DeepSolar++: Understanding residential solar adoption trajectories with computer vision and technology diffusion models

Abstract

•We use computer vision and technology diffusion models to study PV adoption over time •Lower-income communities are delayed in PV adoption onset and saturate more quickly •Performance-based incentives are associated with high PV saturation for low income •Property tax incentives are associated with high PV saturation but not for low income Although the United States has generally experienced a rapid adoption of residential photovoltaics (PV), many communities are lagging behind. To investigate why, we developed a computer vision model that addresses the challenge of low image resolution to identify the installation year of PVs from historical aerial and satellite images. We used the model to construct a granular spatiotemporal dataset of PV deployment across 46 US states and analyzed these data from a technology adoption life cycle perspective. Our analysis of adoption curves and phases showed that low-income communities are not only delayed in their adoption onset but also saturate more quickly at lower levels. We further demonstrated the value of our data via an illustrative analysis of financial incentives and found that performance-based incentives are positively associated with saturated adoption levels—particularly for lower-income communities. Our study highlights the importance of analyzing PV adoption trajectories from dynamic perspectives to inform policy design. Although the United States has generally experienced a rapid adoption of residential photovoltaics (PV), many communities are lagging behind. To investigate why, we developed a computer vision model that addresses the challenge of low image resolution to identify the installation year of PVs from historical aerial and satellite images. We used the model to construct a granular spatiotemporal dataset of PV deployment across 46 US states and analyzed these data from a technology adoption life cycle perspective. Our analysis of adoption curves and phases showed that low-income communities are not only delayed in their adoption onset but also saturate more quickly at lower levels. We further demonstrated the value of our data via an illustrative analysis of financial incentives and found that performance-based incentives are positively associated with saturated adoption levels—particularly for lower-income communities. Our study highlights the importance of analyzing PV adoption trajectories from dynamic perspectives to inform policy design.