Scalable and reproducible mapping and analysis of Arctic landscapes

Possible questions:

• Where are diagnostic permafrost landforms like water tracks located across the Arctic?
• What climate, ecosystem, or geologic factors influence the shape of permafrost landform and drainage networks, and their susceptibility to disturbance?
• In machine learning models for Arctic landscapes, what is the optimal balance of model generalization and additional input data and/or type of data?
• What is the optimal balance in size/nature of training data for model success?
• Can high-resolution training datasets produce accurate, upscaled results on low-resolution prediction datasets?

We've never had better computers, algorithms, and satellites to observe and analyze Earth's surface, especially in the Arctic - but most Earth scientists aren't trained to work with those big datasets and complex software. To overcome this steep learning curve, our project combines geomorphology with advanced computational tools for acquiring, analyzing, and visualizing large interdisciplinary datasets. The Python Computational Geomorphology Software System (PyCoGSS) helps scientists of all experience levels (including and especially undergraduates!) conduct reproducible and scalable analyses of Arctic landscapes. Working with data in a coding environment helps the workflow grow with the scientist's curiosity (scalable) and makes it easy to share your work with others and keep track of how you've manipulated the raw data (reproducible).

We have a grant from the National Science Foundation's Office of Advanced Cyberinfrastructure to build software that helps answer both software and science questions about collecting, automatically mapping, and analyzing permafrost landscape features. In the process, we're also developing tutorials and curricular material so that everyone can learn to code and do computational geomorphology!