, we evaluated the effect of emission inventories on
the accuracy of model simulations through combing GEOS-Chem
Asian-nested modeling and OMI observations (Sect. 3.2.1). Top-down
emissions were developed for both
The main purpose of this subsection is to evaluate the effect of gridded emissions on model performance and figure out the effect of emission estimates and spatial distributions on model performance using satellite observations as a criterion. Therefore, we set up four sensitivity cases of modeling, ECL-case0, ECL-case1, ECL-case2 and MIX. ECL-case0 and MIX form the two basic cases, which apply the ECLIPSE and MIX emissions in the simulation, respectively. ECL-case1 scales China's emissions in ECLIPSE to MIX's value by sectors retaining original spatial distributions. ECL-case2 redistributes the ECLIPSE emissions over China based on the spatial grid ratios of MIX, also on the sector level. The characteristics of the emission inventory used for each case are summarized in Table 4 and shown in Fig. S2. We processed each inventory into model-ready inputs through regridding new emissions, performing VOC speciation and temporal allocation. The speciation factor and monthly profiles by sector of the MIX inventory are used for ECL-case0–case2. We resample the model results based on satellite observations spatially and temporally for consistent comparison as described in Sect. 2.3.
Description of model simulation cases and statistics of
model performance of
Figure 6 compares the
Satellite-based emission inventories were developed following the
finite difference mass balance methodology (Cooper et al.,
2017). Emissions of
Top-down
In spatial distribution, large discrepancies are observed between
bottom-up emission inventories and the top-down ones. For
Comparisons among ECLIPSE, MIX and top-down emissions of
Comparisons among ECLIPSE, MIX and top-down emissions of
Emission changes from 2005 to 2010 were evaluated and presented in
Fig. 10 for
We address several key factors contributing to the differences between
ECLIPSE and MIX for
The source classification for heating plants, fuel conversion and industrial boilers is differently defined between ECLIPSE (GAINS model) and MIX (MEIC model), making the interpretation of comparisons for each source more difficult and to some degree less transparent (Sect. 3.1.1). The source-structure differences are inevitable for emission inventory models designed for estimates on different spatial scales. As a global inventory model, GAINS integrates statistics from international sources (e.g., IEA, FAO). Therefore, the source structure of GAINS is set up in accordance with the international statistics framework. Focusing on emissions on a regional scale, MEIC set up a calculation framework based on statistics from local agencies in China to gain higher specificity in temporal and spatial distribution (e.g., NBS, CPED). As illustrated and analyzed in Sect. 3.1.1, the differences by sector should be interpreted with caution, especially for the power and industry sectors.
The apparent emission uncertainty ratios (the ratio of the maximum
emission discrepancy to the mean value using provincial energy
statistics or national statistics) of
The FGD penetration rate in power plants, as well as assumed removal
efficiencies, significantly affect the
Spatial proxies used in emission inventories are an important factor contributing to the overall accuracy in model simulation. Integration of detailed spatial information that is often included in regional inventories like MIX should be considered as the best way to improve the resolution and spatial allocation of emissions in global products like ECLIPSE.
In this work, moderate negative biases are observed in bottom-up
emission inventories (
We conducted parameter-level comparisons of gridded Chinese emissions
between ECLIPSE and MIX, elucidated the effect on CTM simulations, and
evaluated the inventories based on OMI observations. This work is
important for inventory developers and modelers for understanding the
potential uncertainties in the gridded emission inventory over
China. For inventory developers, the detailed comparisons give
indications on the underlying uncertainties of parameters by source,
including the source classifications, activity rates, emission factors
and technology distributions. For modelers, the comparisons and
validations are important to understand the effect of emissions on
model performance. This work shows that our best inventories appear to
be fit for evaluation of the policies at an aggregated or national
level; more work is needed in specific areas in order to improve
accuracy and robustness of outcomes at the finer spatial and also
technological levels. The main findings are as follows.
In 2010, compared to MIX, the emission estimates of ECLIPSE
are identical for We modeled four sensitivity cases to investigate the effect
of emission estimates and spatial proxies of emission inventories on
model accuracy using GEOS-Chem. The model case using MIX as input
shows the best performance, with mean biases at Both inventories show decreasing trends for
ECLIPSE v5a global emissions developed based on the
GAINS model can be publicly accessed from
The authors declare that they have no conflict of interest.
This article is part of the special issue “Global and regional assessment of intercontinental transport of air pollution: results from HTAP, AQMEII and MICS”. It is not associated with a conference.
This work was supported by the National Key R&D program
(2016YFC0201506), the National Natural Science Foundation of China
(41625020) and IIASA's Young Scientists Summer Program (YSSP)
sponsored by the National Natural Science Foundation of China
(41611140118). We acknowledge the free use of the NASA OMI