A multi-temporal spectral library approach for mapping vegetation species across spatial and temporal phenological gradients

Variability in spectral reflectance due to spatial and temporal gradients in vegetation phenology presents issues for accurate vegetation classification. Phenological variability through space and over time can result in misclassification when spectra from non-representative areas or times are used as training data. Vegetation classification at the species level could benefit from introducing phenological information to spectral libraries, but utilization of this information across multiple dates of imagery will require new approaches to building spectral libraries and to classification. This paper explores an automated method for selecting a single multi-temporal spectral library that can be used to classify vegetation species across multiple dates within an image time series. Iterative Endmember Selection (IES) was used to select spectra from Airborne Visible Infrared Imaging Spectrometer (AVIRIS) data acquired on five dates in the same year. IES selected spectra to maximize species classification accuracy (as measured by Kappa) within a multi-temporal spectral library that included spectra from all image dates. The resulting multi-temporal endmember library was applied using Multiple Endmember Spectral Mixture Analysis (MESMA) to classify vegetation species and land cover across all five images. Results indicate that multi-temporal, seasonally-mixed spectral libraries achieved similar overall classification accuracy compared to single-date libraries, and in some cases, resulted in improved classification accuracy. Several species had increased Producer's or User's accuracy using a multi-temporal library, while others had reduced accuracy compared to same-date classifications. The image dates of endmembers used to map species in each image were examined to determine if this information could improve our understanding of phenological spectral differences for specific species. Multi-temporal endmember libraries could provide a means for mapping species in data where phenology, climatic variability, or spatial gradients are not known in advance or may not be easily accounted for by endmembers from a single date. New missions, such as the proposed Hysperspectral Infrared Imager (HyspIRI) mission, will provide greatly improved access to multi-temporal spectral datasets and new opportunities for mapping vegetation spectral variability on regional-to-global scales. (C) 2015 Elsevier Inc. All rights reserved.