AI Method Enables High‑Resolution Mapping for China's Agricultural Irrigation Water Use Over 20 Years
13 Jul 2026
A research team led by Prof. WANG Shudong from the Aerospace Information Research Institute of the Chinese Academy of Sciences (AIRCAS) has developed a new way to measure agricultural irrigation water use across China by combining satellite observations, physical laws, and artificial intelligence.
Using this framework, the researchers produced a 500-meter resolution dataset of irrigation water use covering the past two decades. It is currently the most detailed and longest-running national-scale record of irrigation water estimates available. The dataset provides scientific foundation for improving agricultural water-saving practices, supporting water resources management, strengthening food security.
Accurately quantifying actual irrigation water in farmland has long been a major challenge in agricultural water resource management. Traditional approaches are largely based on field surveys, administrative reporting, and statistical aggregation. While these methods provide essential baseline information, they are often updated infrequently, costly to maintain, and too coarse in spatial resolution. These limitations make them ill-suited to the growing demands of precision agriculture, regional water allocation, and dynamic monitoring.
This challenge is particularly acute in China, where farmland is highly fragmented and irrigation practices vary widely across regions. As a result, developing accurate and continuous monitoring of irrigation water use has become a critical scientific imperative for ensuring food security and advancing water-saving agriculture.
To address this challenge, the research team developed a new framework that goes beyond conventional statistical or single-model approaches. It integrates multi-source Earth observation satellite data, deep learning techniques, and water balance theory.
The framework operates in three key steps:
First, multi-source satellite observations were combined to improve key hydrological variables such as evapotranspiration, precipitation, and soil moisture.
Second, differences between satellite-based land surface temperature and reanalysis-based temperature are used to identify irrigation periods during the growing season.
Third, water balance relationships identified during non-irrigated periods are extended to irrigated conditions. Artificial intelligence is then used to estimate non-evapotranspiration water loss components that are difficult to observe directly from remote sensing, enabling inverse estimation of irrigation water use.
The method shows strong performance across scales. For key intermediate variables, some models achieved correlation coefficients exceeding 0.9 on the test dataset.
At the field scale, comparisons with agricultural station records yielded a root mean square error of approximately 25.3 mm per year. At the national scale, comparisons with statistical data resulted in a root mean square error of approximately 3.9 cubic kilometers per year, indicating good overall consistency with existing records.
Beyond producing a detailed national map of agricultural water use, this study highlights paradigm broader shift in how Earth system information can be generated.
In this new framework, satellites provide continuous global observations, artificial intelligence extract patterns from complex data, and physical laws ensure scientific consistency and reliability. Together, these elements enable more transparent, updatable, and comparable spatiotemporal information.
The researchers note that such capabilities are increasingly important in the context of climate change, growing food demand, and increasing pressure on water resources. Future agricultural water management, they argue, will require not only historical statistics, but also near-real-time, high-resolution information that can support adaptive decision-making.
This study was jointly conducted by AIRCAS, the Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters at Nanjing University of Information Science and Technology, and the Department of Earth and Environmental Science at the University of Pennsylvania.
Research News
AI Method Enables High‑Resolution Mapping for China's Agricultural Irrigation Water Use Over 20 Years
A research team led by Prof. WANG Shudong from the Aerospace Information Research Institute of the Chinese Academy of Sciences (AIRCAS) has developed a new way to measure agricultural irrigation water use across China by combining satellite observations, physical laws, and artificial intelligence.
Using this framework, the researchers produced a 500-meter resolution dataset of irrigation water use covering the past two decades. It is currently the most detailed and longest-running national-scale record of irrigation water estimates available. The dataset provides scientific foundation for improving agricultural water-saving practices, supporting water resources management, strengthening food security.
The study has been published in the ISPRS Journal of Photogrammetry and Remote Sensing.
Accurately quantifying actual irrigation water in farmland has long been a major challenge in agricultural water resource management. Traditional approaches are largely based on field surveys, administrative reporting, and statistical aggregation. While these methods provide essential baseline information, they are often updated infrequently, costly to maintain, and too coarse in spatial resolution. These limitations make them ill-suited to the growing demands of precision agriculture, regional water allocation, and dynamic monitoring.
This challenge is particularly acute in China, where farmland is highly fragmented and irrigation practices vary widely across regions. As a result, developing accurate and continuous monitoring of irrigation water use has become a critical scientific imperative for ensuring food security and advancing water-saving agriculture.
To address this challenge, the research team developed a new framework that goes beyond conventional statistical or single-model approaches. It integrates multi-source Earth observation satellite data, deep learning techniques, and water balance theory.
The framework operates in three key steps:
First, multi-source satellite observations were combined to improve key hydrological variables such as evapotranspiration, precipitation, and soil moisture.
Second, differences between satellite-based land surface temperature and reanalysis-based temperature are used to identify irrigation periods during the growing season.
Third, water balance relationships identified during non-irrigated periods are extended to irrigated conditions. Artificial intelligence is then used to estimate non-evapotranspiration water loss components that are difficult to observe directly from remote sensing, enabling inverse estimation of irrigation water use.
The method shows strong performance across scales. For key intermediate variables, some models achieved correlation coefficients exceeding 0.9 on the test dataset.
At the field scale, comparisons with agricultural station records yielded a root mean square error of approximately 25.3 mm per year. At the national scale, comparisons with statistical data resulted in a root mean square error of approximately 3.9 cubic kilometers per year, indicating good overall consistency with existing records.
Beyond producing a detailed national map of agricultural water use, this study highlights paradigm broader shift in how Earth system information can be generated.
In this new framework, satellites provide continuous global observations, artificial intelligence extract patterns from complex data, and physical laws ensure scientific consistency and reliability. Together, these elements enable more transparent, updatable, and comparable spatiotemporal information.
The researchers note that such capabilities are increasingly important in the context of climate change, growing food demand, and increasing pressure on water resources. Future agricultural water management, they argue, will require not only historical statistics, but also near-real-time, high-resolution information that can support adaptive decision-making.
This study was jointly conducted by AIRCAS, the Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters at Nanjing University of Information Science and Technology, and the Department of Earth and Environmental Science at the University of Pennsylvania.