Abstract
The Global Positioning System (GPS) ray-shooting model is a self-sufficient observation operator in GPS / MET (Meteorology) data variational assimilation linking up the GPS observation data and the atmospheric state variables. But its huge computations make it impracticable in real data assimilation so far. In order to overcome this default, a parallel version of the GPS ray-shooting model has been developed, and has been running successfully on the PC cluster manufactured under the support of the China National Key Development Planning Project for Basic Research: The Large Scale Scientific Computation Research. High speed-up and Efficiency as well as good scalability are obtained. This is an important step for this GPS observation operator to become practicable.
| Original language | English |
|---|---|
| Pages (from-to) | 1185-1191 |
| Number of pages | 7 |
| Journal | Advances in Atmospheric Sciences |
| Volume | 18 |
| Issue number | 6 |
| DOIs | |
| State | Published - 2001 |
Funding
The Global Positioning System (GPS) ray-shooting model is a self-sufficient observation operator in GPS/MET (Meteorology) data variational assimilation linking up the GPS observation data and the atmospheric state variables. But its huge computations make it impracticable in real data assimilation so far. In order to overcome this default, a parallel version of the GPS ray-shooting model has been developed, and has been running successfully on the PC cluster manufactured under the support of the China National Key Development Planning Project for Basic Research: The Large Scale Scientific Computation Research. High speed-up and Efficiency as well as good scalability are obtained. This is an important step for this GPS observation operator to become practicable. Since the first Low Earth Orbit (LEO) satellite equipped with the GPS receiver was launched in USA in 1995, the research and application of the GPS occultation technique have greatly progressed and become an important subject in remotely sensing the atmosphere (Wang et al., 2000, Zou et al., 2000, Zou et al., 1999, Zou et al., 1995, Kuo et aL, 1997). The variational data assimilation analysis has been paid more and more attention to reconstruct the atmospheric state from the GPS "raw" measurements (Wang et al., 2000, Zou et al., 2000, Zou et al., 1999), although the inversion method is still important. It shows that the inclusion of the information from GPS measurements to model initial conditions might give the numerical weather prediction (NWP) a new look. But it still needs a long time to use the GPS measurements in the operational NWP, because of various difficulties. The researches in this field now are still in a stage of exploring experiments. The GPS / MET observation operator linking up the GPS refraction angle measurements and the atmospheric state variables play an important role in the GPS variational data assimilation. In order to obtain the equivalent horizontal resolution of the conventional global radiosonde observations, at least 1000 GPS occultation data should be included in the GPS data assimilation system that needs more than 40 optimal iterations each time. Suppose 1000 GPS occultation data are included in the system, the GPS/MET observation operator and its adjoint operator will be solved for 1000 times respectively in an optimal iteration. The GPS ray-shooting model is a self-sufficient observation operator in the GPS data assimilation and the old GPS ray-shooting operator is solved by the 4th-order Runge-Kutta method. It takes about 373 seconds on an average to 0)This research was supported by the National Natural Science Foundation of China(Grant No. 49825109) , the National Key Development Planning Project for Basic Research (Grant No. 1999032801) and the CAS Key Innovation Direction Project (Grant No. KZCX2208). Two methods can be used to deal with this problem. The first is to design high efficient numerical method. The 2-order timesaving symplectic scheme is used to solve the GPS ray trajectory equations and saves 75% of the CPU time the old Runge-Kutta method takes. Based on this progress, continuous efforts should be made, because it cannot meet the requirement yet. With the quick development of the parallel computing technique and the frequent appearance of the new advanced parallel computers in recent years, parallel computation with many processors has become one of the best ways to greatly reduce the real computing time of GPS data assimilation. For this reason, the improved GPS ray-shooting model is parallelized with Message Passing Interface (MPI) technique and its parallel computations are implemented successfully on the PC Cluster manufactured under the support of the China National Key Development Planning Project for Basic Research: The large scale Scientific Computation research. High speed-up and Efficiency are obtained. It lies an important base for the GPS / MET data variational assimilation to be used in the operational NWP. Supported by the China National Key Development Planning Project for Basic Research ' The Large Scale Scientific Computation Research (LSSC)', a PC cluster with 128 processors has been made in the State Key Laboratory of Scientific and Engineering Computing (LSEC) of Chinese Academy of Sciences (CAS), which is named ' LSSC'. This cluster is open to all project members. The cluster provides 120 nodes for scientific computing purpose and 4 service nodes. Each computing node has a Pentium III 550E CPU and 512 MB memory and each service node has two Pentium III 550 CPU and 1GB memory. The peak value performance of the 120 computing nodes is 66 Gflops. The measured supreme double-precision Linpack performance is 40.9 Gflops (about 410 hundred million times/seconds), it is almost as fast as the 360th one among the Top 500 computers in the world in November, 1999. The peak value performance of the 4 service nodes is 4.4 Gflops and the peak value performance of the whole system is 70.4 Gflops. The compile system of the LSSC includes GNU Fortran, GNU C/C++, etc. The parallel environment has two different versions of MPI system: MPICH-1.2.0 and LAM MPI 6.3.1.
Keywords
- Efficiency
- GPS ray-shooting
- Parallel computing
- Scalability