Efrat et al. (2025) Modeling seasonal water status and predicting yield in almond orchards using UAV multi-sensor and meteorological data

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-10-10
• Authors: Noam Efrat, Vladislav Dubinin, Shahar Baram, Tarin Paz‐Kagan
• DOI: 10.1016/j.agwat.2025.109868

Research Groups

• French Associates Institute for Agriculture and Biotechnology of Dryland, The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sede Boqer Campus, Israel
• Newe Ya’ar Research Center, Institute of Plant Protection, Agricultural Research Organization – Volcani Institute, Ramat Yishay, Israel

Short Summary

This study developed a scalable framework integrating UAV multi-sensor, meteorological, and irrigation data with Random Forest models to accurately predict seasonal plant water status (Stem Water Potential, SWP; Trunk Growth Rate, TGR) and yield in almond orchards. The research revealed that distinct water stress profiles, identified by fuzzy clustering, are strongly linked to significant yield reductions, offering actionable insights for precision irrigation.

Objective

• To create a scalable, data-driven framework for modeling seasonal dynamics of Stem Water Potential (SWP) and Trunk Growth Rate (TGR) and for predicting yield in almond orchards by integrating UAV-based multi-sensor imagery with physiological, meteorological, and irrigation data.
• Hypothesis: Combining UAV-based thermal and spectral features with field-based physiological indicators will enable accurate prediction of SWP, TGR, and yield, while also revealing phenological periods most critical for differentiating stress responses.

Study Configuration

• Spatial Scale: A 15-year-old commercial almond orchard spanning approximately 18 hectares in the southern Shephelah region of Israel (31°41′22.1″N, 34°46′29.5″E). The study focused on 60 individual trees across three contiguous plots (7.9 ha, 5.3 ha, 4.8 ha) with a planting density of 240 trees per hectare.
• Temporal Scale: Two full growing seasons (2022–2023), with monthly UAV flights and physiological measurements, and daily meteorological and irrigation data collection. Dendrometer data was collected at 30-minute intervals.

Methodology and Data

• Models used:
  • Random Forest (RF) regression models (for SWP, TGR, and yield prediction)
  • Fuzzy C-Means (FCM) clustering (for identifying seasonal SWP and TGR response profiles)
  • FAO-56 Penman-Monteith equation (for Reference Evapotranspiration, ET₀)
  • Magnus formula (for Vapor Pressure Deficit, VPD)
  • Agisoft PhotoScan Standard (for orthomosaic generation)
  • lidR R package (for LiDAR data processing and tree segmentation)
  • Variance Inflation Factor (VIF) and Boruta algorithm (for feature selection)
• Data sources:
  • UAV-based multi-sensor imagery: Multispectral (MicaSense RedEdge-P, Altum-PT, Phantom 4 Multispectral), Thermal (FLIR A655, DJI Zenmuse H20T), and LiDAR (DJI Zenmuse L1).
  • Meteorological data: Daily records from a nearby station (air temperature, relative humidity, wind speed, precipitation, derived Growing Degree Days (GDD), Reference Evapotranspiration (ET₀), Vapor Pressure Deficit (VPD)).
  • Irrigation records: Daily irrigation volume (in mm) per plot.
  • Physiological measurements: Midday Stem Water Potential (SWP) (in MPa) using a pressure chamber, Trunk Growth Rate (TGR) (in µm day⁻¹) and Maximum Daily Shrinkage (MDS) (in µm day⁻¹) from dendrometers.
  • Agronomic data: Dry kernel yield per tree (in kg tree⁻¹), cultivar identity (UEF, 53), and Canopy Nitrogen Content (CNC) (in kg tree⁻¹) from a related study.

Main Results

• Random Forest models demonstrated high predictive accuracy for:
  • TGR: R² = 0.78 (independent test set, Plot A), RMSE = 10.87 µm.
  • SWP: R² = 0.79 (independent test set, Plot A), RMSE = 0.21 MPa.
  • Yield: R² = 0.87 (independent test set, Plot A), RMSE = 0.41 kg tree⁻¹.
• Key predictors for SWP included canopy temperature, Normalized Difference Red Edge Index (NDRE), Crop Water Stress Index (CWSI), and Modified Soil Adjusted Vegetation Index (MSAVI).
• Key predictors for TGR included irrigation amount, air temperature, and spectral reflectance at 475 nm and 842 nm, and NDRE.
• Yield prediction was strongly driven by early-season Canopy Nitrogen Content (June), Trunk Growth Rate (April and June), and Stem Water Potential (April and June).
• Fuzzy C-Means clustering identified three distinct seasonal stress-response groups for both SWP and TGR. Trees in the most stressed cluster exhibited a yield reduction of up to 76 % compared to the least stressed cluster.
• Significant interannual variability was observed: 2023 was warmer and drier, leading to earlier and more intense water stress and a 57% reduction in average yield compared to 2022.
• Cultivar '53' consistently showed higher Maximum Daily Shrinkage (MDS) and more negative SWP values than 'UEF', indicating greater sensitivity to water stress.

Contributions

• Developed a novel, scalable, and data-driven framework for comprehensive monitoring and prediction of tree-level water status and yield in almond orchards by integrating diverse multi-sensor UAV data with physiological and environmental inputs.
• Demonstrated the robust capability of Random Forest models to accurately estimate SWP, TGR, and yield, providing a foundation for precision agriculture applications.
• Identified the most influential spectral, meteorological, and irrigation features driving SWP, TGR, and yield variability, highlighting the critical roles of canopy temperature, red-edge indices, early-season nitrogen, and water status.
• Introduced Fuzzy C-Means clustering to effectively classify trees into distinct physiological stress-response groups based on seasonal SWP and TGR trajectories, and quantified their direct impact on yield outcomes.
• Provided actionable insights for optimizing precision irrigation and resource management strategies, enabling spatially explicit and tree-specific interventions for enhanced resilience and productivity in perennial orchard systems.

Funding

• U.S.–Israel Binational Agricultural Research and Development Fund (BARD), Grant No. IS-5430–21.

Citation

@article{Efrat2025Modeling,
  author = {Efrat, Noam and Dubinin, Vladislav and Baram, Shahar and Paz‐Kagan, Tarin},
  title = {Modeling seasonal water status and predicting yield in almond orchards using UAV multi-sensor and meteorological data},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.109868},
  url = {https://doi.org/10.1016/j.agwat.2025.109868}
}