This project is focused on developing baseline indicators and related maps for several Cities4Forests cities. Selected indicators are in five themes related to urban greenspaces and trees:

1. Health: heat (GRE-1)
2. Health: air quality (GRE-2)
3. Greenspace access (GRE-3)
4. Flooding (GRE-4)
5. Climate change mitigation (GRE-5)

Theme	Indicator core code	Indicator name	Related indicators	Status	Notebook
Heat	GRE-1.1	Expected extreme heat event hazard (expected days above 35C in 2050) and trend		Production	link to notebook
Heat	GRE-1.2	% built land with high (3C+ above built mean) land surface temperature during hot season		Production	link to notebook
Heat	GRE-1.3	% built land with low (below 0.2 albedo) surface reflectivity		Production	link to notebook
Heat	GRE-1.4	% built land without tree cover	SDG 15.1.1, SICB-11	Production	link to notebook
Air quality	GRE-2.1	Level and change in air pollutant emissions by pollutant and sector	SDG 3.9.1	Production	link to notebook
Air quality	GRE-2.2	# of days air pollutants are above WHO standards	SDG 3.9.1	Production	link to notebook
Air quality	GRE-2.3	% population exposed to annual average PM 2.5 concentrations above WHO standards	SDG 3.9.1	Production	link to notebook
Greenspace access	GRE-3.1	% built-up area that is open space for public use	SDG 11.7.1	Production	link to notebook
Greenspace access	GRE-3.2	% population with access to public open space within walking distance (400m)		Production	link to notebook
Greenspace access	GRE-3.3	% population with threshold level (10%+) of tree cover within walking distance (400m)		Production	link to notebook
Flooding	GRE-4.1	% built land cover exposed to coastal or riverine flooding and trend		Production	link to notebook
Flooding	GRE-4.2	Expected extreme precipitation event hazard (expected maximum mm precipitation in one day in 2050) and trend		Production	link to notebook
Flooding	GRE-4.3	% built land cover within less than 10m height above nearest drainage		Production	link to notebook
Flooding	GRE-4.4	% built land cover with impervious surfaces		Production	link to notebook
Flooding	GRE-4.5	% built land cover without vegetation cover	SDG 6.6.1	Production	link to notebook
Flooding	GRE-4.6	% riparian zone with vegetation cover		Production	link to notebook
Flooding	GRE-4.7	% steep slopes without vegetation cover		Production	link to notebook
Climate mitigation	GRE-5.1	Level and share of GHG emissions by pollutant and sector		Production	link to notebook
Climate mitigation	GRE-5.2	Average carbon flux from trees per hectare of city area		Production	link to notebook

Expected extreme heat event hazard (expected days above 35C in 2050) and trend (% change between 2020 and 2050 in number of days meeting 35C threshold )

Understanding the hazard presented by extreme heat and the likely future change of that hazard can inform the importance of planning for an extreme heat event and investing in infrastructure--like trees--that can help mitigate its impacts.

This indicator uses data from the NEX-GDDP climate projections retrieved from Earth Engine (available at the scale of 0.25 degrees, or approximately 28 km), a given location, and two selected years to calculate the expected number of days that have a high temperature of 35 Celsius or greater for each year and the percentage change between years. Methods are further documented in the Data Portal hazards repo.

Percent of built land with high land surface temperature during hot season (3C+ above mean for built land)

Land surface temperature is closely correlated with near surface air temperature, which is how people experience heat. This metric can identify areas of the city exposed to above average heat. Such areas may be good candidates for heat mitigation measures such as tree planting or solar reflective surfaces.

This indicator uses land surface temperature (LST) for each pixel in the area of interest calculated using the methods from Ermida et al. 2020 and Landsat imagery, at 30 meter resolution, as retrieved from Earth Engine. Mean LST is calculated from a mosaic of cloud masked Landsat images from years 2013-2022 selected from the month of the year that had the hottest day during this period as determined from ERA5 DAILY. Average and pixelwise LST is then retrieved for built land cover areas using the built-up class from ESA WorldCover as a mask. Finally, areas that are 3 degrees Celsius or more above the area average are masked to calculate the percent of built area with high land surface temperature.

