A Simple Nonlinear and End-Member-Free Approach for Obtaining Ocean Remineralization Patterns

Abstract

The variability of a biogeochemical property in the ocean is the outcome of both nonconservative (such as respiration and photosynthesis) and conservative (mixing of water masses with distinct concentrations at origin) processes. One method to separate both contributions is based on a multiple regression of the biogeochemical property in terms of temperature u and salinity S as conservative proxies of water masses. This regression delivers the variability related to the conservative fraction and hence allows for identifying the residual as the biogeochemical anomaly. Here, the standard multiple linear regression (MLR) method, which assumes that water masses mix locally and linearly, is compared with a nonlinear polynomial regression (PR) over the entire (u, S) space. The PR method has two important advantages over MLR: allows for simultaneous nonlinear mixing of all water masses and does not require knowing the end-member water types. Both approaches are applied to data along 7.58N in the equatorial Atlantic Ocean, and the biogeochemical anomalies are calculated for humic-like fluorescent dissolved organic matter, apparent oxygen utilization, and nitrate-all of them related through in situ remineralization processes. The goodness of both approaches is assessed by analyzing the linear dependence and the coefficient of correlation between the anomalies. The results show that the PR method can be applied over the entire water column and yet retains the local variability associated with nonconservative processes. The potential of the PR approach is also illustrated by calculating the oxygen-nitrate stoichiometric ratio for the entire 7.58N transatlantic section.

Circumpolar Deep Water and Shelf Sediments Support Late Summer Microbial Iron Remineralization

“…Linear models for remineralization ratios are susceptible to bias as they do not accurately reflect mixing of endmember water masses and ignore external nutrient inputs (such as sediment resuspension or ice melt). Consequently, some studies have shifted toward non‐linear (polynomial) regressions to assess remineralization within defined water masses (De La Fuente et al., 2017). For this study, both a standard Type II regression model and a spline polynomial regression model were fitted to AOU and dFe data (excluding data below detection limit) to reduce bias, however no significant difference was found between the two models ( p = 0.13; Figure S3 in Supporting Information S1).…”

Dynamics of marine dissolved organic matter : ocean metabolism and climate transitions

“…To answer objective 1, a novel statistical modelling approximation was developed and published under the title "A simple nonlinear and end-member-free approach for obtaining ocean remineralization patterns" (De La Fuente et al, 2017). The purpose of this work was to render an objective and simple methodology for resolving the non-conservative fraction of biogeochemical variables.…”