HYDRO-GEOMORPHOLOGICAL DYNAMICS IN ALASKAN ICE WEDGE POLYGONS: A DEEP LEARNING APPROACH

Conventional methods for mapping and analyzing trough networks in Alaskan ice wedge polygons present significant challenges, typically necessitating intensive fieldwork and manual interpretation of satellite imagery. This study proposes a deep learning-based approach to automate and refine the detection and segmentation of these trough networks using high-resolution satellite imagery. Our novel method integrates the YOLO (You Only Look Once) algorithm for initial feature identification with the U-Net architecture for detailed pixel-level segmentation. The YOLO-U-Net combination, proven effective in medical imaging for detailed feature detection and aiding in new disease identification, shows promise for enhancing hydro-geomorphological studies and our understanding of Arctic environments. This innovative approach is anticipated to facilitate the automated detection and delineation of trough networks across diverse environmental conditions, aiming to achieve both accurate and precise results. Beyond detection and segmentation, this research aspires to conduct a speculative analysis of the morphological characteristics of trough networks, particularly focusing on their potential responses to specific environmental factors, such as temperature variations, seasonal changes, and vegetation. and precipitation patterns. This aspect of the study is expected to yield new insights into the adaptability and evolution of these networks in response to shifting climatic conditions. Our ongoing research is set to enhance the understanding of landscape formation and change in Arctic and tundra environments, showcasing the capabilities of deep learning in advancing geomorphological investigations. By establishing a novel standard in methodological approaches for cold region landscapes, this study aims to provide scalable and adaptable solutions for similar geomorphological challenges on a global scale.