Date of Award
Master of Science
Climate change poses a risk to individuals whose livelihoods depend on the health of glacier ecosystems. Monitoring glaciers in the Himalayan Hindu Kush (HKH) region is of high importance especially when we consider the impact of recent climate change on them. Our work aims to provide an automated method to outline glaciers using machine learning techniques and publicly available remote sensing imagery.In this work, we present ways to delineate glaciers from Landsat-7 imagery using various machine learning and computer vision techniques. The multi-step methodology that we present in this work is generalizable across different types of satellite and overhead imagery, lending itself to map other geomorphological features on the Earth's surface. Furthermore, we compare quantitatively and qualitatively the performance of pixel-wise classification using conventional machine learning to a more recent deep learning based architecture, U-Net. Our proposed works consist of integrating conventional computer vision methods with deep learning based approaches to improve the segmentation performance and later generalize across other landcover mapping applications beyond glacier mapping. Despite being faster to train, pixel-wise classification approaches generate segmentation masks that are fragmented. On the other hand, the problem of fragmented prediction masks is visually less apparent when using a U-Net architecture. This could be attributed to the properties of Convolutional Neural Networks which are able to take spatial information into consideration. Specifically, pixels in the predicted segmentation mask using U-Net architecture are computed by taking a neighborhood of pixels in the input image, as opposed to one pixel at a time, resulting in less fragmentation. We also analyze the features of satellite images that are most helpful in classification of glaciers in the HKH region. Based on the domain knowledge, we calculate and add slope, elevation, and spectral indices (i.e. Normalized Difference Snow Index (NDSI), Normalized Difference Vegetation Index (NDVI), and Normalized Difference Water Index (NDWI)) as additional bands on top of the bands from Landsat-7 satellite to help with the segmentation task. We observed that slope, elevation, and NDSI contribute the most towards the final segmentation mask. However, these three channels are not present in Landsat-7 satellite scenes and need to be calculated separately. These findings can change the way people view glaciers and the features associated with it, leading to a better understanding in monitoring them. Overall, we present multiple methods for mapping geomorphological landscape features from overhead imagery. These methods can have major implications in understanding global challenges such as climate change and anthropogenic impacts to ecosystems (i.e. deforestation, urbanization, land use change), particularly due to the large volumes of public freely-available large-scale satellite images made available in recent years. We expect to present a novel method that is optimized for the task of glacier mapping using a combination of deep learning and conventional computer vision methods. We also expect to present an optimal architecture for glacier delineation and deploy it in the form of a tool that will automate the process and also be able to facilitate the delineation of glaciers on satellite images acquired from other sensors.
Received from ProQuest
Aryal, Bibek, "Glacier Segmentation In Satellite Images For Hindu Kush Himalaya Region" (2020). Open Access Theses & Dissertations. 3140.