Substantial terrestrial carbon emissions from global expansion of impervious surface area

The study addresses a key gap in global carbon budgeting: carbon emissions arising from the conversion of vegetated land to impervious surface area (ISA) during rapid urban expansion. While ISA has more than doubled in the past three decades, its direct land-use carbon losses (from biomass and soil) have been omitted from Global Carbon Project assessments and inconsistently represented in DGVMs and bookkeeping models. Prior studies have focused on lost NEP/NPP (unrealized uptake) rather than realized emissions. In contrast, IPCC guidelines require nations to report carbon stock changes from conversion to “settlements,” but independent, satellite-based validation of these reported emissions has been lacking. This paper aims to quantify terrestrial carbon losses (committed emissions) from global ISA expansion over 1993–2018, compare them to national greenhouse gas inventory (NGHGI) reports for Annex I countries, and attribute emission dynamics to socioeconomic and urban drivers, thus informing carbon accounting and mitigation policies.

Previous work has shown substantial global urban and ISA expansion with environmental impacts on heat, biodiversity, and biogeochemical cycles. However, in global carbon budget assessments, transitions to ISA have been excluded from bookkeeping models and inconsistently treated by DGVMs, leaving a missing component in land-use change emissions. Studies quantifying reductions in NEP/NPP over urban areas reflect foregone carbon uptake rather than actual CO2 emissions. IPCC NGHGI guidelines include a broader “settlements” category, but reported carbon changes are dominated by losses linked to ISA, with minimal gains in urban vegetation. Despite this, independent satellite-based estimates to validate NGHGI settlement-related emissions have not been provided. The literature also highlights uncertainty in SOC responses to sealing, limited observational constraints on SOC loss ratios under ISA, and variability in urban land definitions complicating broader urban vegetation carbon accounting.

The authors quantified committed terrestrial carbon losses from global ISA expansion during 1993–2018 using four remote-sensing ISA products: three 30 m datasets (GAUD, GAIA, GISA) and the 300 m ESA CCI product. ISA pixels were assumed to be 100% impervious, ignoring sub-pixel vegetation, consistent with land-cover mapping assumptions. Annual ISA expansion (non-ISA to ISA) was mapped, assuming irreversibility over the study period. Land-cover sources of ISA expansion were derived from the annual ESA CCI land cover (grouped into forest, shrubland, wetland, grassland, cropland, others). To reconcile resolution mismatches and cases where high-resolution ISA fell within ESA CCI urban pixels, 30 m ISA expansions over 300 m urban were redistributed to non-ISA source classes on 5-km grids via a geographical similarity approach. Carbon losses were computed by overlaying annual ISA expansion with carbon stock maps, assuming immediate loss upon conversion. Biomass losses included above- and below-ground live biomass, surface litter, and dead wood for conversions from forest, shrubland, wetland, cropland, and grassland. SOC losses were estimated for all source land covers for the 0–30 cm soil layer. Biomass density inputs combined static maps: Spawn (AGB/BGB, ~2010), ESA CCI BIOMASS AGB (2010), and Harris AGB (~2000), resampled to 300 m; below-ground biomass was derived via root-to-shoot ratios (Spawn). Litter and dead wood fractions were applied by ecological zone following Harris et al. To avoid bias from existing ISA within vegetated pixels, biomass and SOC densities for pixels with >5% ISA were spatially interpolated from nearby valid pixels of the same land-cover type. For SOC, SoilGrids250m v2.0 provided 0–30 cm SOC densities (resampled to 300 m). Two SOC loss ratios represented lower and upper bounds: 20% (IPCC Tier 1 default) and 59.5% derived from a meta-analysis of 22 paired observations comparing SOC under ISA vs urban greens; this synthesis showed depth-independent losses and values consistent with long-term bare fallow experiments. Emissions were aggregated globally and by country and compared to Annex I NGHGI-reported stock changes from land converted to settlements (biomass, litter, dead wood, and top 30 cm mineral soil). Drivers were analyzed using an ISA-driven Emissions Identity analogous to the Kaya identity: E = P × u × r × s × e, where P is total population, u is urbanization rate, r is residential ISA intensity (ISA per urban person), s is ISA expansion speed-up factor (AISA/ISA), and e is emission intensity (E/AISA). Relative change rates (% yr−1) over 1993–2018 were decomposed for global, Annex I (AI), and Non-Annex I (NAI) groups; relationships with GDP per capita were assessed by grouping countries by GDPcap and regressing median change rates against log10(GDPcap). LMDI was used as a robustness check to the Kaya-like decomposition.

