A high-resolution, spatially explicit estimate of fossil-fuel CO 2 1 emissions from the Tokyo Metropolis, Japan

Background: The quantification of urban greenhouse gas (GHG) emissions is an important 11 task in combating climate change. Emission inventories that include spatially explicit 12 emission estimates facilitate the accurate tracking of emission changes, identification of 13 emission sources, and formulation of policies for climate-change mitigation. Many currently 14 available gridded emission estimates are based on the disaggregation of country- or state- 15 wide emission estimates, which may be useful in describing city-wide emissions but are of 16 limited value in tracking changes at subnational levels. Urban GHG emissions should 17 therefore be quantified with a true bottom-up approach. 18 Results: Multi-resolution, spatially explicit estimates of fossil-fuel carbon dioxide (FFCO 2 ) 19 emissions from the Tokyo Metropolis, Japan, were derived. Spatially explicit emission data 20 were collected for point (e.g., power plants and waste incinerators), line (mostly traffic), and 21 area (e.g., residential and commercial areas) sources. Emissions were mapped on the basis of 22 emission rates calculated for source locations. Activity, emissions, and spatial data were 23 integrated, and the results were visualized using a geographic information system approach. 24 Conclusions: The annual total FFCO 2 emissions from the Tokyo Metropolis in 2014 were 25 43,916 Gg CO 2 , with the road-transportation sector (16,323 Gg CO 2 ) accounting for 37.2% of 26 the total. Spatial emission patterns were verified via a comparison with the East Asian Air 27 Pollutant Emission Grid Database for Japan (EAGrid-Japan) and the Open‐source Data 28 Inventory for Anthropogenic CO 2 (ODIAC), which demonstrated the applicability of this 29 methodology to other prefectures and therefore the entire country.


34
Fossil-fuel combustion is a major contributor to increasing atmospheric carbon dioxide (CO2) 35 concentrations [1], with cities worldwide being responsible for more than 70% of the global 36 total fossil-fuel CO2 emissions (FFCO2) [2]. As large sources of FFCO2, cities have great 37 potential for emission mitigation [3]. In response to the need for local climate action, many 38 global approaches, including the purely geographic-based (PB) [9, 10], consumption-based carbon-52 footprint [11,12], and community-wide infrastructure-footprint [13,14] approaches. Such 53 studies estimate emissions for whole cities by monitoring activities at the city level. Due to 54 recent mitigation action, the need for more detailed information on temporospatial 55 distributions of emissions has increased, necessitating the assessment of emission changes at 56 sub-city levels. 57 For spatially explicit emission data, Gurney et al. [15] loosely categorized emission modeling 58 approaches as 'downscaled' or 'bottom-up'. The downscaled approach is used mainly for 59 global-scale greenhouse gas (GHG) EIs, and downscales national total emissions to source-60 related proxies [16,17,18]. Conversely, the bottom-up approach estimates temporospatially 61 explicit emissions based on sectoral activity data derived from socioeconomic sources. These 62 two approaches result in large discrepancies in urban-scale FFCO2 emissions, due to the 63 limitations of downscaled EIs in capturing spatial patterns of complex source activities [19]. 64 Since the bottom-up approach considers local-scale emission sources, emission estimates 65 based on this approach are likely suitable for tracking emission changes in sub-city and local 66 areas. For example, the Hestia Project developed multi-resolution emission data for four US 67 cities This study presents a detailed framework for the direct accounting of local FFCO2 emissions 77 in Japan using a bottom-up approach (i.e., PB approach for direct emissions (Scope-1) [6]) 78 and demonstrates its application to multi-resolution emissions at point, road, building, or 79 mesh levels. As a pilot study, we first consider spatially explicit FFCO2 emission data for the 80 Tokyo Metropolis (population 13.6 M in 2016; area 2,189 km 2 ) for the year 2014. This study 81 is distinguished from the East Asian Air Pollutant Emission Grid Database for Japan 82 (EAGrid-Japan) by the use of a multi-resolution approach and updated information. Here, we 83 describe the various statistical and geospatial data used in estimating and mapping emissions, 84 compare our emission estimates with existing estimates at aggregated city and grid cell 85 levels, and discuss current limitations and future improvements. 86 87

88
Emission definition and modeling framework 89 The focus here is on quantifying FFCO2 emissions from the Tokyo Metropolis using the 90 modeling framework described in Fig. 1 Table S1 for details). 110 The 2014 total emissions from all power plants were calculated by formula: facility. An example of point-source emissions in SG Ward (see Table S2) is shown in Fig.  128 2A.

