The availability of high-resolution, repeated, continent-scale geospatial data sets will enable new science, but only if we have the right tools. To answer geomorphological questions at unprecedented spatial and temporal scales, we need to (a) parse terabyte-scale data sets, (b) perform millions of model realizations to pinpoint the parameters that govern landscape evolution, and (c) do so with statistical rigor, which may require thousands of additional realizations.
In this talk, I will discuss how these goals can be achieved using communication-avoiding algorithms and GPUs, as well as how these techniques can be leveraged to benefit users with limited resources. Finally, I will show several example applications demonstrating the utility of these algorithms.
