Remote Estimation of Sea Surface Nitrate in the California Current System From Satellite Ocean Color Measurements

Sea surface nitrate (SSN) is an important parameter to characterize physical and biogeochemical processes, particularly to quantify oceanic new primary production, yet its remote estimation from satellite has been difficult due to the complex relationships between environmental variables and SSN. In the central and southern sections of the California Current System (CSCCS), this challenge is attempted through modeling, validation, and extensive tests in different oceanic scenarios. Specifically, using extensive SSN datasets collected by many cruises spanning 40 years (1978-2018) and Moderate Resolution Imaging Spectroradiometer (MODIS) estimated sea surface temperature (SST) and chlorophyll-a (Chl), a stacking random forest (SRF) model of SSN has been developed and validated with a spatial resolution of similar to 4 km. The model showed an overall performance of root mean square difference (RMSD) = 0.83 mu mol/kg, with coefficient of determination (R-2) = 0.87, mean bias = -0.11 mu mol/kg, and mean ratio = 1.15 for SSN ranging between 0.05 and 19.90 mu mol/kg (N = 1034). Furthermore, tests of the model with its original parameterization for the upwelling period, oceanic period, and winter period all showed satisfactory performance with an overall RMSD of 1.95 mu mol/kg. The sensitivity of the SRF model to uncertainties of MODIS SST and Chl was examined, with induced uncertainties of <= 2.22 mu mol/kg. The extensive evaluation and sensitivity tests indicated the robustness of the SRF model in estimating SSN in the study area of the CSCCS, and it could serve as a robust approach for other regions once sufficient in situ SSN data are available for model calibration.

Automated summary

The study developed a stacking random forest (SRF) model for estimating sea surface nitrate in the central and southern sections of the California Current System. The model showed high accuracy and robustness, suggesting it could serve as a reliable approach for other regions once in situ SSN data are available for model calibration.