Researchers at the University of Isfahan in Iran claim to have improved the precision of digital elevation models (DEMs) in creating high-resolution annual solar energy potential maps.
DEMs are 3D graphics of the bare earth surface of land shown without trees, buildings or other objects. As for their application in building solar energy maps, these graphics consider elevation, slope, aspect, and shadow effects, among other factors. DEMs such as SRTM, ALOS, ASTER, and Copernicus are available as open-access tools online and are currently being extensively utilized to generate solar maps.
“Due to their extensive spatial coverage, they enable the production of solar energy potential maps in most less developed areas with lower costs,” the scientists explained. “However, the low resolution of these DEMs compared to those obtained from LiDAR data reduces the accuracy of extracting solar energy potential maps, which poses limitations for precise location placement of solar panels, especially in urban areas and surroundings of highways,” they added, noting that the LiDAR method, which measures distances by using a laser on a target, has high costs and significant storage requirements.
The research group proposed to overcome these limitations by using deep learning networks to improve DEMs' resolution. Its work consisted, in particular, of training the DEMs to enhance the spatial resolutions of the solar energy maps and identify the best areas for solar panel deployment in urban areas.
The academics utilized the enhanced deep super-resolution (EDSR) network to improve the resolution of the LiDAR-derived solar map. The EDSR is a convolutional neural network suitable for improving image spatial resolution that is widely employed in computer vision tasks.
The performance of this network in improving the precision of the generated solar energy maps was compared to that of the U-Net network, which is a convolutional network technique used for fast and precise segmentation of images.
The scientists found that the EDSR model was the most accurate and stable in enhancing the accuracy of solar energy potential maps. They then used the model to enhance the resolution of a Copernicus DEM-derived solar energy potential map from 30 m to 6 m.
“A comparative analysis between the Copernicus DEM-based map and the EDSR-enhanced map, using a LiDAR-based reference, demonstrated that the EDSR model not only improved the map's resolution but also enhanced the accuracy of solar energy estimation, particularly in urban areas and along major highways,” the research team emphasized. “Finally, the improved 6-meter resolution map was assessed for its effectiveness in identifying suitable buildings for solar panel installation.”
The details of the modeling are available in the paper “Improving the Resolution of Solar Energy Potential Maps Derived from Global DSMs for Rooftop Solar Panel Placement Using Deep Learning,” published in Helyion.
