A layered approach to parallel computing for spatially distributed hydrological modeling

Abstract

Distributed hydrological simulations over large watersheds usually require an extensive amount of computation, which necessitates the use of parallel computing. Each type of hydrological model has its own computational characteristics and therefore needs a distinct parallel-computing strategy. In this paper, we focus on one type of hydrological model in which both overland flow routing and channel flow routing are performed sequentially from upstream simulation units to downstream simulation units (referred to as Fully Sequential Dependent Hydrological Models, or FSDHM). There has been little published work on parallel computing for this type of model. In this paper, a layered approach to parallel computing is proposed. This approach divides simulation units into layers according to flow direction. In each layer, there are no upstream or downstream relationships among simulation units. Thus, the calculations on simulation units in the same layer are independent and can be conducted in parallel. A grid-based FSDHM was parallelized with the Open Multi-Processing (OpenMP) library to illustrate the implementation of the proposed approach. Experiments on the performance of this parallel model were conducted on a computer with multi-core Central Processing Units (CPUs) using datasets of different resolutions (30 m, 90 m and 270 m, respectively). The results showed that the parallel performance was higher for simulations with large datasets than with small datasets and the maximum speedup ratio reached 12.49 under 24 threads for the 30 m dataset.