There are limitations to these methods and uncertainty regarding the resulting indicator values. These calculations are based fully on remotely sensed data, which may not full capture the reality on the ground. Notably, land surface temperature is not a direct measure of how people in cities experience heat, which is more closely related to near-surface air temperature and heat indices. Additionally this indicator uses measurements from the times of day at which Landsat retrieves images, usually within a few hours of mid-day locally. As a result, this indicator doesn't measure heat in the evening or nighttime, which is important for understanding local heat impacts. Finally, as this indicator uses a image mosaic to remove cloud cover and provide an average for the hot season in the location, it does not reflect any one year, day or a heat event.

Percent of built land with low surface reflectivity (below 0.2 albedo)

This indicator can identify areas of the city with a high share of surfaces that retain excess heat. Surfaces with Low solar reflectivity (albedo) absorb heat and transfer it to immediate surroundings. Areas with a high share of low albedo surfaces may be candidates in which to install measures--such as solar reflective roofs and pavements, or trees--that will reduce the heat retained on surfaces or otherwise cool the immediate area.

This indicator uses pixelwise albedo values derived from Sentinel-1, at 10 meter resolution, as retrieved from Earth Engine using the algorithms defined in Bonafoni & Sekertekin 2020. Annual mean albedo is calculated from cloud-free pixels from 2021. Values for built land cover areas only are derived using the built-up class from the 10-meter resolution ESA WorldCover dataset as a mask. Finally, pixels with albedo values below 0.2 are masked to calculate the percent of built area with low surface reflectivity.

There are limitations to these methods and uncertainty regarding the resulting indicator values. These calculations are based fully on remotely sensed data, which is subject to atmospheric interfere and other challenges in data collection and may not full capture the reality on the ground. As this indicator uses a image mosaic to remove cloud cover and provide an annual average, it does not reflect the situation in any one point in time.

Percent of built land without tree cover

This indicator can identify areas of a city without significant tree cover and therefore lacking in shade and evapotranspiration that can reduce local heat. Areas with low tree cover may be candidates in which to install measures--such as trees, other vegetation, or solar reflective roofs and pavements--that will reduce the heat retained locally or otherwise cool the immediate area.

This indicator uses 10-meter resolution tree cover data for 2020 from the Trees in Mosaic Landscapes dataset as retrieved from Earth Engine. Tree cover for each pixel of built land is derived using the built-up class from ESA WorldCover as a mask of non-zero tree cover pixels. The number of built pixels with tree cover and all built pixels are counted and divided to calculate percent of built land with tree cover. This value for percent tree cover is then inverted to derive a value for percent of built-up land without tree cover.

There are limitations to these methods and uncertainty regarding the resulting indicator values. Limitations and uncertainty related to the tree cover dataset is described in the methods paper for Trees in Mosaic Landscapes.

Level of air pollutant emissions by pollutant species and economic sector

Human activity contributes to air pollution and climate change through emission of gases from fuel combustion, industrial processes, and agriculture. This indicator can help decision-makers and stakeholders identify the most important pollutants emitted locally, the activities responsible for the emissions, and, with multiple years of data, emissions trends.

This indicator is based on the CAMS Global Anthropogenic Emissions dataset of the Copernicus Atmosphere Monitoring Service (CAMS) and ECCAD (Granier et al. 2019). The dataset provides estimates of emissions from 12 sectors of human activity, on a 0.1-degree (approx. 11 km) spatial resolution. The estimates are based on simulations and historical data. The included sectors are: agriculture (livestock); agriculture (soils); agriculture (waste burning); power generation; fugitive emissions; industry; combustion in residential, commercial, and other settings; ships; solvents; solid waste and wastewater; off-road transportation; and on-road transportation.

We used Google Earth Engine for spatial subsetting against cities' administrative boundaries. We report annual, sector-disaggregated emissions for the years 2000 and 2020 of the each pollutant species. We report all emissions as tonnes/year.

In order to allow comparison of emissions of different pollutant species, we convert tonnes to USD based on social costs per tonne as estimated by Shindell 2015 for all pollutant species except non-methane volatile organic compounds, or NMVOC. For the social cost of NMVOC, we used the median value for NMVOC emitted below 100m in van der Kamp 2017. Van de Kamp's estimates are in 2015 euros, which we converted to 2015 USD at 1 EUR = 1.11 USD.

