The backprojection algorithm for synthetic aperture radar (SAR) has the ability of focusing raw data and forming an image without requiring simplifying assumptions for the azimuth-direction phase response of the recorded signal (Zebker et al., 2010). In fact, the usual quadratic phase assumption is not accurate when the recording platform is affected by turbulent motion or when wide variations in incidence angle across the range swath are present in the acquired data (Duersh, 2013). In addition, as we show here the backprojection algorithm is easily extendable to create SAR images of the subsurface in addition to the surface. These reasons make backprojection a well-suited method to create 3D subsurface SAR volume images from spaceborne acquisitions. The ability to form 3D SAR volumes using spaceborne data allows the study of larger areas than those available using to airborne acquisitions (Hélière et al., 2007). In this work, we create volume images by applying a 3D backprojection method to Advanced Land Observing Satellite (ALOS) data recorded over an icy area in Greenland previously studied by Banda et at. (2016). The 3D SAR image volumes created using different acquisition times enables the study of the evolution of the ice structure and properties. To decrease the processing time, we implement our algorithm using graphics processing units (GPUs) by taking advantage of the embarrassingly parallel nature of the backprojection algorithm (Fasih and Hartley, 2010).

Synthetic aperture radar (SAR) multi-temporal techniques have been proposed to improve image resolution, persistent-scatterer detection, and damage-map generation (Bujor et al., 2004; Hooper, 2008; Yun et al., 2012). The usage of multiple scenes diminishes the speckle noise commonly present in a SAR image and increases the overall signal-to-noise ratio (SNR) of dominant radar scatterers. Following the same idea, we propose a new SAR processing workflow based on the back-projection algorithm (Cafforio et al., 1991). We employ the back-projection method’s flexibility to focus the radar pulses at multiple depths shifted with respect to the reference digital elevation model (DEM), yielding a volume of SLCs discretely sampled over a range of elevations/depths. These scenes are then combined to produce a single-look complex (SLC) image, which presents increased SNR and effective resolution. Similar image improvements can be achieved by spatially averaging a SLC image generated at high spatial sampling. However, the proposed method requires a fraction of the processing time of the fine-spatial SLC formation. The proposed approach retains the same benefits of any back-projection algorithm and enables the formation of SLC images corresponding to large Earth surface extents.