Spectral replacement using machine learning methods for continuous mapping of the Geostationary Environment Monitoring Spectrometer (GEMS)

Abstract. Earth radiances in the form of hyperspectral measurements contain useful information on atmospheric constituents and aerosol properties. The Geostationary Environment Monitoring Spectrometer (GEMS) is an environmental sensor measuring such data ultraviolet visible spectral range over Asia–Pacific region. After completion in-orbit test GEMS October 2020, bad pixels are found as one remaining calibration issues resulting obvious spatial gaps measured well retrieved To solve fundamental cause issue, this study takes approach reproducing defective spectra with machine learning models using artificial neural network (ANN) multivariate linear regression (Linear). Here trained defect-free after dimensionality reduction principal component analysis (PCA). Results show that PCA-Linear model has small reproduction errors for a narrower gap less vulnerable to outliers error 0.5 %–5 %. On other hand, PCA-ANN shows better results emulating strong non-linear relations about 5 % except shorter wavelengths around 300 nm. It demonstrated dominant patterns can be successfully reproduced within level radiometric accuracy GEMS, but limitation remains when it comes finer features. When applying retrieval processes cloud ozone, centroid pressure 1 %, while total ozone column density relatively higher variance. As initial step pixels, current provides potential limitations methods improve from geostationary orbit.