Promoting low-carbon land use: from theory to practical application through exploring new methods

The study addresses how to accurately monitor and manage urban carbon sources and sinks to enable low-carbon land use. Urban areas dominate anthropogenic emissions, driven by industry, residential consumption, traffic, and commerce. Concurrently, urban expansion alters land cover, impacting both terrestrial carbon sinks and emissions. Existing spatial carbon monitoring is often conducted at coarse resolutions or administrative units, with significant uncertainties in sink estimates and limitations in emissions spatialization (e.g., night lights miss industrial hotspots; satellite XCO2 offers coarse resolution). Low-carbon land-use optimization commonly operates at the city or land-type scale, neglecting the spatial heterogeneity of sink/source capacity within the same land-use type. The study’s purpose is to: (1) calculate anthropogenic emissions and terrestrial sink/source; (2) produce a high-resolution spatial carbon monitoring map; (3) assess land-use changes and carbon impacts; and (4) optimize land use to increase sinks and reduce emissions, advancing theory into patch-scale practice.

Prior work estimates terrestrial carbon sinks using ecosystem process models, atmospheric inversion, eddy covariance, and inventories, but sink magnitudes remain uncertain and typically resolved at coarse scales (continental, national, provincial). Anthropogenic emissions are computed using top-down and bottom-up approaches; night-light-based spatialization underestimates industrial hotspots, and satellite CO2 retrievals have coarse resolution and weather sensitivity. High-resolution databases like CHRED improve detail but remain limited. Urban-scale or land-type-level low-carbon optimization often uses linear programming or land-change models (CA, CLUE-S, SLEUTH, LTM, ANN, SVM) but rarely operates below city scale or integrates fine-grained sink/source maps. There is a methodological gap in combining high-resolution sink/source monitoring with patch-scale LUCC simulation to guide practical low-carbon land-use planning.

Study area: Shangyu District, Shaoxing, Zhejiang Province, China, an economically developed district with >800,000 population. Land composition: woodland 41.12%, cropland 21.09%, water 18.67%, built-up 17.33%. Data: Energy consumption (China’s Energy Statistical Yearbook), land-use from the third national land survey (detailed classes), Luojia-01 night-light (130 m), industry enterprise point data (location, sector, production/operations), building vector and heights (urban planning), net ecosystem productivity (NEP) following Ye and Chuai (2022) with 500 m NPP, inland water carbon flux (Ran et al., 2021), agricultural inputs and industry products (Shangyu Statistical Bulletin). Emission factors from IPCC 2006 and Zhao et al. (2015). Anthropogenic emissions accounting: - Energy consumption emissions by sector using activity data and fuel-specific coefficients: C = Σ E_i × θ_i. - Industrial process emissions (glass, aluminum, cement, steel, lime, ammonia): C_industry-i = Q_industry-i × V_industry-i. - Agricultural production emissions using input-based coefficients: C_a = Σ T_j × σ_j. Terrestrial sink/source: NEP = NPP − soil heterotrophic respiration (grid-based using satellite and field data; 500 m NPP). Inland water carbon flux capacities incorporated by water type and distribution. High-resolution carbon monitoring (spatial allocation): 1) Assign total sectoral emissions to land-use types (urban residential, traffic, commercial, industrial, cropland, rural residential) using statistical data. 2) Allocate spatially with auxiliary data by land-use type: - Rural residential and cropland: evenly within administrative district (average per patch). - Urban residential and commercial: vector allocation using building footprint area and number of floors (building height). - Traffic land: night-light intensity at 130 m as proxy for traffic activity. - Industrial land: geolocated enterprises; sector-specific emission intensity weights (per output value) from local statistics assigned to enterprise points; spatial distribution simulated considering industrial land extent. - Merge land-type emission layers in GIS to form composite emissions surface; combine with NEP and water flux to derive net sink/source map. Low-carbon land-use optimization (quantity): Linear programming (Lingo v20) minimizes net carbon emissions intensity (total anthropogenic emissions minus terrestrial sink) across land-use categories subject to area constraints for 2030. Decision variables X_i for 12 land-use categories (cropland, woodland, grassland, water, urban green land, rural residential, urban residential, commercial, industrial, traffic, public facility, other built-up). Constraints reflect total area (ΣX_i = 1400 km²), afforestation plans, historical trends, demographic projections and urbanization, consolidation of idle rural residential land, industrial land intensification targets, transport and public service needs, and unused land eliminated by 2030. Bounds specified per category (e.g., woodland 575.81–589.81 km²; traffic 65.1–70.1 km²). Patch-scale spatial simulation: Two-step spatial optimization after quantitative structure is set. - Step 1 (probability-of-occurrence): Integrate traditional spatial drivers (elevation, slope, aspect; distances to highways, trunk roads, railways) with carbon constraints (higher sink reduces conversion probability; higher emission increases it). Use an artificial neural network (ANN) with input-hidden-output layers to estimate per-cell probabilities for each land-use type at 90 m resolution (172,839 grids); 10% training, 90% testing. - Step 2 (allocation): Use FLUS with adaptive inertia to allocate land, combining occurrence probability, neighborhood effects, inertia coefficient, and transfer costs to simulate 2030 land patterns under traditional constraints alone and under added carbon constraints. Compare scenarios to quantify carbon sink gains from patch-scale optimization.

