Figure 2a shows the flow diagram for estimating emissions from road transport. On-road vehicles were classified as motorcycles (MCs), taxis, cars, buses and heavy-duty vehicles (trucks), each of which includes gasoline and diesel vehicles. We calculated annual hot emissions based on the annual number of registered vehicles, average daily vehicle mileage traveled, and emission factors for each vehicle type with the following equation: 1Ehot=∑iVPi⋅DailyVKTi⋅365⋅EFi(Creutzig et al., 2011), where i represents vehicle types, daily VKT is the average daily vehicle kilometers traveled for vehicle type i (km d-1), VPi is the population of vehicle type i and EFi is the hot emission factor of vehicle type i. The daily VKTs of each vehicle type in HCMC (2013) were extracted from the study of Oanh and Van (2015) and were assumed to be constant over the years (Table 3).

The vehicle population data were synthesized by different data sources, such as the statistic of the transport department of HCMC and previous studies about vehicle emissions in HCMC. However, the annual number of registered vehicles in some types was missing, such as population of trucks and buses over the years. The number of trucks was calculated based on the data in 2013 (Oanh and Van, 2015) and proportionally estimated for other years based on annual volume of freight carried that was provided by the HCMC Statistical Yearbook. The bus population and the taxi population in 2015 and 2016 were proportional to the number of cars (Table 4).

Cold emissions from road transport were included for NOx, CO, PM10, PM2.5, BC, OC and NMVOCs by the following equation: 2Ecold=Ehot⋅βi⋅Fi(Ahlvik et al., 1998). Cold emissions were adjusted according to hot emissions using the fraction of distance traveled driven with a cold engine or with the catalyst operating below the light-off temperature (βi) and the correction factor of EFhot for cold-start emissions (Fi) The parameters β and F are functions of average monthly temperature. Equations for β and F and related parameters were taken from the EMEP/EEA EI guidebook 2009 (EEA, 2009). Monthly average surface temperatures in HCMC were adopted from https://www.weather-atlas.com/ (last access: 25 November 2018).

The EFs were extracted from seven different studies conducted in Hanoi and China (Table 5), covering 12 pollutant species. In Eq. (1), EFs and daily VKTs of each vehicle type were assumed to be constant over the years. This usage of constant EFs and daily VKTs is a result of limited availability of public data. Therefore, the annual emissions were mainly driven by vehicle populations.

Figure 2b shows the basic procedure to estimate emissions from the manufacturing industry sector. It focuses on fuel-consumption-based emissions which are considered to be Scope 1 emissions. Scope 1 refers to all direct emissions from sources located within the city boundary. In addition, this study will calculate Scope 2 electricity-consumption-based emissions separately (Sect. 2.3.4). Emissions from fuel combustion were calculated from the following equation, similar to the one applied in REASv2.1: 3EFuel=∑iAi⋅EFi⋅(1-Ri)(Kurokawa et al., 2013; IPCC, 1996), where E is emissions from fuel consumption of manufacturing activities, i is fuel type, A is fuel consumption, EFi is unabated emission factor of each combustion species and Ri is reduction efficiency of the control device. In the case of SO2, the emission factor was estimated from the following equation: 4EFSO2=Si⋅(1-SRi)(IPCC, 1996), where EFSO2 is emissions of SO2 for each fuel type, Si is sulfur content of fuel and SR is sulfur retention in ash. The total fuel consumption in HCMC (2013, 2014, 2015) with the ratio of final fuel consumption by sub-sector and fuel type (Table 6) was provided in the GHG EI compiled by JICA (2017a). Based on this GHG EI, the annual fuel consumption of manufacturing industry and residential sectors, including gasoline, diesel, heavy oil, kerosene, liquefied petroleum gas (LPG) and natural gas, can be estimated for 3 years: 2013, 2014 and 2015. The fuel consumption of the manufacturing industry sector in 5 other years (2009 to 2012 and 2016) was proportionally calculated using annual gross output of industry at current prices by industry activity in HCMC, provided by the HCMC Statistical Yearbook (Table 8) with the assumption that there is a linear relationship between fuel consumption of the manufacturing industry sector and gross output of industry. Unabated emission factors, reduction efficiencies of each pollutant species, sulfur content of fuel and sulfur retention in ash were adopted from the compiled database presented in the Atmospheric Brown Clouds: Emission Inventory Manual (ABCEIM) by Shrestha et al. (2013). ABCEIM has included EFs from several databases as well as available measurement data reported for various sources in Asia (Table 7).

