Skip to main content

Summer Internship Series: Descartes Labs

Ahmad Hojati Malekshah was born in Babol, Iran, and received his Bachelor’s in Surveying Engineering from K.N. Toosi University and his Master’s in Geodesy and Hydrography Engineering from the University of Tehran. While looking for a Ph.D. program  Hojati Malekshah found Dr. Nancy Glenn’s work in remote sensing and noticed that her research group was working on active sensor applications (e.g. lidar) in mapping snow and snow-vegetation interaction as a part of the NASA SnowEX project. The NASA SnowEX project offered a wonderful opportunity to be in a professional remote sensing community and work with high-quality scientists across the United States. “To be effective in these communities you need to develop your skills to handle big data and have knowledge in data science. In this regard, I contacted Dr. Glenn and after her approval applied for the Data Science emphasis in the Computing Ph.D. program at Boise State.”

Hojati Malekshah found out about the Descartes Labs (DL) internship through an email newsletter and was interested in the work they were doing on radar. During the internship, he worked as an applied scientist working in their radar remote sensing division. “My work was a part of research and development. They are providing global and local time series of land surface displacement using interferometric synthetic aperture radar (InSAR) technique. InSAR is a technique that maps subsidence, landslides, dam failures, earthquakes, etc. Therefore, knowing the land surface movement and their evolution in time can be an asset for cities, dams, and mines. DL uses European space agency’s Sentinel 1 A and B data with 10-20 m resolution with 6 days repeat orbit combined. Radar waves could face long-term and short-term decorrelation which means low-quality measurement from the ground if we have dense vegetation, snow or any disturbance that changes ground surface. In terms of long-term decorrelation, almost no information can be retrieved from the region. However, short-term decorrelations could be temporal and/or seasonal (like snow cover in winter) and return high-quality information except that short period. Most InSAR processing methods lose displacement pixels if we have long-term and/or short-term decorrelation. At Descartes Labs, my work was developing and implementing an algorithm that can preserve pixels in case we have temporal-decorrelation.”

When asked what the best part of the internship was, Hojati Malekshah said, “My favorite part was a weekly meeting we had to plan for the work in the following two weeks. Everyone had to add a title and description to Jira and explain what are the steps she/he is taking to complete the project and how much time is required. All the group listened to the steps and description and voted for the number of days it would take to finish a task. It was interesting that all the group together tackled a problem and how you manage your time during the week for completing a task.” He said the experience taught him how to be more effective and organized in his projects, and he appreciated the opportunity to meet people from other backgrounds and universities and network.

Currently, Hojati Malekshah is working on applying machine learning and deep learning techniques on radar and lidar (airborne and terrestrial) data to map the snow depth, pattern, density, and their relationship with the vegetation structures at different spatial scales. Once he completes his Ph.D., Hojati Malekshah intends to work in the private sector specializing in the application of data science in remote sensing of the environment.