Collaborative Research: CAIG: Data Science Frontiers in Advancing Predictive Understanding of Landscapes and Erosional Extremes under Changing Climatic Scenarios

The amplification of extreme events, more frequent flooding and intense fires caused by climate change, combined with escalating anthropogenic alterations like deforestation and changing land-use practices, have led to a period of rapid transformation across landscapes. These transformations unfold over a wide variety of space and time scales. Some landscape transformations such as landslides and debris flows happen very rapidly and generally only affect localized areas. Other processes like the migration of rivers are more gradual but may affect the environment, ecosystem services, and our society across entire regions. Attempting to collect scientific observations of all types of landscape transformations in the field, regardless of size or duration, would be impractical and expensive. This research instead capitalizes on a unique approach that uses artificial landscapes created in a laboratory to create a robust datasets of landscape observations that mimic real-world conditions. Laboratory simulations produces datasets both large enough and realistic enough to be analyzed by AI techniques, which in turn help to understand the environmental conditions that control and trigger extreme erosion events. The overall goal of this research is to advance our predictive understanding of landscapes and identify the most relevant geomorphic variables driving erosion under various climatic forcings, including the emergence of highly erosional events. It is expected that this effort will inform the development of predictive models that can be used for landscape planning and management. The specific objectives of this research are: (1) Extract the local geomorphic transport laws driving landscape evolution under different climatic forcing using the large-scale experimental data and novel physics-informed explainable Artificial Intelligence (xAI) methods, (2) Quantify the imprint of structural connectivity (neighborhood dependence) on emergent behavior and nonlocal transport laws using geomorphologically-inspired Graph Neural Networks that acknowledge the landscape connectivity emerging from the flow accumulation process, and (3) Leverage the transferability of ML models (Transfer Learning) for identification of areas of extreme erosion where fewer observations are available, with a specific application to post-fire hazard assessment. This award by the Division of Research, Innovation, Synergies, and Education within the Directorate for Geosciences is jointly supported by the National Discovery Cloud for Climate initiative within the Office of Advanced Cyberinfrastructure within the Directorate for Computer and Information Science and Engineering. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.