Mahuika-Auckland: A spatially and temporally resolved fossil fuel CO2 emissions data product for Auckland, New Zealand

Abstract

Accurate, high-resolution and sector-specific greenhouse gas emissions information is increasingly needed for the development of local, targeted mitigation policies. We describe a detailed, spatially and temporally resolved CO₂ emissions data product, Mahuika-Auckland, for Auckland, New Zealand, based on Auckland's regional greenhouse gas and air emissions inventories. Emissions are provided at 500 m spatial resolution and at a 1-hr time step, a level of detail not previously available for any New Zealand city. We divide fossil fuel emissions into six sectors that comprise the majority of Auckland Region's CO₂ emissions profile: on-road transport, industrial non-point buildings and point sources, commercial non-point buildings, residential non-point buildings, air transport and sea transport. We also include separate layers representing biogenic CO₂ emissions (primarily waste and wood burning), as these are significant sources in Auckland. We distribute emissions spatially and temporally based on activity data, energy and fuel consumption patterns, and population statistics. The code to generate Mahuika-Auckland has been designed to be flexible so that updated information and/or data from more recent years can easily be incorporated. This data product improves upon New Zealand's current inventories that are only resolved at the regional and annual scale, providing a new level of detail that can be used as a prior estimate for atmospheric inversions, to inform emissions reduction policies and to guide the development of zero carbon pathways.

INTRODUCTION

Cities are responsible for approximately 70%–80% of global total greenhouse gas (GHG) emissions (IPCC, 2014). Consequently, any mitigation efforts to reduce emissions at scale must focus on urban activities and urban form. Accurate, high-resolution and sector-specific information is increasingly needed for targeted mitigation policy (Gurney et al., 2015). In order to formulate effective mitigation strategies and design low-carbon or carbon-neutral urban developments, local authorities need to have detailed, highly resolved information about sources and sinks of carbon.

In addition to national GHG inventories required for international reporting, many cities have developed their own city-wide and/or regional inventories following methods such as the Global Protocol for Community-Scale Greenhouse Gas Inventories (Fong et al., 2021). These inventories provide useful information for cities to begin to understand their emissions, but vary greatly in quality and comprehensiveness, and most do not resolve emissions at a finer scale beyond annual timesteps and city-wide boundaries.

Several gridded global data products exist that downscale anthropogenic GHG emissions from global and national inventories using a combination of energy statistics and fuel consumption data, population density, remote-sensed datasets (e.g. nightlights), power plant databases (CARMA: Ummel, 2012) and/or other proxy information. These include EDGAR (annual, 0.1° grid; Crippa et al., 2019, 2020; IEA, 2019), ODIAC (monthly, 1 km² / 0.083° grid; Oda & Maksyutov, 2011, 2015; Oda et al., 2018), PKU-CO2 (annual, 0.1° grid; Chen et al., 2016; Huang et al., 2015; Liu et al., 2015; Wang et al., 2013) and FFDAS-CO2ff (hourly, 0.1° grid; Asefi-Najafabady et al., 2014; Rayner et al., 2010; Rayner & Gurney, 2014). However, the resolution of most products is still relatively coarse for cities to derive meaningful information for urban-scale planning and specific policy needs. For example, for a city such as Auckland, New Zealand, with an area of approximately 600 km², the 0.1° grid that EDGAR, PKU-CO2 and FFDAS-CO2ff use does not resolve the coastlines or large pieces of infrastructure, which makes detailed sectoral attribution and analysis difficult. In addition, these products are designed to be standardized and applicable to all locations worldwide and thus do not take into consideration all local conditions or processes that contribute to emissions.

High-resolution gridded products have a key role in the independent monitoring and validation of emissions inventories, serving as a priori estimates of the spatial and temporal distribution of CO₂ emissions within an atmospheric inversion framework (e.g. Bréon et al., 2015; Lauvaux et al., 2020; Peylin et al., 2011). Previous inverse modelling studies in New Zealand have been restricted to global data products such as EDGAR (e.g. Steinkamp et al., 2017). This is adequate for national-scale studies but not for the higher-resolution city scale. One of the primary motivations for developing the dataset described in this manuscript is to fill this gap. Several inversion studies comparing the use of priors at different resolutions have shown that the resolution and fine-scale distribution of the prior significantly impacts the final posterior estimate, with the representation of large spatial gradients and sector-specific fine-scale spatial adjustments only possible with information at the sub-kilometre scale (Lauvaux et al., 2016, 2020; Oda et al., 2017; Peylin et al., 2011). Additionally, the finer spatial scale allows for sector-specific attribution and identification of areas dominated by local sources, aiding in the design of observation networks that can validate emissions inventories.

There have been a number of recent efforts to build detailed, high-resolution bottom-up CO₂ emissions products for urban areas and regions (e.g. Northeastern USA (ACES): Gately et al., 2015; Gately & Hutyra, 2017; China High Resolution Emission Database (CHRED): Cai et al., 2017; Cai et al., 2018; Paris, France (Origins): Lian et al., 2022; Ho Chi Minh City, Vietnam: Nguyen et al., 2021; UK NAEI: NAEI, 2013; European municipalities: Moran et al., 2022). The present study builds on one such effort, the Vulcan and Hestia CO₂ emissions products developed for North America (Gurney et al., 2009, 2012, 2020; Gurney, Liang, Patarasuk, et al., 2019; Gurney, Patarasuk, Liang, et al., 2019; Zhou & Gurney, 2010). The Vulcan product provides a gridded estimate of CO₂ emissions from fuel combustion and cement production for the continental USA. It uses a variety of publicly available state and county datasets such as energy-use statistics, traffic volumes and power plant stack monitoring together with fuel-specific emissions factors (EFs) to quantify emissions for 10 different sectors of point and non-point (area) sources. The latest version has a resolution of 1 km² hr⁻¹ for 2010–2015, providing a high-resolution product for every city in the USA (Gurney et al., 2020). Hestia uses Vulcan as a starting point and downscales emissions further to individual cities or urban centres in the USA, such as Indianapolis (Gurney et al., 2012), Los Angeles (Gurney et al., 2020), Salt Lake City (Patarasuk et al., 2016) and Baltimore (Gurney, Liang, O'Keeffe, et al., 2019; Roest et al., 2020), using bottom-up methods to estimate emissions at the road segment, building and point source level at an hourly time resolution. Hestia uses local datasets and tools wherever possible, such as traffic counts, building energy simulation models, electricity production and air pollution reporting.

