Impacts of dust aerosol and adjacency effects on the accuracy of Landsat 8 and RapidEye surface reflectances
The atmospheric correction of satellite data is challenging over desert agricultural systems, due to the relatively high aerosol optical thicknesses (τ550), bright soils, and a heterogeneous surface reflectance field. Indeed, the contribution of reflected radiation from adjacent pixels scattered int...
Published in: | Remote Sensing of Environment |
---|---|
Main Authors: | , |
Other Authors: | , , |
Format: | Article in Journal/Newspaper |
Language: | unknown |
Published: |
Elsevier BV
2017
|
Subjects: | |
Online Access: | http://hdl.handle.net/10754/623849 https://doi.org/10.1016/j.rse.2017.03.013 |
Summary: | The atmospheric correction of satellite data is challenging over desert agricultural systems, due to the relatively high aerosol optical thicknesses (τ550), bright soils, and a heterogeneous surface reflectance field. Indeed, the contribution of reflected radiation from adjacent pixels scattered into the field of view of a target pixel is considerable and can significantly affect the fidelity of retrieved reflectances. In this study, uncertainties and quantitative errors associated with the atmospheric correction of multi-spectral Landsat 8 and RapidEye data were characterized over a desert agricultural landscape in Saudi Arabia. Surface reflectances were retrieved using an implementation of the 6SV atmospheric correction code, and validated against field collected spectroradiometer measurements over desert, cultivated soil, and vegetated surface targets. A combination of satellite and Aerosol Robotic Network (AERONET) data were used to parameterize aerosol properties and atmospheric state parameters. With optimal specification of τ550 and aerosol optical properties and correction for adjacency effects, the relative Mean Absolute Deviation (MAD) for all bands combined was 5.4% for RapidEye and 6.8% for Landsat 8. However uncertainties associated with satellite-based τ550 retrievals were shown to introduce significant error into the reflectance estimates. With respect to deriving common vegetation indices from corrected reflectance data, the Normalized Difference Vegetation Index (NDVI) was associated with the smallest errors (3–8% MAD). Surface reflectance errors were highest for bands in the visible part of the spectrum, particularly the blue band (5–16%), while there was more consistency within the red-edge (~ 5%) and near-infrared (5–7%). Results were generally better constrained when a τ550-dependent aerosol model for desert dust particles, parameterized on the basis of nearby AERONET site data, was used in place of a generic rural or background desert model. This adaptation was particularly pertinent for ... |
---|