In today’s world, the growing demand for renewable energy sources has pushed researchers to explore new ways of harnessing solar power. One promising avenue is rooftop photovoltaic (PV) systems, which use solar panels installed on buildings to generate electricity. While the potential for rooftop solar energy is vast, accurately assessing the available rooftop space for solar installations has been a challenging task, especially in urban areas.
Researchers from the Guangdong University of Technology have recently led a novel study that proposes an innovative method to assess rooftop PV potential using remote sensing images and advanced machine learning techniques. The study addresses a significant issue in the solar energy sector: determining how much rooftop space is actually available for solar panel installations. Measuring the area of rooftops is not a straightforward task. Many rooftops are unsuitable for solar panels due to obstructions such as air conditioning units, water heaters, or irregular shapes. As a result, previous methods of estimating rooftop PV potential have been prone to large errors.
According to lead researcher Dr. Haoliang Yuan, “accurately assessing the available rooftop area for PV installation is crucial for energy planning and policy formulation, especially as we move towards more sustainable energy solutions.”
Traditionally, researchers have relied on methods like sampling, Geographic Information System (GIS) data, or even three-dimensional (3D) modeling to estimate rooftop space. However, these approaches can be time-consuming, costly, or lack accuracy. Annotating remote sensing images manually, for instance, is a labor-intensive task. To overcome these challenges, Dr. Yuan and his team developed a semi-supervised learning (SSL) model that can automatically extract rooftop information from satellite images with minimal human input.
The key innovation of this study lies in its use of SSL, a machine learning technique that allows models to learn from both labeled and unlabeled data. This reduces the need for large amounts of annotated training data, which is often the bottleneck in machine learning projects. By using SSL, the researchers were able to train their model with just a fraction of the data normally required while still maintaining high accuracy.
“Our semi-supervised model significantly reduces the workload of data annotation while ensuring reliable rooftop segmentation results,” explains Dr. Yuan.
After identifying the rooftops, the next step involved classifying them into residential, industrial, or rural categories, as the suitability of solar panels varies based on the type of building. For example, industrial rooftops typically have more open space, while residential rooftops may have more obstructions. The research team developed a method that uses image classification techniques to categorize rooftops based on their features. After classification, the model calculated the ratio of rooftop area available for PV installations (referred to as Ra) for each rooftop category.
They applied the method in Longhu District, Shantou City, Guangdong Province, to test its effectiveness. The results were impressive: the researchers found that out of a total rooftop area of 17.2 square kilometers, 12.7 square kilometers were suitable for solar installations. The study estimated that this would translate into a potential PV installed capacity of 1,849.4 megawatts (MW), which could generate approximately 2,219.3 gigawatt-hours (GWh) of electricity annually.
This is enough to power thousands of homes and significantly reduce the district’s carbon footprint. The implications of this research are wide-reaching. As cities worldwide continue to grow and energy demands increase, accurately assessing rooftop PV potential can help governments and energy companies make informed decisions about where to invest in solar infrastructure. By leveraging free, publicly available remote sensing data, this method offers a cost-effective solution for assessing solar potential across large urban areas.
“This approach can easily scale to assess rooftop PV potential in other regions, helping to accelerate the global transition to renewable energy,” says co-auther, Jinhao Yang.
Furthermore, the limited availability of land for solar farms in densely populated cities could make this technology particularly beneficial. Rooftop PV systems offer a way to generate clean energy without the need for additional land. By maximizing the use of existing infrastructure, cities can reduce their reliance on fossil fuels and contribute to climate change mitigation efforts.
Looking ahead, the researchers plan to refine their model by incorporating additional data, such as rooftop orientation and shading, which could further improve the accuracy of PV potential assessments. They also aim to expand their methodology to encompass various building types, including commercial properties, and investigate the applicability of their approach in regions with varying building styles and climates.
For more, you can visit: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4900382