Auckland Region is the focus of this work because it contains New Zealand's largest city, with an urban population of approximately 1.46 M, compared with New Zealand's total population of 5.12 M (as of June 2021; Stats, 2022). Auckland reported gross CO₂ emissions of 9,447 kt in 2018 (Xie, 2020), which is 27.5% of the national gross total CO₂ emissions in 2018 (34,305 kt CO₂; MfE, 2021). Auckland has set an ambitious target to reduce its greenhouse gas emissions by half by 2030 and to reach net-zero emissions by 2050 (Auckland Council, 2020). To reach this goal, timely and detailed, sector-specific knowledge and monitoring of GHG sources is required.

Here, we describe a new CO₂ emissions data product, ‘Mahuika-Auckland’ (Mahuika-AKL), named for the Māori god of fire. We present results for the base year 2016, the most recent year with detailed sectoral inventories of GHG and air emissions for Auckland Region at the time of development. However, Mahuika-AKL has been designed to be flexible so that updated information and/or data from more recent years can easily be incorporated. We consider only direct emissions that occur in the region of interest, or ‘Scope 1’ emissions (IPCC, 2006). Indirect emissions, sometimes referred to as ‘Scope 2’ (emissions produced elsewhere but consumed within the region boundary, such as power generation plants) and ‘Scope 3’ (embedded emissions from production and transport of products), have been explicitly excluded because our aim is to produce a product that can be directly compared with atmospheric observations (e.g. Basu et al., 2020; Boon et al., 2016; Bréon et al., 2015; Lauvaux et al., 2020; Sargent et al., 2018).

In the following sections, we describe the methods used to construct the data product, compare it to other global gridded products and suggest potential uses and policy applications.

DATA DESCRIPTION CO₂ emission totals

Auckland Council reported 2016 gross emissions of 11,326 kt CO₂e, of which CO₂ contributed 83.1% (AC2016: Xie, 2019). Auckland's GHG inventory has been compiled in accordance with the Global Protocol for Community-Scale Greenhouse Gas Emission Inventories (GPC: Fong et al., 2021), a well-established international standard for city GHG inventory reporting. CO₂ emissions for some sectors were estimated in Auckland's air emissions inventories (AEI), which were undertaken according to the European emission inventory guidebook (EMEP/EEA guidebook: EEA, 2016). Note that the emissions in Xie (2019) are reported in units of CO₂e in accordance with the GPC protocol and include emissions of other GHG species, whereas Mahuika-AKL includes only CO₂ at present. Some totals have been updated or revised since the 2019 report, which we have incorporated here.

We use the information in AC2016 and the sector-specific air emissions inventories to isolate direct (Scope 1) CO₂ emissions and summarize into sectors (Table 1). The CO₂ emissions totals are for the most part derived from regional activity data and fuel consumption statistics (MBIE, 2022). We use the reported totals for each sector as a constraint to downscale emissions to a finer spatial and temporal scale. Note we have not attempted to independently verify the totals reported in AC2016 and the sector AEIs but rather take them and their uncertainties at face value. For further discussion see Appendix S3.

TABLE 1 Summary of data sources for all sectors and space and time distribution method

Sector	Emissions data source	Spatial distribution	Temporal distribution
On-road transport	AC2016a On-road transportation; AEI Transportb	Road segment, road class (OpenStreetMaps); Auckland Transport/Waka Kotahi NZ Transport traffic counts	Auckland Transport/Waka Kotahi NZ Transport traffic counts
Industrial point sources	AC2016 Industrial Processes and Product Uses, Stationary Energy Manufacturing industries and construction; AEI Industryc	Latitude/longitude coordinates	Even across generic business operating hours
Industrial non-point buildings	AC2016 Stationary Energy Manufacturing industries and construction	AUPd light/heavy industry zones, LINZ building outlinese	Even across generic business operating hours
Commercial non-point buildings	AC2016 Stationary Energy Commercial and Institutional buildings and facilities	AUP business zones, LINZ building outlines	Even across generic business operating hours
Residential non-point buildings (ff)	AC2016 Stationary Energy Residential buildings	NZ Censusf, AUP residential and rural zones, LINZ building outlines	NZ household energy consumption surveyg
Residential non-point buildings (bio)	AEI Home Heatingh	NZ Census, AUP residential and rural zones, LINZ building outlines	Auckland monthly wood fuel consumption
Air transport	AEI Transport	Latitude/longitude coordinates, AUP high-noise zone	Generic domestic flight arrivals and departures
Sea transport	AEI Sea Transporti	Latitude/longitude coordinates (Ports of Auckland), AIS (MarineTraffic)	Even across generic year

^aXie, 2019.

^bSridhar & Metcalfe, 2018.

^cCrimmins, 2018.

^dAuckland Unitary Plan (AUP), Auckland Council, 2022.

^eLand Information New Zealand (LINZ).

^fStats NZ.

^gDortans et al., 2018.

^hMetcalfe et al., 2018.

ⁱPeeters, 2018.

