A Framework of Generating 10-m Spectral Surface Albedo Products from Sentinel-2 and MODIS data

Abstract: Surface albedo is a fundamental radiative parameter as it controls the Earth’s surface energy budget and directly affects the Earth’s climate. A new method is proposed of generating 10-m high-resolution spectral surface albedo from Sentinel-2 L1C top-of-atmosphere (TOA) reflectance and MODIS bi-directional reflectance distribution function (BRDF) data. This high-resolution spectral surface albedo generation system will be described and consists of 5 parts: 1) retrieval of Sentinel-2 spectral surface reflectance using the Sensor Invariant Atmospheric Correction (SIAC) algorithm; 2) generation of Sentinel-2 cloud mask using machine learning; 3) extraction of pure pixels and their corresponding abundance values from 20-m Sentinel-2 data using an Endmember Extraction Algorithm; 4) inversion of high-resolution albedo from MODIS_albedo/Sentinel2_BRF ratio matrix; and 5) downscaling retrieved 20-m spectral and broadband albedo to 10-m. The SIAC algorithm is developed by [1], and has demonstrated to vastly improve the accuracy of Sentinel-2 atmospheric correction when compared against the use of in situ AERONET data. The machine learning cloud detection approach CloudFCN [2] is based on a Fully Convolutional Network architecture, and has become a standard Deep Learning approach to image segmentation. The CloudFCN exhibits state-of-the-art performance in picking up cloud pixels which is comparable to other methods in terms of performance, high speed, and robustness to many different terrains and sensor types. The endmember extraction uses N-FINDR along with Automatic Target Generation Process to identify the pure pixels from Sentinel-2 spectral data. The extracted pure pixels are used to relate the albedo-to-reflectance matrix with the abundance values of different pure pixels. The high-resolution albedo values are finally retrieved by solving this over-parameterised matrix. This framework also produces a MODIS BRDF prior based on 20-years of MCD43A1 and VNP43A1 daily BRDF data. This BRDF prior is produced on a daily basis, and will be used to temporally interpolate the high-resolution albedo values over pixels that are covered by clouds. The produced high-resolution albedo data will be validated over different tower sites where long-time series of in situ albedo products have been produced [3].

Keywords: high-resolution, surface albedo, Sentinel-2, SIAC, machine learning, endmember

[1] Yin, F.; Lewis, P.E.; Gomez-Dans, J.; Wu, Q. A sensor-invariant atmospheric correction method: Application to Sentinel-2/MSI and Landsat 8/OLI. EarthArXiv, 21 Feb. 2019 web, doi:10.31223/osf.io/ps957.

[2] Francis, A.; Sidiropoulos, P.; Muller, J.-P. CloudFCN: Accurate and Robust Cloud Detection for Satellite Imagery with Deep Learning. Remote Sens. 2019, 11, 2312. https://doi.org/10.3390/rs11192312.

[3] Song, R.; Muller, J.-P.; Kharbouche, S.; Yin, F.; Woodgate, W.; Kitchen, M.; Roland, M.; Arriga, N.; Meyer, W.; Koerber, G.; Bonal, D.; Burban, B.; Knohl, A.; Siebicke, L.; Buysse, P.; Loubet, B.; Leonardo, M.; Lerebourg, C.; Gobron, N. Validation of Space-Based Albedo Products from Upscaled Tower-Based Measurements Over Heterogeneous and Homogeneous Landscapes. Remote Sens. 2020, 12, 833. https://doi.org/10.3390/rs12050833.