High-resolution monitoring of nutrient concentrations is crucial for assessing water quality in coastal and bay regions. This study introduces a two-step framework, based on the machine learning model AutoGluon, to map dissolved inorganic nitrogen (DIN) and phosphorus (DIP) concentrations with high spatial resolution for Xiamen Bay (XMB). First, DIN and DIP retrieval models (AutoGluonDIN/DIP) were trained using matchups from Sentinel-3 Ocean and Land Color Instrument (OLCI, 300 m) data and in situ measurements. Then, cross-sensor transfer models (AutoGluon-transfer) were developed using matchups between OLCI and Sentinel-2 Multi Spectral Instrument (MSI, resampled to 10 m), which converts MSI data to OLCI- equivalent bands. This allowed the OLCI-trained AutoGluonDIN/DIP models to be applied to MSI data, achieving high spatial resolution mapping of DIN and DIP. Key inputs of AutoGluon-DIN/DIP are the Rayleigh-corrected top-of-atmosphere reflectance (ρ_rc(λ)) at eight common spectral bands between MSI and OLCI. Validation against in situ data demonstrates that AutoGluon-DIN/DIP outperforms other machine learning models, with a root mean squared difference of 0.11 mg L^-1 for DIN and 0.012 mg L^-1 for DIP (N = 636). Rretrieved DIN/DIP also closely matched independent buoy measurements in both magnitude and temporal variability (coefficient of determination R² ~0.6, N = 382). The AutoGluon-transfer effectively converts MSI-measured ρ_rc(λ) to OLCI- equivalent bands (R² > 0.8), producing consistent nutrient maps with that from OLCI in magnitude and spatial pattern (R² ~0.6). Thus, the proposed framework offers a promising solution for high-resolution nutrient monitoring in coastal waters.

Nutrients remote sensing, machine learning, AutoGluon, high spatial resolution, Sentinel-2