Extraction of the vegetation fraction based on a stepwise spectral mixture analysis for the central and eastern area of source region of Yangtze, Yellow and Lantsang Rivers

Abstract

Vegetation cover is an important parameter used in monitoring ecological changes of the source region of Yangtze, Yellow and Lantsang Rivers and understanding human activities. Thus, how to extract the large area's vegetation fraction quickly effectively is an open question. The traditional linear spectral mixture analysis (LSMA) assumes that the spectral reflectance is a mixture of several fixed endmember spectral values, which ignores considerable within-class variability. However, multiple endmember spectral mixture analysis (MESMA) overcomes the disadvantage by allowing the number and types to vary on a per-pixel basis. This paper proposes a stepwise spectral mixture analysis (SSMA) containing two steps of MESMA and adding the endmember fraction rationality rule in each step. The aim of the first step is to detect the pixels that didn't contain vegetation information at all and these pixels would be masked out. In the second step, MESMA is used to unmix the pixels only reserved in previous process. The results show that SSMA is more accurate than LSMA in extracting the vegetation fraction for the Three-Rivers. This means that SSMA is a good substitute for LSMA in studies on ecological changes. The concept of SSMA also can be applied for other large study areas.