Frontal dynamics boost primary production in the summer stratified Mediterranean sea

Abstract

Bio-physical glider measurements from a unique process-oriented experiment in the Eastern Alboran Sea (AlborEx) allowed us to observe the distribution of the deep chlorophyll maximum (DCM) across an intense density front, with a resolution (∼ 400 m) suitable for investigating sub-mesoscale dynamics. This front, at the interface between Atlantic and Mediterranean waters, had a sharp density gradient (Δρ ∼ 1 kg/m³ in ∼ 10 km) and showed imprints of (sub-)mesoscale phenomena on tracer distributions. Specifically, the chlorophyll-a concentration within the DCM showed a disrupted pattern along isopycnal surfaces, with patches bearing a relationship to the stratification (buoyancy frequency) at depths between 30 and 60 m. In order to estimate the primary production (PP) rate within the chlorophyll patches observed at the sub-surface, we applied the Morel and Andrè (J Geophys Res 96:685–698 1991) bio-optical model using the photosynthetic active radiation (PAR) from Argo profiles collected simultaneously with glider data. The highest production was located concurrently with domed isopycnals on the fresh side of the front, suggestive that (sub-)mesoscale upwelling is carrying phytoplankton patches from less to more illuminated levels, with a contemporaneous delivering of nutrients. Integrated estimations of PP (1.3 g C m⁻²d⁻¹) along the glider path are two to four times larger than the estimations obtained from satellite-based algorithms, i.e., derived from the 8-day composite fields extracted over the glider trip path. Despite the differences in spatial and temporal sampling between instruments, the differences in PP estimations are mainly due to the inability of the satellite to measure DCM patches responsible for the high production. The deepest (depth > 60 m) chlorophyll patches are almost unproductive and probably transported passively (subducted) from upper productive layers. Finally, the relationship between primary production and oxygen is also investigated. The logarithm of the primary production in the DCM interior (chlorophyll (Chl) > 0.5 mg/m³) shows a linear negative relationship with the apparent oxygen utilization, confirming that high chlorophyll patches are productive. The slope of this relationship is different for Atlantic, mixed interface waters and Mediterranean waters, suggesting the presence of differences in planktonic communities (whether physiological, population, or community level should be object of further investigation) on the different sides of the front. In addition, the ratio of optical backscatter to Chl is high within the intermediate (mixed) waters, which is suggestive of large phytoplankton cells, and lower within the core of the Atlantic and Mediterranean waters. These observations highlight the relevance of fronts in triggering primary production at DCM level and shaping the characteristic patchiness of the pelagic domain. This gains further relevance considering the inadequacy of optical satellite sensors to observe DCM concentrations at such fine scales.

Appendix

1.1 A Calibration of the optical model

The pragmatic objective of this section is to calibrate, on the basis of the Prov-Bio optical data (Section 2.2), an optical model suited to reconstruct the PAR conditions along the AlborEx coastal glider (SG) transect. For this specific objective, only the profiles obtained between the 26st of May and the 7th of June were considered, when the Prov-Bio path was close to the AlborEx front.

Several candidate optical models were tested, considering a single bandwidth with Chl and CDOM attenuation (PAR₁) or two bandwiths (Zielinski et al. 2002) with attenuation from Chl only (PAR₂) or Chl and CDOM (PAR₃):

$$\begin{array}{@{}rcl@{}} \text{PAR}_{1} & & \left\{\begin{array}{ll} \text{PAR}(z) & = \text{PAR}(0). e^{-{\int\limits_{0}^{z}} k(z^{\prime})\,dz^{\prime}}\\ \text{with } k(z) & = k_{0}+ k_{\text{Chl}}.\text{Chl}(z)+ k_{\text{CDM}}.\text{CDM}(z) \\ \end{array}\right. \end{array} $$

(7)

$$\begin{array}{@{}rcl@{}} \text{PAR}_{2} & & \left\{\begin{array}{ll} \text{PAR}(z) & = \text{PAR}(0).\left[p_{s}.e^{-{\int\limits_{0}^{z}} k_{s}(z^{\prime})dz^{\prime}} + (1-p_{s}).e^{-{\int\limits_{0}^{z}} k_{l}(z^{\prime})dz^{\prime}}\right] \\ k_{s}(z) & = k_{s,0}+ k_{s,\text{Chl}}.\text{Chl}(z) \\ k_{l}(z) & = k_{l,0}+ k_{l,\text{Chl}}.\text{Chl}(z) \\ \end{array}\right. \end{array} $$

