Multi-view RGB phenotyping reveals distinct temporal responses of quinoa to drought and salinity

Plant stress changes over time, but image-based screening often relies on traits selected from one stage of the response. Whether the same traits remain useful as stress develops is less clear. We examined drought and salinity responses in quinoa using multi-view RGB image series from two experiments. For organ-resolved phenotyping, a publicly released pretrained U-Net++ model with a MobileNetV3 encoder was used to identify stem, panicle, and supporting-stick regions, from which leaf and whole-plant masks were derived. In the longitudinal experiment, drought first met the operational discrimination criterion against the control at Day 3, when much of the useful information came from changes in leaf appearance relative to the pre-stress state. Salinity developed a clear visible phenotype later, first meeting the same criterion against the control at Day 10. Drought and salinity first met the criterion at Day 6. The corresponding areas under the receiver operating characteristic curve were 0.804, 0.868, and 0.927 for these three comparisons. The phenotype components associated with treatment separation also changed over time. Early drought differences were concentrated in leaf appearance, while later separation involved projected growth and plant structure more strongly. Salinity-related differences emerged later and were increasingly associated with whole-plant and stem structure. A second drought experiment, based on a different sampling design, showed a similar broad progression. Across 33 shared whole-image traits, drought-effect profiles were positively correlated between experiments, with Spearman correlations of 0.717 and 0.865 in two representative comparisons. The results indicate that stress discrimination from color images depends on the stage of the response and on which part of the phenotype carries the signal. For quinoa stress monitoring, a fixed set of image traits may therefore be less useful than features selected with response stage in mind.