Global photovoltaic solar panel dataset from 2019 to 2022

Area of Science:

• Renewable Energy Systems
• Geospatial Data Science
• Artificial Intelligence in Environmental Monitoring

Background:

• Solar photovoltaic (PV) power is crucial for global energy but lacks precise, up-to-date spatial datasets.
• Existing data limitations hinder effective spatial analysis of PV potential and growth.
• Accurate global PV mapping is essential for energy policy and research.

Purpose of the Study:

• To develop a novel method for global PV panel identification.
• To create a high-resolution (20-meter) annual dataset of PV panels from 2019-2022.
• To provide a detailed and accurate resource for spatial analysis and policy-making.

Main Methods:

• A two-stage PV classification framework combining U-Net and positive unlabelled learning with random forest (PUL-RF).
• U-Net identified PVs from sub-meter Google Earth imagery to generate initial positive samples.
• PUL-RF utilized these samples to classify large-scale Sentinel-2 imagery for global PV mapping.

Main Results:

• Achieved over 90% accuracy using IoU and F1-Score evaluation metrics.
• The generated dataset offers superior precision and spatial detail compared to existing global PV datasets.
• Documented a remarkable global PV capacity increase exceeding 60% between 2019 and 2022.

Conclusions:

• The developed method and dataset provide unprecedented detail for global PV analysis.
• Developing countries are significantly contributing to the rapid growth in solar PV deployment.
• This resource will support future research, energy planning, and climate change mitigation strategies.