Authors: Zhengtao Cui, Baxter E. Vieux, Henry Neeman, Fekadu Moreda
Addresses: School of Civil Engineering and Environmental Science, University of Oklahoma, Norman, OK 73019, USA. ' School of Civil Engineering and Environmental Science, University of Oklahoma, 202 W. Boyd St., Room CEC334, Norman, OK 73019, USA. ' OU Supercomputing Center for Education and Research, Stephenson Research and Technology Center, University of Oklahoma, 101 David L. Boren Blvd., Norman, OK 73019, USA. ' Office of Hydrologic Development, National Weather Service, 1325 East-West Highway, Silver Spring, MD 20910, USA
Abstract: In the solution of streamflow problems, fully distributed hydrologic models (DHMs), which are based directly on governing equations, offer distinct advantages over conceptual rainfall-runoff models, which are derived from empirical observations. However, the primary problem associated with DHMs is that they consume more computational resources than other models, and as a result, they have not been as popular as their capabilities would imply. A parallel DHM software system for solving streamflow prediction problems has been implemented and analysed, and an investigation has been conducted of: the efficiency and scalability of the algorithm; load balancing among processors; interprocessor communication; disk performance. The load balancing algorithms show great promise for the kind of problem addressed. The software exhibits substantial parallel speedup, but the degree of speedup is limited by I/O costs.
Keywords: distributed parallelism; message passing; load balancing; hydrological modelling; flood forecasting; streamflow prediction; distributed hydrological models; rainfall runoff.
International Journal of Computer Applications in Technology, 2005 Vol.22 No.1, pp.42 - 52
Published online: 15 Apr 2005 *Full-text access for editors Access for subscribers Purchase this article Comment on this article