Intrinsic dimensionality as a metric for temporal plant diversity evaluation : case study from the SHIFT campaign

Abstract

Current biodiversity metrics derived from remote sensing data are typically applied to small local areas, require significant training data, and are not easily extensible globally. Here we propose the mathematical concept of intrinsic dimensionality (ID) as a method to quantify terrestrial vegetation variability without a need for in situ training data. We apply this technique to airborne imaging spectroscopy data from the Surface Biology and Geology High Frequency Time series (SHIFT) airborne campaign, with weekly overflights from February to May 2022 over a region in California stretching from Figueroa Mountain in the Los Padres National Forest to Point Conception and adjacent coastal areas. ID is considered in both spatial and temporal context—spatial ID represents spectral variability across a geographical region at a single time step, and temporal ID represents spectral variability over time for a single geographical location. Results show an encouraging and significant correlation between spatially calculated ID and in situ vegetation species richness data despite different spatial scales between the two (p = 0.01). Spatial ID remained largely unchanged at each time step over the course of three months during the spring green-up period when vegetation characteristics and spectral responses were changing rapidly (number of species remains unchanged even though spectra reflect phenological change over time). The temporal ID remained constant for pseudo-invariant surfaces such as parking lots, roofs, and rock, but showed increased ID with time for trees, shrubs, and grasses. This robustness of spatial ID to seasonal change is desirable in any measure of species richness because it is insulated from changes in vegetation condition that are unrelated to plant species richness. Even though the spatial ID is consistent across acquisition dates, when considering the full time series (temporal ID), we find that subweekly sampling may be necessary to spectrally capture the full phenological cycle of certain vegetation types.