A Modified Relative Spectral Mixture Analysis to Extract the Fractions of Major Land Cover Components

Abstract:

As an effective method of extracting land cover fractions based on spectral endmembers, spectral mixture analysis (SMA) has been applied using remotely sensed imagery in different spatial, temporal, and spectral resolutions. A number of studies focused on arid/semiarid ecosystem have used SMA to obtain the land cover fractions of GV, NPV/litter, and bare soil (BS) using MODIS reflectance products to understand ecosystem phenology, track vegetation dynamics, and evaluate the impact of major disturbances. However, several challenges remain in the application of SMA in studying ecosystem phenology, including obtaining high quality endmembers and increasing computational efficiency when considering to long time series that cover a broad spatial extent.

Okin (2007) proposes a variation of SMA, named as relative spectra mixture analysis (RSMA) to address the latter challenge by calculating the relative change of fraction of GV, NPV/litter, and BS compared with a baseline date. This approach assumes that the baseline image contains the spectral information of the bare soil that can be used as an endmember for spectral mixture analysis though it is mixed with the spectral reflectance of other non-soil land cover types. Using the baseline image, one can obtain the change of fractions of GV, NPV/litter, BS, and snow compared with the baseline image. However, RSMA results depend on the selection of baseline date and the fractional components during this date. In this study, we modified the strategy of implementing RSMA by introducing a step of obtaining a soil map as the baseline image using multiple-endmember SMA (MESMA) before applying RSMA. The fractions of land cover components from this modified RSMA are also validated using the field observations from two study area in semiarid savanna and grassland of Queensland, Australia.