Atmospheric correction of metre-scale optical satellite data for inland and coastal water applications
Vanhellemont, Q.; Ruddick, K. (2018). Atmospheric correction of metre-scale optical satellite data for inland and coastal water applications. Remote Sens. Environ. 216: 586-597. https://dx.doi.org/10.1016/j.rse.2018.07.015
In: Remote Sensing of Environment. Elsevier: New York,. ISSN 0034-4257; e-ISSN 1879-0704, meer
|   |
| Author keywords |
Remote sensing; Metre-scale resolution; Water colour; Water quality;Aerosol correction; Automation |
| Abstract |
A new atmospheric correction (AC) method for aquatic application of metre-scale resolution (MR) optical satellite imagery is presented in this article, and demonstrated using images from the Pleiades constellation. MR satellites are typically operated privately and imagery can be costly. However in recent years, the price of individual acquisitions has dropped and their revisit times have improved, making them promising tools for remote sensing of inland and coastal waters. Due to the spatial resolution requirements of these satellites, the bands on the sensors are relatively wide (60-140 nm on Pleiades) in order to achieve an acceptable signal to noise ratio. This bandwidth and the limited number of bands can pose problems for the AC as the water signal may not be negligible in any band, especially over turbid waters. Since the MR sensors have a relatively narrow swath (20 km for Pleiades) the atmosphere can generally be assumed to be homogeneous over a scene or subscene. This assumption allows the atmospheric path reflectance (rho(path)) to be estimated from multiple targets in the scene, which are selected according to the lowest observed top-of-atmosphere reflectances (rho(TO)(A)) in all bands. Rather than using pre-defined "dark" bands (e.g. in the NIR and SWIR) such as is common in other water-focused AC methods, the best band is selected automatically, i.e. the one yielding the lowest rho(path). This criterion avoids unrealistic negative ("overcorrected") reflectances after the AC. Furthermore, for inland waters the NIR bands are usually affected by scattering from adjacent land and vegetation pixels, resulting in unrealistic rho(path) when used in the AC. The spatial resolution of the sensors is used as an advantage here, since ground-level object shadows (e.g. from trees and buildings) can be spatially resolved and are usually the pixels selected by the automated procedure for the determination of rho(path). In fact, it is proposed that using these shadow pixels gives better performance than using any kind of water pixel for these broad-band MR sensors. The method is demonstrated using several Pleiades images, showing good performance in retrieval of the aerosol optical thickness (tau(a)) for an urban (Brussels) and a coastal (Zeebrugge) site. Match-ups with water reflectances measured at the Zeebrugge AERONET-OC station show promising performance, although there is a significant spectral mismatch between the bands on the satellites and the CIMEL radiometer. Pleiades imagery of Zeebrugge reveals a turbid wake associated with the MOW1 measurement station, which opens perspectives of using MR satellites for the characterisation of monitoring and validation sites. Future work includes the application to other MR satellites (e.g. WorldView) and the evaluation for mass processing of open access high resolution (10-60 m) satellite data from Landsat and Sentinel-2. |
|