There are limitations to these methods and uncertainty regarding the resulting indicator values. Importantly, the emissions data used for this indicator only account for direct emission from activities within the boundaries of the city (Scope 1). This data does not account for emissions associated with electricity used in the city but generated elsewhere (Scope 2) or emissions produced elsewhere associated with products or services consumed in the city (Scope 3). The CAMS dataset is modeled data based on an ensemble of multiple emissions models and is subject to the limitation of those models. Description of the methods used to develop the CAMS emissions dataset is described here.

Annual number of days air pollutants are above WHO air quality standards

Exposure to high concentrations of air pollutants increases the probability of developing serious health conditions, reduced lung function, increased susceptibility to respiratory infections, and aggravated asthma. Long-term exposure increases the probability of developing chronic conditions like stroke-susceptibility, heart disease, and cancer. This indicator can help public health officials determine which air pollutants are present at levels dangerous to human health, and for how many days each year the population is exposed to them.

Our data come from the CAMS Global Reanalysis EAC4 dataset of the Copernicus Atmosphere Monitoring Service (CAMS), which combines satellite monitoring of pollutant concentrations with atmospheric modeling to estimate concentrations near the earth's surface. The EAC4 data are provided at approximately 80 km spatial resolution.

We report the number of days each city was estimated in 2020 to have a near-surface concentration of air pollutants that exceeds the World Health Organization's 2021 standards for outdoor air pollutants. The pollutants and standards are given in this table.

Pollutant species	WHO threshold
Nitrogen dioxide (NO2)	25 µg/m3 24-hour average
Sulfur dioxide (SO2)	40 µg/m3 24-hour average
Ozone (O3)	100 µg/m3 8-hour average*
Carbon monoxide (CO)	4 mg/m3 24-hour average
Fine particulate matter (PM2.5)	15 µg/m3 24-hour average
Coarse particulate matter (PM10)	45 µg/m3 24-hour average
* For the ozone calculations, we used data for the 9-hour window 6am-3pm

There are limitations to these methods and uncertainty regarding the resulting indicator values. The CAMS data has uncertainty related to the limitations of atmospheric modeling methods used. Additionally, the low resolution of the dataset does not allow for analysis of sub-city geographic units, and may introduce uncertainty to the city-scale calculations due to aggregation of data from an area larger than the city.

Annual average fine particulate matter (PM 2.5) concentration as percent of WHO standard

Fine particulate matter (PM_2.5) consists of very small particles that are smaller than 2.5 µm in diameter. (A typical human hair has a diameter of 50-70 µm.) PM_2.5 are released in combustion (including fuel combustion in vehicles and power plants, as well as naturally occurring fires), degradation of vehicle tires during use, and some chemical processes that occur in the atmosphere. PM_2.5 is a serious health concern, because the small size allows PM_2.5 particles to travel deep into human lungs, enter the bloodstream, and cause direct damage to internal organs. Long-term exposure to PM_2.5 leads to increased incidence of heart and lung diseases, cancers, and premature death. This indicator can help decision-makers determine how many of a city's residents face long-term exposure to dangerous levels of PM_2.5, and in which neighborhoods these residents live.

We used the global surface PM_2.5 V5.GL.02 dataset from the Atmospheric Composition Analysis Group at Washington University (van Donkelaar et al. 2021). This dataset combines models of atmospheric mixing and chemistry with analysis of imagery from the NASA MODIS, MISR, and SeaWIFS satellite instruments to generate estimates of PM_2.5 concentrations near the earth's surface. The data are provided at a spatial resolution of 0.01 degrees, or approximately 1.1 km. Our indicator is based on annual average concentrations for 2020.

We report each district's 2020 average PM_2.5 concentration as a percentage of the WHO standard. (The average is the annual average, averaged spatially over the area of the district.) For example, an average concentration of 15 µg/m³ would be reported as 300% of the WHO standard.

There are limitations to these methods and uncertainty regarding the resulting indicator values. The PM_2.5 dataset has uncertainty related to the limitations of atmospheric modeling methods used.

Percent of built-up area that is open space for public use

Availability and area of public open space, such as parks, is a key indicator to assess quality of life for residents of a city. Open spaces provide recreation opportunities and can provide ecosystem services and habitat. Indicators of open space are included in the Sustainable Development Goals (Indicator 11.7.1) and the Singapore Index on Cities' Biodiversity.

