Using Image Segmentation to Identify Tundra Vegetation Variability in High Resolution Satellite Images

Abstract:

Arctic tundra ecosystems will play an important role in the global carbon cycle in coming decades and centuries. Amplified climate warming at high northern latitudes has stimulated carbon uptake via plant productivity, while thawing permafrost is releasing carbon to the atmosphere. Accurately quantifying the effect of changing tundra ecosystems on global climate will require detailed understanding of both of these processes. In this context, accounting for the spatial variation of landscape features is critical to creating carbon budgets for ecosystems and regions, and for forecasting the effects of climate change in the tundra. Water tracks and other areas that feature channelized subsurface water flow are landscape features with distinct differences in carbon stocks and fluxes relative to adjacent upland tundra areas. Numerous studies have shown that water tracks and flowpaths have greater water and nutrient availability that leads relatively high carbon stocks and rates of carbon uptake. However a clear understanding of the relative proportion of tundra ecosystems that are comprised of these landscape features is lacking.

Recently developed fine-scale remote sensing technology allows for the spatial analysis of tundra landscapes and specifically water tracks. This study automates process of distinguishing water tracks through the use of image classification and segmentation techniques on high-resolution satellite imagery. Both supervised and unsupervised classification techniques identify water tracks as distinctive and tundra landscape features. Unlike the unsupervised classification, edge detection and region-thresholding algorithms in the supervised classification differentiates water tracks from locations with high productivity by assessing connectivity shape.

Depending on the area of inquiry, water tracks comprise roughly 10%-25% of the landscape. Field observations indicate that water tracks have greater plant productivity, and rates of carbon cycling. Failing to account for these landscape features will lead to errors in regional carbon and methane budgets.