Machine learning offers new way to predict plant water use, Israeli study finds

Tel Aviv [Israel], December 16 (ANI/TPS): A new Israeli study suggests that machine-learning models may soon give growers a far more precise way to predict how much water their crops use each day, while also laying the groundwork for earlier detection of plant stress.
The research focused on daily plant transpiration -- a process by which water evaporates through the leaves and a key indicator of how much water a plant actually consumes. While transpiration is central to irrigation planning, most existing methods of assessing it rely on indirect information such as weather data or soil moisture, rather than the plant's own physiological behaviour.
Led by Shani Friedman and Nir Averbuch under the supervision of Prof. Menachem Moshelion at the Hebrew University of Jerusalem, the study drew on seven years of continuous, high-resolution measurements from tomato, wheat and barley plants grown in semi-commercial greenhouse conditions. Using a high-precision load-cell lysimeter system, the team recorded subtle changes in plant weight in real time, enabling direct and exceptionally accurate measurement of daily transpiration.
That long-term, plant-level dataset enabled a key innovation: training machine-learning models on how healthy, well-irrigated plants actually behave, rather than on indirect environmental proxies. By feeding the data into models such as Random Forest and XGBoost, the team showed that machine learning can reliably predict daily transpiration from environmental conditions and plant characteristics across multiple crops.
In independent tests, the XGBoost model achieved an R² value of 0.82, closely matching measured transpiration even when applied under different climate conditions and in separate research facilities. According to the researchers, this ability to generalise across crops and environments suggests the models are capturing fundamental physiological signals rather than crop-specific noise.
Two variables emerged as especially influential: plant biomass and daily temperature. "These variables consistently shaped how much water plants consumed," Friedman said. "Understanding how a healthy, well-irrigated plant is expected to behave on a given day also allows us to detect when something is off."
That concept represents another novel aspect of the work. Because the model predicts what a healthy plant should be doing, unexpected deviations from the prediction may serve as early warning signs of stress. Such stress could result from drought, salinity, disease, root damage or other environmental pressures, potentially before visible symptoms appear.
"If a plant behaves differently than the model predicts, that deviation can be an indicator of abnormal or unhealthy plant behaviour," Friedman said.
Averbuch, whose research focuses on precision irrigation, said the findings point toward a shift in how data-driven tools could be used in agriculture. "Today, many irrigation decisions still rely on indirect estimates," he said. "Although this model is not yet field-ready, the findings show how future systems could incorporate physiological predictions to support more accurate irrigation scheduling."
While the current approach depends on lysimeter data not typically available to growers, the researchers see it as a conceptual step toward plant-driven decision tools that could eventually be adapted to more practical sensors.
The study also performed well when tested on plants grown in a separate research greenhouse at Tel Aviv University, reinforcing the potential for broader applicability across climates and production systems.
In the near term, the study's approach is most applicable in research and controlled growing environments. By providing a precise physiological baseline for how healthy plants should transpire under given conditions, the model can help researchers benchmark crop water use, validate irrigation algorithms, and improve greenhouse management. Deviations between predicted and measured transpiration may also serve as an early indicator of plant stress in breeding trials or experimental systems, often before visible symptoms appear.
In the longer term, insights from the model point toward more advanced precision agriculture tools for growers supporting better irrigation scheduling and water savings. As similar models are paired with field-ready sensors, they could also form the basis of early warning systems that alert growers to emerging stress caused by drought, salinity, disease, or root damage. (ANI/TPS)