Multitemporal Monitoring of Phenotypic Traits in Wild Tomato Species (S. pimpinellifolium) Using UAV-based Hyperspectral Imagery

Plant phenotyping relies on high quality and timely information. To plan an effective growth strategy, it is important to know the physiological status of the crop and how this changes throughout the crop cycle as well as information to diagnose precisely when, why and where a crop is suffering disease outbreaks, abiotic stress, or other negative impacts. Unmanned aerial vehicles (UAV) are highly flexible platforms for observing agricultural systems with high spatio-temporal resolution and low operational costs. In addition, spectral resolution plays a significant role in the characterization, modeling, and mapping of physiological variables. Hyperspectral UAV-based sensing retrieves biophysical and biochemical metrics, precisely measuring energy at defined wavelengths that are reflected from the plants. A field experiment was conducted to evaluate a diversity panel of wild tomato (Solanum pimpinellifolium), under control and salt stress conditions. A total of 200 plant accessions were monitored by collecting phenotypic in-situ measurements, ground truth spectra (350-2500 nm), and UAV-based hyperspectral data (400 to 1000 nm, comprised of 270 continuous bands), with the purpose of discriminating their salt tolerance features. To account for differences in spectral quantity and quality, accurate reflectance retrievals are required. Despite the potential associated with high spectral resolution data, its reliability is highly dependent on adequate pre-processing. Therefore, radiometric calibration is a critical and non-trivial pre-processing task, which must be performed to gain comparable surface reflectance spectra in time-series imagery. In this study, the empirical line calibration technique together with an optimal denoising strategy is presented as an appropriate method to translate raw data into reflectance. We then demonstrate the value of multitemporal UAV-based reflectance data in capturing complex physiological parameters by developing a multitemporal set of 20 hyperspectral vegetation indices that monitor key dynamic phenotypic traits such as greenness, light use efficiency, and leaf pigments. The approach informs on the usefulness of hyperspectral UAV-based sensing for advanced phenotyping applications in stress detection and diagnostics in precision agriculture.