In Japan, the rapid expansion of solar farms since the Feed-In-Tariff (FIT) policy in 2012 has raised environmental and hazard-risk concerns, necessitating a comprehensive PV inventory for quantitative assessment of those challenges. This study presents a national-scale PV map created through the application of machine learning to remote-sensing imagery, enabling a thorough assessment of environmental changes and hazard risks of existing PV sites. By applying the trained XGBoost model on Sentinel-2 imagery acquired from September to October 2022, we calculated the likelihood of solar PV presence at each pixel followed by noise reduction with morphological filtering. The final solar PV map was produced by applying a threshold on the probabilistic output, and it was further converted into polygons representing solar PV perimeters. The resulting PV map efficiently reduces false positives and could be used to update existing databases both made from manual and machine-learning methods. Comparison with land-cover data revealed significant land-use changes due to PV installations, particularly in forested and agricultural areas. Hazard-risk assessment identified 30.0% of solar farms in flood-prone areas and 3.2% in landslide-prone zones. This study underscores the need for environmental protection and hazard mitigation measures to further advance solar power in Japan through remote-sensing-based national PV mapping and GIS analysis.
Previous Article in event
Previous Article in session
Next Article in event
Next Article in session
Sentinel-2 based solar PV mapping and applications for land-cover change assessment and hazard-risk evaluation in Japan
Published:
16 April 2024
by MDPI
in OHOW 2023 – The 2nd International Symposium on One Health, One World
session Infrastructure Management and sustainable built environment
Abstract:
Keywords: Solar PVs, land-cover mapping, land-cover change, machine-learning, disaster mitigation
Comments on this paper