The groundbreaking study applied artificial intelligence (AI) to study 150 consecutive rice crops grown in the International Rice Institute (IRRI) long-term continuous crop experiments (LTCCE) of the Philippines, which began in 1968 and ended in 2017.
This study was supported by the Japanese government and the Ministry of Agriculture, Forestry and Fisheries (MAFF).
The research team consisted of scientists from GIFU University, Kyoto University, Japan's National Institute of Agricultural Food Research (NARO), the International Fertilizer Association, and IRRI.
According to IRRI, LTCCE cultivated rice three times a year in dry, early wet, and late seasons with different nitrogen fertilizer rates and regular introduction of new varieties.
Scientists combined climate record, agricultural practices, variety turnover with advanced machine learning techniques.
Data over 49 years reveals new insights into long-term crop performance.
“This is a time when machine learning unleashes this complex, long-term interaction between climate, management, and genetics in rice systems,” says Dr. Nishikai of Ili.
Important findings showed that better nitrogen fertilizer use, rapid variety replacement, and solar radiation consistently boosted yields, although results vary with seasons.
Dry season crops thrived at cooler reproductive stage temperatures, while early wet season crops benefited from warm conditions that promote soil nitrogen mineralization.
Late rainy season crops were the most difficult, using the same variety for a long period of time to reduce nitrogen reactivity and increase disease risk.
Unlike previous studies in which linked yields in the 1970s and 80s were reduced primarily against a decrease in nitrogen supply, this analysis shows an increase in nighttime temperatures as an important factor.
“Our results show that maintaining productivity in Asian rice bowls requires not only better management, but also seasonal breeding and more frequent turnover of varieties,” Saito said.
The findings were published in Field Crops Research.
Important pathways for stability
Using the data, researchers were able to identify three key strategies to maintain long-term stability in rice production.
This includes developing climate-sensitive rice varieties, improving adaptation to seasonal changes, and more dynamic varieties rotation.
“Combining 50 years of detailed crop and climate data with modern AI tools gives us a much clearer view of what keeps rice production going, meaning we can design smarter season-specific strategies for farmers.”
IRRI highlighted how projects like LTCCE are “an essential resource for elucidating how rice systems can withstand and adapt through changing circumstances.”
“These insights are important far beyond one experimental site, providing a blueprint for climate-sensitive rice agriculture across Asia's 22 million hectares of irrigation monoculture that feeds billions.
Source: Field crop research. Machine Learning reveals drivers of yield sustainability in 50 years of continuous rice cultivation. Author: Yamaguchi Tomoki et al. https://doi.org/10.1016/j.fcr.2025.110114
