Washington Fire Scar Atlas
A Machine Learning-Based Classification and Interactive Web Visualization of Burn Scars in Washington State using Google Earth Engine and 40 years of LANDSAT Satellite Imagery
The Washington Fire Scar Atlas is a project I designed and authored as a final Senior Research capstone project for the Geography Department at Middlebury College. I am proud to call Washington State my home, but over the past decade, the state has experienced wildfires that have destroyed towns, taken lives, and covered cities in dark smoke. The goal of this project was to build upon existing methodologies for imagery classification to build an algorithm that could identify burn scars from publicly available Landsat imagery. This result could be beneficial to those attempting to model and fight fire as well as those hoping to get a deeper understanding of the extent to which fires have grown in recent years. To do this I trained a random forest machine-learning classifier on hundreds of training points that differentiated burn scars from other types of land cover such as water or bare ground. I then generalized my script such that my algorithm could calculate burn scars for any year between 1984 and 2023, only needing the year as an input. These steps resulted in a Google Earth Engine script that could be applied across many more geographies to estimate burns and patterns of burn. Finally, I created a Google Earth Engine web app to explore some of the history and patterns of fire that were exhibited by my data. I had some experience in Google Earth Engine and making maps of wildfires, but this project tested my ability to dive into new subjects such as machine learning, advanced remote sensing, and web app development. I hope to build on the skills learned in this project to work on more projects related to GIS, remote sensing, climate vulnerability and resiliency, public planning, and science visualization.