New and updated global empirical seawater property estimation routines

Abstract

We introduce three new Empirical Seawater Property Estimation Routines (ESPERs) capable of predicting seawater phosphate, nitrate, silicate, oxygen, total titration seawater alkalinity, total hydrogen scale pH (pH T ), and total dissolved inorganic carbon (DIC) from up to 16 combinations of seawater property measurements. The routines generate estimates from neural networks (ESPER_NN), locally interpolated regressions (ESPER_LIR), or both (ESPER_Mixed). They require a salinity value and coordinate information, and benefit from additional seawater measurements if available. These routines are intended for seawater property measurement quality control and quality assessment, generating estimates for calculations that require approximate values, original science, and producing biogeochemical property context from a data set. Relative to earlier LIR routines, the updates expand their functionality, including new estimated properties and combinations of predictors, a larger training data product including new cruises from the 2020 Global Data Analysis Project data product release, and the implementation of a first-principles approach for quantifying the impacts of anthropogenic carbon on DIC and pH T . We show that the new routines perform at least as well as existing routines, and, in some cases, outperform existing approaches, even when limited to the same training data. Given that additional training data has been incorporated into these updated routines, these updates should be considered an improvement over earlier versions. The routines are intended for all ocean depths for the interval from 1980 to $2030 c.e., and we caution against using the routines to directly quantify surface ocean seasonality or make more distant predictions of DIC or pH T .

Carbonate Parameter Estimation and Its Application in Revealing Temporal and Spatial Variation in the South and Mid‐Atlantic Bight, USA

“…Multiple linear regression models to reconstruct DIC are widely applied in the open ocean, especially in terms of comparison historical and recent data sets and anthropogenic carbon estimations ( Friis et al, 2005;Wallace, 1995 ). Recent studies have developed locally interpolated alkalinity, pH, and nitrate ( Carter et al, 2016( Carter et al, , 2018( Carter et al, , 2021 using Global Data Analysis Project data ( Olsen et al, 2016 ). At the regional scale in ocean margins, MLR models must be able to represent different local processes and variabilities of the coastal carbonate system.…”

“…Field observations of carbonate parameters are relatively limited in space and time compared to basic hydrographic data. To overcome this limitation, researchers have developed MLR models of total alkalinity (TA) (Carter et al., 2016), dissolved inorganic carbon (DIC) (Carter et al., 2019; Friis et al., 2005; Sabine et al., 2008), Ω Ar (Alin et al., 2012; Carter et al., 2021; Juranek et al., 2009; Kim et al., 2010) and pH (Bittig et al., 2018; Carter et al., 2018, 2021; Juranek et al., 2011; Williams et al., 2016) with basic hydrographic data. However, a systematic evaluation of multiple MLR models with broader geographic applications is still desirable and needed.…”

GOBAI-O2: temporally and spatially resolved fields of ocean interior dissolved oxygen over nearly two decades

“…These calculations were made using the Gibbs-SeaWater (GSW) Oceanographic Toolbox for MATLAB (McDougall and Barker, 2011). As was done by Carter et al (2021), longitude was transformed into two separate predictors: cos(Longitude − 20° E) and cos(Longitude − 110° E). Cosine functions were applied to maintain the cyclical nature of longitude as a predictor, and offsets of 20° E and 110° E were intended to shift regions where the cosine function has minimum explanatory power over landmasses.…”

“…The average of FNN and RFR estimates (ENS, for ensemble average) was used as the [O2] estimate for a given set of input data. This ensemble averaging procedure was implemented due to insights from previous work showing that averaging the outputs of multiple ML algorithms or linear regression models often outperforms the output from just one approach on its own (Gregor et al, 2017;2019;Bittig et al, 2018b;Carter et al, 2021;Djeutchouang et al, 2022), likely due to complementary strengths and weaknesses of each approach. For this work, any especially erroneous result from either the FNN or RFR should be mitigated by better results from the other algorithm.…”

“…nitrate and pH, that exhibit distributions more similar to [O2] are good candidates for adoption into the GOBAI mapping scheme. Together with property estimation algorithms for TA (Bittig et al, 2018b;Carter et al, 2021), a mapped ocean interior pH product could be used to resolve the entire ocean carbonate system in four dimensions in near real time.…”