The genetic variability of fruit trees in response to drought stress is scarcely studied. As adaptation of scion cultivars to abiotic constraints constitutes a new challenge for fruit production, in particular where water scarcity is likely to occur, development of high-throughput phenotyping strategies applicable in the field to assess the tree response to soil drought among large populations is needed, overcoming the limitations of usual in-planta measurements. In this research, remotely sensed images were acquired by ultra-light aircraft (ULA) and an unmanned aerial vehicle (UAV) during 4 years in a field trial where an apple progeny (122 hybrids) was studied under contrasted summer irrigation regimes. Ortho-images were simultaneously acquired in visible (RGB), near-infrared (NIR) and thermalinfrared (TIR) bands. After rthorectification, georeferencing and mosaicking, RGB and NIR images were used to compute different vegetation indices over the field trial, while TIR imaging allowed extraction of the vegetation surface temperature (Ts), which was calibrated at ground level by using hot and cold reference targets. The Morans'f water deficit index (WDI), which combines the surface minus air temperature (Ts-Ta) and the normalized difference vegetation index (NDVI), was used as a stress phenotypic variable. WDI estimates the ratio of actual to maximal evapotranspiration (WDI=1.ETact/ETmax) in discontinuous plant covers. Like the Ts-Ta variable, it significantly discriminated the tree water statuses and genotypes. On the basis of different plant- and image-based indices, individual tree behaviour trends (isohydric vs. anisohydric) can be distinguished among the progeny, irrespective of tree vigour. This opens potential applications for plant breeding, and genetic bases of apple tree response to water stress are currently investigated through quantitative trait locus (QTL) detection. Making use of ULA with flights performed at 40-60 m altitude made it possible to strongly improve the TIR image resolution ('10 cm) and to limit the number of vegetation/soil mixed pixels. However, it will require careful image posttreatment, possibly including classification and/or segmentation.