Study On The Parallehzation Of A Physics-based Hydrologic Mode InHM Utilizing Computer Cluster

Due to the shortage of physics-based hydrologic model and parallel computing of these models, in the presented study, a complex physics-based hydrologic model InHM was revised to conduct a regional scale hydrologic simulation utilizing a PC-cluster parallel system (named as ParInHM). The ParInHM has been tested to simulate100years’soil erosion for the Hawaiian Island of Kaho’olawe. Both validation and parallel efficiency have been tested for ParInHM. The main contents of this dissertation are as follows:(1) InHM is the first fully physics-based hydrologic model that has ever been developed. The principle goal of InHM is to integrate surface and subsurface flow and transport processes in one coherent framework, eliminating iterative coupling between surface and subsurface models through the use of physics-based first-order flux relationships.(2) A physics-based hydrologic model InHM utilizing cluster parallel computing (ParInHM) has been developed. ParInHM utilizes task decomposition method based on drainage characteristics. The simulation task is divided into several sub-tasks and then the sub-tasks are sent to a group of networked computers for parallel computing. One sub-task is corresponding to a single catchment. This parallel method can decrease the message passing, making it more suitable for PC cluster. The mesh used in InHM is based on real terrain, which means that both flat area and depressions are acceptable in the model. For this reason, a drainage extraction algorithm from triangulation irregular networks (TINs) DEM is developed according to geomorphologic principles, without pre-processing of flat area and depressions. The drainage extraction method can keep the real terrain and TIN structure.(3) Based on ParInHM cluster enviroment, a detailed description of ParInHM parallel module development is given, which includes:message passing based parallel mechanism; task devision and taks property; parallel scheduling method like longest processing time (LPT) first scheduling method and ordinal scheduling method. A hybrid algorithm utilizes both static and dynamical scheduling is introduced in ParInHM and gives out better scheduling results. In the last, the drainage data synchronization, distributed data storage, parallel monitor and system exception handling are discussed.(4) ParInHM has been tested in a computer cluster to simulate long-time hydrologic-response and landscape evolution for the Hawaiian Island of Kaho’olawe. Soil erosion and catchment change shows the importance and advantage of dynamically updating catchment boundary. Compared with InHM that simulates the island in one single mesh, the parallel efficiency and validation of ParInHM indicate that the parallel system is reasonable and acceptable in the presented study.