Accelerating a fluvial incision and landscape evolution model with parallelism

Abstract: Solving inverse problems and achieving statistical rigour in landscape evolution models requires running many model realizations. Parallel computation is necessary to achieve this in a reasonable time. However, no previous algorithm is well-suited to leveraging modern parallelism. Here, I describe an algorithm that can utilize the parallel potential of GPUs, many-core processors, and SIMD instructions, in addition to working well in serial. The new algorithm runs 43x faster (70s vs. 3,000s on a 10,000x10,000 input) than the previous state of the art and exhibits sublinear scaling with input size. I also identify methods for using multidirectional flow routing and quickly eliminating landscape depressions and local minima. Tips for parallelization and a step-by-step guide to achieving it are given to help others achieve good performance with their own code. Complete, well-commented, easily adaptable source code for all versions of the algorithm is available as a supplement and on Github.