Characterization and mapping of photovoltaic solar power plants by Landsat imagery and random forest: A case study in Gansu Province, China.

Numbers and sizes of photovoltaic solar power plants have grown unprecedentedly over the last few years in China, which aims to achieve a carbon emission peak by 2030 and carbon neutrality by 2060. Thus, timely and accurate monitoring of photovoltaic solar power plants is crucial to the design and management of renewable electricity systems in China. Random forest algorithm has been used to map photovoltaic solar power plants at multiple scales, however, it always causes several salt-and-pepper noises, limiting its application at larger spatial scales. Here we first develop a photovoltaic solar power plant mapping method through integrating time series Landsat imagery, random forest, and morphological characteristics. Then we apply this method in Gansu Province, which has abundant solar and wind energy resources and provide large amounts of potential lands for photovoltaic development, and generate the annual photovoltaic maps from 2015 to 2020. We further analyze the spatial-temporal dynamics of sizes and areas of photovoltaic solar power plants and major land cover conversion of expansive photovoltaic regions. Finally, we discuss the reliability, uncertainties, implications, and future development of our improved methods. We find our photovoltaic mapping method can remove most of salt-and-pepper noises effectively, and the resultant maps in Gansu for 2020 have very high accuracies with user's and producer's accuracies of 97.57% and 99.22%, respectively. There are 165.29 km2 photovoltaic solar power plants in Gansu for 2020, and most of which are located in the northwestern Gansu. In addition, the photovoltaic with patch size > 1 km2 and ≤ 2 km2 (53.4 km2, 32.3%) has largest patch number (39, 15.7%). The improved photovoltaic mapping methods and further analysis in this study provide critical information for accurate and automatic classification of photovoltaic solar power plants in the future, as well as the environmental and sustainable development of solar energy in China. • Improved random forest mapping algorithm of photovoltaic solar power plants. • An updated annual map of photovoltaic solar power plants in Gansu Province with improved accuracy. • Detailed spatial-temporal dynamics and major land cover conversion of photovoltaic solar power plants in Gansu. [ABSTRACT FROM AUTHOR]