Satellite Retrieval of Oceanic Particulate Organic Nitrogen Vertical Profiles

Accurate satellite retrieval of oceanic particulate organic nitrogen (PON) vertical profile is essential for understanding global biogeochemical processes; however, no dedicated retrieval models currently exist. This study developed a novel PON profile retrieval model using the eXtreme Gradient Boosting (XGBoost) algorithm, based on a comprehensive global dataset that includes in situ PON measurements, MODIS-Aqua bio-optical data, and 3D reanalysis physical data. The XGBoost-retrieved PON profiles were compared with those derived from Copernicus particulate backscattering coefficient (bbp) profiles and were further used to estimate the euphotic-zone PON stocks through an optimally performing regression model. The results showed that the proposed model significantly outperformed models constructed without physical inputs, achieving R2 of 0.83, RMSE of 1.49 mg m−3 and MAPE of 18.07%. Compared to the bbp-based profiles, the XGBoost-retrieved profiles exhibited higher accuracy. The model also provided reliable estimates of euphotic-zone PON stocks, with R2 of 0.76, RMSE of 200.31 mg m−2 and MAPE of 15.09%. These findings demonstrate the potential of the proposed retrieval model for investigating oceanic nitrogen dynamics and biogeochemical cycles.