(8)

$$\begin{array}{@{}rcl@{}} \text{PAR}_{3} & & \left\{\begin{array}{ll} \text{PAR}(z) & = \text{PAR}(0).\left[p_{s}.e^{-{\int\limits_{0}^{z}} k_{s}(z^{\prime})dz^{\prime}} + (1-p_{s}).e^{-{\int\limits_{0}^{z}} k_{l}(z^{\prime})dz^{\prime}}\right] \\ k_{s}(z) & = k_{s,0}+ k_{s,\text{Chl}}.\text{Chl}(z)+ k_{s,\text{CDM}}.\text{CDM}(z) \\ k_{l}(z) & = k_{l,0}+ k_{l,\text{Chl}}.\text{Chl}(z)+ k_{l,\text{CDM}}.\text{CDM}(z) \\ \end{array}\right. \end{array} $$

(9)

For the sake of simplicity and since all profiles were taken at the same hour of the day in a 12-day interval, the incoming surface radiation was considered to be identical for all profiles and was tuned as a single parameter. Only two profiles were excluded for the calibration (corresponding respectively to the 1st and 5th of June) as they presented obviously affected incoming surface radiation, for instance due to cloud cover (Fig. 10).

Fig. 10

Optical parameters recorded by the Prov-Bio floats near the AlborEx front

The parameters of models PAR₁, PAR₂, and PAR₃ were calibrated to reproduce at best the corresponding Prov-Bio PAR profiles, when applied on the concurrent Chl and CDOM data. The skill associated with each model is given as the root of the PAR mean squared residuals evaluated for all the selected profiles (i.e., 11 profiles consisting of ∼ 215 measurement each) and are provided in Table 2.

Table 2 Number of parameters and model skill evaluated for the optical models

The consideration of two band widths in models PAR₂ and PAR₃ enhances the model skills. The consideration of CDOM in PAR₃ does not appear beneficial in what regards the model skill and poses an additional question in terms of parameter identifiability.

We finally retained model PAR₂, with parameters PAR₀ = 1532 μE m⁻² s⁻¹; p _s = 0.806; k _s,sw = 5.295 10⁻²m⁻¹; k _l,sw = 3.189 10⁻⁶m⁻¹; k _s,Chl = 3.328 10⁻²m² mg Chl⁻¹; k _l,Chl = 7.23m² mg Chl⁻¹;

The probability distribution around those values, as well as the dependencies between different parameter estimates, are depicted on Fig. 11 showing the distribution of parameter values retained in a Monte Carlo Markov Chain procedure (Soetaert and Petzoldt 2010). The pairwise relationships between successful parameter sets indicate a strong correlation between the long-wave band attenuation coefficients for seawater (k _l,sw) and Chl (k _l,Chl). In other terms, the good matching between simulated and observed PAR profiles is somewhat equivalent whether the long-wave band is attenuated by seawater or Chl. We retained the best parameter values indicated above, which gives a large weight to Chl for the long-wave attenuation, but we checked carefully that the PP estimates obtained from SG data were only marginally affected when using a parameter set in which long-band attenuation was driven by seawater.

Fig. 11

Marginal parameter distributions (diagonal), pairwise relationship (upper panels), and correlation coefficients (lower panels) between parameters of the optical model PAR₂, obtained by applying a Markov Chain Monte Carlo procedure as described in (Soetaert and Petzoldt 2010). Note the strong relationship between the calibrated seawater and chlorophyll attenuation coefficient, in particular for the long-wave light band

The PAR₂ model provides a representation of the PAR profiles suitable for the next steps of this study (Fig. 12, with percentage residuals always below 50% and usually well below 25% in the upper 60 m, a depth below which PAR is always lower than 5% of the surface incoming radiation).

Fig. 12

PAR profiles reproduced by applying the PAR₂ function on Prov-Bio profiles. First and second rows present (wide light line) modeled and (thin dark line) observed PAR profiles on different scales. Third and fourth rows present the corresponding residuals. The fifth row indicates the relative residuals, i.e., $\frac {\text {PAR}_{\text {model}}-\text {PAR}_{\text {obs}}}{\text {PAR}_{\text {obs}}}.100$

As the model calibration was restricted to AlborEx Prov-Bio input data, we do not pretend that our conclusions concerning the optical model suitability apply, for instance, to the entire Mediterranean Sea, over which the concentrations of optically relevant water constituents vary on ranges much larger than those encountered here.