Parametric assessment of urban rooftop photovoltaic potential: Threshold calibration and spatial differentiation by city scale classification

Assessing the potential of urban rooftop photovoltaic (URPV) is fundamental for promoting energy transition and urban sustainability. However, existing research is often confined to potential assessments, failing to reveal the driving mechanisms and critical parameter thresholds at the city scale. To address this gap, this study systematically calculates the URPV application potential for 330 cities using a potential estimation framework, incorporating sequential corrections for solar irradiation at optimal tilt angles, rooftop availability factor, ambient temperature, and rooftop area density. Multiple linear regression is applied to deconstruct factor influences on annual generation, while XGBoost is introduced to diagnose whether linear model failure for small cities stems from nonlinearity or scale confounding. A parametric method calibrates thresholds of key factors, leading to an optimized generation model tailored to city size, validated on five representative cities. Results indicate that the total URPV power generation potential is approximately 70,691 GWh, with a levelized cost of energy (LCOE) of 0.36 CNY/kWh and a net carbon emission reduction (NCER) of about 7.83 × 10 8 t. Rooftop area density is the most critical driver across all city sizes. More importantly, city scale systematically moderates the effects of all drivers, with XGBoost confirming that small-city linear failure is attributable primarily to confounding by urban area, not inherent non-linearity. This moderation further manifests as distinct parameter threshold patterns. The optimized model, incorporating these scale-specific thresholds, yields an average midpoint relative error of 8.92% across the five validation cities, confirming internal consistency. The study provides a scientific methodological reference and data support for estimating power generation in cities of different sizes.