AI Detection of Post-Fire Permafrost Degradation
Naughton Fellow & REU Researcher - Rocha Lab (Notre Dame, Indiana & Toolik Field Station, Alaska)Problem: Arctic fires accelerate permafrost thaw and ice-wedge degradation. Ice-wedge troughs are narrow, linear features that are difficult to map manually. Models have been built to tackle this, but have not been tested fully.
Solution: Evaluated and optimised a U-Net Convolutional Neural Network to identify ice wedge troughs in various kinds of satellite imagery data from the 2007 Anaktuvuk River Fire, the largest recorded fire in the Arctic.
Read Technical Breakdown
1. Computational Research (University of Notre Dame, Indiana): Annotated ice-wedge troughs satellite imagery for ground truth data and experimented with various parameters to optimise AI models.
2. Arctic Field Expedition (Toolik Field Station, Alaska): Conducted fieldwork at Toolik Field Station, America's premier Arctic Research Base in the Arctic Circle. Worked with the Lamont-Doherty Earth Observatory at Columbia University taking thaw depth measurements, measuring soil gas fluxes and chlorophyll fluorescence of leaf samples. Successfully led a helicopter campaign to collect data from multiple remote sites at the 2007 Anaktuvuk River Fire while up against time, fuel and weather restrictions.