- ISA expansion averaged 15,913 ± 3331 km² yr−1 over 1993–2018, cumulating 0.41 Mkm² and increasing global ISA by 112% (0.37 to 0.78 Mkm²). Sources of ISA expansion: ~65% cropland, 12% forest, 11% grassland, 4% shrubland, 1% wetland. - ISA-driven terrestrial carbon losses (biomass + SOC, 0–30 cm) averaged 45.8 ± 8.2 Tg C yr−1 (lower SOC loss ratio, 20%) to 74.9 ± 13.7 Tg C yr−1 (upper SOC loss ratio, 59.5%). SOC contributed 32–59% of emissions; biomass 41–68%. Biomass emissions averaged 31.0 ± 5.5 Tg C yr−1. - Cumulative ISA-driven emissions over 1993–2018 were 1.19–1.95 Pg C, representing 3.7–6.0% of concurrent global land-use change emissions; using 0–1 m SOC depth raises this share to 6.0–12.9%. - Emissions by source land cover: 49–55% from cropland conversion, 25–31% forest, 11–12% grassland, 4–5% shrubland, ~1% wetland. - Trends: global emissions show a small positive trend of 0.36–0.78 Tg C yr−2; NAI countries exhibit significant increases (0.53–1.01 Tg C yr−2), while AI countries show a small, non-significant decline (−0.23 to −0.17 Tg C yr−2). - Concentration among emitters: the top 15 countries/regions (EU27 as one) account for ~81–82% of global emissions and ~79% of ISA growth. AI countries contributed 10.1% of global urban population growth but 39.2% of ISA expansion and 51.1–54.5% of emissions, indicating high per-capita ISA intensity. - Validation with NGHGIs (Annex I): biomass losses from this study (18.2 ± 3.9 Tg C yr−1) align with NGHGI (22.9 ± 0.9 Tg C yr−1). NGHGI SOC losses (14.8 ± 2.2 Tg C yr−1) lie between this study’s lower (6.8 ± 1.5) and upper (20.1 ± 4.3) bounds. Total NGHGI emissions (37.8 ± 2.9 Tg C yr−1) are close to this study’s upper bound (38.3 ± 8.2). Excluding the USA improves agreement in area and emissions. - Decomposition of drivers (global average relative change rates, % yr−1): r(E) ≈ 0.92% (midpoint of 0.79–1.05%); positive contributions from population (P, +1.25%), urbanization (u, +0.93%), and residential ISA intensity (r, +0.64%); offset by declining speed-up factor (s, −0.99%) and emission intensity (e, −1.00%). Emission intensity declines reflect lower biomass/SOC densities in later-expanding regions and a shift from woody to non-woody sources and toward subtropical latitudes. - Developmental pattern: r(E) decreases with log10(GDPcap); emissions growth is high at low GDPcap and turns negative when GDPcap exceeds ~US$10,000 (2012$ adjusted to 2022 price context), consistent with faster population/urbanization growth in NAI versus higher per-capita ISA intensity but stronger deceleration in AI.

The findings demonstrate that ISA expansion constitutes a previously uncounted but material component of anthropogenic land-use CO2 emissions. The magnitude (1.19–1.95 Pg C, 1993–2018) and consistent validation against Annex I NGHGI settlement-related stock changes underscore the need to incorporate ISA-driven emissions into global carbon budgets and national inventories. The decline in emission intensity over time stems from shifts in the geography and land-cover sources of ISA expansion—toward regions and land types with lower carbon density—while total emissions still rise globally due to population growth, urbanization, and increasing ISA per capita. Contrasting dynamics between AI (declining) and NAI (rising) countries align with economic development stages: as GDP per capita increases, population and urbanization growth slow, residential ISA intensity rises, and the relative ISA growth rate decelerates, causing emissions growth to peak and then decline. This provides a predictive framework for future ISA-related emissions under socioeconomic development. Policy-wise, integrating satellite-based ISA change into NGHGIs enhances transparency and verification for the Paris Agreement and Global Stocktake. The emission intensity metric (E per area of new ISA) highlights a lever for mitigation: allocate new ISA to low-carbon-density lands. Given projected further urban expansion, sustainable urban planning that minimizes conversion of carbon-dense ecosystems can meaningfully reduce land-use emissions.

This study quantifies and attributes a substantial, previously overlooked source of land-use CO2 emissions from global ISA expansion, providing independent, satellite-based evidence consistent with Annex I NGHGI reports. It establishes that ISA-driven emissions contributed 3.7–6.0% of land-use change emissions (potentially up to 12.9% when considering deeper soils) from 1993–2018, and reveals robust socioeconomic controls on their evolution with development stage. The work supports integrating ISA-driven emissions into global carbon budgets, enhances NGHGI verification, and informs mitigation via siting strategies that prioritize low-carbon-density lands. Future research should: (1) expand field observations of SOC loss under ISA, particularly in underrepresented regions and climates; (2) refine SOC loss parameterizations (depth profiles, time dynamics) for inventories; (3) better quantify potential offsets from urban vegetation and greenspaces with harmonized urban definitions; and (4) develop forward-looking scenarios that embed sustainable urban planning to limit conversions of high-carbon ecosystems.

- Assumption of 100% ISA coverage within mapped ISA pixels ignores sub-pixel vegetation; conversely, sub-pixel ISA within vegetated pixels is addressed via interpolation, but residual biases may remain. - Exclusion of urban vegetation (non-ISA) means potential carbon sinks within urban greens are not quantified; current large-scale estimates are uncertain due to inconsistent urban definitions. - SOC loss ratios: limited paired observations (n=22), primarily from the Northern Hemisphere, introduce uncertainty and regional bias; the study brackets losses using 20% (IPCC Tier 1) and 59.5% (meta-analysis) but lacks spatially explicit variation. - Use of static biomass maps and assumed immediate biomass, litter, and dead wood losses; temporal biomass dynamics post-conversion are not modeled. - Irreversibility assumption for ISA expansion (minimal observed re-greening) may overlook rare ISA-to-vegetation transitions. - Resolution mismatch and redistribution of high-resolution ISA over 300 m ESA CCI urban pixels via geographical similarity is an approximation. - SOC quantified for 0–30 cm; deeper soil changes are not directly estimated in the main results, though sensitivity suggests larger totals if 0–1 m is used. - Potential under-detection of minor roads and some ISA in source products could bias area and emissions low.