Yu et al. (2026) Toward practical irrigation decision support: Predicting grapevine stem water potential from soil moisture and atmospheric demand
Identification
- Journal: Agricultural Water Management
- Year: 2026
- Date: 2026-11-12
- Authors: Runze Yu, Jesse Anastacio, Sho Murakoshi, Seiya Nishimura, Gen Sakoda, Takashi Takinami, Hiroyuki Okita
- DOI: 10.1016/j.agwat.2026.110929
Research Groups
- Department of Viticulture and Enology, California State University-Fresno, Fresno, CA, USA
- Sony Semiconductor Solutions Corporation, Kanagawa, Japan
- Sony Group Corporation, Tokyo, Japan
- Crossing Labs Inc., Tokyo, Japan
- Sensiphia, Inc., Tokyo, Japan
Short Summary
This proof-of-concept study evaluated the feasibility of estimating grapevine stem water potential (Ψstem) using a practical combination of soil moisture sensors and local weather data, demonstrating its potential for irrigation decision support in hot, water-limited vineyards.
Objective
- To determine whether grapevine stem water potential (Ψstem) can be estimated across multiple growing seasons using a practical combination of root-zone soil moisture and weather variables.
- To identify a stable modeling framework suitable for irrigation decision support.
Study Configuration
- Spatial Scale: Primary experiment conducted in a Barbera (Vitis vinifera L.) vineyard in Fresno, California, USA, with vines spaced at 2.12 m × 3.66 m. The experiment included four irrigation regimes and monitored 36 data vines. Independent external validation was performed in a French Colombard (V. vinifera L.) vineyard in Five Points, California, USA (approximately 23 km distant), with vines spaced at 1.83 m × 3.05 m.
- Temporal Scale: Data collected over three growing seasons (2023, 2024, 2025) for model development and one independent growing season (2026) for external validation. Soil moisture and temperature were recorded every 30 minutes, weather variables hourly, and midday Ψstem was measured weekly (2023) or twice weekly (2024, 2025, 2026).
Methodology and Data
- Models used: Support Vector Regression (SVR), Gaussian Process Regression (GPR), Random Forest (RF), Light Gradient Boosting Machine (LightGBM) (with default parameters and Bayesian optimization), eXtreme Gradient Boosting (XGBoost) (with default parameters and Bayesian optimization), Partial Least Squares (PLS) regression (linear and nonlinear terms), Multiple Regression (linear and nonlinear terms). Permutation Feature Importance (PFI) was used for predictor selection.
- Data sources:
- Soil Moisture: Spatial Frequency Domain Transmissometry (SFDT) sensors (20 cm, 37.5 cm, 75 cm depth) and Time-Domain Reflectometry (TDR) sensors (20 cm, 37.5 cm, 75 cm, 150 cm depth) at the primary site. TDR sensors (37.5 cm, 75 cm depth) at the external validation site.
- Soil Temperature: TDR sensors (20 cm, 37.5 cm, 75 cm, 150 cm depth).
- Weather: California Irrigation Management Information System (CIMIS) stations (Fresno State #80 and FivePoints #2). Variables included total precipitation (mm), average solar radiation (W m⁻²), average vapor pressure (kPa), average air temperature (°C), average relative humidity (%), average dew point (°C), average wind speed (m s⁻¹), wind run (km), and reference evapotranspiration (mm).
- Ground-truth (Plant Water Status): Midday stem water potential (Ψstem) measured using a pressure chamber.
- Soil Characterization: Soil organic matter (%), total nitrogen (%), total organic carbon (%), and soil texture (sand, silt, clay %) using NAPT methods.
Main Results
- Permutation Feature Importance (PFI) analyses consistently identified deep soil moisture (75 cm) and air temperature as the most influential predictors of grapevine Ψstem across various algorithms and datasets. Relative humidity provided limited additional explanatory value.
- Partial Least Squares (PLS) regression with linear terms was selected as the most stable and interpretable modeling approach for irrigation decision support.
- For the combined three-year (2023–2025) dataset, the final PLS regression models achieved:
- Two-variable model (soil moisture at 75 cm + air temperature): R² = 0.629, Root Mean Square Error (RMSE) = 0.141 MPa.
- Three-variable model (soil moisture at 75 cm + air temperature + relative humidity): R² = 0.639, RMSE = 0.139 MPa.
- The derived models, when applied without recalibration to an independent commercial vineyard in 2026, retained moderate predictive relationships with measured Ψstem, showing R² values ranging from 0.515 to 0.619 and RMSE values from 0.182 to 0.232 MPa.
- The final two-variable model equation (2023-2025) was: Ψstem = −0.7035 + 0.02852 SM75 −0.02751 Tair.
- The final three-variable model equation (2023-2025) was: Ψstem = −0.38275 + 0.02744 SM75 −0.03327 Tair −0.003615 RH.
Contributions
- Demonstrated the practical feasibility of estimating grapevine Ψstem using a minimal sensor package combining deep root-zone soil moisture and local weather station data.
- Identified deep soil moisture (75 cm) and air temperature as consistently dominant and robust predictors for Ψstem, providing a physiologically relevant basis for vine water status.
- Developed and validated a stable and interpretable linear PLS regression framework for Ψstem prediction that performs consistently across multiple growing seasons and irrigation regimes.
- Showcased the transferability of the developed models to an independent commercial vineyard without site-specific recalibration, highlighting the potential for broader adoption.
- Offers a practical, plant-based irrigation decision support framework that interprets soil moisture in the context of atmospheric demand, moving beyond reliance on fixed soil moisture thresholds.
Funding
- Sony Semiconductor Solutions
- Farm Food Future Innovate (F3i) grant
- California State University Agricultural Research Institute (CSU ARI) Campus grant (Project number: 25-02-104)
- American Society for Enology and Viticulture (ASEV) scholarship
- Ag One Foundation
Citation
@article{Yu2026Toward,
author = {Yu, Runze and Anastacio, Jesse and Murakoshi, Sho and Nishimura, Seiya and Sakoda, Gen and Takinami, Takashi and Okita, Hiroyuki},
title = {Toward practical irrigation decision support: Predicting grapevine stem water potential from soil moisture and atmospheric demand},
journal = {Agricultural Water Management},
year = {2026},
doi = {10.1016/j.agwat.2026.110929},
url = {https://doi.org/10.1016/j.agwat.2026.110929}
}
Original Source: https://doi.org/10.1016/j.agwat.2026.110929