Mingjie Ma stood in his test field under the Xinjiang sun, watching a small drone buzz overhead, its multispectral camera quietly capturing data that could change how farmers—and energy producers—think about water. Ma, from the College of Hydraulic and Civil Engineering at Xinjiang Agricultural University, isn’t just studying maize—he’s helping build a smarter way to manage one of the planet’s most precious resources. And for industries from agriculture to energy, that matters more than ever.
Most irrigation monitoring systems focus on either soil moisture or plant health, treating them as separate problems. But in reality, water stress in a maize field isn’t just about dry soil or wilting leaves—it’s about the balance between the two. “We were trying to see the whole picture,” Ma explains. “Not just how much water is in the soil, or how much the plant is using, but how the two interact under real field conditions.”
To do this, Ma and his team used a drone equipped with multispectral sensors—cameras that detect light beyond what the human eye can see. Over four key growth stages in 2023, they flew the drone over drip-irrigated maize fields, collecting data that revealed subtle changes in plant and soil conditions. They then combined soil moisture measurements with plant water status using two mathematical methods: the coefficient of variation (CV) and the entropy weight method (EWM). The result was a single index—the Composite Moisture Index (CMI)—that acts like a health score for the entire soil-plant system.
What makes this approach groundbreaking is its ability to show not just where water is, but how effectively it’s being used. “Previous models treated soil and plant as separate entities,” Ma notes. “But in a drip-irrigated field, water moves from soil to plant in a continuous cycle. Our index reflects that dynamic.”
The team then tested several machine learning models to predict CMI from the drone data. Among them, a new hybrid model called SSA-RFE-XGBoost stood out. By combining swarm optimization (SSA), recursive feature elimination (RFE), and an advanced gradient boosting algorithm (XGBoost), the model achieved the highest accuracy across all growth stages, with R² values between 0.511 and 0.714—solid results for real-world field conditions.
The practical implications are significant, especially for energy-intensive agricultural regions like Xinjiang, a major grain-producing area in northwest China. Drip irrigation already saves water compared to flood irrigation, but knowing exactly where and when water stress occurs can help farmers fine-tune irrigation schedules, reduce energy use in pumping, and avoid over-application of water and fertilizers. That means lower electricity demand for irrigation systems and reduced strain on local water supplies—both critical for sustainable energy and water management in arid regions.
Perhaps most importantly, the study produced high-resolution spatiotemporal maps of CMI across the field. For the first time, researchers could see not just averages, but fine-scale patterns of water stress—revealing hotspots where plants might be struggling even if the overall field looks healthy. “This kind of fine detail is what smart farming needs,” says Ma. “It lets farmers act before problems become visible to the naked eye.”
The research, published in *Agricultural Water Management* (《农业水管理》), represents a shift from monitoring single components to understanding the entire soil-plant-water system. For energy planners, this could translate into more efficient water and power use in agriculture, a sector that accounts for a significant share of water and energy consumption globally.
As climate change intensifies droughts and water scarcity, tools that help farmers do more with less—and do it smarter—will be essential. Ma’s work offers a glimpse of that future: a world where drones, data, and machine learning come together not just to observe the land, but to care for it—and in doing so, support both food security and energy resilience.