Analysis of Spectral Simulation Using Different Spectral Libraries

Spectral simulation technology is used to simulate new spectral bands from available observed bands. In this study, we aim to evaluate the performance of a spectral simulation using traditional four-band (i.e., blue, green, red and near-infrared) multispectral satellite images to new spectral bands such as the coastal and short-wave infrared (SWIR) bands. In the proposed scheme, nonlinear support vector regression (SVR) is performed to simulate different reflectances from the spectral libraries, and the accuracy of spectral simulation is discussed. Two different spectral libraries (i.e., ECOSTRESS and USGS splib07a) are used to establish the SVR model. The experimental results indicate that the spectral library can be used to simulate the reflectances of satellite images, and the performance of ECOSTRESS is marginally superior to that of USGS in the spectral simulation between Formosat-2 (FS-2) multispectral images and Landsat-8 (LS-8) coastal and SWIR band images.