Climate change necessitates global carbon mitigation efforts. China, a major CO2 emitter, aims for carbon peaking by 2030 and carbon neutrality by 2060. Its power sector, heavily reliant on fossil fuels, is a key target for decarbonization. Solar photovoltaics (PV), particularly distributed PV (DPV) systems, offer a promising solution due to their broad applicability and ease of implementation. Rooftop photovoltaics (RPVs), a significant component of DPV, are especially relevant in densely built-up cities. While the potential of RPVs is substantial, estimating their carbon mitigation potential at a national scale presents challenges, particularly in accurately assessing rooftop area. Existing studies often rely on indirect methods using national statistics with limitations in accuracy and city-level resolution. This study addresses this data gap by employing a novel machine learning-based approach using multi-source geospatial data to estimate the RPV carbon mitigation potential across 354 Chinese cities, covering 88% of the country's area. The study further investigates the geographical heterogeneity of this potential and projects future changes under varying scenarios of urban land expansion and power mix transformation.

Several studies have explored PV application at the city level, often focusing on case studies with limited generalizability. Challenges in accurately assessing RPV potential stem from the decentralized nature of deployment and the difficulty in delineating and calculating building rooftop areas. Previous research either uses high-resolution remote sensing data, which is costly and unavailable for most cities, or relies on national statistics like floor area and land use, leading to lower accuracy. The existing national-level studies often lack detailed city-level assessments due to data aggregation at higher administrative levels. This paper addresses this gap by leveraging advancements in urban information acquisition and the use of artificial intelligence, specifically, a high-accuracy machine learning approach to provide refined urban measurements. Previous work on developing a large-scale vector building rooftop area dataset (covering 16% of China) provided a foundation for this study.

This study used a machine learning-based regression model to estimate the rooftop area of 354 Chinese cities in 2020. The model utilized a previously developed vectorized rooftop area dataset covering 16% of China's area (containing cities with diverse economic, political, and geographical characteristics) and multisource geospatial data including road length, built-up area, population size, and nighttime light intensity. The dataset provided measured rooftop areas for 86 cities, which served as the sample area for model training. The random forest algorithm was employed due to its superior accuracy compared to other models. A 10-fold cross-validation was performed to tune hyperparameters. The trained model was then applied to the remaining 268 cities to extrapolate rooftop area predictions. To validate the extrapolation results, high-resolution satellite images were processed using a deep learning semantic segmentation method for 18 cities representing China's six geographic regions. The accuracy of the extrapolation was assessed through the comparison of extrapolated rooftop areas with ground truth data from the validation dataset, resulting in a low mean absolute error (MAE) at both cell and city levels. The conversion of rooftop area to solar potential used a high-resolution (10 km) surface solar radiation dataset. The carbon mitigation potential was calculated using a carbon mitigation factor derived from baseline emission factors of China's regional power grids, representing the CO2 reduction achieved by RPVs replacing electricity from conventional power plants. The calculation incorporated assumptions of rooftop availability (35%), PV panel conversion efficiency (20%), and overall RPV system efficiency (80%). K-means++ clustering was performed to group cities based on rooftop area, solar radiation, and grid emissions, facilitating analysis of geographical heterogeneity. Future RPV carbon mitigation potential was projected for 2030 considering scenarios of urban land expansion (9%–14%) and power mix transformation based on China's carbon neutrality roadmap. Finally, a regional analysis compared theoretical RPV potential against existing installed capacity, energy consumption, and carbon emissions.

The study estimated a total RPV carbon mitigation potential of 4 billion tons (BT) for 354 Chinese cities in 2020, nearly 70% of electricity and heat sector emissions. This potential ranged from 0.04 to 52 million tons (MT) per city, with significant regional variation. Cities in the southeastern regions (Clusters 1-3), characterized by large populations and abundant building stock, contributed 89% of the total potential. Cluster 1, including major cities like Shanghai, Beijing, and Guangzhou, showed the highest average potential (29 MT). Several cities in Cluster 1 exhibited RPV potential comparable to the carbon mitigation of the Three Gorges Dam. Cities in Cluster 4 (western regions) had the lowest average potential (6 MT) despite high solar radiation, due to limited rooftop area. Per capita and per GDP carbon mitigation potential decreased from southeast to northwest. Future projections for 2030 under different scenarios of urban land expansion and power mix transformation suggest the potential could remain around 3-4 BT. Under a scenario of rapid power sector decarbonization, the carbon mitigation potential of RPVs could reduce by more than 20% despite the increase in installed capacity. Regional analysis reveals that most provinces/municipalities have developed less than 1% of their RPV potential, highlighting a considerable gap between theoretical potential and current deployment. Many provinces possess RPV power generation potential exceeding half of their electricity consumption, while some regions require more substantial RPV deployment to meet local energy demands.

This study provides a comprehensive city-level assessment of RPV carbon mitigation potential in China, addressing critical data limitations in previous research. The findings highlight the substantial contribution of RPVs to China's carbon mitigation goals, particularly in densely populated southeastern regions. The regional analysis identifies areas with high and low potential, informing targeted deployment strategies. The projections for 2030, while considering uncertainty related to urban expansion and power mix transformation, show the continued significance of RPVs in the nation's carbon peaking and neutrality targets. The considerable gap between theoretical potential and actual deployment emphasizes the need for policies and initiatives that promote RPV adoption, improving public awareness, and addressing factors limiting wider implementation. The methodology developed provides a valuable framework for similar assessments in other countries.

This study provides a comprehensive national-level assessment of the carbon mitigation potential of rooftop photovoltaics (RPVs) in China. The findings reveal substantial untapped potential, particularly in densely populated areas. The study's methodology, combining multi-source geospatial data with machine learning, offers a robust and scalable approach for assessing RPV potential elsewhere. The substantial gap between theoretical potential and current installations indicates a need for policy interventions to promote wider adoption and address limitations such as grid integration and energy storage requirements. Future research could focus on more detailed economic analyses, the integration of RPVs with energy storage and other renewable energy sources, and refinement of rooftop area prediction using higher-resolution data.

The study's carbon mitigation estimates assume a constant 35% rooftop availability and 20% PV panel conversion efficiency, which may vary among cities. Future research should investigate city-level variations in these parameters. The study's projection for 2030 relies on assumptions about urban land expansion and power mix transformation. The actual figures may differ, affecting the precise carbon mitigation potential. Economic factors such as initial investment costs and levelized cost of energy were not considered, though the study cites relevant literature to this end.