A forecasting system tailored to small watersheds has prevented weather-related disasters, including a flash flood in the city of Yuyao, Zhejiang province, in recent years.

By analyzing real-time monitoring data and rainfall patterns, the meteorological team in Yuyao built a machine-learning model in 2023 that incorporates multiple variables, such as cumulative rainfall, hourly peak intensity and three-hour rain rates. The platform, called "One Strategy for One Gully", can produce dynamic thresholds that reflect the specific behavior of each gully in the city, according to Jin Xufeng, a forecaster at the Yuyao Meteorological Bureau.

"It displays terrain, sensors and forecast models for every drainage gully. When a threshold is breached, the alert is immediate and clearly visible," Jin said.

When Typhoon Co-May intensified and stalled off the Zhejiang coast in July, torrential rain quickly engulfed the mountainous city of Yuyao. Within hours, all 1,153 people living in designated risk zones were safely evacuated thanks to efforts in precision meteorological forecasting and warning systems, local authorities said.

The landscape in Yuyao is often described as five parts mountain, two parts water and three parts farmland. The terrain funnels extreme rainfall into narrow valleys, resulting in flash floods.

"The landscape is higher in the south and lower in the north, and that means rainfall in the mountains can be more than twice what we see in the plains," Jin said.

The city has faced several major tests over the past years. Typhoon In-Fa in 2021 delivered 1,048.2 millimeters of rainfall to the township of Dalan, a record high in Zhejiang. In 2022, Typhoon Muifa pushed the Yaojiang River over guaranteed flood levels by 1.07 meters, submerging neighborhoods and farmlands.

"Flash floods come fast, peak fast and recede fast. The window for action is very short," said Gu Jinyin, an official with the Yuyao water resources bureau.

To close the gap, meteorologists and water resource officials built a joint consultation mechanism to assess risk and coordinate action before storms arrive. Alert poles in high-risk zones are now equipped with automated audio and visual alarms, and are triggered when water levels exceed danger thresholds, giving residents up to 30 minutes to evacuate.

The city has also built a more robust monitoring network, including X-band radar and millimeter-wave cloud radar. As of last year, monitoring stations covered every village and community. With data refreshed every 2.3 kilometers, forecasters receive real-time rainfall and water level data, enhancing prediction accuracy.

During Typhoon Kong-rey in late 2024, rainfall in Yuyao exceeded 70 millimeters in three hours, prompting red alerts issued through precision messaging tools. Within minutes, alerts reached every household. A total of 1,238 residents were evacuated without injuries or losses, said Hu Donglin, an agricultural official in Dayin township.

Jin from the Yuyao meteorological bureau said the city is seen as a model for Zhejiang in terms of small watershed forecasting. "We will further expand alert coverage, combining advanced technology with a community-based emergency system to safeguard lives amid intensifying weather extremes," he said.