Quantifying the Spatiotemporal Inundation Patterns of Tundra Polygonal Ground and Its Relationship with Arctic Climate
Restricted (Penn State Only)
- Author:
- Di Domenico, Nicolle
- Graduate Program:
- Geography
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- June 27, 2023
- Committee Members:
- Shujie Wang, Thesis Advisor/Co-Advisor
Manzhu Yu, Committee Member
Brian King, Program Head/Chair
Elizabeth Herndon, Special Signatory - Keywords:
- geomorphology
Arctic
polygonal ground
geography
remote sensing
machine learning
climate change - Abstract:
- Polygonal ground (≤5-30 m diameter), a common Arctic tundra landscape, exerts an important control on small and large scale tundra hydrology. Spatially complex polygonal ground networks facilitate unique hydrology, affecting the quantity and type of carbon emitted from soils. The two main types of polygonal ground are considered dry and wet end members: high-centered polygons (HCPs) and low-centered polygons (LCPs), respectively. During the Arctic wet season, HCPs are assumed to remain dry and promote the release of soil carbon as carbon dioxide, and LCPs are assumed to remain wet and promote the release of soil carbon as carbon dioxide and methane. However, their inundation dynamics are poorly understood over space and time. Characterizing the spatial and temporal variance of HCP and LCP inundation is critical to understand current carbon emissions and project how emissions may change as the Arctic rapidly warms. We use WorldView-3 satellite imagery over an area of Utqiagvik, Alaska to map HCPs and LCPs at a regional scale using a convolutional neural network deep learning algorithm. The regional LCP and HCP classification model yielded 96% classification accuracy and identified 15.9 km2 of regional LCPs and 27.43 km2 HCPs. Sentinel-2 multispectral imagery of 24 time stamps spanning six wet seasons (2016-2021) are used to investigate LCP and HCP inundation trends over space and time. It is shown that LCPs experience more inundation both spatially and temporally, while HCPs experience less inundation on average. ERA5-Land climate reanalysis data are combined to understand the connections of inundation to key climate variables, indicating that evaporation and precipitation may have more influence on both LCP and HCP inundation than air temperature. These results are at an optimal scale for inclusion in terrestrial carbon models and can provide a path forward for better constraining global carbon and climate models as Earth’s climate rapidly changes.