Improving Atmospheric Correction for Visible/Short Wave Infrared (VSWIR) Imaging Spectrometers with Iterative Fitting of Absorption By Three Phases of Water

Abstract:

Airborne imaging spectrometers like the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) offer valuable insight into the Earth’s terrestrial and ocean ecosystems, mineralogy, and land use. Estimating surface reflectance requires accounting for atmospheric absorption, which is sensitive to the local abundance of water vapor. Analysts typically estimate water vapor concentrations using the depths of absorption features, which can be inaccurate by up to 50% over surface features containing liquid water or ice. This can bias the retrieved water vapor maps and create atmospheric artifacts in reflectance spectra. A new retrieval method offers significant accuracy improvements over plant canopies or ice by estimating the path lengths of all three phases of water simultaneously, adjusting absorptions to best fit the measurement over a broader spectral interval. This paper assesses the remaining sources of error for the three-phase retrieval technique. We analyze retrievals for synthetic data when the 940 and 1140 nm wavelength features are fitted, for initial vapor path estimates ranging from 0 to ±50% accuracy. These tests indicate that most error comes from inaccuracy in the initial path estimate used to obtain vapor absorption coefficients. We evaluate a modified algorithm that uses multiple iterations to refine this estimate. Error is found to approach a constant value, demonstrating improved robustness to initialization conditions. We also assess the new iterative method using corrected AVIRIS data over various environments. The iterative method yields significantly better water vapor maps, reducing spurious correlations between vegetation canopy water and vapor estimates. The new iterative method offers accuracy improvements over traditional Visible/Short Wave Infrared (VSWIR) atmospheric correction methods, at modest computational cost.