Decomposition of Landsat 8 OLI Images by Simplified Spectral Patterns for Land Cover Mapping

Abstract

Decomposition of multispectral image data has recently become the focus of remote sensing scientists because it improves efficiency of land cover mapping by data reduction. The three main traditional techniques used for decomposition are principal component analysis, linear spectral mixing, and pattern decomposition. This paper presents a new method of pattern decomposition using simplified spectral patterns. Spectral characteristics of land cover are analyzed using the shape of their spectral reflectance curves, leading to decomposition of an image into component images with pixels that are characterized by spectral reflectance curves of the same shape. The author proposes a method for converting a spectral reflectance curve from an analog to a simplified digital form consisted of 15 digits of zeros, ones, and twos. This form of the spectral reflectance curve is called simplified spectral pattern (SSP). One hundred Landsat 8 OLI scenes of 14 global biomes were collected for the research. The study found that 34 dominant simplified spectral patterns represent land cover types in selected scenes, covering 96.53% of the total 4,150,671,513 pixels. SSP correlates with single land cover types such as water, ice, and snow or with multiple land cover types such as vegetation, minerals, rock, and soil.