129
Civil aviation emissions included those from passenger and cargo aircraft during landing and 130 take-off (LTO) at an international airport, four helipads, and six domestic airports (Table S1). 131 The 2014 total emissions from LTO movements were calculated as follows: Emissions from the waterborne navigation sector included emissions from fuel consumed by 147 vessels during round trips to ports in Tokyo ( = 15; Table S1). The 2014 total emissions 148 from vessels were calculated as follows: (3) 150 where represents the total annual emissions by vessels (Gg CO2); , is the emission 151 intensity of fuels (tonne per vessel) for type vessels (including merchant vessels, car ferries, 152 evacuation vessels, fishing vessels, and other vessels) at mode (travelling, cargo loading 153 and unloading, and mooring); , is the load factor of type-vessels in mode ; , is the 154 fuel consumption time of type-vessels in mode (h); , is the annual number of type-155 vessels travelling to port ; and is the emission factor (Gg CO2 tonne -1 ) for type-vessels 156 consuming heavy oil or light oil. The , , , , and , values and the average travel speed 157 for these vessels were obtained from technical reports [49,50]. As shown by the parameters 158 listed in Table S3, the mean travel distance was assumed to be 1 km in travelling mode to 159 derive the travel time. The emissions from travelling mode were considered for fishing 160 vessels but all mode were considered for the other types of vessels. The , values were 161 obtained from statistical data on vessels in ports [51,52]. The data were extracted from 162 the official guideline [34]. As waterborne vessels are mobile sources, representative points on 163 port buildings were used as their point-source locations. 164 Emissions from incineration plants do not contribute to FFCO2 emissions. However, this 165 study included their emissions because the emission intensity is significant. Emissions from 166 incineration plants for municipal solid waste (MSW, = 46) and industrial waste ( = 15) 167 (Table S1) included those from the combustion of wastes containing carbon (e.g., papers, 168 plastics, textiles, rubbers, and oil) and the combustion agent (CA, "city gas" comprising 169 liquid petroleum gas and natural gas). Emissions from MSW waste combustion were 170 calculated as follows: [53] were used for all MSW incineration plants. 184 Emissions from industrial waste combustion were calculated as follows: general national highways, major regional roads (prefectural roads and designated city roads), 207 and general regional roads. Emissions on minor roads that were not covered by the census 208 were not considered. 209 Road transportation emissions were calculated for single road segments ( = 45,564) as 210 follows:  with an elevation accuracy within 0.7 m (standard deviation) [67]. buildings. 272 Emissions from the residential sector were calculated for all of the population census areas in 273 Tokyo ( = 5,578) by formula: farmland, using a land-use map at a 10 × 10 m spatial resolution based on remote-sensing data 296 for the 2006-2011 period [74]. Finally, the agricultural emissions mapped in each municipality 297 [59] were sorted into a 10 × 10 m mesh for mapping based on the two types of farmland. 298 299 Data integration 300 Emission calculations and spatial emissions mapping/modeling were integrated using ArcGIS 301 v. 10.4. The world geodetic system (1984) was used for mapping all of the emission sources, 302 and a symbol tool was used here for visualizing the emissions on maps. A 3D map of the 303 emission sources in SG Ward is shown in Fig. 2D as an example, allowing visualization of 304 the emissions from local facilities, road segments, and buildings. 305 All of the data used, their versions or editions, and sources are summarized in Table S4. More 306 than two million building polygons were used to produce emission maps around 500 MB in 307 size. The emission maps were not gridded products since a multi-resolution approach was 308 adopted. The original maps were converted to a 1 km mesh size (Fig. 3) for convenience in 309 data handling. 310 311

