f Scalable, parallel constrained inversion of large AEM datasets

Over the past decade the typical size of AEM datasets has been growing rapidly, while at the same time targeting new applications that rely on advances in terms of resolution and accuracy. Approximate inversions and data transform techniques have previously defined the norm for interpretation of huge surveys, but rarely pose attractive solutions for modern applications such as aquifer mapping, uranium exploration, integrated modeling etc. For these applications high-resolution full system modeling techniques provide the only acceptable solution, but their refinement comes at the expense of significant added computational complexity. Further applying spatial constraints to the inversion of multiple 1D soundings facilitate a resulting quasi-3D model, however, there are severe intrinsic issues in effectively solving the underlying systems of linear equations. Here, we describe how we have attacked scalability issues of the Spatially Constrained Inversion (SCI) formalism and optimized our code to handle arbitrarily large problems on parallel multi-processor computers.