Colorado is home to several magnificent outdoor sites. This picture was taken in March 2023 by the frozen Sprague Lake in the Rocky Mountains National Park.

Hi, I'm Rafael. I’m from Brazil and my background is in geophysics, specifically in geological mapping and exploration. Since joining Morteza Karimzadeh’s team as Postdoctoral scientist in June 2022, I have been mainly working on developing new algorithms to map sea ice in polar regions using remote sensing data, with a particular focus on Synthetic Aperture Radar and machine learning techniques. 

My current research has important implications for understanding the dynamics of sea ice, its impact on climate and ecosystems in polar regions, and for safer navigation in ice-infested waters. Leveraging my expertise in geophysics, I bring a unique perspective to my work on this project. I am passionate about using spatial data to better understand the Earth's natural systems and visualize them in new and innovative ways. 

In addition to my research, I also enjoy mentoring and collaborating with students in my lab group to help them develop their skills and gain new insights. I believe that working with students is an essential part of being a researcher, and I find it extremely rewarding to learn from their fresh perspectives and develop new approaches to solving complex problems. 

I am especially excited about the outcome of my work as the co-investigator on a NASA proposal that focuses on using ICESat-2 lidar data to map sea ice in polar regions. We recently received notification that this project will be funded by NASA, helping us explore an alternative and novel approach to mapping sea ice, overcoming significant uncertainty in traditional ice charts and SAR measurements.   

Overall, I feel grateful to be part of the Department of Geography and work alongside such a talented group of researchers and scholars. I look forward to continuing to explore new frontiers in geospatial data analysis and making meaningful contributions to our understanding of the world around us.

Convolutional neural networks (CNNs) are useful for processing data organized in grids, such as remote sensing images. CNNs are composed of multiple filters that are iteratively updated to highlight different patterns in the data in a process called “training”. The image above is a representation of a collection of filters that are randomly initialized (left) and become more organized as training evolves (center). These filters are useful to separate ice from water (right).

The image on the left shows an RGB composition of Synthetic Aperture Radar from Sentinel-1 acquired on the East Coast of Greenland. Sea ice services use these types of data to manually interpret sea ice properties for large regions of the Arctic. The brighter “cyan” colored ice on the upper left contrasts to the darker water on the lower right. The image on the right shows the ice probability generated by an artificial intelligence algorithm. Yellow shows high ice probability and blue identifies low probability regions. The white dots show the boundary associated with ice interpreted by a sea ice analyst. We are developing algorithms that accelerate sea ice interpretation.

Off

Rafael Augusto Pires de Lima
Post-Doc, Department of Geography
University of Colorado Boulder

In Person:
GUGG 205
Feb 24, 2023, 3:35 PM - 5:00 PM

Or Join Zoom Meeting:
Zoom login required (free account available at )

Abstract

Earth Scientists have used machine learning for at least three decades and the applications span is large, from remote sensing to analysis of well log data, among many others. Although machine learning is becoming more popular in different fields of Earth Sciences, some concepts of convolutional neural networks may be vaguely understood by non-practitioners. In this presentation, I show some of the key concepts on the foundations of convolutional neural networks, some techniques that can help us understand them, and strategies that can be used to train these types of models. The objective of the presentation is not to focus on mathematical and implementation details, but to help build the intuition necessary to use and analyze the outputs of convolutional neural network models. The case studies include microfossil classification and core lithofacies classification, both using colored photographs. The final case study shows how convolutional neural networks can be used for sea ice segmentation using synthetic aperture radar. 

Bio

Rafael has a bachelor in Geophysics from the University of São Paulo, a MSc. in Geophysics, a second MSc. in Data Science and Analytics, and a Ph.D. in Geophysics from the University of Oklahoma. His research interests include developing machine learning algorithms for different geospatial and geoscience applications. He also has experience with advanced image processing techniques for seismic attributes development, as well as data analysis for several geophysical methods. Rafael's current research is funded by a National Science Foundation grant focused on developing machine learning models for sea ice mapping using Synthetic Aperture Radar images and other remotely sensed data. 

Watch the Presentation

[video:https://vimeo.com/802891538]

Off