We categorize CO₂ emissions into the following sectors: on-road transport, industrial point source, industrial non-point buildings, commercial non-point buildings, residential non-point buildings, air transport and sea transport. They are largely the same as those defined in Gurney et al. (2020). ‘Non-point’ refers to diffuse area sources (e.g. an entire building or group of buildings), as opposed to point sources that come from a specific, discrete location (e.g. a chimney stack at a manufacturing plant). In the case of the residential and industrial sectors, we have separated fossil fuel CO₂ and biogenic CO₂ (from residential wood burning and industrial waste) because they are typically reported separately and can be distinguished with atmospheric observations (Turnbull et al., 2009). We refer to these as CO₂ff and CO₂bio, respectively.

Off-road transport and railroad sectors are excluded from Mahuika-AKL because these sectors are estimated to contribute only 0.1% of total CO₂ emissions (Xie, 2019). Electricity generation point sources are not considered because there are no fossil fuel electricity generation plants within Auckland Region. We have also excluded emissions from the agriculture, forestry and land use category in AC2016 because most of these emissions originate from non-CO₂ GHGs from sources located in rural areas.

The spatial domain of Mahuika-AKL is the political boundary of Auckland Region (Figure 1), which encompasses the Auckland metropolitan area and surrounding rural areas. A single unitary council oversees this domain, and AC2016 represents the entire domain. This extends to the sea domain as well.

FIGURE 1. (a) Map of New Zealand. Area outlined with orange box shown in (b), with Auckland Region's political boundary outlined in black. Image © Google Earth 2022

The references for the total emissions and the methodology for distributing emissions in space and time for each sector are summarized in Table 1 and described in more detail in Sections. 2.3–2.8. Each sector total is downscaled according to the native resolution of the underlying data for that sector. For the purposes of combining sectors and providing a standardized data product, all emission sectors have been re-gridded to a common 0.5 km² grid in python and are available in netCDF format.

Overall uncertainty

There are two distinct components to the uncertainty in Mahuika-AKL: (a) uncertainty in the total CO₂ emissions in AC2016 and AEIs that derive from the underlying activity data and the emissions factors that were used to convert activity data to CO₂ emissions; and (b) uncertainty from the spatial and temporal disaggregation. Common methodologies for assigning uncertainties in similar studies include comparison to atmospheric inversion solutions and/or global gridded products. For example, Hestia-Indianapolis was found to agree with a whole-city inverse estimate from the INFLUX project (Lauvaux et al., 2016) to within 3.3% (CI: −4.6% to +10.7%; Gurney et al., 2017). This was used to assign a whole-city uncertainty to Hestia-LA, rounded up to 11% (Gurney et al., 2020). Hestia-LA was assigned a grid-scale uncertainty of 25% based on independent electricity production estimates in the United States and uncertainty in CO and CO₂ emissions factors (Gurney et al., 2020). Gately and Hutyra (2017) calculated an uncertainty of ±8.6% for ACES by comparing it to global (EDGAR, ODIAC and FFDAS) and national (Vulcan) gridded products. Other inventories and data products rely on expert judgement to assign uncertainties, or uncertainties are only described qualitatively (e.g. Moran et al., 2022).

Our calculation of uncertainties follows the guidelines in EEA (2016) and IPCC (2006). Uncertainties for sector- and gas species-specific emissions factors are given in MfE (2020). Uncertainty reported here represents a 95% confidence interval (CI). We assume that sources of uncertainty are uncorrelated and combine them according to standard error propagation rules, although we acknowledge that this assumption might not be valid in all cases. Total uncertainty for CO₂ff combined in square quadrature and weighted by emissions amount from each sector is estimated at ±13.4%. This is similar in magnitude to the whole-city uncertainty in Hestia (Gurney et al., 2020). The individual sector uncertainties that make up this estimate and sector-specific disaggregation uncertainties, as far as the available data permit, are described in Appendix S3. Work is underway to compare Mahuika-AKL to atmospheric observations, which will provide an independent measure of its accuracy and uncertainties.

On-road transport

Total CO₂ff from on-road transport was sourced from AC2016 and Sridhar and Metcalfe et al. (2018), which calculated total emissions from motor vehicles using the Auckland Regional Transport model version 3.2 (Auckland Transport). The model output, vehicle kilometres travelled (VKT), was combined with emissions factors and vehicle fleet statistics to arrive at total annual CO₂ff. We have distributed the emissions spatially and temporally using traffic count data and geo-located road segments.

We used vehicle count data from two sources: Auckland Transport (for arterial and local roads) and Waka Kotahi NZ Transport Agency (for state highways). These data comprise 13,410 on-road vehicle counts conducted on 5,432 unique street segments. Each count dataset reports the number of vehicles passing the location during each hour of the monitoring period (typically 7 to 14 days). The hourly counts are also categorized by vehicle speed, direction of travel and the US Federal Highway Administration's (USFHA) 15 vehicle classifications (Hallenbeck et al., 2014). Following Sridhar and Metcalfe et al. (2018), we have reduced the original 15 vehicle classifications into three types: passenger cars, light commercial vehicles (LCV) and heavy commercial vehicles (HCV). We considered vehicle count datasets obtained from July 15, 2012, to November 18, 2019. We excluded counts obtained during 2020 and 2021 in order to omit emissions reductions caused by reduced activity during the COVID-19 pandemic. We distinguish between three street classes: primary, secondary and tertiary. We divide the week into working days (Monday–Friday) and non-working days (Saturdays, Sundays and New Zealand public holidays).

To build an hourly vehicle emissions dataset for all roads in Auckland Region, we (a) geo-located each count dataset to a street segment; (b) aggregated each street segment's count dataset(s) by hour, street type, vehicle class and day type; (c) interpolated the count datasets in time and normalized to 2016 to allow for trends in Auckland traffic volume during the decade that the count datasets span; (d) interpolated the count datasets in space to estimate counts for street segments with no direct observations; (e) apportioned the total vehicle emissions per vehicle type from Sridhar and Metcalfe et al. (2018) to each street segment and each hour in proportion to its contribution to total VKT. These steps are further detailed in Appendix S1. The temporal distribution of emissions by hour for both working and non-working days is shown in Figure 2.