Figure 2b shows the flow diagram for estimating emissions from the residential sector. This sector covers all fuel combustion activities in households, including domestic cooking and use of fireplaces. Kerosene, liquefied petroleum gas (LPG) and natural gas are used for cooking, while kerosene is used for lighting in the residential sector in many regions. Coal and biomass fuels such as wood are used mostly for domestic cooking and heating stoves in rural areas. Similar to the manufacturing industry sector, the annual emissions from the residential sector were calculated using Eqs. (3) and (4). Fuel consumption in 2013, 2014 and 2015 was provided by GHG EI compiled by JICA (2017a) (Table 6). The fuel consumption in other years was proportionally calculated using population provided by the HCMC Statistical Yearbook (Table 8) with the assumption that there is a linear relationship between fuel consumption of the residential sector and population of the city.

The uncontrolled emission factors and reduction efficiencies for the residential sector, sulfur content of fuel and sulfur retention in ash were adopted from ABCEIM by Shrestha et al. (2013) (Table 7).

Apart from Scope 1 emissions, this study also considered CO2 Scope 2 emissions that are from purchased energy generated upstream from the city, mainly electricity. Consumption-based emissions encompass those emissions produced by consumption within those same boundaries, regardless of the origin of those emissions. Local governments often include Scope 2 emissions when they do not have electric generating plants within their boundaries but still wish to evaluate the impacts of electricity use in the community. CO2 emissions from electricity consumption of manufacturing industries and residential sectors were calculated from the simple equation 5Eelectricity=∑A×EFelectricity, where E is emissions from electricity consumption, A is activity data which here is the amount of electricity consumption from each sector and EFelectricity is grid emission factor, specific for each region. In GHG EI compiled by JICA, the electricity consumption in 2013, 2014 and 2015 by sub-sectors was collected from Vietnam Electricity (EVN) using data collection forms. The electricity consumption consists of five sub-sectors (residential buildings, commercial and institutional buildings and facilities, manufacturing industries and construction, energy industries and agriculture, and forestry and fishing activities) (Table 9). This value includes emissions from both consumption of grid-supplied energy consumed within the city boundary and transmission and distribution loss from grid-supplied energy.

The significant linear relationships during 2013–2015 between electricity consumption of the industry sector and residential sector with annual gross output of industry and annual population were found (Fig. S1 in the Supplement). Thus, electricity consumption in other years was proportionally calculated using the same proxies applied in the fuel consumption part.

The manufacturing industry and construction sector used annual gross output of industry at current prices by industry activity with the assumption that there is a linear relationship between electricity consumption of the manufacturing industry sector and gross output of industry.

Figure 2c shows the methodology for spatial allocation applied in this study. Current EIs like REASv2.1 use population datasets to allocate their emissions from area sources to grid cells. Similarly, in the studies of Ho (2010) and Ho et al. (2019), the spatial distribution of industrial and residential emission sources is estimated by using the population density in each cell with the justification that the industry in HCMC is mainly located in residential areas. However, the level of detail required by local emissions inventories cannot be met sufficiently if these approaches are applied. To overcome such limitations, we prepared original datasets that allowed spatial allocation at 1 km grid nets. This advantageous method, as summarized below, can benefit the compilation of other community-scale EIs in the future.

In order to estimate gridded emissions from the road transportation sector, the road density from the road network downloaded from Open Street Map (OpenStreetMap contributors, 2019, distributed under a Creative Commons BY-SA License) was applied for spatial disaggregation. Here, a gridded network was created whereby road density was estimated for each “grid square”, with different weights for three types of roads: 2 for primary roads, 1 for secondary roads and 0.5 for tertiary roads. These weights were derived from modeled road capacity in HCMC (2016) by the HOUTRANS project, JICA (2004), in which the assigned traffic volume on primary roads is over 85 000 passenger car units (PCU) per day, on secondary roads 44 000 to 85 000 and the smallest roads under 44 000 PCU per day (JICA, 2004).

