Mapping spatial distribution, percent cover and biomass of benthic vegetation in optically complex coastal waters using hyperspectral CASI and multispectral Sentinel-2 sensors

Abstract This work assessed the capability of Compact Airborne Spectrographic Imager (CASI) and satellite multispectral Sentinel-2 image data for mapping the distribution, percent cover (%cover) and biomass of submerged aquatic vegetation (SAV) in optically complex coastal waters of the Baltic Sea. As a first step, the distribution maps of SAV were created for brown macroalgae, green macroalgae and higher plants classes. Secondly, %cover maps were retrieved by building class level relationships between in situ estimated %cover and image reflectance. Thirdly, statistical models were built for estimating class specific SAV biomass as a function of SAV %cover. Finally, developed biomass models were applied to class specific %cover maps derived from the step 2 for landscape scale biomass estimation. CASI sensor had higher classification accuracy (78%) compared to Sentinel-2 sensor (69%). CASI also outperformed Sentinel-2 in the %cover assessment showing R2 values in the range of 0.55–0.73, while R2 values in the range of 0.36–0.49 were retrieved for Sentinel-2. However, both sensors provided similar distribution and %cover patterns of benthic vegetation. The %cover-biomass models showed a very good fit explaining 66–82% of variance of different SAV classes. Comparison of biomass estimates from both images revealed that the total dry biomass (t) was underestimated by Sentinel-2 by 10.6%. However, if biomasses were retrieved per unit area (t/km2), then both instruments resulted in nearly identical total SAV biomasses.