FIGURE 2. On-road transport CO2ff emissions by hour of day and day of week. Weekend = Saturday/Sunday and working day = M–F (except public holidays). Shading represents 95% CIs from bootstrapping all grid cells in Auckland region.

Industrial Industrial CO₂ff point sources

Emissions from industrial point sources are not tabulated individually in AC2016 but rather are included in several aggregate categories: ‘Stationary Energy – Manufacturing industries and construction’ and ‘Industrial processes and product uses’, which include non-point sources as well. These totals were compiled using energy consumption statistics and standard emissions factors (MBIE, 2022). In order to identify individual point sources for Mahuika-AKL, we used information from Crimmins (2018), in which emissions from all industrial businesses with air pollution consent within Auckland Region were tabulated using air pollutant discharge monitoring data from chimney stacks. Resource consent and monitoring is required for air pollutant discharge over permitted thresholds, according to the specific activity rules in the Auckland Unitary Plan (AUP; Auckland Council, 2022). This includes both emissions from industrial processes (e.g. steel manufacture and glass making) and stationary energy combustion. There are 110 individual CO₂ff sources that were included in this inventory. Geographic latitude/longitude coordinates were either provided with the emissions dataset or looked up on Google Maps using the business name and address and checked against aerial imagery where possible. The total annual emissions for each source were allocated to a single point corresponding to this location, and all sources were combined into a geopandas (Jordahl et al., 2021) point feature data frame in python.

Of note is the Glenbrook Steel Mill (New Zealand Steel Limited), which comprises almost 25% of Auckland's total CO₂ff emissions on its own. We have separated emissions from the Glenbrook Steel Mill into its own layer because it is overwhelmingly larger than any other source (larger than the second-greatest point source by a factor of 50), and it is located relatively far from the city centre in a semi-rural area. The layer can easily be re-combined with the other point sources if required.

Annual emissions were distributed evenly in time across assumed business hours of Monday–Friday 7:00 a.m.–7:00 p.m. New Zealand public holidays were excluded. As manufacturing operating hours are considered confidential information, we were not able to refine this further. Seasonal variations were also not taken into account due to the lack of data.

Industrial CO₂bio point sources

CO₂bio emissions from seven individual point sources (wastewater treatment plants and landfill gas engines and flares) were tabulated in Crimmins (2018). These were geo-located and combined as described in Section 2.4.1 to form a separate layer to distinguish them from fossil sources. We have assumed that all of these emissions are biogenic, although a small proportion could be derived from the decomposition of fossil sources (e.g. plastics). Note that we only include direct CO₂ emissions, not CO₂e. Most GHG emissions from waste are methane and nitrous oxide, which are not included in Mahuika-AKL.

Emissions were distributed evenly across all hours of the year, since biogenic sources are likely to be continuously emitting and are not tied to business operating hours or working days. Although there is most likely some seasonal and diurnal variation, we did not attempt to resolve this further due to the lack of data.

Industrial CO₂ff non-point buildings

The CO₂ff emissions reported in AC2016 under the ‘Stationary Energy – Manufacturing industries and construction’ category that are unaccounted for by consented point sources are assumed to originate from non-point area sources (buildings). These largely consist of small-scale heat-generating combustion sources, such as natural gas or diesel-fired boilers and engines, that are not large enough to require resource consent for air discharge (Crimmins, 2018). This amounted to approximately half of the AC2016 total, 450 kt year⁻¹. Annual total emissions were distributed spatially in proportion to the area of building outlines captured from aerial imagery (Golubiewski et al., 2019; LINZ, 2020) in light and heavy industry-zoned areas (Auckland Council, 2022) and allocated to the polygons associated with each building in a geopandas polygon feature data frame. (The methodology for allocating emissions to building outlines is described in more detail in Appendix S2). We acknowledge that this almost certainly misses some areas where the emissions occur, as construction activity can happen in many different zone types and is not necessarily associated with existing buildings. However, without any further information, we assume that the majority of emissions do occur in existing industrial areas and conservatively assign a large uncertainty (see Appendix S3).

Similar to the industrial CO₂ff point sources, emissions were distributed in time evenly across generic business hours of Monday–Friday 7:00 a.m.–7:00 p.m., excluding public holidays.

Commercial CO₂ff non-point buildings

Total CO₂ff emissions for the commercial sector are taken directly from AC2016 under the ‘Stationary Energy – Commercial and Institutional Buildings and Facilities’ category. The total is based on fuel consumption statistics for the sector and standard emissions factors (MBIE, 2022). CO₂ emissions from natural gas consumption account for over 80% of the total for this sector. Emissions were spatially distributed in proportion to the area of building outlines within areas zoned for business (Auckland Council, 2022) and allocated to the building polygons in a geopandas polygon feature dataframe, as was done with industrial non-point sources (Section 2.4.3 and Appendix S2). Research has shown that commercial energy use is correlated to floor area (Amitrano et al., 2014), making the building outline area a reasonable first-order proxy for CO₂ emissions, although we acknowledge that it is not always representative of the floor area of, for example, multi-story office space.

Emissions were distributed evenly in time between assumed business hours of 7:00 a.m.–9:00 p.m., 7 days a week. Although the commercial sector encompasses a diverse range of activities, we have assumed that many businesses that are large users of natural gas, such as restaurants and hospitality, operate during mornings and nights beyond standard business hours of Monday–Friday 9:00 a.m.–5:00 p.m. For the same reason, no distinction between weekdays, weekends and public holidays has been made.

Residential Residential CO₂ff non-point buildings

The CO₂ff emissions total for the residential sector is taken directly from that reported in AC2016 under the ‘Stationary Energy – Residential Buildings’ category. This emissions estimate is based on fuel consumption data from the sector (MBIE, 2022) and includes direct emissions from the combustion of fossil fuel (mostly natural gas) for residential energy needs such as heating and cooking. Note that it does not include emissions associated with electricity supplied by the grid, as these are Scope 2 emissions.

