Aerosol mapping over land with imaging spectroscopy using spectral autocorrelation

S. Bojinski, D. Schlapfer, M.E. Schaepman, J. Keller, K.I. Itten

Research output: Contribution to journalArticleAcademicpeer-review

8 Citations (Scopus)


A new method for aerosol retrieval over land is proposed that makes explicit use of the contiguous, high-resolution spectral coverage of imaging spectrometers. The method is labelled Aerosol Retrieval by Interrelated Abundances (ARIA) and is based on unmixing of the short-wave infrared sensor signal by region-specific endmembers, assuming low aerosol radiative influence in this spectral region. Derived endmember abundances are transferred to the visible part of the spectrum in order to approximate surface reflectance where aerosol influence is generally strongest. Spectral autocorrelation of surface spectra is a precondition for ARIA and demonstrated using a reference spectrum database. The re-mixed surface reflectance is used as input quantity for the inversion of aerosol optical depth tau(a) at 0.55 mum wavelength on a pixel basis. Except for the choice of endmembers and the atmospheric vertical profile, no a priori assumptions on the image scene are required. The potential of the presented method for aerosol retrieval is demonstrated for an Airborne Visible/ Infrared Imaging Spectrometer (AVIRIS) scene, collected in California in 2000. Comparisons with existing aerosol retrieval methods showed encouraging results in terms of achieved spatial smoothness and degree of uncertainty of aerosol optical depth across the scene.
Original languageEnglish
Pages (from-to)5025-5047
JournalInternational Journal of Remote Sensing
Issue number22
Publication statusPublished - 2004


  • atmospheric correction algorithm
  • satellite imagery
  • optical-thickness
  • vegetation index
  • ocean color
  • eos-modis
  • retrieval
  • aviris
  • space


Dive into the research topics of 'Aerosol mapping over land with imaging spectroscopy using spectral autocorrelation'. Together they form a unique fingerprint.

Cite this