JAXA Supercomputer System Annual Report February 2025-January 2026

Report Number: R25EER20252

Subject Category: Space Technology

PDF (to be added)

• Responsible Representative: Osamu Ochiai, Director, Earth Observation Research Center, Space Technology Directorate I
• Contact Information: hirayama sota(hirayama.sota@jaxa.jp)
• Members: Takeo Tadono, Sota Hirayama

Abstract

The Space Technology Directorate 1 aims to capture domestic and international nature-based credit markets as part of its co-creation initiative by leveraging satellite data. To achieve this goal, efforts are being made to develop forest carbon sink inventories and establish effective methodologies for assessing natural capital. In this context, it is essential to obtain wide-area, high-precision information from earth observation satellites, including forest extent, tree species, tree height, and changes in land-use and land-cover (LULC). In particular, LULC information serves as a fundamental base map for natural capital products such as biomass maps, and improving its accuracy is critical for promoting its utilization across various sectors, including biodiversity, agriculture, forestry and fisheries, environmental management, disaster risk reduction, and public health. However, existing global LULC datasets still face challenges in terms of category definitions and accuracy, and there is a strong demand for more precise and higher-resolution global datasets.

The EORC/ALOS group has developed high-accuracy, high-resolution LULC maps (HRLULC) primarily for Japan and Vietnam using data from ALOS-2/PALSAR-2 and related sensors, and has also conducted experimental efforts toward global expansion, particularly in South Asia. In the Fifth Medium- to Long-Term Plan, HRLULC is positioned as a fundamental dataset supporting key priority themes such as natural capital, high-precision 3D topographic information, and the water cycle. While HRLULC has already been widely utilized across various fields in Japan, extending this level of accuracy to a global scale has become an urgent requirement. Based on this background, this project aims to develop a global high-accuracy, high-resolution HRLULC dataset.

Reference URL

Please refer to https://www.eorc.jaxa.jp/ALOS/en/dataset/lulc_e.htm .

Reasons and benefits of using JAXA Supercomputer System

Currently, the generation of HRLULC products for Japan and other regions is based on machine learning–driven algorithms. However, producing a global-scale product using existing in-house computational resources would require approximately 213.2 days per processing run. In particular, this project requires the release of high-accuracy products, and achieving this necessitates the experimentation and validation of multiple classification schemes. To support this, a computational environment equipped with multiple high-performance GPUs is essential. However, procuring new GPU servers and establishing the required environment within the current fiscal year, along with completing the classification process, is time-prohibitive.

Under these circumstances, JSS plays a critically important role in the execution of this project. By leveraging the high-performance computational resources of JSS3, it becomes feasible to conduct the necessary classification experiments within a realistic timeframe, thereby making a significant contribution to the achievement of the Fifth Medium- to Long-Term Plan objectives.

Achievements of the Year

In this fiscal year, preprocessing of input data such as satellite observations was conducted, classification categories were defined, and training and validation samples were collected. To ensure a sufficient number of experimental iterations, classification trials were performed using reduced-resolution products. In addition, the algorithm was updated for global HRLULC classification, and the trial results achieved an overall accuracy of 84.61%. The classification scheme consists of 23 categories (see Figure 1).

Fig.1: Preliminary Results of Global HRLULC (Categories: Water Bodies, Built-up Areas, Single Paddy Field, Multi Paddy Field, Single Cropland, Multi Cropland, Orchard, Grassland, Deciduous Broad-leaved Forest, Deciduous Needle-leaved Forest, Evergreen Broad-leaved Forest, Evergreen Needle-leaved Forest, Rubber Tree Plantations, Palm Tree Plantations, Mangrove Forest, Shrubland, Bare, Solar Panel, Wetland, Greenhouse, Moss and Lichen, Glaciers and Perennial snow patches, Aquaculture Areas)

Publications

N/A

Usage of JSS

Computational Information

• Process Parallelization Methods: MPI
• Thread Parallelization Methods: N/A
• Number of Processes: 1 - 10
• Elapsed Time per Case: 10 Hour(s)

JSS3 Resources Used

Fraction of Usage in Total Resources*1(%): 0.15

Details

Please refer to System Configuration of JSS3 for the system configuration and major specifications of JSS3.

*1: Fraction of Usage in Total Resources: Weighted average of three resource types (Computing, File System, and Archiver).

*2: Fraction of Usage：Percentage of usage relative to each resource used in one year.

ISV Software Licenses Used

*2: Fraction of Usage：Percentage of usage relative to each resource used in one year.

JAXA Supercomputer System Annual Report February 2025-January 2026