The CO₂ff emissions total was first distributed spatially in proportion to population reported in the New Zealand 2013 and 2018 Census ‘usually resident’ population count for each Statistical Area (SA1), the smallest spatial unit for reporting census data (Stats NZ, 2020). The SA1 boundaries are publicly available as a polygon feature shapefile (Stats NZ, 2020). To obtain a population count for 2016, a year in which there was no formal census, we linearly interpolated the counts from 2013 and 2018. Other studies have shown a high correlation and a nearly linear relationship between population and residential sector CO₂ff emissions (e.g. Patarasuk et al., 2016), indicating that this is a reasonable first-order estimation. Further spatial refinement within each SA1 was done using the building outlines dataset described in Section 2.4.3 (LINZ, 2020). Emissions assigned to each SA1 were divided among building polygons located in residential and rural zoned areas (Auckland Council, 2022) in proportion to the area of the building outline. The results were combined into a geopandas polygon feature data frame. When overlaying the SA1s with the building outlines, approximately 4% of the recorded population counts occurred in SA1s that did not contain any building outlines. These emissions were redistributed evenly over the building outlines in the rest of the SA1s. This slightly biases the spatial emissions distribution towards older housing developments, and we have added this to the uncertainty (see Appendix S3). For more detail on the methodology and an example calculation, refer to Appendix S2.

Emissions were allocated in time based on a recent report of residential household energy demand in New Zealand (Dortans et al., 2018), which provides data on seasonal and diurnal variation in household energy usage. The dataset largely consists of electricity usage, and while it is a reasonable proxy for household CO₂ff emission temporal patterns, it might not accurately represent other fuel usage such as natural gas, particularly since gas is often connected to specific appliances that do not have the same seasonal usage pattern (e.g. a gas oven or water heater would be used consistently year-round, as opposed to a space heater which would be in use in winter only). Nonetheless, this is the best available information on seasonal and diurnal variations in energy demand for New Zealand at the time of writing (note that it is not specific to Auckland). Seasonal variations are shown in Table 2 and Figure 3; total emissions for the year were first divided into seasons and then allocated each hour of the day according to the proportions shown. Emissions peak in winter and are at a minimum in summer, with average emissions in winter approximately five times those in summer (Table 2), since winter heating is a substantial source and summer air conditioning is not widely used. The daily emissions are broken into four time intervals (Table 2 and Figure 3), corresponding to the peak and off-peak intervals defined in Dortans et al. (2018). Emissions are highest in the morning and evening intervals, when most people are at home and active, following intuitive patterns. No distinction between weekdays and weekends was made due to the absence of underlying data, although there are likely to be differences in a typical household, with a more even distribution of emissions throughout the day on weekends.

TABLE 2 Seasonal and diurnal fractional distribution of CO₂ emissions for residential sector, based on Dortans et al. (2018)

Season	Annual	10:00 p.m.–6:00 a.m.	6:00 a.m.–10:00 a.m.	10:00 a.m.–5:00 p.m.	5:00 p.m.–10:00 p.m.
Spring (SON)	0.223	0.091	0.424	0.121	0.364
Summer (DJF)	0.088	0.167	.0.167	0.222	0.444
Autumn (MAM)	0.196	0.103	0.414	0.138	0.345
Winter (JJA)	0.493	0.068	0.438	0.11	0.384

Abbreviations: DJF, December–January–February (austral summer); JJA, June–July–August (austral winter); MAM, March–April–May (austral autumn); SON, September–October–November (austral spring).

FIGURE 3. Residential sector CO2ff diurnal and seasonal cycle

Residential CO₂bio non-point buildings (wood burning)

A separate layer was constructed for CO₂ emissions from wood burning (CO₂ bio), which may be treated differently in mitigation efforts since the underlying fuel is nominally renewable. New Zealand has a high proportion of older dwellings with wood burners and/or fireplaces; the 2013 New Zealand census reported 21% of households in Auckland Region using wood for home heating (Metcalfe et al., 2018), and thus, CO₂ emitted from wood burning used to heat homes is significant. Auckland's total CO₂bio emissions from wood burning for 2016 were estimated in Metcalfe et al. (2018) using wood fuel consumed, household heating surveys and corresponding emissions factors. Fuel consumption rates come from regular home heating surveys of Auckland Region (Auckland Council, 2014; Auckland Regional Council, 2010).

Spatial disaggregation was performed using the New Zealand 2013 and 2018 Censuses (Stats NZ, 2020), which contain a survey question that asks what fuel types are used to heat occupied, private dwellings (choices are electricity, gas, wood, coal, home heating oil, solar power, no fuel and other; selection of multiple items is possible). We distributed the total CO₂ emissions proportionally to each SA1 according to the number of households that reported using wood as a fuel type and then further by area of building outlines, as was done for CO₂ff (see Section 2.6.1). Population was not used as a proxy in this case because we judged population to be a poor predictor of wood burning, that is the most densely populated urban areas are more likely to consist of newer, multi-household dwellings (apartments and townhouses) that do not rely on wood burning for heating. The New Zealand Census data support our assumption that wood use is more common in sparsely populated, semi-rural areas. For more detail see Appendix S2.

Emissions were varied temporally according to reported monthly percentage of annual wood and pellet fuel consumption given in Metcalfe et al. (2018), peaking in mid-winter and falling to zero in the summer months (Table 3). Note that these numbers are based on a home heating emissions inventory from 2002 (Wilton, 2002), and its accuracy for more recent years is unknown. The diurnal variation follows the proportions listed under summer in Table 2, with a single peak in the evening, since this agrees with anecdotal evidence that people are more likely to light fires in the evening when they are home for the night, and less likely in the morning before they leave for work or school. However, this estimate would benefit from more data.

