Machine Learning Estimates of Global Marine Nitrogen Fixation

Abstract

Marine nitrogen (N2) fixation supplies “new” nitrogen to the global ocean, supporting uptake and sequestration of carbon. Despite its central role, marine N2 fixation and its controlling factors remain elusive. In this study, we compile over 1,100 published observations to identify the dominant predictors of marine N2 fixation and derive global estimates based on the machine learning algorithms of random forest and support vector regression. We find that no single environmental property predicts N2 fixation at global scales. Our random forest and support vector regression algorithms, trained with sampling coordinates and month, solar radiation, wind speed, sea surface temperature, sea surface salinity, surface nitrate, surface phosphate, surface excess phosphorus, minimum oxygen in upper 500 m, photosynthetically available radiation, mixed layer depth, averaged photosynthetically available radiation in the mixed layer, and chlorophyll‐a concentration, estimate global marine N2 fixation ranging from 68 to 90 Tg N/year. Comparison of our machine learning estimates and 11 other model outputs currently available in literature shows substantial discrepancies in the global magnitude and spatial distribution of marine N2 fixation, especially in the tropics and in high latitudes. The large uncertainties in marine N2 fixation highlighted in our study argue for increased and more coordinated efforts using geochemical tracers, modeling, and observations over broad ocean regions.

Data‐Driven Modeling of the Distribution of Diazotrophs in the Global Ocean

“…In addition, although the quantification of diazotrophs in coastal oceans is scarce in our dataset, diazotrophic hot spots are predicted in coastal waters (e.g., eastern American coast, coast of islands in the western Pacific), in line with the recent studies of Tang, Wang, et al (2019) and Mulholland et al (2019). More broadly, the RF-estimated distributions of diazotrophs combining Trichodesmium, UCYN-A, UCYN-B, and Richelia match the modeled distribution of N 2 fixation by Tang, Li, and Cassar (2019). To the best of our knowledge, only one other study has estimated the distribution of diazotrophs other than Trichodesmium (Monteiro et al, 2010).…”

“…For example, only 55 locations have concurrent surface observations of the explanatory variables used in the model construction (temperature, salinity, nitrate, and phosphate) and volumetric Trichodesmium abundance. The processes of data matching and correlation analyses are shown in a workflow chart ( Figure S6) and in Text S2 in the supporting information (Boyer et al, 2013;de Boyer Montégut et al, 2004;Kalnay et al, 1996;Luo et al, 2014;Moore et al, 2013;Tang, Li, & Cassar, 2019;Taylor et al, 2012).…”

Optimality-Based Non-Redfield Plankton-Ecosystem Model (OPEMv1.0) in the UVic-ESCM 2.9. Part II: Sensitivity Analysis and Model Calibration

“…sensitive to φ dia(Figure 2).No single parameter dominates the sensitivity of N 2 fixation in the simulations(Figure 1), which resembles the result ofTang et al (2019) that no single environmental property predicts global N 2 fixation, even with a data-based machine-learning method.…”

Reconstructing Global Chlorophyll-a Variations Using a Non-linear Statistical Approach

“…SVR also uses a robust error norm based on the principle of structural risk minimization, where both the error rates and the model complexity should be minimized simultaneously. Because SVR can efficiently capture complex non-linear relationships, it has been used in a variety of fields, and more specifically for oceanographic, meteorological and climate impact studies (Aguilar- Martinez and Hsieh, 2009;Descloux et al, 2012;Elbisy, 2015;Neetu et al, 2020), as well as in marine bio-optics (Kim et al, 2014;Hu et al, 2018;Tang et al, 2019). Predictors and Chl are normalized by removing their respective average and dividing them by their standard deviations.…”