Identifying polygonal ground in arctic regions using GLCM texture features for support vector machine classification
Abstract
Permafrost thaw due to climate change is increasing greenhouse gas (CO2 and CH4) emissions from arctic soils. In polygonal ground, the magnitude and relative abundance of CO2 and CH4 release are influenced by redox conditions that vary across polygon features (i.e., trough, rim, and center), which are highly heterogeneous and complex on small scales (10-30 m diameter). These small-scale features can be problematic to identify using satellite imagery, but their abundance and extent in the arctic must be accounted for in global climate models to accurately scale-up landscape emissions and predict carbon budgets. Therefore, this study tests a novel polygonal ground classification model to identify low-centered and high-centered polygons and differentiate them from non-polygonal ground in the arctic tundra. Here, we analyzed Worldview-3 (WV3) panchromatic (PAN) imagery of a study site (~1.5 km2) near the Barrow Environmental Observatory (BEO) on Alaska's North Slope. Image textural features including contrast, correlation, homogeneity, and variance were calculated using gray-level co-occurrence matrices (GLCM) for an object-based image analysis (OBIA) approach suitable for very high-resolution imagery. GLCM outputs served as predictor variables for polygonal ground classification in a support vector machine (SVM) machine learning algorithm. This OBIA approach is applicable to any panchromatic imagery, uses open-source software (R, QGIS), and does not require advanced computer processing capabilities. We determined that optimal classification occurs with GLCM displacement of 0.93 m, or about the scale of polygon features, and with gray quantization level of 64 levels. GLCM outputs at this scale and quantization level were distinct amongst high-centered polygons, low-centered polygons and various non-polygonal ground types, and were used in a SVM model to accurately differentiate regions of polygonal ground from non-polygonal ground. This approach is valuable for classifying different greenhouse gas emitting landforms and can also be used to evaluate landscape change, e.g. degradation of low-centered polygons into high-centered polygons, as permafrost thaw progresses.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMB019.0012D
- Keywords:
-
- 0428 Carbon cycling;
- BIOGEOSCIENCES;
- 0439 Ecosystems;
- structure and dynamics;
- BIOGEOSCIENCES;
- 1615 Biogeochemical cycles;
- processes;
- and modeling;
- GLOBAL CHANGE;
- 1622 Earth system modeling;
- GLOBAL CHANGE