Iksha Gurung: Leading IMPACT’s Machine Learning Team

Having grown up in Nepal a mere few hundred kilometers from the base of Mount Everest, Iksha Gurung knows a thing or two about mountains. Although he now lives half a world away from Earth’s tallest peak, Iksha is still surrounded by mountainous terrain. However, these peaks aren’t forged from rock and capped with snow. In his role as a UAH computer scientist and leader of the IMPACT development and machine learning (ML) team, Iksha routinely scales mountains made of data, using algorithms to blaze trails for Earth science researchers.

As an undergraduate at Kathmandu University, Iksha was intrigued by the power of complex ML algorithms and the potential they have to reveal patterns and insights within massive datasets. He graduated in 2013 with a bachelor’s degree in computer engineering and began working as a software engineer for CloudFactory, a data management company with offices in Nepal. During his time at CloudFactory, Iksha helped design, develop, and maintain the platforms used to create labels for datasets that trained ML models. He often interacted with the company’s data science team, which deepened his interest in ML capabilities.

Image

After working in industry for a few years, Iksha became interested in expanding his ML skills through additional education. He applied to the master’s program in computer science at the University of Alabama in Huntsville (UAH) and soon began the next chapter of his life on the other side of the world from his native country.

Iksha started his coursework in 2016 and also joined the UAH Data Science Informatics Group as a graduate research assistant. While completing his master’s degree, he worked on web and mobile app development and also gained his first experience with applying machine learning to Earth observation data, using ML to pinpoint hailstorms from radar data. When the Interagency Advanced Implementation and Concept Team (IMPACT) was formed in 2018, Rahul Ramachandran, the project manager of IMPACT, invited Iksha to join the team as a research associate. Iksha accepted and soon began applying his ML skills to Earth science data by designing and developing cloud-based applications and applying algorithms to make datasets more accessible and searchable. His determination to build effective, open-source tools led him to quickly become an integral part of IMPACT’s mission to promote open science. Iksha claims that he just “fell into working with Earth data,” but admits that he takes pride in the influence that his ideas and efforts have had on many NASA Science Mission Directorate (SMD) projects.

Image

In 2020, Iksha was promoted to lead developer of the IMPACT development and ML team and has continued to add valuable insight and leadership to all IMPACT teams. When all 90+ members of IMPACT meet biweekly to review accomplishments and challenges for their array of projects, Iksha’s name is invoked numerous times as the guy responsible for proposing and spearheading solutions for everyone’s data needs. He is constantly propelling himself and others forward in both a leadership role and as a scientist who is excited to explore advancements in technical architecture.

Iksha describes his and other IMPACT team members’ work as being “on the leading edge for datasets and data systems.” The cloud-native applications they implement are novel in their ability to scour enormous Earth science datasets. For example, the recently released Similarity Search tool uses advanced self-supervised learning (SSL) algorithms, which are not reliant on labeled training datasets, to automatically detect visible atmospheric phenomena and ground features from satellite imagery. The tool leverages Moderate Resolution Imaging Spectroradiometer (MODIS) datasets provided by the Level-1 Atmospheric Archive and Distribution System Distributed Active Archive Center (LAADS DAAC) accessed through NASA’s Global Imagery Browse Services (GIBS). Similarity Search can rapidly compile image collections of everything from dust storms to coral reefs and allows users to expedite discovery of relevant satellite images within the archive by using a sample image for comparison. This creative application of SSL supports the Earth science research community by reducing the considerable time and effort required to sift through satellite imagery.

Image

Another opportunity to apply machine learning to Earth observations emerged during the COVID-19 pandemic. Iksha and his team realized that massive shifts in global human and industrial behaviors in response to the pandemic could be tracked through analysis of satellite imagery. They soon became involved in constructing the COVID-19 Dashboard, a tool that streamlines access to NASA satellite, airborne, and ground-based observation data for several major geographical regions. Now featured on NASA’s Earthdata website, the dashboard allows researchers and the public to explore pandemic-related effects on our planet and its atmosphere. Iksha and the IMPACT ML team focused on applying AI to identify and track ships, producing insights into how the shipping industry and global economy were disrupted by travel restrictions.

Image

While Iksha’s workflow often involves developing tools used for probing Earth observation data, he also leads endeavors to assess and improve data quality and access. The Analysis and Review of the CMR (ARC) project managed by IMPACT team members is an ongoing effort to assess NASA’s metadata records contained in the Common Metadata Repository (CMR). This centralized metadata repository catalogs records for tens of thousands of Earth observation data products. Iksha and his team supported this project by creating pyQuARC (pronounced “pie-quark”), an automated assessment tool for evaluating metadata. This open-source tool identifies ways to enhance contextual metadata information, which ultimately helps researchers find and use data products.

Iksha was also involved in designing and building an open-source tool called ImageLabeler. Earth science researchers often require accurately classified images in their investigations, but compiling such data can be difficult and time-consuming. ImageLabeler allows users to tag images that contain certain visible features, and the images can then be used to train image-based ML models. Researchers in any field who rely on labeled image datasets to extract information and patterns could benefit from ImageLabeler’s capabilities. This achievement in particular highlights how Iksha and his team aspire to accelerate the pace of scientific discovery by advancing ML methods.

When asked to share what he enjoys most about his work, Iksha says:

"I enjoy being part of a team that is utilizing the latest and greatest of technologies, especially artificial intelligence and machine learning, along with the experience and knowledge of the extremely talented team members to make science available to all."

Iksha’s achievements and talents were recently recognized by the UAH Earth Systems Science Center (ESSC) when they presented him with the ESSC Meritorious Service Award for 2022. The award description emphasized how Iksha is remarkable in his ML-based approaches to meeting “one of the greatest challenges in science — to discover something new and interesting from an ocean-sized bucket of data.”

Image