JAXA Supercomputer System Annual Report April 2016-March 2017

Report Number: R16E0090

• Responsible Representative: Kinji Furukawa(Space Technology Directorate I, GPM/DPR Project)
• Contact Information: Takahiro Minami(minami.takahiro@jaxa.jp)
• Members: Yuichiro Kitayama, Takanori Kikuchi, Tomomi Nio, Toshiyuki Konishi, Kaya Kanemaru, Takahiro Minami, Masaya Torii, Tadahiro Yamamoto, Koki Ishimaru, Sachiko Kawase, Daichi Obinata, Shun Ito, Yasuhiro Kawashima
• Subject Category: Space(Satellite utilization)

Abstract

In recent years, worldwide interest has been increasing about the necessity of grasping the global environmental change. To deal with such problems, various approaches using observation technology from space have been carried out by artificial satellites.

Global Precipitation Measurement (GPM) mission, as follow-on and expansion of the TRMM satellite, is an international mission to achieve highly accurate and frequent global rainfall observation. it is carried with multiple satellite, one primary satellite with Dual-frequency Precipitation Radar (DPR) jointly developed by JAXA and NICT, and with GPM Microwave Imager (GMI) developed by NASA, and another 8 constellate satellites with Microwave Imager.

Goal

Global Precipitation Measurement (GPM) mission is follow-on and expansion of the TRMM mission. It aims to provide satellite observation data that contributes to elucidation of global environmental problems and improvement of meteorological forecast accuracy. To surpass the results achieved by TRMM, the GPM mission aim to realize the; 1) expansion of observation coverage; 2) increase of observation frequency; and 3) improvement of observation accuracy.

Objective

In the GPM mission, one primary satellite and eight constellation satellites will produce 3-hour global precipitation maps that will be delivered to users. Real-time monitoring of 'water' using satellites seeks to provide immediate and practical information for weather prediction, land management, agriculture, fisheries and, possibly, disaster prevention through improved weather forecasting and warning.

References and Links

Please refer 'JAXA | Japan Aerospace Exploration Agency'.

Use of the Supercomputer

Processing of earth observation data includes 'routine processing' performed routinely and 're-processing' performed once a year or so for several year data. The porpuse of re-processing is to correspond with version-up of computing model and algorithm performed periodically. The amount of observation data grows year by year. Then, we need more and more time to complete reprocessing of all archived observation data. On the other hand, quick distribution of re-processed earth observation data to users is neccessary for measures for disasters and better understanding of Global Change. By using supercomputers, the calculation time is greatly shortened, and the period until the provision of the processing result is shortened.

Necessity of the Supercomputer

In the processing computer for GPM at TUKUBA, it is necessary to preferentially execute the 'routine processing'. Therefore, 're-processing' using free time takes considerable time such as months to 1 year. In order to solve these problems, we decided to use the JAXA supercomputer System Generation 2 (JSS2). The data during 2.5 year period has been processed and 4 times faster latency was achieved. Considering ever-increasing observation data, the use of JSS2 is indispensable.

Achievements of the Year

In the improvement of the algorithm (DPR V5) implemented this fiscal year, it was possible to estimate precipitation particle information and discriminate ground snowfall by improving the calibration accuracy and improving the precipitation estimation algorithm. Fig.1 and Fig.2 show the results of comparing precipitation distributions before and after improvement. The blue area shows snowfall, but clear differences can be confirmed from the Eurasian Continent to the North American Continent. Re-processing of the observation data of the past 2.5 years by the new algorithm was started from June in 2016 using JSS2, and it is completed in about 17 hours. As the processing time was greatly shortened, we were able to speed up the pace of data evaluation and improvement of algorithms for higher order processing. Consequently Re-process calculation is ongoing with a plan to complete higher order processing of L2/L3 level within the fiscal year.

Fig.1:Precipitation map in old version(2015/01)(Snowfall on Eurasia continent ~ North American continent can not be detected in old version)

Fig.2 :Precipitation map in latest version(2015/01)(Snowfall can be detected in the latest version)

Publications

N/A

Computational Information

• Parallelization Methods: Serial
• Process Parallelization Methods: n/a
• Thread Parallelization Methods: n/a
• Number of Processes: 30
• Number of Threads per Process: 30
• Number of Nodes Used: 30
• Elapsed Time per Case (Hours): 0.0167
• Number of Cases: 31058

Resources Used

Total Amount of Virtual Cost(Yen): 515,369

Breakdown List by Resources

Note: Virtual Cost=amount of cost, using the unit price list of JAXA Facility Utilization program(2016)

JAXA Supercomputer System Annual Report April 2016-March 2017