China-developed Knowledge-guided AI Algorithm Offers Promising Results in Global Soil Moisture Management
09 May 2026
Professor Zeng Jiangyuan during an interview with China.org.cn in Beijing, April 29, 2026. [Photo by Liao Jiaxin/China.org.cn]
Researchers from the Aerospace Information Research Institute under the Chinese Academy of Sciences (AIRCAS) have developed a knowledge-guided artificial intelligence algorithm that significantly improves the resolution of global soil moisture observation.
The findings were published in the journal IEEE Transactions on Geoscience and Remote Sensing on March 3.
The new tool, known as the Heterogeneity-Aware Downscaling Algorithm (HADA), addresses a long-standing limitation in remote sensing: the loss of detail when soil moisture is averaged across large, diverse landscapes. Currently, most satellite-derived soil moisture data has a resolution of roughly 25 km. HADA enhances that data to 5 km resolution. This breakthrough has the potential to be applied in agriculture, water management, and disaster early warning.
"Take a 625-square-kilometre area with a river running through it — dry land on one side, wetlands on the other," said Zeng Jiangyuan, a professor at AIRCAS who leads the team. "At 25 km resolution, you see only a blurry average. But when you break that grid into 25 smaller ones at 5 km resolution, you can actually see which side is dry and which is wet. That is what improving resolution does."
The algorithm is particularly vital for regions with sparse ground-level monitoring. According to the World Meteorological Organization, surface station density in Africa is far below global standards.
"Our product can play a useful role in these regions," Zeng said. "It does not require any ground station data and is not limited to any specific microwave satellite sensor. It can provide them with high-resolution, high-frequency soil moisture information."
Extreme droughts and floods are becoming more frequent worldwide. Zeng said that forecasting such events often relies on climate models, where soil moisture is a key input because it regulates land-atmosphere interactions. Higher-resolution soil moisture, he added, helps models produce more precise forecasts for disaster prevention.
According to Zeng, higher-resolution soil moisture data can also support more targeted irrigation advice, reducing water use while maintaining crop yields, particularly in regions such as for fragmented landscapes, southern China's hilly terraced fields.
Soil moisture directly determines crop growth and yields, he said, noting that many crop yield models rely on it as a key input. "With coarse resolution such as tens of kilometers per value, you only know a region is roughly dry. You cannot tell which field is dry," Zeng said. "With higher resolution, you can tell farmers exactly which plots need water." Better soil moisture information means better irrigation decisions, saving water while increasing yields.
The technology also aligns with China's No. 1 Central Document for 2026, which prioritizes smart agriculture and calls for integrating AI and data-driven technologies into farming.
Zeng recalled a particularly encouraging moment during the research: the team compared HADA with existing products and downscaling algorithms using all available soil moisture data from the International Soil Moisture Network (ISMN). "We found that our downscaling results outperformed these products and algorithms. It was a very encouraging process that proved our algorithm is reliable," he said.
The team plans to apply the algorithm to Chinese satellite missions soon. Their next goal is to further increase the resolution to 1 km. "I think it can be achieved within one to two years," Zeng said.
Research News
China-developed Knowledge-guided AI Algorithm Offers Promising Results in Global Soil Moisture Management
Professor Zeng Jiangyuan during an interview with China.org.cn in Beijing, April 29, 2026. [Photo by Liao Jiaxin/China.org.cn]
Researchers from the Aerospace Information Research Institute under the Chinese Academy of Sciences (AIRCAS) have developed a knowledge-guided artificial intelligence algorithm that significantly improves the resolution of global soil moisture observation.
The findings were published in the journal IEEE Transactions on Geoscience and Remote Sensing on March 3.
The new tool, known as the Heterogeneity-Aware Downscaling Algorithm (HADA), addresses a long-standing limitation in remote sensing: the loss of detail when soil moisture is averaged across large, diverse landscapes. Currently, most satellite-derived soil moisture data has a resolution of roughly 25 km. HADA enhances that data to 5 km resolution. This breakthrough has the potential to be applied in agriculture, water management, and disaster early warning.
"Take a 625-square-kilometre area with a river running through it — dry land on one side, wetlands on the other," said Zeng Jiangyuan, a professor at AIRCAS who leads the team. "At 25 km resolution, you see only a blurry average. But when you break that grid into 25 smaller ones at 5 km resolution, you can actually see which side is dry and which is wet. That is what improving resolution does."
The algorithm is particularly vital for regions with sparse ground-level monitoring. According to the World Meteorological Organization, surface station density in Africa is far below global standards.
"Our product can play a useful role in these regions," Zeng said. "It does not require any ground station data and is not limited to any specific microwave satellite sensor. It can provide them with high-resolution, high-frequency soil moisture information."
Extreme droughts and floods are becoming more frequent worldwide. Zeng said that forecasting such events often relies on climate models, where soil moisture is a key input because it regulates land-atmosphere interactions. Higher-resolution soil moisture, he added, helps models produce more precise forecasts for disaster prevention.
According to Zeng, higher-resolution soil moisture data can also support more targeted irrigation advice, reducing water use while maintaining crop yields, particularly in regions such as for fragmented landscapes, southern China's hilly terraced fields.
Soil moisture directly determines crop growth and yields, he said, noting that many crop yield models rely on it as a key input. "With coarse resolution such as tens of kilometers per value, you only know a region is roughly dry. You cannot tell which field is dry," Zeng said. "With higher resolution, you can tell farmers exactly which plots need water." Better soil moisture information means better irrigation decisions, saving water while increasing yields.
The technology also aligns with China's No. 1 Central Document for 2026, which prioritizes smart agriculture and calls for integrating AI and data-driven technologies into farming.
Zeng recalled a particularly encouraging moment during the research: the team compared HADA with existing products and downscaling algorithms using all available soil moisture data from the International Soil Moisture Network (ISMN). "We found that our downscaling results outperformed these products and algorithms. It was a very encouraging process that proved our algorithm is reliable," he said.
The team plans to apply the algorithm to Chinese satellite missions soon. Their next goal is to further increase the resolution to 1 km. "I think it can be achieved within one to two years," Zeng said.
Source: China.org.cn