The authors have developed a comprehensive global dataset of multispectral satellite imagery of solar panel farms, with the goal of facilitating the development of advanced machine learning models for monitoring the growth and distribution of solar energy infrastructure worldwide.
The dataset construction process involved three key steps:
Gathering an initial training dataset from diverse sources, including historical global ground truth data, manually annotated US ground truth data, and Sentinel-2 multispectral satellite imagery spanning 2017-2023.
Training state-of-the-art (SOTA) machine learning models on the initial dataset, with the top-performing model achieving an impressive 96.47% Intersection over Union (IoU) and 98.2% F-score in accurately segmenting solar panel areas.
Employing the top-performing models to predict and label data for the global dataset covering 2021-2023, followed by a rigorous quality control process to ensure the dataset's precision and reliability.
The resulting dataset, named GloSoFarID, comprises a total of 13,703 data samples, spanning various global regions and covering the years 2021 to 2023. It features multispectral satellite images with 13 spectral bands and a spatial resolution of 10 meters per pixel, providing a rich and comprehensive resource for solar energy research and analysis.
The authors have also evaluated the dataset using benchmark semantic segmentation models, including Fully Convolutional Network (FCN), U-Net, and Half-UNet, establishing a performance baseline and enabling fair comparative evaluation in future studies.
This dataset represents a significant contribution to the solar energy research community, as it addresses the limitations of existing resources and offers a more diverse, up-to-date, and multispectral dataset for accurately mapping and analyzing the global expansion of solar panel farms. The insights gained from this dataset can inform decision-making and guide the transition towards a sustainable energy future.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Zhiyuan Yang... at arxiv.org 04-09-2024
https://arxiv.org/pdf/2404.05180.pdfDeeper Inquiries