This indicator uses polygon data on categories of open space as retrieved from OpenStreetMap in August 2022. The OpenStreetMap tags used to retrieve these areas are 'park', 'nature_reserve', 'common', 'playground', 'pitch', and 'track' in the 'leisure' category and 'protected_area' and 'national_park' in the 'boundary' category. Status as open space or non-open space for each pixel of built land are derived using the built-up class from ESA WorldCover as a mask. Finally, the count of masked pixels of open space is use to calculate the percent of built area that is open space.

There are limitations to these methods and uncertainty regarding the resulting indicator values. Importantly, OpenStreetMap (OSM) is a crowd-sourced dataset with a diverse user community and its completeness, accuracy and standards of use vary considerably. Some regions lack data on all of their open spaces in OSM. Additionally, use of OSM taxonomic system of tags to categorize features is used differently by different local contributing communities. In selecting the tags used in our methods we considered the most commonly used tags to designate open spaces available for public use (parks, athletic fields, etc.), while attempting to exclude tags used primarily for private, limited access or indeterminate open spaces. However, as the OSM tagging system is not designed to make this distinction, this may result in the exclusion of data on some open spaces that are available for public use as well as inclusion of some non-public open spaces for some cities. Finally, OSM is being constantly edited, potentially improving the relevant data for these cities, so our download of data from one point in time may quickly become outdated.

Percent of population with access to public open space within walking distance (400m)

Availability of open space in a city is also a function of its ease of access and who does or does not have easy access. As a result, spatial distribution of open space across the city and its alignment with the location of population, and its accessibility within walking distance (commonly defined at 400 meters, including in the Singapore Index) is a critical indicator to understand how many city residents are well served by open space.

This indicator makes use of gridded population at 100-meter resolution from the WorldPop project as accessed on Earth Engine. The open space polygons retrieved from OpenStreetMap are buffered by 400 meters to derive recreation catchment areas. The population within those recreation catchment areas is calculated and then converted to percent by dividing that value by the total population of the area of interest.

There are limitations to these methods and uncertainty regarding the resulting indicator values. The limitations of OpenStreetMap data previously described are also relevant to this indicator. Additionally, our approach for calculating access to open space relies on Euclidean distance for simplicity. However, this straight line measurement is not an accurate characterization of how people travel within cities as it doesn't account for the orientation of streets or barriers to pedestrian travel. Because of this reality on the ground, in most cases real pedestrian travel of 400 meters will enable access to a smaller area than our methods present.

Percent of population with threshold level (10%+) of tree cover within walking distance (400m)

Access to tree cover is also an important indicator of quality greenspace, whether it is public or private space. Privately maintained trees also provide a variety of public benefits from clean air to heat mitigation and shade. This indicator considers all trees within walking distance (400 meters) for each resident of the city as an indicator of the quality of greenspace that they interact with in an average day.

This indicator uses 10-meter resolution tree cover data for 2020 from the Trees in Mosaic Landscapes dataset and gridded population at 100-meter resolution from the WorldPop project as retrieved from Earth Engine. A neighborhood reduction method using a circular kernel of 400 meters is applied to the tree cover layer to determine the percentage tree cover within 400 meters of each pixel in the area of interest. This results is then used to mask the population layer to only include pixels with at least 10% tree cover within 400 meters. The population of this masked population layer is then calculated and then converted to percent by dividing that value by the total population of the area of interest.

There are limitations to these methods and uncertainty regarding the resulting indicator values. The limitations previously described related to the use of Euclidean distance and the Trees in Mosaic Landscapes dataset are also relevant to this indicator.

Percent of built land exposed to coastal or riverine flooding

Most cities are built near coasts or along rivers. These natural assets for economic development can also present a hazard when they cause flooding in built-up areas. Prevalence of these floods is increasing globally as a result of sea level rise and extreme precipitation caused by climate change.

We use projections from Aqueduct Floods, with a resolution of 30 arc-second (approximately 1km), retrieved through Earth Engine to characterize the coastal and riverine flood hazards for each city in year 2050 in terms of meters of inundation depth. The scenario variables we use are a 100 year return period, the Business As Usual/Pessimistic (RCP8.5) climate scenario, high sea level rise, no subsidence, and the GFDL-ESM2M riverine projection model. We then compare locations of either projected coastal or riverine flooding to built-up areas in the city in 2020 as defined by the built-up class from ESA WorldCover. For each city and city district we calculate the number of built-up pixels and the number of built-up pixels projected to experience a flood inundation depth greater than zero. Then we calculate the percent of built-up pixels with flood exposure.

