Carbon mitigation potential afforded by rooftop photovoltaic in China

China, one of the largest CO2 emitters, has pledged to peak carbon emissions by 2030 and achieve carbon neutrality by 2060. The power sector contributes about half of national energy-related emissions and photovoltaics (PV) are central to decarbonization. Distributed PV (DPV), particularly rooftop PV (RPV), is increasingly favored but nationwide city-level assessment is constrained by the lack of accurate rooftop area data. Existing studies often focus on single cities or rely on coarse national statistics, limiting accuracy and transferability. This study aims to accurately quantify city-level RPV carbon mitigation potential across 354 Chinese cities, understand its geographical heterogeneity, and project future changes to 2030 under urban expansion and power mix transformation scenarios, providing a basis for targeted deployment toward China’s dual-carbon goals.

Prior work has assessed PV potential at local or provincial scales, often using high-resolution remote sensing for rooftop delineation at substantial cost or indirect proxies such as floor area and land use, which suffer from aggregation and accuracy limitations. Distributed PV growth in China has accelerated, yet city-level comprehensive assessments remain scarce. Studies highlight that rooftop availability factors (orientation, slope, obstacles) and grid carbon intensity critically affect mitigation outcomes. Advances in AI, deep learning, and geospatial big data enable improved urban measurements, including vectorized rooftop datasets and satellite-derived solar radiation, motivating scalable, accurate national assessments.

• Study scope: 354 Chinese cities (excluding Hong Kong, Macau, Taiwan, and Tibet due to unavailable baseline grid emission factors), covering >8 million km². Measured rooftop data available for 86 cities from a previously developed vectorized rooftop dataset (overall accuracy 98%, F1 83%), covering 16% of China.
• Extrapolation of rooftop area: Study area partitioned into 1,045,022 grid cells of 10 km². Four explanatory variables per cell: road length (OpenStreetMap), built-up area proportion (Esri Landcover, 10 m), population (WorldPop, 100 m), and night-time light intensity (EOG VIIRS, 500 m). A random forest regression model was trained (90% train/10% test; hyperparameters via 10-fold cross-validation). The model outperformed alternatives (R²=0.83; lower MAE). Validation used 18 cities (across six regions), with ground truth rooftops extracted via deep-learning semantic segmentation on high-resolution imagery.
• Validation results: Per 10 km² cell MAE 0.06 km²; most cell errors within −0.05 to 0.05 km². At city level in validation area: cumulative error −26 km² over 2641 km² (−1% relative error). City-level relative errors within ±20%.
• Solar resource and grid emissions: Annual surface solar radiation at 10 km resolution; grid baseline emission factors (combined margin with weights WOM=0.75, WBM=0.25) used as carbon mitigation factors.
• RPV system assumptions: Rooftop availability 35%; PV panel conversion efficiency 20%; overall system efficiency 80%; rated power 200 W/m²; panels assumed horizontally fixed. Focus on operational-stage emissions only (life-cycle not included).
• Calculations: Installed capacity derived from available rooftop area and panel power density. Annual generation computed from rooftop solar radiation, panel efficiency, and system efficiency. Carbon mitigation computed as generation multiplied by grid combined margin emission factor.
• Clustering: K-means++ on three indicators (rooftop area, solar radiation, grid emission) to classify cities into four clusters; ANOVA confirmed significant between-cluster differences (94% of pairwise comparisons significant at 0.01 level).
• Scenario analysis to 2030: Urban land expansion scenarios based on SSP-derived projections (9% low, 14% high rooftop growth) and power mix transformation scenarios: STEPS (10% grid EF decline 2020–2030) and APS (30% grid EF decline 2020–2030). Effects estimated on installed capacity and carbon mitigation potential.

