POSTER Session 3

DEVELOPING A SENSOR-AGNOSTIC MODEL FOR CHLOROPHYLL RETRIEVAL FROM MULTIPLE SATELLITE SENSORS USING TRANSFORMERS

Satellite-derived chlorophyll products are critical for monitoring and understanding aquatic ecosystems. However, the accuracy and utility of these products are limited by sensor-specific algorithms that necessitate adjustments for each satellite sensor. This sensor dependency arises due to variations in spectral bands availability and response functions across different sensors, creating inter-sensor bias and inconveniences for users. To address these challenges, I developed a sensor-agnostic model employing a transformer-based machine-learning architecture. This innovative approach enables the retrieval of chlorophyll concentrations from four different sensors—MODIS, MERIS, VIIRS, and OLCI—using a single unified model. My method leverages the capabilities of transformers to handle the variable-length input effectively and to generalize across different spectral inputs by training on a diverse dataset covering multiple sensors. This study highlights the potential of advanced machine learning techniques to revolutionize the field of ocean color remote sensing by overcoming traditional limitations imposed by sensor heterogeneity.

Guangming Zheng, University of Maryland College Park, and NOAA National Environmental Satellite, Data, and Information Service (NESDIS) Center for Satellite Applications and Research, USA, [email protected], https://orcid.org/0000-0003-4624-7976

Code of Conduct
Privacy Policy