ABSTRACT
Within-orchard transfer of European hazelnut water-stress classification from a single reference station

Stanley Best1*, José Maldonado1, Cristóbal Campos2, Jae Seung Noh3, and Francisco Altimiras4
 
Soil water content measured at a single depth or location may not represent irrigation need across a heterogeneous orchard. The objective was to determine whether a classification model developed at one reference station could identify water-stress states at three other stations of a commercial European hazelnut (Corylus avellana L.) ‘Tonda di Giffoni’ orchard without recalibration. Three management zones were delineated by k-means clustering of Sentinel-2 imagery acquired from 2023 to 2025. Station-specific irrigation thresholds (0.190 to 0.238 m3 m-3, all reached at a stem water-potential tension of 1.28 to 1.32 MPa) were defined from soil-water depletion and pressure-chamber measurements. An L2-penalized logistic model combined the reference-station Surface-to-Profile Water Extraction Ratio (SPWER), local PlanetScope canopy dynamics, and vapor pressure deficit. Development used 46 station-dates with 12 stress events in 18 irrigation cycles, and leave-one-cycle-out cross-validation yielded an area under the receiver operating characteristic curve (ROC-AUC) of 0.971. Applied without recalibration to 141 station-dates at three held-out stations, the model reached ROC-AUC 0.874 (95% confidence interval [CI] 0.814 to 0.928; lower limit above the 0.500 no-discrimination value), balanced accuracy 0.779, sensitivity 0.612, and specificity 0.946. A reduced model without SPWER reached ROC-AUC 0.849 and balanced accuracy 0.810, and a paired cycle bootstrap did not establish an incremental benefit of SPWER (difference 0.025; 95% CI -0.018 to 0.069). Twenty-three of 24 errors occurred within 0.02 m3 m-3 of the local threshold. The model separated within-orchard stress states, whereas the contribution of a single reference station was not resolved.
Key words: Irrigation management, management zones, precision agriculture, remote sensing, soil-plant-atmosphere continuum.
1Instituto de Investigaciones Agropecuarias, INIA Quilamapu, Chillán, Chile.
2Fundación Observatorio Ñuble, Chillán, Chile.
3National Institute of Agricultural Sciences, Rural Development Administration, Field Farming Mechanization Division, Jeonju-si, Republic of Korea.
4Universidad de Las Américas, Facultad de Ingeniería y Negocios, Santiago, Chile.