312
Total emissions from the Tokyo Metropolis 313 Tokyo is one of 47 prefectures in Japan and comprises 23 central city wards and multiple 314 cities, towns, and villages (Table S2). The three highest point-source gridded emissions in 315 2014 occurred in SG, OT, and MN Wards at 6,183, 907, and 253 Gg CO2 km -2 , respectively 316 ( Fig. 3A), due to two large power plants and a major airport being located within these areas. 317 The highest line-source emissions occurred in KT, OT, and EG Wards at 155, 146, and 144 318 Gg CO2 km -2 , respectively (Fig. 3B). The highest gridded emissions for area sources (Fig.  319 3C) occurred in CD, CO, and SJ Wards at 173, 168, and 164 Gg CO2 km -2 , respectively. 320 These high emissions are primarily due to the high floor numbers and large building areas for 321 residential, industrial, and commercial use concentrated in these areas. A total emissions map 322 is given in Fig. 3D, with the three highest emissions being 6,210 in SG Ward, 1,058 in OT 323 Ward, and 295 Gg CO2 km -2 in MN Ward, respectively. 324 The estimated total 2014 FFCO2 emissions from Tokyo were 43,916 Gg CO2 ( Gg CO2 (21.6%); major regional roads (1,625 km) 4,761 Gg CO2 (29.2%); and general 345 regional roads (1,614 km) 2,128 Gg CO2 (13.0%). 346 The highest area-source emissions from the industrial and commercial sector for 2014 were 347 recorded in the inner-city areas in CD (172.4), CO (167.0), and SJ (162.0 Gg CO2 km -2 ) 348 Wards (Fig. 6A), respectively. The industrial and commercial emissions counted from 349 economic census areas were shown in Fig. 6B. Those from the residential sector were in KT 350 (10.0), TS (9.9), and TT (9.5 Gg CO2 km -2 ) Wards (Fig. 7A), respectively. The residential 351 emissions counted from population census areas were shown in Fig. 7B. Those from the 352 agricultural sector (Fig. 8A) were recorded in MS (0.45) and NK Cities (0.36), and EG Ward 353 (0.33 Gg CO2 km -2 ), respectively. The agricultural emissions counted for 62 municipalities 354 (Fig. 8B) were finally allocated for high-spatial-resolution map (Fig. 8C) Spatial distributions of the emissions between the two EIs were compared at a 1 × 1 km 404 resolution by scaling the total EAGrid emissions to our 2014 EI (Fig. 10). The difference in 405 the point sources (Fig. 10A) shows that some gridded emissions of this study were lower than 406 those in EAGrid. To map the gridded values of EAGrid, the counted total emissions from 407 each airport and port were allocated according to the number of persons engaged in the 408 related industry groups (Table S5), with the number of point sources being higher than those 409 in this study. Other differences are due to the EAGrid EI, which does not include recently 410 constructed major sources, such as Shinagawa power plant and Haneda airport domestic 411 terminal 2. As shown in Fig. 11A, the correlation of the gridded emissions of point sources 412 between the two sets of results is very low (R 2 = 0.42). 413 Line-source differences (Fig. 10B)  burning, and other emissions as area sources, whereas this study only considers residential, 421 industrial and commercial, and agricultural sectors as area sources. The area-source 422 emissions in the present EI were 3,292 Gg CO2, lower than those of the EAGrid. As shown in 423 Fig. 11B-C, the correlations of the gridded emissions for line and area sources between the 424 two sets of results are high (R 2 = 0.74 for line sources and 0.71 for area sources). 425 Differences in total emissions vary between -700 and +4,500 Gg CO2 km -2 (Fig. 10D), with 426 differences being smaller in the western mountain and forest areas and larger in the inner-city 427 areas (eastern Tokyo). As shown in Fig. 11D, the correlation of the total gridded emissions 428 between the two sets of results is moderate (R 2 = 0.69). The number of cells in the present EI 429 is much greater than that in EAGrid in the 0-10 Gg CO2 km -2 emission range (Fig. 12), with 430 the present EI therefore including more low-emission areas than the EAGrid, while greater 431 10-50 Gg CO2 km -2 emissions are included in the latter. The numbers of cells are consistent 432 for the other emission ranges. Thus, we could conclude that even the number of cells in some 433 emission ranges and the total annual emissions between the two sets of results seem to be 434 close but the distributions of the source emissions are different. 