• Rooftop area and national potential: Identified 65,962 km² of rooftop area (2020) across 354 cities. Under assumptions, total national RPV carbon mitigation potential is ~4 billion tons CO2 (2020), nearly 70% of electricity and heat sector emissions.
• City-level variation: City potentials range from 0.04 to 52 million tons (MT), average 11 MT. Clusters 1–3 (southeast) contribute ~89% of national potential due to larger populations and building stock. Cluster 4 (west) has highest solar radiation (avg 1667 kWh/m²) but lowest rooftop area (avg 80 km²) and lowest average potential (6 MT).
• High-potential cities: Cluster 1 includes megacities and economic hubs; average potential 29 MT. Top cities include Weifang (52 MT), Chongqing (47 MT), Linyi (46 MT), each with potentials comparable to roughly half of the Three Gorges Dam’s 2020 CO2 reduction (94 MT).
• Model performance: Random forest regression achieved R²=0.83; per-cell MAE 0.06 km²; validation city-level relative errors within ±20%.
• 2030 projections: Urban land expansion increases rooftop area by 5937–9235 km² and installed capacity by 416–646 GW (2020–2030). Under STEPS (10% EF decline): carbon mitigation potential in 2030 is 3730 MT (low expansion, −72 MT vs 2020) to 3901 MT (high expansion, +99 MT). Under APS (30% EF decline): 2901–3034 MT (−901 to −768 MT vs 2020). Despite capacity growth, mitigation declines with faster grid decarbonization.
• Deployment status gap: As of 2018, most provinces/municipalities had developed <1% of their rooftop potential. In 2020, theoretical RPV generation could exceed half of local electricity consumption in ~80% of provinces/municipalities. Greater mitigation per kWh in northern and northeastern grids with higher coal dependence.
• Per capita/GDP metrics: RPV mitigation potential per capita and per GDP increase from southeast to northwest, indicating different strategic priorities across regions.

The study addresses the challenge of accurately estimating city-level RPV mitigation potential across a large, diverse country by combining measured rooftop data with machine learning and high-resolution geospatial inputs. Findings show substantial national-scale mitigation potential (4 BT in 2020), concentrated in densely populated southeastern cities, highlighting the value of leveraging building rooftops where land is scarce. However, the effectiveness of RPV in mitigating carbon declines as the grid decarbonizes, emphasizing the time sensitivity of RPV deployment and its transitional importance. Regional clustering clarifies where rooftop area, solar resources, and grid emissions align for maximal impact: eastern/southeastern cities for distributed deployment; sparsely populated northwestern regions for centralized PV; coal-dependent industrial provinces for prioritized DPV to accelerate local grid transformation. Analyses of volume and intensity guide both top-down policy and bottom-up investment decisions. The gap between theoretical potential and current deployments indicates major untapped opportunities, and the datasets provided can inform targeted policies and integration with broader decarbonization plans.

This work delivers a national, city-level assessment of RPV carbon mitigation potential in China, producing high-resolution rooftop and mitigation datasets via machine learning and multisource geospatial data. It quantifies a 4 BT annual mitigation potential in 2020, elucidates pronounced regional heterogeneity, and projects 2030 outcomes under urban expansion and power mix scenarios, showing capacity growth but potential mitigation declines under rapid grid decarbonization. The results support prioritizing RPV deployment in high-potential urban clusters, especially in coal-dependent regions, and underscore the need for complementary measures such as storage, renewable integration, and targeted policies to bridge the theory–practice gap. Future research should refine rooftop availability estimates at city level, incorporate higher-resolution and additional predictors, evaluate operational flexibility solutions (storage, sector coupling), and conduct detailed techno-economic and life-cycle assessments to translate technical potential into achievable abatement.

• Rooftop availability assumed uniformly at 35% nationally; actual city-level availability varies with orientation, slope, obstacles, structure, and use.
• Panels assumed horizontally fixed; tilt/azimuth optimization, shading dynamics, and structural constraints not fully resolved.
• Life-cycle emissions (manufacturing, construction, recycling) excluded; only operational-stage mitigation considered.
• Grid emission factors for Hong Kong, Macau, Taiwan, and Tibet unavailable; these regions excluded.
• Scenario analysis considered only urban land expansion (as proxy for rooftop growth) and grid decarbonization; other drivers (policy shifts, technology changes, economic factors, building codes) not fully modeled.
• Economic feasibility not assessed in detail at city scale; financial, regulatory, and market constraints may limit realizable potential.
• PV intermittency and grid integration challenges noted but not quantitatively modeled (e.g., storage sizing, curtailment, temporal matching).