This article investigates the performance of four data-driven machine learning models for making large-scale maps of surface albedo from Sentinel-2 satellite data at a 10-meter resolution. Albedo maps are valuable for things like climate modeling, agriculture, energy management, and urban heat island management. The four models evaluated in this study are the random forest, the neural network, total variation, and the gradient boosting algorithm. The evaluation process compared each model against the ground truth albedo values taken from Landsat-8 satellite data.
MDPI (Multidisciplinary Digital Publishing Institute) is an online open-access scientific publishing platform that is a leader in scholarly and scientific journals. Founded in Berne, Switzerland, in 1996, MDPI publishes more than 200 diverse open access journals. It has offices in Basel, China, and Taiwan. MDPI provides open access to its articles through the use of Creative Commons licenses, which allows the free distribution and re-use of content throughout the world.
This article’s author is research professor Jing Su, from the School of Geography at Wuhan University in China. Professor Su has dedicated his career to researching data-driven machine learning models in a variety of contexts, including this paper’s focus on satellite image-making for albedo estimation. His research also includes hydrology, climate change, extreme rain, and mountain wetland ecologies. With extensive expertise in modeling and analysis, Su’s interests span both academia and industry, leading to a wide variety of research papers, articles, and collaborators.
