POSTER SESSION 2
10:30–12:30, Tuesday, September 15
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
ABSTRACT 1224 | POSTER 254
FROM EMPIRICAL TO INTELLIGENT: A CROSS-DATASET MATCHUP ANALYSIS OF EMPIRICAL, SEMI-ANALYTICAL AND MACHINE LEARNING ALGORITHMS FOR SATELLITE CDOM ESTIMATION IN EUROPEAN WATERS
Aiming to enhance the applicability of satellite retrieval of Colored Dissolved Organic Matter (CDOM) in European waters, a comprehensive matchup analysis was conducted of empirical, semi-analytical and machine learning algorithms using multi-source data. The study region encompasses sub-basins of the European seas characterised by diverse optical water types, with the predominant types being high CDOM such as Baltic Sea (BLTS), turbid waters with high suspended matter content, and complex coastal transitional waters exemplified by the Adriatic Sea (ADRS). The integration of two in-situ measurements of HyperBOOST and COASTS-BioMaP was conducted, in addition to one hyperspectral simulation Hyper_SD. The key findings are presented: (1) B21 attains the optimal overall performance on COASTS-BioMaP (Bias = -0.0015, RMSE=0.1281, QAA-V6-X method has the best performance on HyperBOOST (Bias=0.0001, RMSE=0.1433). (2) The relationship between bbp440 and ad440, which is a prerequisite for CDOM calculation, follows with the power-law trend, with a R² ranging from 0.010 (EMED) to 0.878 (ADRS), yielding an overall R² of 0.544 of COASTS-BioMaP and 0.8628 of Hyper_SD. (3) The relative error was used to indicate the retrieval reliability. QAA-derived methods in overestimation, while empirical underestimates in BLTS. The B21, conversely, yields the smallest relative errors. Furthermore, validation with HyperBOOST shows that most algorithms exhibit relative errors between –150% and 200%. Our results demonstrate that integrating multi-source in situ data with neural network effectively handles regional bio-optical diversity, marking a transition from empirical to intelligent CDOM retrieval.
Mengjie Zhao*, Institute of Science Marine, National Research Council(CNR-ISMAR), [email protected], https://orcid.org/0009-0007-6751-3248
Vittorio Brando, CNR-ISMAR, [email protected]
Carolina Cantoni, CNR-ISMAR, [email protected]
Simone Colella, CNR-ISMAR, [email protected]
Chenqian Tang, University of Rome, [email protected]
Emmanuel Boss, University of Maine, [email protected]
Chiara Santinelli, CNR-IBF, [email protected]
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
