Optimizing Water Quality Parameters Retrieval from PRISMA data: A Comparative Study of Atmospheric Correction Algorithms

Hyperspectral remote sensing provides relevant information for monitoring biophysical water parameters [1]; an appropriate atmospheric correction (AC) is crucial, especially considering its impact on the estimation of the composition of phytoplankton functional types. In this context, for this preliminary study, 31 match-ups have been considered, using data from the hyperspectral Italian mission PRISMA over three Italian lakes: Trasimeno (meso-eutrophic), Varese (eutrophic) and Garda (meso-oligotrophic). The first aim was to validate PRISMA L2C standard reflectance products and PRISMA L1 data atmospherically corrected using ACOLITE, ACOLITE-T-Mart, iCOR, MIP, hGRS, POLYMER, 6SV processors, with in situ measurements, provided by the WISPStation and ad-hoc field campaigns. In accordance with the uncertainties assessed through statistical analysis, the second aim was to provide the most accurate water quality maps in terms of Chlorophyll-a (Chl-a), Total Suspended Matter and Phycocyanin, and bottom substrate characterization maps, using the bio-optical model BOMBER [2], the Semi-empirical band ratio [3], and a machine learning algorithm (MDN) [4]. Overall, results identified different performances of AC algorithms across the study areas and each spectral region (R2 greater than 0.80 for Lake Trasimeno, R2 below 0.75 for Lake Varese, and higher accordance in Lake Garda’s shallow waters compared to deep waters). Figure 1 reports the analysis of the image acquired over Lake Trasimeno on 12/08/2022 as an example. Comparing the AC processors, PRISMA L2C shows a Spectral Angle (SA) of 8° and an R² of 0.95, while other algorithms, like MIP, shows better performance with SA equal to 3° and 0.98 as R². When generating Chl-a maps, PRISMA L2C shows a 16% bias, whereas MIP shows only 1% with in situ data, indicating that more performant AC improves the accuracy of biophysical parameter retrieval, and that PRISMA L2C may not always provide the most accurate results. Although a more comprehensive dataset and further analysis are recommended to improve PRISMA's performance across different optical water types and atmospheric and geometric conditions, this study demonstrates the importance of AC for biophysical parameters estimation, and highlights hyperspectral sensors as a suitable technology for water quality mapping.