MIT Department: Aeronautics and Astronautics
My name is Luiz Fernando Leal, I am from Rio de Janeiro, Brazil, and I am a rising senior student in Aerospace Engineering with a minor in Computational Mathematics at Florida Institute of Technology. I plan to pursue a Ph.D. in Geosciences and Planetary Sciences to become a professor who will lead major scientific expeditions throughout Earth and the solar system. I am primarily interested in Planetary Hydrology and Paleogeology, along with Data Science and Artificial Intelligence (AI) to understand how past activities influenced the present-day realms of the Earth and other celestial bodies. I want to advance data interpretation gathered from satellites, rovers, and astronauts to expand the frontier of knowledge of how Earth and other planets were formed, and use that knowledge to inform decision-makers about space exploration.
2019 Research Abstract
Finding Anthropogenic Dark Earth in the Amazon Basin Using Fully Convolutional Networks
Luiz Fernando Leal1,2, Morgan J. Schmidt2, Samuel Goldberg2, J. Taylor Perron2, Dorothy Hosler3, Michael Heckenberger4, Wetherbee Dorshow5, Bruno Moraes6, Eduardo Kazuo Tamahana7, Kumessi Waura8, Hulke Kuikuro8, Wate Kuikuro8, Afukaka Kuikuro8
1Department of Aerospace, Physics and Space Sciences, Florida Institute of Technology
2Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology
3Department of Materials Science and Engineering, Massachusetts Institute of Technology
4Department of Anthropology, University of Florida
6Department of Anthropology, University of Pittsburgh
7Mamirauá Institute for Sustainable Development
8Kuikuro Indigenous Association of the Upper Xingu (AIKAX)
Amazonian Dark Earth (ADE) is anthropic soil formed by indigenous peoples in the Amazon Basin beginning at least 5,000 years ago. The remarkable fertility and elevated carbon content of ADE make it relevant to global concerns about sustainable landuse, tropical deforestation, and climate change. Although numerous ADE sites have been identified, the geographic distribution and total extent of ADE are unknown. The size and terrain of the Amazon Basin make it impractical to conduct an inventory of ADE through systematic field surveys. We aim to develop an automated procedure for detecting ADE locations by applying machine learning techniques and spatial statistics to remote-sensing data. Our approach is based on field observations indicating that vegetation growing on patches of ADE has characteristic features in terms of shape, color, and spatial distribution. We first developed a proof-of-concept workflow designed to identify indigenous village locations, which also have characteristic geometric features. Beginning with satellite images (acquired from Planet Labs Inc.) covering an area of interest in the upper Xingu River Basin, we manually constructed a mask to distinguish village and non-village areas. Image chips were extracted and used to train a fully convolutional network (FCN) model. Image augmentation was applied to increase the accuracy of the model. Even with a small training dataset, this approach correctly identifies more than 75% of villages in the study area. Drawing on a set of known ADE sites in the same region based on field mapping, we extend the workflow to the detection of ADE. The ultimate goals of our mapping efforts are to understand the role of ADE in Amazonian societies, improve estimates of long-term carbon storage in anthropic soils, and explore modes of sustainable farming in tropical regions.