Advancing cyanobacteria biomass estimation from hyperspectral observations: Demonstrations with HICO and PRISMA imagery

Retrieval of the phycocyanin concentration (PC), a characteristic pigment of, and proxy for, cyanobacteria biomass, from hyperspectral satellite remote sensing measurements is challenging due to uncertainties in reflectance (?Rrs) resulting atmospheric correction instrument radiometric noise. Although several individual algorithms have been proven capture local variations biomass specific regions, their performance has not assessed on images sensors. Our work leverages machine-learning model, Mixture Density Networks (MDNs), trained large (N = 939) dataset collocated situ chlorophyll-a concentrations (Chla), PCs, (Rrs) estimate PC all relevant spectral bands. The developed model demonstrated via maps produced select Hyperspectral Imager for Coastal Ocean (HICO) Italian Space Agency's PRecursore IperSpettrale della Missione Applicativa (PRISMA) using matchup dataset. As input MDN, we incorporate combination widely used band ratios (BRs) line heights (LHs) taken existing multispectral algorithms, that both Chla estimation, as well novel BRs LHs increase overall estimation accuracy reduce sensitivity ?Rrs. When random half dataset, MDN achieves 44.3%, which less than viable optimized algorithms. notably better at preventing overestimation low (<10 mg m?3) PC. Visibly, HICO PRISMA show wider dynamic range can be represented by MDN. available satellite-derived Rrs matchups measured demonstrate robustness estimating reduced impact ?Rrs medium-to-high (>10 m?3). According our extensive assessments, anticipated enable practical products HICO, therefore promising planned missions, such Plankton Aerosol Cloud Ecosystem (PACE). This advancement will enhance complementary roles radiometry low-altitude platforms quantifying monitoring harmful algal blooms spatial scales.