November 21, 2025
Japan International Research Center for Agricultural Sciences (JIRCAS)

JIRCAS has developed a new soil diagnostic technology that uses AI to precisely estimate multiple soil properties simultaneously from full-wavelength data obtained through ICP analysis.

As food demand grows, particularly in developing regions, and the impacts of climate change intensify, it has become increasingly important for agricultural producers to correctly understand and sustainably manage soil fertility. However, conventional soil analysis requires combining multiple measurement methods and often takes several days to weeks to produce results. High analysis costs also hinder adoption among smallholder farmers and in regions such as Sub-Saharan Africa. As a result, proper fertilizer management becomes difficult, which can lead to declines in soil fertility caused by over-fertilization or nutrient deficiencies, ultimately impacting yields and environmental conservation.

In this study, researchers adopted a new approach by training AI on the full-wavelength emission spectrum obtained via Inductively Coupled Plasma Atomic Emission Spectroscopy (ICP-AES)¹. Whereas conventional ICP analysis uses only a limited number of wavelengths corresponding to specific elements, this research focused on analyzing the vast amount of wavelength data that had previously gone unused. Using conventional analysis values as training data, a deep learning model was constructed. Approximately 2,000 soil samples collected from seven countries across Asia and Africa—representing diverse soil types and environmental conditions—were used, enabling the development of a highly versatile AI model.

As a result, it became possible to simultaneously and accurately predict 12 items, including CEC (cation exchange capacity), exchangeable Ca, Mg, K, and Na, pH, available phosphorus, total carbon, total nitrogen, and soil particle size distribution. Many items achieved a coefficient of determination (R²) of 0.9 or higher. This technology significantly shortens analysis time while enabling integrated diagnosis of multiple items. It is expected to have wide-ranging impacts both domestically and internationally, such as accelerating fertilizer formulation and field management, reducing environmental load through decreased chemical use, and enabling utilization in regions lacking soil diagnostic infrastructure.

The results of this research have been published as an open-access article in the online edition of Scientific Reports (November 20, 2025, JST).
 

Related Information

Funding

Operating Expenses Grant Project: “Development of soil and crop management technologies to stabilize upland farming systems of African smallholder farmers (Africa upland farming system)”

Patent

Patent No. 7464284: “Soil diagnostic method using plasma emission spectroscopy”

Authors

NAKAMURA Satoshi, IMAYA Akihiro (Currently at the Forestry and Forest Products Research Institute), IKAZAKI Kenta

Paper Title

Deep learning using inductively coupled plasma spectroscopy spectra accurately predicts various soil physicochemical properties for soil diagnosis

Journal Title

Scientific Reports
DOI: https://doi.org/10.1038/s41598-025-24274-3

JIRCAS President: KOYAMA Osamu

Program Director:

FUJITA Yasunari

Research Staff:

NAKAMURA Satoshi (Project Leader, Crop, Livestock and Environment Division)
IKAZAKI Kenta (Senior Researcher, Crop, Livestock and Environment Division)

Press Coordinator:

OMORI Keisuke (Head, Information and Public Relations Office)
Press e-mail: koho-jircas@ml.affrc.go.jp