Modeling Spatial and Temporal pCO₂ Variability at 49°N/16.5°W and 56.5°N/52.6°W in the North Atlantic Ocean

Global oceans provide large uptake potential for anthropogenic atmospheric CO₂. This is of great importance considering the oceanic stabilization effect on atmospheric CO₂. The oceanic uptake of excess of inorganic carbon from the atmosphere has been shown to account for approximately 48% of the total anthropogenic CO₂ emissions within the past 200 years [1]. Current estimates of global oceanic CO₂ sink as well as spatial variability and trends in pCO₂ in the oceans vary significantly [2-7]. Observational studies and models suggest that in some regions of the Atlantic Ocean, there may be a weakening trend of oceanic CO₂ sinks in response to either climate change or natural variability [7-11]. However, a switch from natural carbon storehouses to sources is highly unlikely in the foreseeable future.

The North Atlantic is considered to represent the most intense, per unit area, ocean sink for atmospheric CO₂ in the Northern Hemisphere on interannual to decadal timescales [12]. On the global scale, the oceans are the Earth’s primary sink that offsets anthropogenic CO₂ emissions and the North Atlantic is the primary gate for CO₂ entering the global ocean. However, according to the recent IPCC Global Carbon Budget 2013 assessment report, the ocean sink for anthropogenic CO₂ is estimated to be approximately equal to land sink fraction [13]. The subpolar North Atlantic is a complex and highly variable oceanic region both in space and time, and has in recent studies been confirmed as a significant sink based on automated surface ocean pCO₂ measurements [6,8,14].

Numerical ocean biogeochemical models have been and are employed in understanding and capturing physical-chemical and biological response patterns and driving mechanisms (processes) of the uptake and release of CO₂ in several ocean studies, usually in conjunction with in situ pCO₂ measurements. These coupled physicalbiogeochemical models are efficient, fast, compact and powerful scientific tools that are employed in understanding the past and present and predicting future cycles and changes in ocean carbonate chemistry system and climate, while being conscious of the ocean and climate system’s response to this forcing [15] have outlined the imperativeness of accurately capturing both the pCO₂-T (thermal) and pCO₂-nonT (non-thermal) in order to understand the biogeochemical dynamics of the oceanic pCO₂ cycle.

Modeling the oceanic response typically involves the interaction between physical (stratification, temperature, salinity, mixing, solar radiation, etc.), chemical (dissolved inorganic carbon (DIC), total alkalinity, particulate inorganic carbon (PIC), pH, oxygen, phosphate, silicate and iron), and biological (zooplankton, phytoplankton and dissolved organic carbon (DOC), dissolved organic matter (DOM) and particulate organic matter (POM) parameters and processes [7,16,17].

The fundamental objective of this study is the characterization and evaluation of the annual/seasonal patterns of pCO₂ variability as well as a deconvolution of associated physical and biological forcing’s using MITgcm biogeochemical model and in situ data from two North Atlantic time series sites. While we seek to improve our ability to quantify the North Atlantic sink as well as understand the surface ocean pCO₂ variability and governing processes, we took advantage of observed data from the porcupine abyssal plain (PAP) and the K1 central labrador sea (K1 CELAS) observatories, and attempt to investigate how well a physical-biogeochemical-ecosystem model can capture in situ observations of surface ocean pCO₂ in the North Atlantic western and eastern oceanographic locations.

Sources of observed data and deployments

The details of the sampling strategy, analytical methods, data quality control and calibration procedures employed for taking the diverse set of biogeochemical and physical measurements and recovered observed data at the KI CELAS and PAP observatories have been reported by [18,19] and in the Euro SITES data documentation [http://www.eurosites.info/]. The observed data used in this research were obtained during three consecutive mooring deployments in July 2003-March 2005 and September 2004-July 2005 for the northeast PAP (49°N/16.5°W) and northwest K1 CELAS (56.5°N/52.5°W) locations, respectively (Figure 1 and Table 1). The deployments at the Porcupine Abyssal Plain time series site were designated PAP2, PAP3 and PAP4 for observations made during the second, third and fourth deployments respectively.