Scope 1 urban emissions from manufacturing industries, commercial places and residences were allocated spatially based on the urban land use morphology. As spatial distribution of land use is not available for HCMC, annual urban land use maps were created with the help of remote sensing datasets for the period 2009–2016 in HCMC following Misra et al. (2019). Urban morphology maps include three land use types most commonly associated with urban emissions: residential, commercial and industrial land. In this study, the industrial emission sector is considered an area source instead of point source like in previous studies. Identification of the three land use type areas (residential, commercial and industrial) was based on the hypothesis that each land use typology generally follows a distinct morphology with regards to the height of structures and nighttime artificial lighting. Therefore, urban morphological maps were prepared at 30 m spatial resolution by classifying digital building heights and nighttime light over each pixel into the three land use types.

Digital building heights were extracted from publicly available ALOS World 3D (AW3D30) digital surface model (DSM) data at 30 m resolution. A DSM is a representation of visible geological Earth terrain and any other features (tree and crop vegetation, built structures, etc.) occurring over the ground terrain. The AW3D DSM was generated using images acquired from PRISM's (Panchromatic Remote-Sensing Instrument for Stereo Mapping) front, nadir and backward-looking panchromatic bands aboard ALOS (Advanced Land Observing Satellite) between 2008 and 2011. It is publicly available at 1 s (30 m) horizontal resolution from JAXA (http://www.eorc.jaxa.jp/ALOS/en/aw3d30/, last access: 18 September 2017). The AW3D DSM generally meets the 5 m root-mean-square target height accuracy as per its producers (Tadono et al., 2015). To extract the height of features that do not form part of the terrain (known as normalized digital surface model or nDSM), first a continuous ground terrain (known as digital terrain model or DTM) needs to be constructed, which can then be differenced from the DSM (Eq. 6). A multidirectional processing and slope-dependent (MSD) filtering approach was used for DTM extraction and is further described in Misra et al. (2018). Accordingly, the MSD filtering technique requires four parameters to generate a DEM: the Gaussian smoothing kernel size, the scan line filter extent, the height threshold and the slope threshold. Each DSM pixel was checked to determine whether it should be considered ground by comparing it with other pixels within the predefined neighborhood scan line filter extending in eight directions. If the pixel was identified as a ground pixel in more than five directions, it was labeled as a terrain pixel by the majority voting method. To draw the comparison, a local reference terrain slope was first generated by 2D Gaussian smoothing. Then, the pixel's height was compared with the lowest elevated pixel within the scan line filter extent. If this height difference was more than the height threshold parameter, the pixel was classified as a non-ground pixel. Then, if the slope difference between the current and the successive pixel in the scan line direction was greater than the slope threshold, it was labeled as a non-ground pixel. If the slope was positive and less than the slope threshold, then that pixel was given the same label as its previous pixel. Otherwise, that pixel was labeled as ground. 6nDSM=DSM-DTM To ascertain that the height of the extracted features was indeed from the built-up structures and not features like trees, a built-up class binary mask was generated and multiplied with the corresponding pixels in the nDSM raster to generate nDSM for built-up area (subsequently referred to as digital building height). A time series of Landsat imagery (Landsat 7 for 4 years: 2009 to 2012 and Landsat 8 for 4 other years: 2013 to 2016) was classified to generate the urban built-up extent for 2009 to 2016. A Mahalanobis-distance-based supervised classification was performed to identify five classes (including built-up, vegetation, fallow land, lake and river).

Nighttime light was obtained using the VIIRS (Visible Infrared Imaging Radiometer Suite) DNB (Day–Night Band) monthly images for the year 2014. The VIIRS DNB was freely obtained from https://ngdc.noaa.gov/eog/viirs/ (last access: 28 October 2017); its spatial resolution of approximately 15 s (450 m) was resampled to 30 m. An annual DNB image was prepared by considering median radiance of monthly composites of the DNB product. It consists of light from persistent sources, but the original data have not been filtered for forest fires or any other activity that may generate light from natural sources. Thereafter training samples were collected for the digital building height and the nighttime light over the residential, commercial and industrial pixels they were classified as using the random forest classifier.

