A Model-Based ELM for Atmospheric Correction over Case 2 Water with Landsat 8

摘要

The Landsat 8 satellite, recently launched (February 2013), carries the next generation of Landsat sensors and extends over 40 years of continuous imaging acquisition. Landsat 8, with its improved spectral coverage and radiometric resolution, has the potential to dramatically improve our ability to simultaneously retrieve the three primary Color Producing Agents (CPAs) (chlorophyll, colored dissolved organic matter, and suspended minerals) from water bodies and considering its 30-meter resolution should be especially useful for studying the nearshore environment. In the Case 2 water problem, accurate atmospheric correction is essential, yet remains a significant source of water-constituent retrieval error particularly since the sensor-reaching signal, due to water, is very small compared to the signal from atmospheric effects. Furthermore, the standard black target assumption commonly used for open ocean studies is not valid in turbid water due to the presence of water-leaving signal in the near infrared (NIR). In this work, a modified version of the traditional empirical line method (ELM) has been developed, which utilizes reflectance from both an in-water radiative transfer model (Hydrolight) and a reflectance product (Landsat surface reflectance product) to atmospherically correct Landsat 8 images. This method employs pseudo-invariant feature (PIF) pixel extraction to mask urban landscape from the reflectance product for the bright pixel determination. For the dark pixel, Hydrolight is used to obtain the field spectra that replaces ground-truth measurements normally used in the traditional ELM. The radiance values for the dark and bright pixels are extracted from the corresponding regions in the Landsat 8 image. Initial results of this method are compared to results obtained from a traditional ELM for validation.