AI can now predict rice harvest dates — within 3 days

"Should I harvest before the rain, or wait a few more days?" That age-old question for rice farmers may soon have a data-driven answer. An AI-powered program that analyzes growing regions and weather data to predict the optimal harvest window has been developed — and when applied to 9,565 rice plots this year, its predictions matched actual harvest dates to within three days.

The National Seed Institute announced Wednesday that it has jointly developed an "AI-based seed harvest timing prediction program" with the Korea Electronics Technology Institute and is making it available to government seed production sites, local governments and private seed companies.

Timing is critical in rice harvesting, particularly for seed rice. Harvesting too early leaves the grain insufficiently ripened, while harvesting too late risks rain damage or pre-harvest sprouting — a condition in which seeds germinate while still on the stalk — both of which reduce germination rates and seed quality.

Climate change has made the challenge harder. More frequent extreme weather events, including heat waves and heavy downpours, have made it increasingly difficult to predict when rice will ripen. Even the same variety can have different harvest windows depending on the growing region and that year's weather. Farmers have long relied on experience and field observation, but sudden typhoons and harvest-season rainfall have exposed the limits of that approach.

The new program was developed to address those challenges. Farmers enter basic information — growing region, variety and heading date — and the AI analyzes 46 years of accumulated temperature data (1980–2025) alongside current-year weather information. It then tracks the ripening progress of the rice and predicts the optimal harvest date for each plot, displaying the results as a graph.

Field results have been encouraging. When the institute applied the program to 24 rice varieties across 9,565 plots currently being grown as government-distributed seed stock this year, predicted harvest dates matched actual harvest dates within plus or minus three days. The institute said the results confirmed the program's practical applicability.

The program is not limited to government seed production. It is also being provided free of charge for higher-grade seed categories — such as breeder seed and foundation seed — produced by local governments, as well as for private food crop seed production. The aim is to allow institutions and companies to assess harvest timing using a common standard without having to develop their own tools.

The National Seed Institute and the Korea Electronics Technology Institute plan to improve prediction accuracy further by incorporating additional harvest and weather data. They also intend to link the program with agricultural satellite data to enable more precise monitoring of seed field conditions.

"Rice and cereal crop producer organizations or institutions that wish to use this program can contact the National Seed Institute to receive it along with a user manual, free of charge," said Yang Ju-pil, director of the National Seed Institute. "As climate change makes it increasingly difficult to determine the right harvest time based on experience alone, we expect this program to help producers respond proactively to weather changes and make harvest timing decisions more easily and accurately."