In this study, we relied on only one building height data source (extracted from AW3D30) in 2011 and calibrated VIIRS DNB night light radiance in 2014 to prepare land use maps for 8 years. This usage of constant building height is a result of limited availability of public DSMs. Therefore, any land use transitions among the residential, commercial and industrial land use types were assumed to be negligible. However, the changes in built-up land cover over 8 years as identified in Landsat images help in accounting for horizontal urban growth. Ultimately this resulted in eight annual urban morphological maps for HCMC from 2009 to 2016. These land use maps were used for spatial distribution of manufacturing industry and residential emissions into the same grid net – 1 km resolution with transportation sector. Based on the spatial matching, the total emissions of three key sectors are simply the sum of grid nets of manufacturing industry, residential and transportation emissions. Note in this study the industrial emission sector is considered to be an area source instead of point source like in previous studies.

Table 10 summarizes emissions for each species in HCMC from on-road transportation for 8 years, from 2009 to 2016. Figure 3 shows the relative contribution of vehicle types to total transportation emissions. On the whole, all 12 pollutants expressed the same gradual growing trend over 8 years. The reason is that the increase in emissions of all species was driven by the same dataset of vehicle population, and VKT and emission factors were assumed to be constant. However, the mix of contributions from different vehicle types was different for each pollutant species.

Total CO emissions in 2016 were 682 Gg, increasing by 312 Gg (+98 %) compared to 2009. Over 95 % of CO emission from transportation was accounted for by MCs. In HCMC, the growth rate of MCs reached 180 % over 8 years (from over 4 million to over 7.2 million vehicles). Although the increase rate of personal cars was higher (+130 % from 2009 to 2016), MCs had constantly shared around 90 % of total vehicle population during that period (Fig. 3). Furthermore, the CO emission factor of MCs was by far the highest one among five types of vehicles (12.592 g km-1 per vehicle). This is almost 6 times higher than that of personal cars and taxis.

Over this period, NOx presented a different pattern, despite the same monotonically increasing trend with CO. Total NOx in 2016 was 84.8 Gg (+157.4 %) for HCMC. The majority of NOx emissions were from heavy-duty vehicles (HDVs, trucks; 50 %–61.9 % during the period 2009–2016), followed by personal cars (19 %–21 %). The fact is that HDVs use diesel engines which emit a higher amount of PM and NOx (Reşitoğlu et al., 2015). In addition, the truck fleet in HCMC is quite old (the average age is 11.7 years), leading to a high sharing ratio of outdated engines (75 % of trucks used Euro 2 engines) (Oanh and Van, 2015). As a result, the NOx emission factor of trucks is the highest one among five types of vehicles. Although emission factors of trucks (17 g km-1 per vehicle) and buses (16.954 g km-1 per vehicle) are roughly equal, the dominant population of trucks made it the largest contributor to total NOx emissions from transportation.

Total SO2 emission from on road traffic in 2016 was 6.773 Gg. The emission values more than doubled from 2009 to 2016 (+155.78 %). Again, the contribution was dominated by trucks (39 %–48 %), followed by personal cars (38 %–42 %). Different from CO emissions, MCs were not an important source for SO2. This common vehicle accounts for a modest proportion (7.4 %–10.5 %) compared to others. This trend reflects the highest SO2 emission factor of trucks which use diesel engines (1.06 g km-1 per vehicle).

Total emissions of PM10 and PM2.5 in 2016 from the transportation sector in HCMC were 20.4 and 5.07 Gg (+143 % and 157 %). Showing a similar pattern to NOx, trucks made the highest contributions to emissions of particulate matter (38.4 %–49.5 % and 54.7 %–66.9 % for PM10 and PM2.5, respectively). The reason is that diesel engines emitted much more fine particles than gasoline engines that are mainly found in MCs (Reşitoğlu et al., 2015). EFs of HDVs used in this study are 3.28 and 1.1 g km-1 per vehicle for PM10 and PM2.5, respectively, which are almost 35 and 61 times higher than those of MCs. Consequently, although MCs shared over 90 % of total vehicle population, the dominant emission factors of vehicles using diesel engines made them the main emission sources of PM.

