[Reconstruction of Water Hyperspectral Remote Sensing Reflectance Based on Sparse Representation and Its Application].

Multispectral satellite sensors have several limitations with respect to capturing the target's spectral information due to their band setting and number of bands. The hyperspectral reconstruction technique is an effective method to obtain hyperspectral information from multispectral data. In this study, we propose a hyperspectral reconstruction algorithm based on the sparse representation of water remote sensing reflectance. The proposed algorithm was validated for five ocean color sensors (Sentinel-2A MSI, MERIS, MODIS Aqua, GOCI, and ⅦRS) using measured above-water remote sensing reflectance. The mean absolute percentage error (MAPE) and root mean square error (RMSE) of the reconstructed and measured spectra for five ocean color sensors were less than 10% and 0.005 sr, respectively. Compared with the spectra reconstruction algorithm based on multi-variable linear regression, the proposed algorithm can obtain the features of complex water remote sensing reflectance without using -measured reflectance for algorithm tuning. In addition, the accuracy of the proposed algorithm is better than the spectra reconstruction algorithm based on multi-variable linear regression. Two spectra reconstruction algorithms were applied to five ocean color sensors to test the applicability of the remotely estimated water constituent concentration. The statistical results for the reconstructed spectral factors and water constituent concentration suggest that the reconstructed reflectance derived by the proposed algorithm has a performance similar to that of -measured hyperspectral reflectance. The reconstructed reflectance derived by the proposed algorithm performs better than the spectra reconstruction algorithm based on multi-variable linear regression. Finally, the proposed algorithm was applied to GOCI data to remotely estimate the chlorophyll-a and total suspended matter concentrations. The accuracy of the water constituent concentration estimated from reconstructed images is better than that using original multispectral images. For the estimation of the chlorophyll-a concentration, the MAPE improved from 80.6% to 51.5% and the RMSE improved from 12.175 μg·L to 7.125 μg·L. For the estimation of total suspended matter, the MAPE improved from 19.1% to 18.8% and the RMSE improved from 29.048 mg·L to 28.596 mg·L.