There are limitations to these methods and uncertainty regarding the resulting indicator values. The Aqueduct Floods dataset with at a 1km resolution may be too coarse for some applications related to assessing flood risk to individual blocks and buildings. Additionally, because this indicator considers 2020 built-up in comparison to projected 2050 inundation, it may underestimate built area exposed to the hazard as a result of future land development. Note that this is an indicator of flood hazard and it does not consider how mitigation measures such as flood protection infrastructure (levees, etc.) may impact risk of inundation from flooding under the scenario described. This indicator is limited in scope to river and coastal flooding hazards and does not describe hazards from flash flooding resulting from precipitation events.

Expected extreme precipitation event hazard (expected maximum mm precipitation in one day in 2050) and trend (% change between 2020 and 2050 in mm of precipitation on the most extreme day)

Understanding the hazard presented by extreme precipitation and the likely future change of that hazard can inform the importance of planning for extreme precipitation events and investing in infrastructure--like trees and other vegetation--that can help mitigate their impacts.

This indicator uses data from the NEX-GDDP climate projections retrieved from Earth Engine, a given location, and two selected years to calculate the expected millimeters of precipitation expected on the day with the largest amount of precipitation for each year and the percentage change between years. Methods are further documented in the Data Portal hazards repo.

There are limitations to these methods and uncertainty regarding the resulting indicator values.

Percent of built land cover within 1 meter height above nearest drainage

Land near natural drainage (e.g., streams, rivers) is at risk for pluvial or flash flooding hazards. Development closer in elevation to natural drainage, especially near drainage for large land areas, is at greater risk. Risk for these type of floods can be difficult to predict, and beyond height above drainage, other factors--such as amount of precipitation, soil type, distance from natural drainage and the presence and efficacy of built drainage systems--are also important influences.

This indicator uses 30 meter resolution data on height above drainage channels from Global 30m Height Above the Nearest Drainage to estimate the land area 1 meter or less above nearest drainage, for drains with a flow accumulation area of at least 1 km2. We then compare these areas to built-up areas in the city as defined by the built-up class from ESA WorldCover. For each city and city district we calculate at 10 meter scale the number of built-up pixels and the number of built-up pixels within 1 meter of nearest drainage. Then we calculate the percent of built-up pixels that are 1 meter or less above nearest drainage.

There are limitations to these methods and uncertainty regarding the resulting indicator values. While this indicator is an important factor related to flash flooding hazards, it does not include other important factors that go into assessing flash flood hazards (soil type, artificial drainage systems, etc.) comprehensively and should not be interpreted as a hazard map.

Percent of built land that has impervious surface

Impervious surfaces (in urban areas, typically buildings, roads and other pavement) prevent water from soaking into the ground on-site and contribute to decreased overall water infiltration and increased runoff and can increase the prevalence of localized flooding. This type of flooding also lowers the quality of receiving waters, impacting the biodiversity of aquatic and marine systems. Building cities to minimize use of impervious surfaces (with alternative construction materials or intentional greening) allows for greater natural water infiltration and can enable cities to "sponge" up excess water, decrease flooding and protect groundwater resources.

We obtain the estimated extent of impervious areas in 2018 from Tsinghua Annual maps of global artificial impervious area. This 30-meter resolution dataset defines a pixel with at least 50% impervious surface as impervious. We then compare these areas to built-up areas (human settlement areas) in the city as defined by the built-up class from ESA WorldCover. For each city and city district, we calculate at 10 meter scale the number of built-up pixels and the number of impervious built-up pixels. Then we calculate the percent of built-up pixels that are impervious.

There are limitations to these methods and uncertainty regarding the resulting indicator values. These calculations are based fully on remotely sensed data, which may not full capture the reality on the ground. The two datasets compared are in different resolutions and assess different years and therefore present difficulties for precise comparison.

Percent of built land cover without vegetation cover

Vegetation cover provides many benefits in urban areas. Importantly for flood mitigation, greater vegetation cover can increase water infiltration into the ground, reducing water stagnation and runoff on the surface. Absence of vegetation cover can increase the prevalence of localized flooding.

To estimate vegetation cover we used a 10-meter annual greenest pixel mosaic of Sentinel-2 imagery to calculate normalized difference vegetation index (NDVI) with a minimum threshold of 0.4. We then compare these areas to built-up areas in the city as defined by the built-up class from ESA WorldCover. For each city and city district, we calculate at 10 meter scale the number of built-up pixels and the number of vegetated built-up pixels. Then we calculate the percent of built-up pixels that are vegetated.

