“…A number of other studies have successfully developed empirical predictions of carbonate chemistry variables using methods that vary from relatively simple to significantly more complex and range from regionally specific algorithms to global assessments. The most common strategy for predicting carbonate chemistry variables is a multi‐linear regression approach, either predicting dissolved inorganic carbon (DIC) and/or total alkalinity (TA) and calculating the carbonate chemistry variable of interest (e.g., pH, saturation state, pCO 2 ), or directly predicting variable of interest (e.g., Alin et al., 2012; Bostock et al., 2013; Carter et al., 2018, 2021; Davis et al., 2018; Evans et al., 2013; Hales et al., 2012; Juranek et al., 2009; Kim et al., 2015; Lee et al., 2006; McGarry et al., 2021; Millero et al., 1998; Turk et al., 2017; Vance et al., 2022; Velo et al., 2013; Williams et al., 2016). More recently, some studies have incorporated machine learning techniques such as neural networks or random forest regression (Bittig et al., 2018; Broullon et al., 2019; Chen et al., 2019; Fourrier et al., 2020; Li, Bellerby, Ge, et al., 2020; Li, Bellerby, Wallhead, et al., 2020; Lohrenz et al., 2018; McNeil & Sasse, 2016; Sasse et al., 2013; Sauzède et al., 2017; Velo et al., 2013; Xu et al., 2020).…”