Skip to content Skip to sidebar Skip to footer

Recent studies have shown that our oceans are warming at rates greater than initially forecast and increasing the negative impacts of global warming and sea level rise as polar ice melts.

Unfortunately, melting ice and increased ocean temperatures are not the only water-related threats we are facing. Scientists have also discovered the release of methane from thawing permafrost in arctic regions across the globe. Yet, it is difficult to quantify such emissions due to spatiotemporal factors. Nevertheless, a research team from the University of Alaska Fairbanks, in the U.S., and Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, in Germany, have developed a novel approach to calculate such emissions based on Trimble eCognition’s Object-based Image Analysis (OBIA) tools. They have presented their results in a recent paper entitled “An Object-Based Classification Method to Detect Methane Ebullition Bubbles in Early Winter Lake Ice” published in MDPI’s remote sensing journal.

The researchers discovered that the methane bubbles released from the thawing permafrost are visible in lake ice surfaces at high latitudes when the bubbles emerge from the lake bed and become trapped by the downward growing ice. “This temporary preservation provides an opportunity to quantify the spatial distribution and characteristics of methane bubbles and to correlate seep fluxes with trapped bubble size”. The bubbles are visible in high-resolution (9-11 cm) optical aerial images, making a remote sensing-based approach attractive due to the environmental challenges that come with field work.

The authors organized the analysis into a 4-level image object hierarchy to take advantage of domain-based processing, giving them increased flexibility and transferability within class descriptions as well as more efficient processing.

Level I simply split the image object domain between 2 broad classes: land and lake. A combination of spectral, spatial and size criteria were used to make a robust classification, especially on shoreline areas.

Level II examines the lake area in further detail. The lake area is split into 3 classes: (1) lake ice with vegetation, (2) lake ice without shadow, and (3) lake ice with shadow. Again, a multiresolution segmentation is used further segment the lake area class at finer scale. Subsequently, a mix of spectral and contextual features were used to discriminate the 3 classes from one another.

Level III is used to focus the analysis on the lake ice class, specifically looking to differentiate between dark ice and white ice based on the spectral characteristics.

Level IV then analysed the dark and white ice classes to identify ebullition bubble patches trapped in specific lake ice zones. Bubble objects were initially divided in 3 stages representing the bubble edges, bubble centers and finally a merger of bubble edge and center objects. A Canny edge detection layer was calculated and objects enclosed by bubble edge objects were classified as bubble center. Finally, a combination of segmentation and classification algorithms were used to refine the bubble objects and isolate single bubble – a wonderful example of image object refinement through the use of the contrast split segmentation and pixel-based operations.

The results performed very well when compared to manual interpretation – unfortunately, ground truth data could not be collected at the time of data acquisition to the ice thickness (<5 cm). “The multi-level segmentation and classification rule sets that we adopted performed very well to identify the final target objects, ebullition bubble patches. Bubble patches were mapped with an overall accuracy of 95.1% and 98% for the year 2011 and 2012, respectively”.

The authors conclude that “optical remote sensing-based lake ice classifications will be helpful to better estimate methane emission from cold-climate lakes at regional scales” but also not that the use of aerial images is limiting due to the crucial temporal characteristics of the ice conditions.

I would be curious see if the use of UAVs could give the authors more freedom for data acquisition. The spatial resolution could also be improved – could this also help the researchers find additional seeping at a finer scale?

This paper illustrates yet another fascinating use of our Trimble eCognition software. It is amazing to see the different application fields our users continually find!

Follow Us on Social Media

For more news on eCognition