An Object-Based Classification Method to Detect Methane Ebullition Bubbles in Early Winter Lake Ice

Abstract

Thermokarst lakes in the Arctic and Subarctic release carbon from thawing permafrost in the form of methane and carbon dioxide with important implications for regional and global carbon cycles. Lake ice impedes the release of gas during the winter. For instance, bubbles released from lake sediments become trapped in downward growing lake ice, resulting in vertically-oriented bubble columns in the ice that are visible on the lake surface. We here describe a classification technique using an object-based image analysis (OBIA) framework to successfully map ebullition bubbles in airborne imagery of early winter ice on an interior Alaska thermokarst lake. Ebullition bubbles appear as white patches in high-resolution optical remote sensing images of snow-free lake ice acquired in early winter and, thus, can be mapped across whole lake areas. We used high-resolution (9–11 cm) aerial images acquired two and four days following freeze-up in the years 2011 and 2012, respectively. The design of multiresolution segmentation and region-specific classification rulesets allowed the identification of bubble features and separation from other confounding factors such as snow, submerged and floating vegetation, shadows, and open water. The OBIA technique had an accuracy of >95% for mapping ebullition bubble patches in early winter lake ice. Overall, we mapped 1195 and 1860 ebullition bubble patches in the 2011 and 2012 images, respectively. The percent surface area of lake ice covered with ebullition bubble patches for 2011 was 2.14% and for 2012 was 2.67%, representing a conservative whole lake estimate of bubble patches compared to ground surveys usually conducted on thicker ice 10 or more days after freeze-up. Our findings suggest that the information derived from high-resolution optical images of lake ice can supplement spatially limited field sampling methods to better estimate methane flux from individual lakes. The method can also be used to improve estimates of methane ebullition from numerous lakes within larger regions.

A synthesis of methane dynamics in thermokarst lake environments

“…The 14 C-CH4 age (up to 28,500 ±140 yrs BP) and δD (-383 to -413‰) of ebullition bubbles emitted from Vault Lake reflect Pleistocene-yedoma carbon and hydrogen (from Pleistocene ice wedges) as the atomic components of the CH4 molecules, as opposed to more modern (Holocene) sources which have younger/heavier CH4 isotopic values (Brosius et al, 2012). This agrees with ebullition bubble mapping (Lindgren et al 2016(Lindgren et al , 2019Walter Anthony et al 2016;Walter Anthony et al, submitted) and modeling studies (Kessler et al, 2012) of thermokarst lakes that suggest recently-thawed former permafrost at and directly above the thaw boundary is a source of high CH4 production. The high CH4 production at the thaw boundary may be due to high bioavailable substrate potential.…”

Integrating Perspectives to Understand Lake Ice Dynamics in a Changing World

“…RGB sensors and multispectral sensors offer an opportunity to study ice phenology and extent in both lakes and rivers (Alfredsen et al, 2018). Researchers at arctic regions have used UAVs to study methane emissions from arctic lakes (Lindgren et al, 2019) during early winter lake ice. UAV‐mounted LiDAR (Light Detection and Ranging) can be used to measure snow depth from differencing coregistered lidar maps (one on bare terrain and one when snow covered), similar to those acquired from airborne and on‐the‐ground instruments (Deems et al, 2013) in terrestrial environments.…”

Unpiloted Aerial Vehicle Retrieval of Snow Depth Over Freshwater Lake Ice Using Structure From Motion

“…UAVs have been utilized for a wide range of applications when monitoring ice, including lake ice phenology, ice jam on rivers (Alfredsen et al, 2018), ice sheet velocity (Chudley et al, 2019), and to monitor methane emissions (Lindgren et al, 2016;Schlobies et al, 2016;Lindgren et al, 2019). The direct observation of snow on ice requires a detailed understanding of the ice dynamics, including the variability in thickness and freeboard to quantify snow depth retrievals accurately, as sea ice features are a complex mix of first year ice, ridges, leads, and hummocky multi-year ice (Tan et al, 2020).…”