Multi-temporal analysis of community scale vegetation stress with imaging spectroscopy

Abstract

Research focuses on the development of simple techniques to identify and model multi-temporal red-edge geometry in disparate vegetation communities at the Jasper Ridge Biological Preserve, Palo Alto, California. This study examines the performance of selected measures designed to monitor components of vegetation stress from early spring to late autumn. Defining complex red-edge symmetry through a range of geometric and statistical measures is also examined for application to seasonal and long-term vegetation change monitoring. The vegetation red-edge, centred at the largest change in reflectance per wavelength change, is located between two of the most widely used wavelength regions used for broad band vegetation studies, the red trough and the NIR plateau. Statistical analyses show that more than 90% of the spectral information of a plant canopy are contained in red and near-infrared bands (Baret et al., 1988a, 1990; Sheffield, 1985). The red-edge therefore provides the focus for vegetation stress analyses. The vegetation red-edge for Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data comprises eight contiguous spectral bands between 0.6852 - O.7523μm. This research identifies hyperspectral indices used to accurately measure, monitor, and model terrestrial ecosystems at the community scale. A selection of five recently described biochemical, biophysical, and geometric indices are used as important indicators of vegetation stress at the community scale. The five vegetation indices measuring disparate components of vegetation stress are: NDVI, Nitrogen, Lignin, MSI, and PRI. Pearson's product-moment correlation analysis is used to examine relationships between vegetation indices as community gradients. Community gradients are constructed from ranked multi-temporal R-values to reduce the complexity of correlation variables, maintain inter-community relationships, and reduce computational requirements. Furthermore, FR-value calculations provide a measure of relative position along multi-temporal gradients for each community. Identification of communities with comparable multi-temporal index value trends as community clusters is an effective technique for identifying communities responding to broad changes in phenology, physiology, and environmental influences. Analyses based on multi-temporal index values can also be used to identify important boundaries between community groups along environmental gradients. Defining complex red-edge symmetry through a range of geometric and statistical measures is identified as having great potential for monitoring seasonal and long-term vegetation change. A selection of statistical and spectral transformations are used to examine red-edge reflectance to identify simple techniques that can be applied to investigations of vegetation stress. Both mean and median reflectance values are best used to describe the central tendency of multi-temporal community trends. Reflectance range appears more suitable for monitoring tree communities rather than those containing predominantly annual/perennial species. Kurtosis provides a suitable measure to extrapolate late season trends in tree communities towards the subsequent winter and early spring where no data exist. The Pearson's Coefficient of Skewness provides one of the best statistical methods for interpreting multi-temporal vegetation phenology. As most vegetation exhibits near-linear reflectance change between the red trough and the NIR plateau, second-derivative transformations (t") provide an effective technique to highlight subtle absorption/reflection features and multi-temporal trends in red-edge spectra. s" values are essential to the identification of upper red-edge concavity that cannot be readily identified in reflectance spectra. The upper red-edge displays the c"max waveform propagating to shorter wavelengths with increasing stress. Continuum removal provides an effective technique for normalising reflectance spectra to allow comparison of individual absorption features from a common baseline. Absorptions in continuum removed spectra are used to identify differences between communities related to vegetation phenology. The design of red-edge measures to highlight disparate components of vegetation stress is directed towards providing additional information to support interpretations of complex community and index relationships. Large positive values at the second-derivative 0.7330m band (m"733) indicate low stress curves, whereas large negative values indicate high stress curves. The 0.733vm band provides a particularly sensitive and simple indicator of the presence or absence of upper red-edge concavity in second-derivative spectra. Positive values indicate non-stress asymptote curves, whilst negative values indicate a stress response characterised by upper red-edge concavity. Positive NSI values are calculated from red-edge spectra with steep upper red-edge slopes relative to lower slopes, and negative NSI values from low angle upper red-edge slopes relative to lower slopes. High angle upper red-edge slopes are associated with low stress vegetation spectra, whereas low angle red-edge slopes are associated with high stress vegetation spectra. Identifying red-edge spectral shifts through derivative analysis permits the comparison of seasonal intra- and inter-community change. The identification of i enables the tracking of wavelength displacements as spectral shifts independent of reflectance. Multi-temporal blue-shifts in Mi are associated with increasing environmental stress, whilst red-shifts indicate a reduction in environmental stress. The RVSI was developed in this study to identify inter- and intra-community stress trends based on spectral changes in upper red-edge geometry. Multi-temporal analyses show that RVSI responds to a component of vegetation stress independent of all other hyperspectral indices. Positive RVSI values are returned from concave upper red-edge spectra, and indicate increased relative stress at the community scale. Asymptote and near-linear red-edge curves are generally fully developed by late spring, whilst concave upper red-edge spectra occur during summer and early autumn. Upper red-edge concavity is correlated to seasonal changes in precipitation and temperature associated with periods of increasing environmental stress. The RVSI performed satisfactorily as an indicator of changing phenology and environment-induced vegetation stress. Hysteresis analysis is developed and trialed as a promising new technique for modeling multi-temporal vegetation stress from high spectral resolution datasets. Patterns of community hysteresis based on multi-temporal changes in: reflectance, spectral shifts, the RVSI, and other important vegetation indices are examined. Hysteresis plots identify the most appropriate index to use at specific times of the year for each community. hysteresis analysis highlight subtle differences between indices, and both intra- and inter-community relationships. Hysteresis trajectories can also be used effectively to estimate the timing of seasonal events that, influence red-edge geometry. Defining the general direction of community hysteresis trajectories highlight important differences between multi-temporal stress dynamics for related indices. The hysteresis component of this research attempts to provide a new technique to model change between selected spectral measures, and to improve interpretation and presentation of complex multi-temporal trends relating to phenology at the community scale. Key Words: AVIRIS, community scale, hyperspectral remote sensing, spectral hysteresis, imaging spectroscopy, Jasper Ridge Biological Preserve, multi-temporal, red-edge, vegetation indices, vegetation stress.