Emission of the aerosols BC and OC showed almost opposite tendencies. Total emissions of BC and OC in 2016 in HCMC were 0.222 and 0.982 Gg (+85 %/+87 %), respectively. MCs made the most considerable contribution to these primary aerosol emissions (92 %–94 %). The remaining transportation types accounted for very small shares (1 %–5 %). This is because we applied emission factors of BC and OC from updated emission factors of air pollutants from vehicle operations in GREET (Greenhouse Gases, Regulated Emissions, and Energy Use in Transportation) using MOVES (MOtor Vehicle Emission Simulator) (Cai et al., 2013). According to this database, MCs are the most common source of BC and OC emissions (0.004 and 0.0178 g km-1 per vehicle).

Regarding greenhouse gases, total emissions of CO2, CH4 and N2O in 2016 in HCMC were 21 999 Gg (+103 %), 6.601 Gg (+100 %) and 0.292 Gg (+92 %), respectively. In the cases of CO2 and N2O, MCs and personal cars are considered the main sources for emissions. A total of 50 %–57 % of CO2 emissions were from MCs, and their proportion decreased by 7 % over 8 years. Personal cars ranked second but their share of the total rose from 32 % to 36.8 % from 2009 to 2016. A similar trend was seen in N2O, although MCs had a larger share for this species than CO2 (74 %–79 %). The contribution of personal cars slowly grew from 18 % to 22 % of total N2O emission from transportation. In terms of CH4, which is important in reducing short-lived climate forcers, the share of MCs was by far the highest (95 %–97 %). The very small share of CH4 emissions in the transportation sector was from other vehicle types. These estimations are in line with other previous studies that claimed diesel engines emit less CO2 and greenhouse gases than similar gasoline ones (Reşitoğlu et al., 2015).

Table 11 presents annual emissions from fuel consumption in two other key sectors: manufacturing industries and residential buildings. Figure 4 shows the comparison among three key sectors for each species in HCMC from 2009 to 2016. It should be noted that only Scope 1 emissions that occur within the boundary of the city are considered in this sector. Generally speaking, both of these emission sources expressed a much smaller amount of emissions and slower growth paces than transportation.

In terms of manufacturing industries in HCMC, total SO2 emissions from this sector in HCMC increased monotonically from 1.092 to 2.36 Gg (+116 %) over 8 years. However, these amounts are still modest compared to emissions from transport activities, and the gap between the two sectors increased with time. In 2009, emissions of manufacturing industries were less than a half of traffic emissions. About 8 years later, this proportion decreased to 34.8 %. Normally, the main source of sulfur dioxide in the air is industrial activity that processes materials that contain sulfur. However, the explosion of transport activities in HCMC led to the dominant contribution of this sector to SO2 emissions. SO2 is a component of great concern because controlling SO2 emission may have the important co-benefit of reducing the formation of particulate sulfur pollutants, such as fine sulfate particles. NOx emission from manufacturing industries in 2016 was 9.739 Gg, increasing by 120 % over 8 years. Similar to SO2, this accounted for only a very small fraction (almost a ninth) of traffic emissions. This is reasonable because NOx is produced from the reaction of nitrogen and oxygen gases in the air during fuel combustion. In large cities, the highest amount of nitrogen oxide emitted into the atmosphere as air pollution is usually from road transport. This distance is even more profound in the case of CO. In 2016, CO emission from manufacturing industries was 4.152 Gg. Meanwhile, the transportation sector emitted 682.613 Gg, around 160 times higher than manufacturing industries. In addition, emission of CO from manufacturing industries showed a moderate growth rate compared to NOx and SO2, 75 % over 8 years.

Regarding primary particle matter, both PM10 and PM2.5 emissions almost doubled from 2009 to 2016. Total emissions of PM10 and PM2.5 in 2016 were 0.6 and 0.33 Gg (+96 % and 93 %). Again, these amounts of emissions are relatively insignificant compared to emissions from transport activities. However, the sharing ratio among sectors changed for the case of BC. BC emission were still mainly from road transport but the contributions of manufacturing industries to total emissions of this short-lived climate pollutant cannot be neglected. In 2016, the industry sector in HCMC emitted 0.14 Gg (+129 %) BC into the atmosphere, which was equal to 63 % of BC emission from transportation, and this proportion tended to increase. OC emissions differed from BC. The total OC emissions from manufacturing industries in HCMC (2016) were 0.0071 Gg (+86.8 %), which was equal to only 7.2 % of the emissions from transportation activities. Organic carbon has a cooling effect as it reflects light. Meanwhile, black carbon absorbs light. If the ratio of warming particles is higher, sources may have less of a cooling effect. This implies that apart from transportation, reducing short-lived climate forcers cannot disregard the share of manufacturing industries in HCMC.

