Deciphering the Spectra of Flowers to Map Landscape-scale Blooming Dynamics

Like leaves, the coloration of flowers is driven by inherent optical properties, which are determined by pigments, scattering structure, and thickness. However, establishing the relative contribution of these factors to canopy spectral signals is usually limited to in-situ observations at a flower scale. Modeling flowering dynamics (e.g., blooming duration, spatial distribution) at the landscape scale may reveal insights into ecological processes, diversity of plants and pollinators, and phenological adaptations to environmental changes. Multitemporal visible to shortwave infrared (VSWIR) imaging spectroscopy aerial and satellite-based observations are especially suited for such efforts. Reflectance in this spectral range is sensitive to major flower pigments, flowering phenology traces, and biophysical differences between flowers and other plant parts. We explored how flowers contribute to spectral signals using a time series of imagery from the Airborne Visible InfraRed Imaging Spectrometer - Next Generation (AVIRIS-NG) collected as part of the SBG High-Frequency Time Series (SHIFT) campaign as a case study. Airborne data were collected weekly during the spring of 2022 and in situ samples of flowering plants were sampled weekly in conjunction with these flights across two natural reserve areas in California. Field spectra were gathered from blooming plots at leaf, flower, and canopy levels at two time points during the campaign. The processed data was used to investigate flowering species' spectro-temporal variation and spatial distribution using Spectral Mixture Residual, Gaussian clustering techniques, and a proposed narrow-band flowering index. Linear spectral unmixing allowed the computation of the weighted contribution of four major low-variance endmembers (leaves, flowers, soil, dark) and high-variance residual signal that comprises subtle spectral features used to track biophysical processes. The reflectance residual was projected on a low principal component basis to characterize flowering clusters' variation and spatial distribution based on the Gaussian mixture model, providing an uncertainty metric to assess the results. Mapping flowering events from modeling spectro-temporal dynamics throughout the season, from pre-blooming to post-flowering stages, allowed us to identify gradient variations in spectral features within the visible and shortwave infrared spectral ranges linked to flowering pigments and water content changes, respectively. Time series of the Modified Enhanced Blooming Index and the Red-Edge Normalized Difference Vegetation Index revealed specific flowering cycles and emergence/senesce sequence phenophases across the two main species (e.g., Giant Coreopsis and California Sagebrush) in the flowering areas. Overall, our approach opens opportunities for future satellite monitoring of floral cycles at broader scales.