Figure 1: Map of the North Atlantic ocean showing the Porcupine Abyssal Plain (49°N, 16.5°W) and K1 (Labrador Sea) (56.5°N, 52.5°W) sites.

Table 1: PAP and K1 CELAS observatory mooring deployments and the physico-biogeochemical parameters measured.

The physical-biogeochemical-ecosystem ocean model

Details of the MIT ocean general circulation model (MITgcm) used for the comparison of observational and model data in this study, which is regionally configured for the North Atlantic between 20°S and 81.5°N with a horizontal resolution of 0.5° latitude and 0.5° longitude have been described in previous papers and the model has been applied to the North Atlantic carbon sink by [7,20-22]. However, the distinctive characteristics of the physical part of the biogeochemical model are shown in Table 2, while a schematic presentation of the ecosystem part is illustrated in Figure 2.

Figure 2: Schematic diagram of ecosystem model showing the cycling of nutrients (Phosphate, P; Silica, Si; Iron, Fe) within two classes of phytoplankton, one class of zooplankton, and the two pools of dissolved organic matter (DOM) and particulate organic matter (POM) [16].

Table 2: Key features of the physical model.

The ecosystem model is moderately complex taking into account the cycling of carbonates and nutrients [16,23]. It also features a pelagic ecosystem with one zooplankton class and two phytoplankton classes (diatoms and small phytoplankton) [7,15]. The atmospheric pCO₂ forcing in the model is incorporated from Mauna Loa Observations [24], and takes into account the annual and seasonal pCO₂ cycle. The dissolved and particulate carbon pools are resolved, and parameterisation of calcium carbonate cycling is included, while the sea-air exchange of CO₂ in the biogeochemical model is parameterized according to [25], with surface ocean pCO₂ determined as discussed by [23]. A positive flux is described as a net CO₂ exchange from sea to the atmosphere (outgassing) while a negative flux value implies a net CO₂ invasion from air to the sea (uptake).

Calculation of pCO₂ thermal and non-thermal components

The exchange of CO₂ between the atmosphere and ocean surface is primarily controlled by the air-sea difference in partial pressure of CO₂ (ΔpCO₂), which provides the thermodynamical driving force and is regulated by naturally occurring physical, chemical and biological processes [26-31]. Considering the thermodynamics of the marine CO₂ system, the thermal component of surface-water pCO₂ variability is comparable in magnitude but opposite in direction to the biological effect [32]. However, it has been demonstrated that that the temperature (thermal) and biological (non-thermal) effects on seasonal pCO₂ amplitude can be derived from the total concentration of CO₂ effect, and the relative contributive impact of each processes for regulating the seasonal amplitude in pCO₂ assessed over the oceans [6]. In this synthesis, this approach was adopted in separating the thermal and non-thermal components of surface ocean pCO₂ variability.

In order to remove the temperature effect from the observed pCO₂, the observed pCO₂ values are normalized to the annual mean temperature of seawater at the observatory, using the equation:

(1)

and the effect of temperature changes on pCO₂ at a given observed temperature, T_obs is computed using the equation:

(2)

where T is the temperature in °C, pCO₂ is the partial pressure of carbon dioxide in μatm, and the subscripts “" and “obs” indicate annual mean and observed values respectively, [6].

Data/model pCO₂ comparison at the NE PAP observatory

In this section, the seasonal cycles of pCO₂, pCO₂-T, and pCO₂- nonT from the biogeochemical model and the PAP (PAP2 - PAP4) moored data are presented. For the most part and within specific surface mixed layer, the ocean model appears to be highly consistent and compares favorably with the PAP observed pCO₂ data and derived thermal/non-thermal components (Figures 3a-3c). However, the disagreement between the in situ PAP observations and model results at certain mixed layer depth may be partly attributed to the model resolution, and this obviously presents a major limitation of the MITgcm biogeochemical model.

Figure 3a: Mixed layer comparison of model and measured pCO2 at the NE Atlantic PAP site.

Model results for the seasonal pCO₂ cycle relative to the observations indicate that the model captures the cycle reasonably well although with significant underestimation of the pCO₂ amplitude especially within the 35-75 m depth mixed layer during summer/early wintertime of the 2^nd (PAP2) and 3^rd (PAP3) deployments. As described below, the model fails to capture the full magnitude of the biological signal (nutrient drawdown) observed at the PAP site. However, the model captured the characteristic subpolar time series observed pCO₂ summertime minimum, wintertime maximum.

