Corporate News: Development of a Machine Learning Model for Paddy Water Management and Publication of Research Paper

— Aiming to contribute to greenhouse gas reduction (climate change mitigation) by utilizing JAXA’s ALOS-2 satellite data —

Sagri Co., Ltd. (Headquarters: Tamba City, Hyogo Prefecture; Representative Director: Shunsuke Tsuboi) has developed a high-precision automated detection model for paddy water status (flooding/ponding). This model utilizes high-resolution images from JAXA’s Advanced Land Observing Satellite-2 “DAICHI-2” (ALOS-2) and machine learning (AI) technology, covering not only flatland paddies but also those in hilly and mountainous areas (where rice fields are often irregularly shaped).

We are pleased to announce that the research paper summarizing these findings has been published in the international remote sensing journal, Remote Sensing Applications: Society and Environment.

Research Paper

https://www.sciencedirect.com/science/article/abs/pii/S2352938526002272

Background and Social Significance

Water management practices in rice cultivation, such as mid-season drainage and Alternate Wetting and Drying (AWD), are attracting global attention for their role in climate change mitigation and the carbon credit market, as they can significantly reduce methane emissions—a potent greenhouse gas—from paddy fields.

However, these practices have historically been labor-intensive, relying on manual field inspections or water level sensors. In particular, accurately grasping water management status using satellite data has been considered difficult for the small, complex-shaped paddy fields often found in Japan’s hilly and mountainous regions.

Research Results and Key Points

In this study, we developed a machine learning model to determine paddy status (flooded, wet, or non-flooded) by combining field observation data from June 2022 to October 2024 in central Hiroshima Prefecture with L-band Synthetic Aperture Radar (SAR) images from ALOS-2.

1. Capability for Hilly and Mountainous Regions: By using SAR data, which allows for surface observation unaffected by cloud cover, we successfully detected water status with practical precision even in the complex paddy fields of hilly and mountainous areas.
2. High Prediction Accuracy: Through the optimization of machine learning algorithms (such as XGBoost) and our proprietary resampling techniques, we were able to accurately reproduce the transition of water management according to the rice growth stages.

Future Outlook

The establishment of this technology enables the “visualization” of paddy water management status over wide areas. We anticipate this will contribute to visualizing greenhouse gas reduction effects in the agricultural sector, promoting eco-friendly farming, and facilitating the creation of carbon credits. Sagri remains committed to leveraging cutting-edge technology to contribute to sustainable agriculture and the preservation of the global environment.

Paper Details

• Journal: Remote Sensing Applications: Society and Environment (Impact Factor = 4.5)
• Title: A machine-learning modelling study on the surface water detection using ALOS-2 L-band SAR images: Irrigation status in a hilly-mountainous region, Japan
• Authors: Itsuki C. Handoh, Rie Sakai, Keita Wakabayashi, Kenshi Kobayashi, Wataru Yasuhara, Tomoko E. Yano, Takashi S.T. Tanaka