TABLE 3 Monthly distribution of CO₂bio emissions for residential sector from wood burning, based on Metcalfe et al. (2018)

	Jan	Feb	Mar	Apr	May	Jun	Jul	Aug	Sep	Oct	Nov	Dec
Percentage of annual fuel consumption (%)	0.0	0.0	1.0	1.0	10.5	22.0	30.0	28.0	6.5	1.0	0.0	0.0

Air transport CO₂ff

The total annual CO₂ff emissions in 2016 for the aviation sector is taken directly from Sridhar and Metcalfe et al. (2018). Note the total was not taken from AC2016 because it includes only the portion of emissions from departing flights attributable to Auckland residents (here defined as Scope 3 emissions rather than Scope 1). Sridhar and Metcalfe et al. (2018) include all activity from arriving and departing flights below 3,000 ft (taxi, take-off, initial climb out, final approach and landing; European Environmental Agency (EEA), 2016). The total includes only activity from Auckland International Airport, as it is by far the largest airport in the region and accounts for approximately 97% of emissions (Sridhar & Metcalfe, 2018). Two smaller airports, Ardmore Airport and North Shore Aerodome, were excluded because they service smaller aircraft and account for a very small proportion of total emissions. The process for going from activity data to CO₂ emissions follows the ‘Tier 2’ method from the EMEP/EEA guidebook (EEA, 2016), using information on the landing and take-off phases and emissions factors for each aircraft type. The full methodology and underlying data are described in Sridhar and Metcalfe et al. (2018).

Emissions were allocated in space using Auckland Council's designated ‘high-noise’ zone over the main runways at Auckland Airport (Figure 5 inset), which aligns with typical take-off and landing patterns. It is available as a polygon feature shapefile as part of Auckland Council's Unitary Plan (Auckland Council, 2022). The northern-most portion of the zone was removed because it describes a planned expansion of the runway and was not in use in 2016.

Emissions were allocated evenly through time using normal domestic flight operation hours, 6:00 a.m.–10:00 p.m. The majority of flights arrive and depart during these hours, but we acknowledge that this misses some late-night international departures, which are typically large, long-haul aircraft. We did not have any information on typical flight schedules in 2016 and aircraft departures and arrivals to be able to translate daily emissions to a diurnal or seasonal pattern, so we did not attempt to distinguish between weekdays, weekends and holidays or peak hours of the day.

Sea transport CO₂ff

Sea transport CO₂ff emissions for Mahuika-AKL were adapted from Peeters (2018) and GIS shapefiles provided by S. Peeters (pers. comm.). All ocean-going vessel (OGV) tracks were geo-located using 2016 Automatic Identification System (AIS) data, supplied by MarineTraffic and purchased by Auckland Council. At-berth emissions were derived from the Ports of Auckland Ltd recorded hostelling hours. Emissions from fishing activity were processed separately and based on data from the Ministry for Primary Industries. It does not include emissions from military vessels, recreational vessels and dredging activity due to lack of data; however, these vessels make up a small proportion of overall sea traffic in Auckland.

Emissions were separated into vessels at berth (point locations) and moving OGV (line segments). Emissions for each event were calculated by assuming the fuel type used in each vessel category and multiplying by an emissions factor determined from a 2002 Entec study for the European Commission (Entec, 2002; USEPA, 2009).

Each emission event was marked with a timestamp specific to the year of the inventory (2016). For the purposes of this data product, we have redistributed the emissions evenly across all hours of each day. This smooths spikes inherent in the nature of sea transport and will not accurately portray the magnitude of emissions for a specific time. However, we feel it is better not to tie Mahuika-AKL to the arbitrary timing of events in a particular year in order to be more generally applicable.

RESULTS Mahuika-Auckland CO₂ff and CO₂bio emissions dataset

Total annual CO₂ff for all sectors (except sea transport) for the base year of 2016 is shown in Figure 4. CO₂ff and CO₂bio total annual emissions for each sector are shown in Figures 5–10 and Table 4. The dataset, divided into six sectors, is available at https://doi.org/10.5281/zenodo.6395226 (Keller et al., 2022) as a set of 18 netCDF files at 500 m spatial resolution and annual and hourly temporal resolution.

FIGURE 4. Total CO2ff annual emissions on a 500 m grid from all sectors, excluding sea transport (left out for visual clarity). Note log scale

FIGURE 5. On-road sector total CO2ff annual emissions on a 500 m grid

FIGURE 6. Industrial sector total CO2ff (a) and CO2bio (b) annual emissions. Glenbrook steel mill is marked with a blue circle in (a). Note log scale. CO2ff is shown on a 500 m grid. CO2bio shown on a 1.5 km grid for visual clarity.

FIGURE 7. Commercial sector total CO2ff annual emissions on a 500 m grid

FIGURE 8. Residential sector total CO2ff (a) and CO2bio (b) annual emissions on a 500 m grid

FIGURE 9. Air transport sector total CO2ff annual emissions on a 500 m grid. Inset: ‘High-noise’ zone over Auckland International Airport. The area outlined in red was used as the area to distribute emissions. The northern part of the shape is a planned runway extension and was not in use in 2016.

FIGURE 10. Sea transport total CO2ff annual emissions on a 500 m grid. Note the log scale and the expanded area to include all of the Auckland Region sea domain.

TABLE 4 CO₂ emissions totals for 2016 in Auckland region, by sector

Sector	CO2ff (kt)	Percent of total (%)	CO2bio (kt)	Percent of total (%)
On-road transport	3,183.0	44.5
Industrial non-point buildings	478.5	6.7
Industrial point sourcesa	494.2	6.9	78.3	22.11
Glenbrook Steel Mill	1,770.8	24.8
Commercial non-point buildings	421.0	5.9
Residential non-point buildings	211.1	3.0	276.0	77.89
Air transport	455.3	6.4
Sea transport	134.5	1.9
Total	7,148.3		354.3

Note: CO₂ff point source emissions from the industrial sector exclude Glenbrook steel mill, which we have separated into its own category because it dominates the sector and is well outside the Auckland City Centre.