The model pCO₂-T cycle (Figure 3b) has a pattern that clearly captures the cycle shown by the PAP data especially within the 20 - 55m mixed layer, although the peak data pCO₂-T amplitudes during late summer of the second deployment (PAP2) were higher than the model pCO₂-T cycle. Interestingly, the model pCO₂-T cycle at 55-75 m mixed layer depth was obviously inconsistent with the seasonal data pCO₂-T pattern. The amplitude of the model pCO₂-T (55-75 m) cycle was relatively higher in magnitude than the data pCO₂-T at this depth range. In general, the model and data results reveal that the summer months were characteristically warmer with 2004 recording comparatively higher sea surface temperature values. Surface temperature values were however low at the subsurface euphotic depths especially during the winter months.

Figure 3b: Mixed layer comparison of observed pCO₂-T and model pCO₂-T at the PAP site.

Model non-thermal forcing pattern (pCO₂-nonT) consistently agree with the data pCO₂-nonT cycle at 20-75 m depth mixed layer on a seasonal scale, and the amplitude broadly aligns with or comes within the range of the observed pCO₂-nonT. However, the model pCO₂-nonT cycle at 55-75 m mixed layer underestimates the biology signal clearly imprinted by the data pCO₂-T cycle during the summertime of the 2^nd deployment (Figure 3c).

Figure 3c: Mixed layer comparison of observed and model pCO₂-nonT at the PAP site.

A combination of the seasonal patterns of pCO₂-T and pCO₂-nonT cycles creates the model pCO₂ cycle, which indicates a large increase in surface pCO₂ during the winter and a significant decrease during the summertime. The enhanced surface pCO₂ is due to vertical mixing of CO₂ (DIC) enriched water from the subsurface depths (mesopelagic zone) to the surface of the ocean in winter [12,18]. However, it is obvious that the model actually estimates the surface mixed layer pCO₂ levels of the PAP site, and could be considered realistic in this regard.

The modeled pCO₂ performance during the summertime corresponding to the period of the second deployment suggests a temperature-controlled cycle when actually it is the non-thermal forcing that is dominantly regulating the total pCO₂ pattern. From non-thermal component attributes of the observed/model output, it can be proposed that the pCO₂ variability of the PAP site may not be governed primarily by seasonal temperature changes only, but also by biological forcing fueled by carbon fixation and nutrient availability, as well as deep vertical mixing of dissolved inorganic carbon enriched water from the bottom mixed layer depth to the ocean surface.

A striking observation is that the model evidently captures observed pCO₂ summertime minimum and wintertime maximum respectively except for 2003, and also accurately depicted that pCO₂ is governed by contrasting effect of temperature and biology. The model captured observed cycles of pCO₂, thermal and biology components reasonably well within surface 20-35 m with caveats at deeper layers. More so, the model underestimated the amplitude of annual temperature cycle at 35 m depth and below, suggesting shallow depth of seasonal thermocline.

Data/model pCO₂ comparison at NW K1 central labrador sea

In this section, the pCO₂, pCO₂-T and pCO₂-nonT model outputs to observed data at the K1 Labrador Sea time series are presented (Figure 4). At site K1, model pCO₂ has a rather peculiar jump during wintertime (mid-Feb. in 2004, and mid-Jan. in 2005), followed by a rather long period of flat value. It seems the model grossly underestimated both the physical and biological activities at this site during this period. The model pCO₂ is obviously out of phase with observed pCO₂ by about 3 months, peaking in November instead of February. More so, pCO₂ peak-to-peak amplitudes between the observed data and model output in early fall (October 2004) and in wintertime (January- March) were out of phase by about 60 μatm. This difference might be considered very significant, thereby indicating a moderate flaw of the model. Discrepancies might be due to model being too cold in the fall, and thus not allowing biological activity to drawdown pCO₂ as much as observed. Further, although both model and observed data show a dominance of biological activities over pCO₂ variability, caution must be made since modeled seasonal patterns were quite different from the data.

