Extracting nutrient - ocean state relationships from oceanic biogeochemistry simulations for macroalgae mariculture

Macroalgae mariculture has the potential to mitigate harmful algal blooms in nutrient rich coastal waters and provide a sustainable biofuel feedstock, as is being evaluated within the Advanced Research Projects Agency–Energy Macroalgae Research Inspiring Novel Energy Resources (MARINER) program. The evaluation of viable mariculture field sites requires a detailed understanding of nutrients and ocean conditions on site. The fundamental challenge is that existing observational and reanalysis data products do not have nutrient information for these purposes. Nutrients computed within the US Department of Energy’s Energy Exascale Earth System Model (E3SM) are at global to regional scale, but mariculture applications require nutrient information at field scale. To bridge this gap, we build a nutrient prediction model based on calibrated ocean biogeochemstry output from E3SM using a Random Forest Regression (RFR) algorithm. The E3SM output includes the ocean state variables of surface current velocities, temperature, salinity and biogeochemical variables like nutrients at a three hour frequency. The RFR algorithm is trained by daily data to estimate nitrate for the rest of the variables. The optimized RFR model, fine-tuned by grid search technique, shows great performance in predicting the spatio-temporal nitrate distribution, with an Out-of-Bag score of 0.97 in model calibration and Goodness-of-Fitting of R2 = 0.96 in model verification. These results highlight key factors responsible for nutrient spatio-temporal variability and will be used to approximate nutrient concentrations for remote sensing data and in higher resolution reanalysis products, e.g., HYCOM ocean model forecasts.