Since 1950, supercomputing has been playing an important role in numerical modeling for weather and climate forecast. Over recent years, volume and diversity of weather and climate data have increased exponentially and this presents a great challenge to transmission, management and storage, and also requires more powerful computation capability to simulate and make these data use. Currently, common weather and climate simulation can only be completed when tens of thousands of processor cores operate in parallel, which means that a large bandwidth and low latency Internet network and high-performance storage devices are required to meet the requirements.
Examples of Typical Software Applications
Numerical atmospheric forecast model
WRF, MM4, MM5, GRAPES, CAMS, AREMS, LAPS, ARPS
Atmospheric circulation model
GAMIL, SAMIL, GCMs, AGCM
CCSM, RegCMS, PNNL2RCM, RIEMS
Application Performance Features
Weather and climate simulation software applications have different demands for computational resources. We analyzed software characteristics in the weather and climate field by taking the weather data from the bureau of certain provinces as an example. In this example, the medium scale forecast model WRF is used to perform the short-term forecast simulation. The system forecasts the weather and the climate four times a day, the forecast time is 84 hours and it requires to complete parallel computation of the WRF main model within 90min. The data about operational characteristics of WRF as below are provided by TEYE.
Number of lattice points
From data provided by TEYE monitoring, it can be seen that the operation of the WRF model contains grid reading, numerical computation, data writing, assimilation computation, etc., in which the utilization of CPU resources is different. Memory bandwidth, network and I/O throughput in the application execution are increased by optimizing the parallel mode, input parameters and I/O mechanism of the software. This example shows that most of weather and climate simulation software applications are computation intensive, memory bandwidth intensive and network intensive, so a high frequency CPU, high memory bandwidth and a high-speed network can help to improve the computational performance.