Figure 4: Comparison of model and observed pCO₂, pCO₂-T and pCO₂-nonT at the K1 central labrador sea time series site.

Model pCO₂-T of K1 CELAS site looks good between November through June although it appears too cold in late summer/fall. However, it warms up late by three months (three months phase offset). K1 CELAS data pCO₂ trend may be considered to be in-phase with model pCO₂-T pattern, although there is a significant digression during summertime. The MITgcm fairly captures the K1 CELAS wintertime pCO₂-nonT pattern (Figure 4). However, it partially fails in capturing the summertime biological signal observed at the K1 site within the 35-55 m depth. It appears the model pCO₂-nonT is not warming enough in summer hence its failure to capture the observed pCO₂-nonT drawdown and biological productivity.

Anomalies of pCO₂, pCO₂-T and pCO₂-nonT

The patterns of the averaged daily anomalies of pCO₂, pCO₂-T and pCO₂-nonT calculated from observed data at the PAP and K1 CELAS observatories and model outputs are presented in Figures 5 and 6 respectively. Measurements at PAP site shows a much elevated pCO₂ during summer 2003. However, the model grossly over-estimated the biological drawdown during this period, resulting in much lower pCO₂ (Figure 5). Relatively high positive anomalies of pCO₂ (61 μatm), pCO₂-T (58 μatm), and pCO₂-nonT (77 μatm) were obtained based on average daily data for PAP pCO₂, temperature and non-temperature components of the total pCO₂ considered. More so, the computed anomalies based on daily averages indicated negative deviations of 54, 58 and 84 μatm for pCO₂, pCO₂-T and pCO₂-nonT respectively. For K1 CELAS site, positive anomalies of pCO₂ (35 μatm), pCO₂-T (67 μatm), and pCO₂-nonT (57 μatm) were obtained based on average daily data, with corresponding negative deviations of 61, 25 and 96 μatm for pCO₂, pCO₂-T and pCO₂-nonT respectively (Figure 6).

Figure 5: Observed and model pCO₂, pCO₂-T and pCO₂-nonT anomalies for PAP site.

Figure 6: Observed and model anomalies for pCO₂, pCO₂-T and pCO₂-nonT for K1 CELAS observatory.

In this paper, the seasonal distribution of surface ocean observational data and associated thermal and non-thermal factors controlling the variability and processes of pCO₂ at the PAP (Northeastern) and K1 Central Labrador Sea (Northwestern) observatories in the North Atlantic Ocean have been investigated using a modestly highresolution ocean general circulation model. This work indicates that in the subpolar North Atlantic on a seasonal timescale the observed as well as the modeled pCO₂ distribution is marked by minimum and maximum pCO₂ for the summertime and wintertime respectively. The seasonal cycles of model pCO₂, pCO₂-T, and pCO₂-nonT of the biogeochemical model compared to observational data revealed that the thermal component of surface water pCO₂ is comparable in magnitude but opposite in direction to the biological effect. For the most part and within specific depth of the mixed layer, the ocean biogeochemical model appears to reproduce seasonal pCO₂, pCO₂-T, and pCO₂-nonT phases and amplitudes favorably consistent with the seasonal PAP pCO₂, pCO₂-T, and pCO₂-nonT observations.

However, few but significant disagreements between the in situ PAP observations and model results at certain mixed layer depths may be partly attributed to the model resolution amongst various factors. In general, model results for the seasonal pCO₂, pCO₂-T, and pCO₂-nonT cycles relative to the observatory data indicates that the model captures the pCO₂, pCO₂-T, and pCO₂-nonT cycles fairly well although with significant underestimation of the pCO₂ and pCO₂-nonT amplitudes especially within the 35 - 75m mixed layer depth during summer/early wintertime of the 2^nd and 3^rd deployments. In order words, the model fails to capture the biological signal (nutrient drawdown) observed at the PAP site.

The EuroSITES Project data was used for this research. The contributions of the principal investigator and other scientists involved in the PAP project are acknowledged. The first author is particularly thankful to his Fulbright host Prof. G. A. McKinley for guidance. The authors would like to thank anonymous reviewers for their comments and suggestions that much improved the original manuscript.