There are limitations to these methods and uncertainty regarding the resulting indicator values. These calculations are based fully on remotely sensed data, which may not full capture the reality on the ground. An annual mosaic is used to assess the greenest point in the year for each individual pixel, as a result this assessment does not assess the state of vegetation at any point in time, but rather the maximum state of vegetation during the year.

Percent of riparian areas without vegetation or water cover

Riparian areas--the spatial interface between land and water, particularly rivers and streams--can provide natural infrastructure for flood control. The presence of vegetation cover in the riparian zone is a primary factor in determining its flood mitigation effectiveness, as vegetation will slow down water flow, increase its absorption into land and its transpiration into the air. Riparian areas are also critical habitat to support biodiversity.

This indicator uses data on elevation and drainage channels from Global 30m Height Above the Nearest Drainage and on the location of other waterbodies from JRC Global Surface Water to estimate the location of lakes, rivers and streams. To estimate riparian zones, we buffered these waterways by 144 meters (the size of riparian areas needed to preserve bird diversity) and excluded the area of the water channel itself. To estimate vegetation and water cover we used an annual greenest/bluest pixel mosaic of Sentinel-2 imagery to calculate normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) with a minimum threshold of 0.4 and 0.3 respectively. For the buffered riparian areas in each city or sub-city unit we calculated at 30 meter scale the count of all riparian area pixels and the count of riparian area pixels with vegetation or water cover. Then we calculated the percent of riparian area pixels without vegetation or water cover.

There are limitations to these methods and uncertainty regarding the resulting indicator values. These calculations are based fully on remotely sensed data, which may not full capture the reality on the ground. In particular, the general definition of riparian zone used may not match the situation on the ground as characterized with additional local data.

Percent of steep hillside slopes without vegetation cover

Steep slopes are vulnerable to landslides, especially during incidence of extreme precipitation of flooding where soil can become saturated with or washed away by water. Vegetation is critical for stabilizing slopes as roots hold soil in place and plant absorb water from the soil. Steep slopes without vegetation coverage are particularly vulnerable to landslides.

To estimate slopes we apply the Earth Engine slope algorithm to the global NASA NASADEM 30 meter digital elevation model. The slope layer is then masked to only include high slope areas of 10 degrees or greater (the threshold at which landslide susceptibility starts to grow quickly). Vegetation cover is estimated using Sentinel-2 imagery to calculate greenest pixel normalized difference vegetation index (NDVI) with a minimum threshold of 0.4. For the high slope areas in each city or sub-city unit we calculated at 30 meter scale the count of all high slope pixels and the count of high slope pixels with vegetation cover. Then we calculated the percent of high slope pixels without vegetation cover.

There are limitations to these methods and uncertainty regarding the resulting indicator values. These calculations are based fully on remotely sensed data, which may not full capture the reality on the ground. The digital elevation model is based on data collected in 2000, which may be outdated if there have been significant changes in local slopes. Additionally, its relatively low resolution (30 horizontal meters) may not capture important smaller slope details.

Level and share of GHG emissions by pollutant and sector

Human activity contributes to air pollution and climate change through emission of gases from fuel combustion, industrial processes, and agriculture. This indicator can help decision-makers and stakeholders identify the most important pollutants emitted locally, the activities responsible for the emissions, and, with multiple years of data, emissions trends.

Similar to GRE-2.1: Air pollutant emissions, this indicator is based on the CAMS Global Anthropogenic Emissions dataset of the Copernicus Atmosphere Monitoring Service (CAMS) and ECCAD (Granier et al. 2019). The dataset provides estimates of emissions from 12 sectors of human activity, on a 0.1-degree (approx. 11 km) spatial resolution. The estimates are based on simulations and historical data. The included sectors are: agriculture (livestock); agriculture (soils); agriculture (waste burning); power generation; fugitive emissions; industry; combustion in residential, commercial, and other settings; ships; solvents; solid waste and wastewater; off-road transportation; and on-road transportation.

We used Google Earth Engine for spatial subsetting against cities' administrative boundaries. We report annual, sector-disaggregated emissions for the years 2010, 2015, and 2020, of the each pollutant species. We also report emissions of the greenhouse gases, as total CO₂e aggregated over sectors for each of the three years. We calculated CO₂e based on 20-year and 100-year global warming potentials, as listed in the table. We report all emissions as tonnes/year.

