COS 244-6 - HYPER-PHYS: HYPERspectral Plant HYdration Status

Abstract: Over land, most of the water transported from soil to atmosphere travels through the vascular system of plants. During drought, the vascular system can fail—leading to declines in plant water status, loss of productivity, and in extreme cases result in plant death. Thus, sensing plant water status is critical to understand the function and fate of plants in our warming planet. One widely measured indicator of plant water status and drought tolerance is the leaf water potential at wilting (turgor loss point, TLP). The most common method to estimate TLP is the pressure-volume curve (PV curve). This curve describes the relationship between water potential (WP) and relative water content (RWC) through repeated measurements of a drying leaf. Current methods for quantifying PV curves are slow and require constant monitoring by an expert hand, which limits our capacity to assess plant water status across large scales due to cost and labor constraints. Here we aim to develop HYPERspectral Plant HYdration Status (HYPER-PHYS), a novel and scalable method for sensing plant function, dysfunction, and death by leveraging hyperspectral imaging and automated image analysis. To automate the measurement of TLP, we developed models to estimate plant water potential and relative water content using spectral reflectance (450-2450 nm) and tested and validated them against the gold-standard pressure-volume method. We further compared the accuracy of estimating TLP by two different approaches: PV curve analysis spreadsheet tool (PVAST) and segmented regression. Our partial least-square regression models successfully predicted both WP and RWC (R² > 0.8, slope > 0.9, < 11 percent root mean square error). TLP values were not significantly different between our HYPER-PHYS method and PVAST or segmented methods. PVAST generated an average TLP of -1.75 ± 0.12 MPa, spectrally-predicted TLPs fit with PVAST were –1.70 ± 0.10 MPa, segmented regression a TLP of -1.42 ± 0.08MPa, and spectra values fitted to segmented regression were -1.44 ± 0.10 MPa. Species-specific models performed better than multi-species models, likely because of anatomical differences among species. We are presently conducting intensive anatomical surveys to include structural traits alongside physiological ones in our predictive models to identify which anatomical traits influence variation in model performance. Our findings indicate the potential to develop “Spectral Ecophysiology” measurements to automatize the determination of plant water status.