^aExcludes Glenbrook Steel Mill.

Most of the emissions are concentrated in the Auckland Central Business District (CBD), the motorways and industrial zones on the outskirts of the city centre. The on-road transport and industrial sectors dominate overall spatial patterns, since together they represent 83% of CO₂ff emissions in Auckland Region. Emissions from residential and commercial buildings are dispersed widely and make up only 3% and 6% of total emissions, respectively, so there are many grid cells throughout the region with relatively small emissions. The grid-cell distribution of total annual land-based emissions is shown in Figure 11. The box and whisker plot in Figure 12 shows median and first–third quartile range of emissions by grid cell. The distribution of all sectors shows the contribution of two distinct clusters: one at the lower end of the scale originating mostly from the residential sector, and the other centred at a higher value associated with on-road transport and industrial sources. The industrial sector has the widest spread of values (Figure 12 and Figure S5), reflecting the diversity of industrial sources and the emission-intensive processes that occur at some sites. The industrial, on-road, and sea transport sectors also include a considerable number of outliers (not shown) at the top end of the range, demonstrating the disproportionately high amount of emissions coming from a few point sources, motorways and ports. More detail is available in Appendix S4.

FIGURE 11. Histogram of grid-cell distribution of annual CO2ff emissions for all land-based sectors. Note log scale on the x-axis

FIGURE 12. Box and whisker plot of grid-cell distribution of annual CO2 emissions. Orange line shows the median. The box extends from the first quartile (Q1) to the third quartile (Q3) of the data, and whiskers extend to 1.5× the interquartile range. (Data points / outliers outside of this range not shown for visual clarity). Inset shows residential and sea distributions zoomed in. (Air transport not shown because there are too few grid cells).

Evaluation: Comparison to global products

One means of evaluating Mahuika-AKL is to compare it to similar, global gridded data products. Figure 13 shows four different global CO₂ff emissions products for the Auckland domain: ODIAC, EDGAR, PKU-CO2 and FFDAS (Crippa et al., 2020; Oda & Maksyutov, 2015; Rayner & Gurney, 2014; Wang et al., 2013). Comparison with Mahuika-AKL (Figure 4) reveals that the finer spatial resolution allows for more realistic spatial gradients and direct attribution of emissions to the originating processes and structures. We have left EDGAR, PKU-CO2 and FFDAS at their native resolution (0.1^°) to highlight the difference; all are too coarse to even resolve the coastline of New Zealand, effectively placing some land-based emissions over the ocean.

FIGURE 13. CO2ff emissions for Auckland region from global products: ODIAC (a), EDGAR (b), PKU (c) and FFDAS (d). ODIAC is shown on a 1 km2 grid and has been masked to the same land region as Mahuika-AKL, while EDGAR, FFDAS and PKU are shown in their native 0.1° resolution (unmasked). EDGAR, FFDAS and PKU shown on the same colour scale; ODIAC is shown on a different scale to reflect the higher resolution and smaller per-grid-cell emissions

ODIAC has a comparable spatial resolution (1 km²) to Mahuika-AKL. Summing the 2016 ODIAC dataset over the Auckland Region domain gives total annual emissions of 1,831 kt CO₂ff. Compared with the Mahuika-AKL annual total of 6,563 CO₂ff (excluding sea and air transport, which are not included in ODIAC; Table 4), this is 72% lower. We note that ODIAC does not contain specific information about Auckland's industrial point sources (such as Glenbrook Steel Mill) because industrial sources are not included in the global power plant point source database used to construct ODIAC. If we additionally exclude Glenbrook Steel Mill from Mahuika-AKL, giving a total of 4,792 kt CO₂ff year⁻¹, the ODIAC total is still 62% lower and well outside Mahuika-AKL's uncertainty range of ±13.4%. From visual inspection of Figure 13, ODIAC emissions are spread diffusely over the whole urban area rather than being assigned to discrete structures (e.g. roads and buildings), as we have done with Mahuika-AKL. The grid-cell distribution of ODIAC (re-gridded to 500 m resolution; Figure S6) is much more strongly centred around a mid-range value and does not contain as many outliers at either the low or high end of the scale as in Mahuika-AKL (Figure 11). The mean grid-cell value of Mahuika-AKL is around two-thirds larger (480 t year⁻¹) than ODIAC (293 t year⁻¹; Table S2), whereas the medians are much closer together (43 vs. 55 t year⁻¹, respectively), again indicating that Mahuika-AKL contains more outliers at the high end of the scale. The standard deviation of Mahuika-AKL (15,194 t year⁻¹) is also much larger than ODIAC (645 t year⁻¹), underscoring the fact that ODIAC does not capture the overall spatial heterogeneity of the emissions or the sharp gradient between adjoining cells. A cell-by-cell spatial comparison of the two products (Figure 14) reveals that the largest discrepancies between ODIAC and Mahuika-AKL are at locations of large point sources and roads. We judge Mahuika-AKL to be in general more accurate than ODIAC because we are able to resolve the structures where emissions actually occur and to include local knowledge of point sources and infrastructure. It is likely that ODIAC's main proxy (nightlights) misses some sources that are not illuminated at night, which explains the large underestimate compared with Mahuika-AKL.

FIGURE 14. Cell-by-cell comparison of Mahuika-AKL and ODIAC annual CO2ff total. ODIAC has been re-gridded onto Mahuika-AKL's 500 m grid. Colour scale shows absolute difference (Mahuika-AKL–ODIAC) in t year−1. Some grid cells exceed the maximum on the colour scale. Note that the total for Mahuika-AKL shown here does not include air and sea transport, in order to more closely match ODIAC's definition.