Total CO2 emission in 2016 from manufacturing industries in HCMC was 3861.89 Gg (+114.7 %). CO2 emissions generally reflect the energy consumption, infrastructure buildup and economic growth. The dominance of energy consumption from traffic activities was confirmed again by the gap in CO2 emission between manufacturing industries and the transportation sector. Emissions of CH4 and N2O from manufacturing industries in 2016 were 0.1512 Gg (+116 %) and 0.03 Gg (+115 %), respectively.

The fuel consumption of the residential sector in big cities mainly includes heating, cooling, lighting, water heating and consumer products. In the GHG emission inventory prepared by JICA, the energy consumed by households in HCMC included only kerosene and liquefied petroleum gas (LPG). Because of tropical climate in HCMC, this energy consumption is generally for cooking/household stoves; the heating and water heating can be excluded. This explains quite trivial shares from the residential sector in total emissions of each pollutant species, although the population explodes in the study area over 8 years (+27.1 %). For example, CO emission from residential buildings in 2016 in HCMC was 0.4189 Gg, which is equivalent to a 1/10 of manufacturing industry emissions. In addition, our estimation implied that the evolution of household emissions is much slower compared to other sectors. In parallel with 27 % of the population growth in HCMC over 8 years is a 34.5 % increase in greenhouse gases emitted from the residential sector. Meanwhile, CO2 emissions from transportation soared by 104 % during the same period. In fact, the shares in household energy consumption are incommensurate. Particularly in tropical regions, where fuel consumption is mainly used for cooking purposes, the largest contributor is often household electricity consumption, which belongs to Scope 2 emissions. The following section discusses this in more depth.

The transportation emissions estimated in this study were compared with four previous studies about vehicle emissions in HCMC (Table 13). The study of Oanh and Van (2015) applied the same equation to estimate emissions of on-road traffic. Activity data included the number of active vehicles, divided by five vehicle types, and daily VKTs of each vehicle type. In addition, this research considered the daily number of startups per vehicle category and average speed to estimate detailed emission factors. EFs were separated into start-up EFs and running EFs. Their output is annual emission in HCMC (2013) for CO, VOCs, NOx, SO2, PM, BC, OC, CO2, N2O and CH4.

The second study is the GHG emission inventory of JICA. This study applied a different method: activity data were fuel consumption of transportation sector (mainly gasoline and diesel), and EFs were extracted from the 2006 IPCC Guidelines.

Another GHG emission inventory for HCMC was the study of Le et al. (2018). This author used activity data that were vehicle counts, by type of vehicle, and daily VKTs were from the study of Oanh and Van (2015). Their vehicle counts were derived from field measurements and vehicle registry data. Regarding the CO2 emission factor, they used country-specific EFs from the COPERT model instead of EFs from the 2006 IPCC guidelines. Noticeably, they divided the vehicle fleet into four types only: MCs, buses, diesel cars and gasoline cars.

Another highly detailed vehicle emission inventory in HCMC was prepared by Ho (2010). The activity data were hourly traffic counts, including five categories, namely cars, light-duty trucks, heavy-duty trucks, buses and MCs. EFs were extracted from literature review and were assumed to be constant in each street category and constant in time. The output is hourly emissions from the vehicle fleet in HCMC. The contribution percentage of each vehicle type for each pollutant in this study was compared with the estimation in this study.

Our CO2 emission in 2013 was quite close to the calculation of JICA, although they applied different approaches to estimate emissions from transportation. But our finding is higher than the estimation of Oanh and Van (2015), around 4000 Gg. We applied the same numbers of active vehicles and daily VKTs used in the research of Oanh and Van (2015), but they used EFs in much more detail. As a result, the different emission factors are likely the reason for this gap. In comparison with the findings of Le et al. (2018), our CO2 emissions in 2016 are double. As mentioned before, this author classified vehicle fleet in HCMC into only four types, without considering trucks. In addition, the difference in daily VKTs, vehicle population and emission factor is likely to contribute to the inconsistency with our calculation.