- Time series (2000–2020): Anthropogenic emissions rose from 51 × 10^4 t to 265 × 10^4 t, with industry ~80% of total throughout; traffic emissions increased rapidly; urban and rural residential, commercial, and cropland emissions also rose. Terrestrial ecosystems were net sinks with interannual variability; peak sink in 2019 was 13 × 10^4 t; average sinks offset ~3% of emissions. - Spatial pattern (2019): Highest emission intensities in the northeast industrial park (~3.7 × 10^4 t/km²), with additional urban hotspots of lower intensity. Urban residential and commercial ~0.4 × 10^4 t/km²; rural residential ~0.12 × 10^4 t/km²; cropland ~0.02 × 10^4 t/km²; negligible anthropogenic emissions from other green land. Net sink concentrated in the south (~33% of area), while the north is a net source. - Land-use change (2000–2020): 14.5% (193 km²) of land changed type. Cropland transfer-out was largest (96.44 km²; 50% of transfers), mainly to built-up land (rural residential 32.21 km²; other built-up—industrial and traffic—33.7 km²). Water was occupied by 73.54 km², predominantly converted to cropland (60.98 km²). Woodland loss totaled 16.06 km². - Carbon impact of transfers (annualized, 2000–2020): Net increase of 77.72 × 10^4 t/year due to land transfers; 76.28% attributable to cropland transfer-out. Key contributors: cropland → other built-up (+50.08 × 10^4 t/year), water transfer-out (+10.34 × 10^4 t/year), woodland transfer-out (+7.39 × 10^4 t/year). Minor decreases from urban and other built-up transfer-out. - Land-use structure optimization (2030 vs. 2020): Net carbon increase limited to 7154 t/year under optimized structure. Area changes raise ecological lands (cropland, woodland, grassland, urban green), with emissions reductions from woodland (−2058 t/year), grassland (−307 t/year), water (−499 t/year), urban green (−113 t/year). Major reduction from rural residential consolidation (−18,515 t/year). Increases arise from built-up expansions: urban residential (+10,465 t/year), traffic (+11,244 t/year), commercial (+4249 t/year), public service (+1518 t/year). Industrial land held or slightly constrained (0 t change at 58.05 km² scenario reported). - Patch-scale optimization benefit: Incorporating carbon constraints in spatial allocation further enhances the carbon sink by 129.59 t C/year relative to the scenario using only traditional spatial constraints; avoided sink loss is largest for woodland (−166.78 t C), and added sink from transfers-in is largest for cropland (+100.59 t C).

The study demonstrates that integrating high-resolution anthropogenic emissions spatialization with terrestrial sink mapping enables practical, patch-scale low-carbon land-use management, directly addressing the gap between coarse monitoring and actionable planning. Combining top-down and bottom-up inventories leverages accessible statistics while improving accuracy for dominant sectors (industry, traffic). The refined spatialization using building heights (vector-level for residential/commercial), high-resolution night lights for traffic, and enterprise POIs with sectoral weights for industry outperforms approaches relying solely on night lights or coarse satellite XCO2 retrievals. Results confirm industrial activity as the primary emission source and show limited offset capability of the terrestrial sink, indicating strong mitigation pressure. Optimized land structures, alongside patch-scale allocation guided by carbon constraints, substantially curb the growth of emissions compared to historical trajectories and yield additional sink gains. The approach advances LUCC modeling by fusing ANN-optimized occurrence probabilities (including carbon) with FLUS allocation, making low-carbon targets operational at the land-patch scale. Policy implications include industrial restructuring toward low-emission, high-efficiency enterprises, intensification of industrial land use, transport decarbonization (public transit, EVs), clean energy transitions, and consolidation of underutilized rural residential land. The framework supports urban and land spatial planning by enabling fine-grained, carbon-informed land adjustments.

This work contributes a county-scale, high-resolution carbon monitoring framework and a low-carbon land-use optimization model operable at the land-patch scale. Key outcomes: (1) Emissions and sinks both increased from 2000–2020, but sinks offset only ~3%, leaving a large gap to neutrality; (2) Net emissions dominate in the north (industrial hotspots), while the south is a net sink; (3) Cropland-to-built-up conversions are the principal driver of carbon increases; (4) Land-use structure optimization to 2030 sharply reduces the growth of emissions compared to 2000–2020, and patch-scale optimization further enhances sinks; (5) Significant potential exists to reduce emissions through land-use control by combining structural adjustments and patch-level allocation informed by carbon maps. Future directions include improving data (traffic flows, enterprise-level accounting, residential vacancy, crop-specific inputs), refining urban vegetation mapping for NEP, further disaggregating land-use categories (industrial subsectors, road classes), incorporating efficiency and suitability into transfer rules, and adopting multi-objective optimization to handle trade-offs among carbon, economy, and biodiversity. Recommendations call for integrating low-carbon goals throughout planning cycles, strengthening ecological land protection, and penalizing high-emission, low-efficiency land uses.

Limitations include potential biases from top-down allocation relative to bottom-up accounting for certain sectors; relatively coarse spatialization for transport and cropland compared to vector-level monitoring; and an optimization that constrains transfers based on absolute carbon sink/source without integrating land-use efficiency or multi-objective trade-offs (economic benefits, biodiversity). Data confidentiality limited public sharing of certain datasets. Future work should incorporate vacancy rates, field-based cropland emissions, traffic flow data, enterprise-level carbon accounting, finer land-use subclasses, and multi-objective optimization frameworks.