LB 31-263 - Disentangling Flowers Spectra to Map Landscape-scale Blooming Phenology

Modeling flowering dynamics (e.g., blooming) at the landscape scale is relevant for revealing hints on ecological processes such as: understanding phenotypic pollinator selections, monitoring floral adaptations to environmental changes, and shifts in spatio-temporal plant phenology and diversity. Depending on their chemical structures, flower pigments absorb light along the spectral range between the ultraviolet to near-infrared wavelengths (below 800 nm), which can be measured by image spectroscopy across different spatio-temporal scales. However, it is challenging to understand flower coloration's contribution to canopy spectral signals. Our research investigates methods that account for the relative contributions of flowers, leaves, and soil to typical hyperspectral scenes to characterize the spectral variability within a pixel, map flowering areas, and reveal specific phenophases across species. Weekly time series imagery from the airborne imaging spectrometer AVIRIS-NG (427 spaced 5nm-bands from 0.38 to 2.51μm and 8m spatial resolution) and field spectra were collected as part of the NASA SBG High-Frequency Time Series (SHIFT) campaign carried out in The Nature Conservancy's Jack and Laura Dangermond Preserve, Santa Barbara, CA, during 2022's spring (March-May)—acknowledged as one of the most biodiverse and dynamic land and marine environments in North America. Radiance swaths were atmospherically corrected with NASA-JPL's ISOFIT, and ground spectra from flowering plots were spectrally resampled to match the AVIRIS-NG response. The processed data was input to investigate flowering species' spectro-temporal variation and spatial distribution using the Spectral Mixture Residual, Bayesian clustering techniques, and a proposed hyperspectral flowering index. Spectral unmixing allowed the computation of the within-pixel contribution of leaves, flowers, and soil, retrieving residual signal that comprises subtle spectral features linked to plant functional traits. Principal components of the mixture residual spectra reduced the dimensionality of the data to identify flowering clusters based on a Gaussian mixture model and retrieve the uncertainty of the results. Mapping flowering events from modeling spectro-temporal dynamics throughout the season —from pre-blooming to post-flowering stages—allowed us to identify spectral gradient variations linked to flowering pigments and water content changes. Time series of the Modified Enhanced Blooming Index (MEBI) and the Red-Edge Normalized Difference Vegetation Index (RENDVI) revealed specific flowering and emergence/senesce phenophases across the two predominant native species (e.g., Coreopsis gigantea and Artemisia californica) in the flowering areas. Our approach contributes to the challenge of adopting imaging spectroscopy for monitoring floral cycles and biodiversity at broader scales, illustrating the use of current and future hyperspectral data collected by airborne and satellite missions (e.g., NASA-SBG).