This project focused on the automated detection and analysis of boulders across a large study area in the Atacama Desert. High-resolution drone imagery was used to capture detailed information about the surface and the spatial distribution of individual boulders. Because manually mapping such a large number of objects would be extremely time-consuming, machine-learning-based image segmentation was used to automate the detection process.

I developed and refined a Python-based workflow to process the segmentation results and convert the detected boulders into geospatial data. The workflow enabled hundreds of thousands of individual boulders to be mapped and analyzed consistently. For each detected boulder, geometric properties including size, major and minor axes, aspect ratio, orientation, roundness, and spatial position were calculated. The resulting dataset was integrated with high-resolution digital elevation models to extract information about the surrounding terrain. GIS and Python-based spatial analysis were then used to investigate relationships between boulder characteristics, slope, orientation, and topographic conditions. The automated results were also evaluated against manually mapped boulders to assess the reliability and limitations of the detection approach.

Statistical and spatial analyses were applied to identify patterns within the large boulder population. The resulting workflow provides a scalable method for studying boulder distributions and their relationship with geomorphological processes. Ultimately, the project uses automated image analysis as a tool to better understand surface evolution and possible earthquake-driven boulder movement in the hyper-arid Atacama environment.