Object-Based Aerial Photos Analysis for Arctic Sea Ice Melt Ponds and Pressure Ridges

Abstract:

High resolution aerial photographs can provide detailed distribution of sea ice features so as to extract physical parameters to refine, validate, and improve climate models. For example, melt ponds play an important role in Earth's radiation balance since they strongly absorb solar radiation rather than reflecting it as snow and ice do. Furthermore, no previous studies have ever considered shadow in sea ice detection, which is ubiquitous in the aerial photographs especially in multi-year ice regions and during late melting phase. Based on our previous study, an object-based classification scheme is used to extract sea ice features including melt ponds and shadow from 163 selected aerial photographs taken during the Chinese National Arctic Research Expedition (CHINARE 2010). The classification algorithm includes three major steps as follows. (1) Image segmentation groups the neighboring pixels into objects according to the similarity of spectral and texture information; (2) random forest ensemble classifier can distinguish the following objects: water, general submerged ice, shadow, and ice/snow; and (3) polygon neighbor analysis can further separate melt ponds from general submerged ice according to the spatial neighboring relationship. Finally, the shadows are used to estimate the sea ice ridge distribution based on local solar illumination geometry. Our results illustrate the spatial distribution and morphological characters of melt ponds and ridges in different latitudes of the Arctic Pacific sector. This method can be applied to massive photos and images taken in past years and future years, in deriving the detailed sea ice feature distribution and changes through years.