In terms of other pollutants, estimations were lower by factors of 2–10 in comparison with the study of Oanh and Van (2015). The vehicle populations and daily VKTs were the same for both studies. The smallest gap was observed for NOx emissions, while BC and OC emissions showed significant distinctions. The EF dataset used in the study of Oanh and Van (2015) was not clarified in their publication, but it is expected to explain the gap between the two studies. Their study applied the International Vehicle Emissions (IVE) model to produce the EFs that are relevant to the local driving conditions and local fleet composition, and it considers the engine technology distribution in the vehicle fleet. Meanwhile, this study applied constant emission factors from different previous research about transportation emissions in HCMC, Hanoi and China.

Regarding the sharing ratio of emissions from MCs and personal cars (PCs) in this study and previous studies for 2010 and 2013, our results are relatively consistent with the research of Ho (2010) (Table 14). MCs are responsible for over 94 % of CO emission from on-road traffic, 29 % and 15.6 % of NOx emissions in the study of Ho (2010) and in this study, respectively. In comparison with the study of Oanh and Van (2015), the sharing proportion of MC from my estimation is higher, but the contribution from personal cars is lower in terms of CO. A significant gap was recorded in the sharing percentage of MCs for NOx emission. According to their calculation, the MC fleet accounts for 80 % of total NOx emission from transportation in HCMC. This ratio was much higher than the findings of Ho (2010) (29 %) for vehicle emissions in 2010. As mentioned before, this study applied the same number of active vehicles and the same daily VKT data as the study of Oanh and Van (2015) for emissions in 2013. As a result, this gap is expected to come from inconsistent EF datasets between two studies.

The general information of REASv2.1 and the data sources applied for three sectors, transportation, manufacturing industry and residential, are mentioned in the Supplement (Tables S1 and S2). REAS2.1 mainly used the national statistical data for their activity data. Then the spatial allocation was based on road network or population data to create grid maps with 0.25∘ resolution. Table 15 shows the comparison between the estimations in this study for 2009 and the estimation of REAS 2.1 for three key sectors in Kurokawa et al. (2013). The transportation emissions from REAS for 2008 were much lower than our calculation for 2009, except BC, by factors of 1.5 to 10, depending on pollutant species. It is worth noticing that the data sources applied in two studies were not the same. REAS was based on vehicle numbers, annual distance traveled and emission factors to estimate vehicle emissions. Their vehicle population was the national one, and then total emissions were allocated to HCMC using the road network. Thus it is likely that the calculation underestimated the emissions from a traffic hotspot such as HCMC. In addition, the gap between their annual VKTs and our daily VKTs could be a cause.

Conversely, the estimation of industry and residential emissions of REASv2.1 surpassed the findings in this study by factors of 3–104. The difference for the residential sector is more significant than for industry. In this comparison, only Scope 1 emissions of HCMC were compared with REAS emissions. So both studies applied fuel consumption as activity data. But REAS fuel consumption data were national data provided by the International Energy Agency (IEA) energy balance database, and the data applied in this study are annual sale data provided by the HCMC Department of Industry and Trade (DOIT) and fuel companies. Apart from different sources of activity data, Table 16 compares the EFs of fuel consumption in these two sectors that were applied in two studies. In addition, for countries which do not have their own emission inventories, REAS adopted emission factors from 1980 to 2003 from many sources, including Asian emission inventories. Meanwhile, this study applied EFs provided by ABCEIM from Shrestha et al. (2013). REAS 2.1 applied EFs of oil and gas only. In terms of the industry sector, EFs of NOx and NMVOC were pretty similar. But EFs of CO and SO2 from REAS are much higher. The CO emission factor was over double that applied in this study. Regarding oil, the SO2 emission factor was over 10 times higher than our EF. This trivial consistency was seen in EFs used in the residential sector as well (Table 17).

The large discrepancies can be seen in the sum of emissions from fuel consumption of three key sectors. The gap of under 25 % between our EI and REASv2.1 was recorded only in the cases of CO, NOx and CO2. According to the big gaps drawn from this analysis, the limitation when comparing a regional emission inventory with a local emission inventory can be implied. This inconsistency is expected due to the differences in activity data and EF databases. Again, the limitations of downscaling a regional EI to the community scale can be seen here. The regional scale is likely to underestimate the most profound emission sector like transportation and show the overestimation of other sectors when applying population as the only spatial allocation index.