Development and validation of the Landsat-8 surface reflectance products using a MODIS-based per-pixel atmospheric correction method

Landsat satellites have the longest history of making global-scale Earth observations at medium spatial resolution of any series of satellites and have been widely used in various remote-sensing fields. However, many remote-sensing applications, including large-area or long-term land-cover monitoring, need Landsat reflectance data that have had accurate atmospheric correction carried out. In this research, a MODIS-based per-pixel atmospheric correction procedure was developed and employed to produce the surface reflectance (SR) product. A total of 510 Landsat-8 Operational Land Imager (OLI) scenes covering the whole of China in 2013 were collected and processed. The mean relative differences between the surface and top-of-atmosphere (TOA) reflectance for China, composited and expressed as percentages, were found to be 67, 47, 18, 13, 4, 4, and 7% for Landsat-8 OLI bands 1, 2, 3, 4, 5, 6, and 7, respectively. Then, the accuracy of MODIS atmospheric products was validated using ground-based sun-radiometer observation network data, including Sun/sky-radiometer Observation Network (SONET) and Aerosol Robotic Network (AERONET) data collected from 14 SONET/AERONET stations. The validation results showed that the MODIS atmospheric products are reliable for China, with an R-2 value of 0.78 and a root mean square error (RMSE) value of 0.12 for aerosols, and an R-2 value of 0.98 and an RMSE of 0.25 for water vapour. Third, the SR product using our per-pixel atmospheric correction method was evaluated by comparison with the MODIS daily surface reflectance product (MOD09GA) and the United States Geological Survey (USGS) provisional Landsat-8 SR product, with a mean R-2 of 0.93 and an RMSE of 0.02 for MOD09GA; and with a mean R-2 of 0.97 and an RMSE of 0.01 for the USGS SR product. Finally, the advantage of our per-pixel atmospheric correction method over the perscene method was investigated by analysis of the spatial variation of the atmospheric parameters within one Landsat scene (about 1.5 degrees x 1.5 degrees), with a mean standard deviation value of 0.03-0.09 for aerosol. When such aerosol variation was omitted as the per-scene atmospheric correction method, the SR absolute error due to aerosol optical thickness (AOT) spatial variation was about 0.027, 0.018, 0.005, 0.003, 0.002, 0.0007, and 0.003 for the seven reflectance bands of Landsat-8. Therefore, use of Landsat-8 SR products over China with our per-pixel atmospheric correction was proved reliable, and more promising than the per-scene method, especially for the short-wavelength bands.