We do not attempt a quantitative comparison between Mahuika-AKL and EDGAR, PKU-CO2 or FFDAS, as the spatial and temporal resolution, sector definitions and underlying datasets are very different, and going into these details is beyond the scope of this paper. However, even a qualitative comparison shows the potential benefit of utilizing local information and datasets to disaggregate to the sector-specific and finer spatiotemporal scale of Mahuika-AKL.

CONCLUSIONS AND POTENTIAL USAGE

We have developed a spatiotemporally resolved CO₂ emissions data product for Auckland Region for six sectors at an hourly, 500 m resolution, a level of detail not previously available for any New Zealand city. The methodology builds upon well-established protocols for GHG inventories and uses activity data and proxies to downscale the regional inventory in space and time to the individual point source, building and road level. The code to generate Mahuika-AKL has been designed to be flexible so that updated information and/or data from more recent years can easily be incorporated.

This product was designed first and foremost for use as a prior estimate of CO₂ emissions for a city-scale atmospheric inversion. When combined with atmospheric observations, the dataset will allow us to verify the emissions totals and assumptions that went into the regional inventory (such as emissions factors). We have collected flask measurements at approximately 30 sites around Auckland over a period of four years, beginning in 2017, for which we have measured total CO₂ and ¹⁴CO₂ (Young, 2022). We also have a growing in-situ network, with four sites currently active (Brailsford et al., 2021). These observations, in combination with an atmospheric transport model, will provide spatiotemporal information that can be used to refine the bottom-up estimates and fill in some of the data gaps that currently exist, especially in temporal distributions.

This data product can also assist in the development of targeted policies for emissions reduction and monitoring progress towards reduction goals (e.g. Lauvaux et al., 2020; Patarasuk et al., 2016). Auckland's goal of net-zero carbon by 2050 (Auckland Council, 2020) will require dedicated, sustained emissions reductions across a broad range of activities; specific information on sectors and spatial distribution of sources that Mahuika-AKL provides can help set priorities and inform decisions on where investment might be most cost-effective or could make the most difference. Mahuika-AKL can provide information useful for urban planning, such as how the layout of the city affects GHG emissions, quantifying the effect of, for example, housing density and distance to commercial services on emissions intensity (Patarasuk et al., 2016). Recognizing that not all local governments have the resources to compile a detailed regional emissions inventory like Auckland, we plan to expand this product to all of New Zealand so that all cities and urban areas can access a CO₂ emissions inventory compiled using consistent, ground-truthed methodology. We are working closely with local and national government agencies to ensure that Mahuika-AKL and future extensions are policy relevant.

AUTHOR CONTRIBUTIONS

Elizabeth D. Keller: Conceptualization (equal); data curation (lead); formal analysis (lead); methodology (equal); software (equal); supervision (lead); visualization (equal); writing – original draft (lead); writing – review and editing (lead). Timothy W. Hilton: Conceptualization (supporting); data curation (supporting); formal analysis (equal); methodology (equal); software (equal); visualization (supporting); writing – original draft (supporting); writing – review and editing (supporting). Adrian Benson: Data curation (supporting); methodology (supporting); software (equal); writing – review and editing (supporting). Sapthala Karalliyadda: Data curation (supporting); methodology (supporting); software (supporting); writing – review and editing (supporting). Shanju Xie: Data curation (supporting); formal analysis (supporting); methodology (supporting); writing – review and editing (supporting). Kevin R. Gurney: Conceptualization (equal); methodology (equal). Jocelyn C. Turnbull: Conceptualization (equal); funding acquisition (lead); project administration (lead); writing – original draft (supporting); writing – review and editing (supporting).

ACKNOWLEDGEMENTS

We would like to thank Auckland Council and Auckland Transport for their generous provision of data for this work. We thank Jeremy Parry-Thompson for suggesting the name ‘Mahuika’ for the data product. The name was chosen to both pay homage to the Hestia project and to reflect New Zealand's culture and values. We would also like to thank two anonymous reviewers whose comments have improved this manuscript.

FUNDING INFORMATION

Funding for this work was provided by the Government of New Zealand (CarbonWatch-NZ: MBIE Endeavour Research Programme #C01X1817).

CONFLICT OF INTEREST

The authors declare that they have no conflict of interest.

OPEN RESEARCH BADGES

This article has earned an Open Data badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at [Doi: https://doi.org/10.5281/zenodo.6395226]. Learn more about the Open Practices badges from the Center for Open Science: https://osf.io/tvyxz/wiki

DATA AVAILABILITY STATEMENT

The Mahuika-Auckland v1.0 data product is available at https://doi.org/10.5281/zenodo.6395226 (Keller et al., 2022). Auckland's GHG and air emissions inventories available from Auckland Council: https://www.aucklandcouncil.govt.nz/ (last access: March 6, 2022). New Zealand building outlines dataset available from Land Information New Zealand (LINZ) at https://data.linz.govt.nz/layer/101290-nz-building-outlines/ (last access: January 4, 2022). Auckland Council's Unitary Plan and associated maps are available at https://www.aucklandcouncil.govt.nz/plans-projects-policies-reports-bylaws/our-plans-strategies/unitary-plan/Pages/default.aspx (last access: January 4, 2022). Map data copyrighted OpenStreetMap contributors and available from https://www.openstreetmap.org (last access: March 20, 2022). The New Zealand census dataset and SA1 boundaries are available from Stats NZ: https://www.stats.govt.nz/topics/census (last access: March 20, 2022).

CODE AVAILABILITY

Python code used to construct the model is on a private GitHub repository and is available on request. We analysed the emissions data using the OSMnx (Boeing, 2017), geopandas (Jordahl et al., 2021), pandas (Reback et al., 2021), pyKrige (Murphy et al., 2021) and scipy.stats.bootstrap (Virtanen et al., 2020) packages for Python, all freely available.