The effects of vegetation phenology on endmember selection and species mapping in southern California chaparral

Image classification typically utilizes single-date imagery and does not take into account seasonal variation in the spectral characteristics and separability of image spectra. While global vegetation classifications have relied on seasonal changes in multitemporal data, seasonal vegetation dynamics have seldom been explored at higher spatial and spectral resolutions. In particular, investigations of vegetation phenology in the hyperspectral domain have been limited. This paper uses endmember average root mean square error (EAR), a method for selecting endmembers for multiple endmember spectral mixture analysis (MESMA), to explore temporal changes in the spectral characteristics of the selected endmembers. The images modeled by these endmembers demonstrate temporal changes in the confusion between vegetation species in southern California chaparral. Endmembers were selected from five Airborne Visible Infrared Imaging Spectrometer (AVIRIS) images of the area surrounding Santa Barbara, CA, USA. The selected endmembers demonstrated spectral changes that were consistent with an increase in nonphotosynthetic vegetation (NPV) as soil water balance decreased. Polygon level modeling accuracies for soil water surplus images ranged between 59% and 90%, while accuracies for soil water deficit images ranged between 52% and 81%. Variation in NPV content in the water deficit images resulted in increased confusion between chaparral species. Increasing the number of endmembers used to model each land cover class significantly increased accuracy in the water deficit images. Seasonal variations in the spectral response of chaparral are important for determining the separability of chaparral species.