Evaluation of atmospheric correction methods for low to high resolutions satellite remote sensing data

Abstract

The objective of atmospheric correction is to retrieve surface reflectance from the top of atmosphere (TOA) reflectance. However, estimating surface reflectance from the TOA reflectance satellite data requires knowledge about the state of the atmosphere (e.g., water vapor and ozone) and the contribution of aerosols to the atmospheric path radiance. Thus, obtaining precise measurements of these parameters, which is very difficult, is crucial for accurate estimation of surface reflectance. The SREM (Simplified and Robust Surface Reflectance Estimation Method) is a physical-based atmospheric correction method based on the Radiative transfer (RT) equations of the second simulation of the Satellite Signal in Solar Spectrum (6SV). Essentially the SREM is a simplified version of 6SV which does not require Aerosol Optical Depth (AOD), aerosol type, water vapor, and ozone. An initial study showed accuracy comparable to the Landsat operational Surface Reflectance Products (SRProd) which is generated through different RT models using AOD, water vapor, and ozone data. To further validate the SREM under varying atmospheric conditions and at different spatial resolutions, an independent Reference Surface Reflectance (SRRef) dataset was generated using the AERONET (Aerosol Robotic Network) measurements as input to the 6SV RT model. The surface reflectances estimated by SREM (SRSREM) and SRProd from Planet Scope (PS, at 3 m spatial resolution), Sentinel-2 AB (S2AB) Multi-spectral Instrument (MSI, at 10 to 60 m spatial resolution), and Landsat-8 (L8) operational Land Imager (OLI, at 30 m spatial resolution) were validated against SRRef. Results showed that SRSREM performed similar to the SRProd of PS, S2AB MSI, and L8 OLI against SRRef. An inferior performance (R of 0.35 and 0.57) of L8 OLI's SRProd in the coastal blue (SB1) and blue (SB2) bands was observed, compared to SREM. The comparison of SRSREM with SRProd reveals the robustness of SREM, without using AOD, water vapor, and ozone data, for estimation of surface reflectance for all RT models tested. For some dates, SRRef and the SRProd under-corrected and produced higher values than the TOA reflectance, even when the atmosphere was clear but this was not the case for SREM. Analysis of surface reflectance estimation in shadowed areas revealed that the SRRef and SRProd had mainly negative values in coastal blue and blue bands for L8 OLI, while no negative SR value was observed for SREM in any band. These results recommend the utilization of SREM for the provision of surface reflectance products across a range of sensors