Abstract: Green tides have emerged as a significant marine ecological disaster worldwide, rendering the accurate detection and area estimation of green algae crucial. To accurately estimate the coverage area of green algae communities in the monitoring of green tides based on low-resolution satellite images, this study proposed a dictionary learning-based method for estimating the area of green algae using hyperspectral images. The proposed method involves deriving the endmember spectrum database that is closest to the unknown surface feature spectra via online robust dictionary learning, obtaining the abundance map of green algae through sparse coding, and calculating the coverage area of green algae. It was verified through the experiment using the spectral images acquired by the geostationary ocean color imager (GOCI) on June 25, 2016, and June 21, 2020. The experimental results reveal that the calculated coverage areas of green algae on the two days were highly close to the approximate measured results, with a minimum error of only 2.15 %, suggesting that the proposed method outperforms traditional index-based hard thresholding algorithms. Independent of the pure pixel hypothesis, the proposed method can effectively address the mixed pixel problem and enhance area estimation accuracy in the absence of a pre-estimated number of endmembers or prior spectral information, thereby achieving high-precision subpixel-level area estimation of green algae.