Estimates of Water-Column Nutrient Concentrations and Carbonate System Parameters in the Global Ocean: A Novel Approach Based on Neural Networks

Abstract

A neural network-based method (CANYON: CArbonate system and Nutrients concentration from hYdrological properties and Oxygen using a Neural-network) was developed to estimate water-column (i.e., from surface to 8,000 m depth) biogeochemically relevant variables in the Global Ocean. These are the concentrations of three nutrients [nitrate (NO 3 − ), phosphate (PO 4 3− ), and silicate (Si(OH) 4 )] and four carbonate system parameters [total alkalinity (A T ), dissolved inorganic carbon (C T ), pH (pH T ), and partial pressure of CO 2 (pCO 2 )], which are estimated from concurrent in situ measurements of temperature, salinity, hydrostatic pressure, and oxygen (O 2 ) together with sampling latitude, longitude, and date. Seven neural-networks were developed using the GLODAPv2 database, which is largely representative of the diversity of open-ocean conditions, hence making CANYON potentially applicable to most oceanic environments. For each variable, CANYON was trained using 80 % randomly chosen data from the whole database (after eight 10 • × 10 • zones removed providing an "independent dataset" for additional validation), the remaining 20 % data were used for the neural-network test of validation. Overall, CANYON retrieved the variables with high accuracies (RMSE): 1.04 µmol kg −1 (NO 3 − ), 0.074 µmol kg −1 (PO 4 3− ), 3.2 µmol kg −1 (Si(OH) 4 ), 0.020 (pH T ), 9 µmol kg −1 (A T ), 11 µmol kg −1 (C T ) and 7.6 % (pCO 2 ) (30 µatm at 400 µatm). This was confirmed for the eight independent zones not included in the training process. CANYON was also applied to the Hawaiian Time Series site to produce a 22 years long simulated time series for the above seven variables. Comparison of modeled and measured data was also very satisfactory (RMSE in the order of magnitude of RMSE from validation test). CANYON is thus a promising method to derive distributions of key biogeochemical variables. It could be used for a variety of global and regional applications ranging from data quality control to the production of datasets of variables required for initialization and Sauzède et al. Nutrients and Carbonate System from T/S/O2validation of biogeochemical models that are difficult to obtain. In particular, combining the increased coverage of the global Biogeochemical-Argo program, where O 2 is one of the core variables now very accurately measured, with the CANYON approach offers the fascinating perspective of obtaining large-scale estimates of key biogeochemical variables with unprecedented spatial and temporal resolutions. The Matlab and R codes of the proposed algorithms are provided as Supplementary Material.

Biogeochemical sensor performance in the SOCCOM profiling float array

“…This process is modeled on the procedures used to adjust Argo salinity data [ Owens and Wong , ]. As an independent check on the correction process, the adjusted nitrate concentrations were also compared to the predictions of the CANYON neural network based system [ Sauzède et al ., ]. Mismatches between the MLR and CANYON estimates were generally less than 1 µmol kg −1 .…”

“…The corrections are validated over longer time periods by comparing the corrected data to estimations of neural network systems [ Sauzède et al ., ] and observations in the GLODAPv2 data set. The accuracy of the sensor data for nitrate and pH should not degrade appreciably in time because the correction process is the same as that used to make the initial corrections that produce the favorable comparison to independent deployment profiles (Table ).…”

Temporal and Vertical Variations of Particulate and Dissolved Optical Properties in the South China Sea

“…As there were no nitrate sensors on the floats, two depth values are derived as the proxy for the nitracline: (i) the 22 °C isotherm ( z T22 ) which was used by Zhang et al () as the proxy for nutricline and could be expected to covary with the depth where [NO] 3 reaches 0–5 μmol/kg in SCS based on a regional empirical relationship (Chen et al, ); and (ii) the depth which the neural‐network model CANYON finds [NO 3 ] = 2 μmol/kg ( z NO3 ; Sauzède et al, ). CANYON estimates [NO 3 ] based on the date, location, temperature, salinity, and dissolved oxygen.…”

“…Symbols Used in This StudyChen et al, 2006); and (ii) the depth which the neural-network model CANYON finds [NO 3 ] = 2 μmol/kg (z NO3 ;Sauzède et al, 2017). CANYON estimates [NO 3 ] based on the date, location, temperature, salinity, and dissolved oxygen.…”

Annual nitrate drawdown observed by SOCCOM profiling floats and the relationship to annual net community production

“…The MLR equations are derived from laboratory measurements on water samples collected throughout the Southern Ocean [ Williams et al ., ]. The adjustments are verified by comparing the corrected data estimates generated from a neural network‐based procedure [ Sauzède et al ., ] and data from GLODAPv2 [ Olsen et al ., ]. The adjustment process produces a data set that matches samples collected at stations where floats are deployed to better than 1 µmol kg −1 throughout the water column [ Johnson et al ., ].…”