Pollutant species	Included as GHG	GWP (20-year)	GWP (100-year)	GWP source
Black carbon (BC)	yes	460	1504	global values reported in IPCC AR5 Table 8.A.6
Methane (CH4)	yes	84	28	IPCC AR5 Table 8.A.1
Carbon monoxide (CO)	yes	7.65	2.49	midpoints of global values reported in IPCC AR5 Table 8.A.4
Carbon dioxide (CO2)	yes	1	1	definition of GWP
Nitrogen oxides (NOx)	yes	19	-10.34	global values reported in IPCC AR5 Table 8.A.3
Sulfur dioxide (SO2)	no	NA	NA	NA
Organic carbon (OC)	yes	-240	-64.86	global values reported in IPCC AR5 Table 8.A.6
Ammonia (NH3)	no	NA	NA	NA
Non-methane volatile organic compounds (NMVOC)	yes	14	4.23	global values reported in IPCC AR5 Table 8.A.5

There are limitations to these methods and uncertainty regarding the resulting indicator values. Importantly, the emissions data used for this indicator only account for direct emission from activities within the boundaries of the city (Scope 1). This data does not account for emissions associated with electricity used in the city but generated elsewhere (Scope 2) or emissions produced elsewhere associated with products or services consumed in the city (Scope 3). The CAMS dataset is modeled data based on an ensemble of multiple emissions models and is subject to the limitation of those models. Description of the methods used to develop the CAMS emissions dataset is described here.

Average annual carbon flux from trees (2001-2021) per hectare of city area (Mg CO2e/ha)

Trees are critical contributors to a balanced climate system. Healthy, growing trees remove carbon from the atmosphere and sequester it, while trees that are cut or die emit carbon. Cities can work to keep forests and tree healthy and invest in expanding tree cover locally, regionally and globally to increase carbon removals and reduce their contributions to climate change.

This indicator is calculated from version: 1.2.2 of the Net Carbon Flux from Forests dataset per Harris et al. 2021 as accessed on Google Earth Engine. This 30-meter resolution data layer provides an estimate of total carbon flux from trees from the years 2001 through 2021 inclusive in units of Mg CO2e/hectare. To calculate the mean annual carbon flux for each area of interest, we first unmask the dataset so pixels where the dataset shows no carbon flux are given a value of zero and included in subsequent calculations, apply a mean reducer to the area, and divide the resulting value by 21 to annualize the 21 year dataset. The resulting value estimates the average carbon emissions or removals per hectare from the area of interest during each year in the 21 year period. Negative numbers represent greenhouse gas removals and positive values indicate emissions.

There are limitations to these methods and uncertainty regarding the resulting indicator values. The carbon flux model used for the calculations was designed for forests and is based on tree cover as detected by Landsat; it does not pick up or measure carbon flux from sparse tree cover. At 30-meter resolution, Landsat and products developed from it do not pick up isolated trees. Additionally, the model is limited to pixels with tree canopy density >30% and trees of >5 meter height in 2000 or subsequent tree cover gain, so low density or short tree cover is not included.

---

Granier, C., S. Darras, H. Denier van der Gon, J. Doubalova, N. Elguindi, B. Galle, M. Gauss, M. Guevara, J.-P. Jalkanen, J. Kuenen, C. Liousse, B. Quack, D. Simpson, K. Sindelarova The Copernicus Atmosphere Monitoring Service global and regional emissions (April 2019 version)

Aaron van Donkelaar, Melanie S. Hammer, Liam Bindle, Michael Brauer, Jeffery R. Brook, Michael J. Garay, N. Christina Hsu, Olga V. Kalashnikova, Ralph A. Kahn, Colin Lee, Robert C. Levy, Alexei Lyapustin, Andrew M. Sayer and Randall V. Martin (2021). Monthly Global Estimates of Fine Particulate Matter and Their Uncertainty Environmental Science & Technology, 2021, doi:10.1021/acs.est.1c05309.

Shindell, D.T. The social cost of atmospheric release. Climatic Change 130, 313–326 (2015). https://doi.org/10.1007/s10584-015-1343-0

van der Kamp, J. Social cost-benefit analysis of air pollution control measures. (2017) Karlsruhe Germany: KIT Scientific Publishing. doi:10.5445/KSP/1000072046