Atmospheric correction for a Landsat and Sentinel-2 product over water surfaces

Abstract

In the last decade, satellite derived standard land products have increasingly been produced for medium resolution satellites such as Landsat and (more recently) Sentinel-2. These mostly involve estimating surface reflectance and surface temperature. The products generally remove or standardise atmospheric effects with some normalizing for surface bidirectional reflectance distribution function (BRDF) and terrain illumination effects to provide consistent time series and mosaics. The products have been used in various land surface applications, for instance, land cover, fractional cover, water identification, flooding mapping, crop monitoring and other time series analysis. However, the products are generally not immediately sufficient for applications over persistent water areas, such as estimating water quality, benthic cover, sediment transport, erosion and shallow water bathymetry. These need additional corrections with different physics that are not included in standard land products. In this paper, a method is proposed that treats persistent water areas separately within the standard product and includes corrections not generally applied to the land. The processing has been designed to be fully consistent between water and land in atmospheric correction and to provide comparable definition of reflectance factors so that they can be combined in the same time series and form mosaics. The first step in this process is the use of an effective and up-to-date classification to separate the persistent water and land. The water areas are then atmospherically corrected in the same way as the land without applying BRDF or shading effects. For the water areas, adjacency effects are more significant near the water-land interfaces due to the large contrast between land and water. Water surface effects have also different physics from land surfaces. Therefore, the extra corrections currently include correction for adjacency effects, regional sun glint and sky radiation effects. The water mask and these corrections have been added to the current existing atmospheric, BRDF and terrain corrected surface reflectance product (standard product) from Geoscience Australia (GA). However, at the scale of the Landsat and higher resolution satellite images, residual local surface and bidirectional effects still occur and are discussed in this paper. Results from the new processing strategy have been compared with GA standard products in test images of Canberra and the North Queensland coast near Ingham and used as a basis to discuss the likely residuals of surface and atmospheric effects and options for the inclusion of methods to overcome them in a standard product. The results show that: • Both inland and sea water signatures behave as expected from other data and models. • Adjacency correction seems most useful where a water-Land interface is close to the water body. • Sky glint removal is sometimes too much in the Canberra site when water is shielded by local terrain. • Sun and sky glint correction greatly improves the coast and deep sea water signatures.