Supraglacial lake bathymetry from high-resolution airborne imagery using stereophotogrammetry and structure from motion

Abstract. Understanding the hydrological system of the ice sheet of Kalaallit Nunat (Greenland, GrIS) requires observational knowledge on a range of scales with varying spatial and temporal resolutions and coverage. In general, process-related studies require high-resolution observations over regional scales spanning hundreds of kilometers. The spatial resolution of spaceborne imaging is continually improving, however, airborne imaging can provide complementary information at higher resolution and can combing multiple sensors with coincident data acquisition. Here we use regional-scale, high-resolution (10×10 cm) aerial imagery collected by NASA’s Airborne Topographic Mapper instrument suite (ATM) to derive bathymetry of supraglacial lakes and streams using Structure from Motion (SfM) and stereo processing techniques. SfM uses multiple overlapping images and iterative bundle adjustment to solve and optimize camera positions and parameters. Stereophotogrammetry requires highly accurate camera geometry and calibration. Supraglacial lakes appear as distinct blue features in natural-color imagery of the ice sheet with often clearly visible features at the lake bottom, making them ideal for SfM and stereo processing. The ATM airborne data also includes two coincident green (532 nm) small footprint lidar providing a rare opportunity to compare lidar and imagery-based bathymetry methods.

We use the NASA Ames Stereo Pipeline (ASP), a powerful open-source processing toolbox with a long legacy, a broad user community, and active tool development. We compare the ASP stereo bathymetry with lake depth estimates from a commercial SfM package and bathymetric estimates from our coincident lidar data. Our results show that supraglacial lake bathymetry can be mapped at high spatial resolution using stereophotogrammetry and SfM reconstruction of topobathymetric digital elevation models (DEMs). The stereographic method we use accurately accounts for refraction in water. The SfM approach uses an iterative refraction correction for SfM to accurately determine the lake bottom topography. We find two main limitations for image-based bathymetry estimates: thin lake ice cover can obscure lake bottom features making feature tracking between images impossible. The second limitation comes from caustics at the lake bottom caused by refraction of sunlight on moving surface waves. These caustics are not stationary patterns and can move in the 0.5 secs between image acquisitions ruling out static feature tracking necessary for SfM or stereophotogrammetry. Each method has its own limitations and biases. By analyzing changes in the shape of lidar waveforms between survey targets with no differential penetration and supraglacial lakes we have shown for the first time that green photons can penetrate below the lake bottom surface resulting in a range and therefore elevation bias.