435 The Open-source Data Inventory for Anthropogenic CO2 (ODIAC) [18] provides emissions 436 with less detailed patterns from inner-city areas (eastern Tokyo) to the western mountains and 437 forested areas in Tokyo in 2014 (Fig. 13A). Such characteristic emission distributions are 438 also reported in other urban areas [15,25,82]. Differences in gridded annual emissions 439 between our study and ODIAC ranged from about -800 to +3,300 Gg CO2 cell -1 (Fig. 13B), 440 with differences being greater in inner-city areas. The blue-color grids (Fig. 13B) indicate the 441 three most negative values in densely populated areas. The ODIAC 2014 estimated higher 442 total emissions over these areas, while our estimates were lower. The red-color grids (Fig.  443 13B) indicate the three highest positive values, where two power plants (the Shinagawa and  444 Oi plants, with 3,219 and 2,965 Gg CO2, respectively) and an international airport (Tokyo 445 Haneda Airport, with 940 Gg CO2) are located (Table S1). It is clear that ODIAC 2014 does 446 not include such large point sources. We emphasize that local activity data are critical in 447 capturing spatial patterns of local emissions in urban areas. 448 449 Current limitations and future perspectives 450 Uncertainties associated with emission factors, activity data, and emission spatial modeling 451 introduce uncertainties in the final emission estimates [e.g., 24, 83]. We refer to the 452 uncertainties on the basis of activity data and emission factors (Table S6) using IPCC  453 guidelines [27,84]. The total uncertainty is estimated to be ±3.57%, equivalent to 43,916 ± 454 1,568 Gg CO2. 455 Uncertainties introduced from emissions calculations and mapping processes are likely to be 456 large due to the assumptions and approximations used. For example, the operation ratio of 457 power plants varies with individual plants; however, this study applied averaged utilization 458 efficiency for the whole plants in the calculation process. This approach reduces the 459 variability in emissions at each power plant, leading to poor representation of emissions with 460 higher temporospatial resolution than we applied here. The road segments that are not fully 461 covered by the census contribute over 4,205 km in our calculation. We substituted the 462 average traffic conditions for the road segments to estimate the emissions. This approach 463 could overestimate the traffic quantities and emissions for the segments. 464 In mapping processes, this study treated the mobile emissions of aircraft and vessels as point 465 sources. This means that the whole emissions over their moving paths were aggregated to a 466 point, leading to an overestimate of the point-source emissions. The building emissions were 467 estimated using TFAs of buildings in each census area. In this estimate we used DSM data 468 with a spatial resolution of 30 m, but this spatial resolution is insufficient to calculate the 469 heights and TFAs for individual buildings. Additionally, our downscale approach did not 470 distinguish occupied and vacant houses. All of these limitations should be improved in the 471 next study. As in previous studies (e.g., Hestia [23]), better data availability for emissions 472 calculations and mapping should greatly improve the accuracy of estimates. 473 We plan to update our emission estimates once updated activity data become available.

Availability of data and materials 534
The data used in this study are either presented in this manuscript or available from the data 535 source indicated. The authors plan to make the data product developed in this study publicly 536 available with a DOI. 537