A sensor-invariant atmospheric correction method: application to Sentinel-2/MSI and Landsat 8/OLI

This is a Preprint and has not been peer reviewed. This is version 3 of this Preprint.


Download Preprint


Feng Yin, Philip E Lewis, Jose Luis Gomez-Dans , Qingling Wu


Mitigating the impact of atmospheric effects on optical data is a critical for monitoringland processes. Consistent approaches to different sensors, which also quantify uncertainty, are required to combine surface reflectance observations from heterogeneous sensors. This paper provides a sensor agnostic approach to atmospheric correction, called SIAC. It exploits operational global datasets on (i) coarse resolution spectral surface bi-directional reflectance distribution function (BRDF) and (ii) coarse resolution atmospheric composition. The method infers aerosol optical thickness (AOT) and total columnar water vapour (TCWV) from top of atmosphere (TOA) reflectance observations, using a Bayesian framework that exploits the MODIS MCD43 BRDF descriptor product and the Copernicus Atmosphere Monitoring Service (CAMS) operational forecasts of AOT and TCWV to provide an \emph{a priori} estimate. Spatial smoothness constraints are assumed for AOT and TCWV, and efficient statistical approximations (so-called emulators) to atmospheric radiative transfer (RT) codes are used to efficiently invert the parameters. BRDF descriptors are used to provide an estimation of surface directional reflectance (SDR) at a coarse resolution, and linear spectral mappings to convert to the target sensor spectral configuration. The method is demonstrated on Sentinel 2 and Landsat 8 data. AOT retrieval for both S2 and L8 shows a very high correlation to AERONET estimates ($r^2 > 0.9, \, RMSE < 0.025$ for both sensors), although with a small underestimate of AOT. TCWV is accurately retrieved from both sensors $(r^2>0.95,\, RMSE < 0.02)$. Comparisons with \emph{in situ} surface reflectance measurements from the RadCalNet network show that the proposed method provides accurate estimates of surface reflectance across the entire spectrum, with $RMSE$ mismatches with the reference data between 0.005 and 0.02 in units of reflectance, both for Sentinel 2 and Landsat 8. For near-simultaneous Sentinel-2 and Landsat-8 acquisitions, there is a very tight relationship ($r^2>0.95$ for all common bands) between surface reflectance acquired from both sensors, with no negligible biases.




Earth Sciences, Environmental Monitoring, Environmental Sciences, Physical Sciences and Mathematics, Planetary Sciences


Atmospheric Correction, Data Fusion, aerosol optical thickness, analysis ready data, landsat-8, sentinel-2, total column water vapour


Published: 2019-02-21 11:18

Older Versions

CC BY Attribution 4.0 International

Add a Comment

You must log in to post a comment.


There are no comments or no comments have been made public for this article.