Assessment of the Black Sea observing system. A focus on 2005-2012 Argo campaigns

Abstract

An observing system in the Black Sea combining remote sensing data such as sea level anomalies from altimetry, sea surface temperature from satellite radiometer and data from Argo floats has been analyzed with the aim to quantify the contribution of different information sources when reconstructing the ocean state. The main research questions are: (1) do Argo float measurements substantially impact the quality of estimates, (2) what is the dependence of this quality upon the data and sampling used, and (3) are there specific Black Sea issues? Numerical model output and statistical analysis were used for this purpose. It has been demonstrated that the statistical method performs in a consistent way reproducing known geophysical patterns. Maximum footprints of sea level, salinity and temperature were illustrated, most of them clearly connected with specific thermohaline conditions and the general circulation. Reduced analysis capabilities were identified as associated with a low level of dynamical coupling between the shelf and the open ocean, mesoscale dynamics and representation of diapycnic processes in the models. The accuracy of Argo pressure measurements appeared very important to resolve the extremely sharp stratification in the upper layers. The present-day number of Argo floats operating in the Black Sea of about 10, seems optimal for operational purposes.

Appendices

Appendix 1: Altimetry

The standard deviation (STDV) of the altimeter data during 2005-2012 (Fig. 13a), shows well known spatial patterns connected to prominent hydro-dynamical and eddy propagation features of the Black Sea (Grayek et al. 2010). The comparison with the amplitude of oscillations in earlier periods (Stanev et al. 2000) demonstrates that the basic sub-basin scale patterns consisting of the Batumi and Sevastopol eddy and the large meander south of the Kerch Strait are very robust at all multi-year time intervals. The pattern of the formal mapping error (percent of the standard deviation) follows the one presented by (Stanev et al. 2000) therefore it is not shown here. It is below 3 cm (Fig. 13c) with minimum of ∼0.8 cm along the satelite tracks and maximum of ∼4 cm near the coasts. The averaged for the entire basin mapping error varies between 2 cm and 2.4 cm, with only several spikes reaching ∼3 cm, thus it is relatively stable.

Fig. 13

Standard deviation of SLA during 2005-2012 in cm (a) and basin mean formal mapping error of SLA times the standard deviation (b)

Appendix 2: Sea surface temperature

The standard deviation of SST during 2005-2012 (Fig. 14a) states that the largest variations are found in the northern part of the continental shelf region. This is explained by the shallow depths which enhances the amplitude of seasonal signal. This region extends to the south along the western coast and re-directs in the south-western corner of the basin to the interion part. The larger amplitudes in the basin interior in comparison to the ones in the coastal zone of the southern and northern Black Sea are explained by the shallower pycnocline there. In conclusion: maximum amplitudes of SST are either supported by the shallow bottom or by the shallow pycnocline. The relatively smaller SST amplitudes along the coastal zone in the south and north where the abyssal plain almost reaches the coast are explained by the intensive downwelling there. A prominent area with low variability is the region of formation and propagation of the Sevastopol eddy. Its counterpart from the eastern Black Sea, the Batumi eddy is not visible in Fig. 14a because this eddy occurs usually only in summer. The temporal mean error of the SST product (Fig. 14b) is substantially lower than the standard deviation with a minimum of about 0.18 ^∘C in the basin interior. Maximum errors of ∼0.25 ^∘C are located in a very narrow coastal band. The temporal evolution of spatial mean absolute errors (Fig. 14c) reveals a distinct seasonal characteristic. In summer the mean values are ∼ 1.5 ^∘C, while the values in winter are more than twice as high (∼ 3.5 ^∘C).

Fig. 14

Standard deviation of SST during 2005-2012 in ^∘C (a), absolute error of SST for 2005-2012 in ^∘C (b) and basin mean SST error (c)

Appendix 3: Argo floats

All profiles used in the following analysis were real-time quality controlled (QC) and flagged with a QC-level of ’1’. Overall 1724 profiles from fourteen floats were available. Fig. 1a displays trajectories and profile positions of the used floats. Nearly all profiles are positioned within the deeper inner part of the basin. There are only few observations close to the coast and no observations within the shallow shelf region. Profile cycle time, parking and maximum profile depth differed between individual floats. In the first period the maximum profile depth was ∼ 1500 m for every cycle. During the second period fewer profiles with a maximum depth below ∼ 700 m were taken. The Table 6 shows specification of the individual floats programming used in the study. The instrument errors for temperature and salinity are small (0.002 ^∘C, 0.01 PSU, see ARGO (2000) and Argo data management (2013)).

Table 6 Inventory of the used floats

The floats presented in Table 6 are different; some older model floats used ARGOS communications technology. The vertical resolution of the data from some of these floats was likely coarser than 5 m, which is probably not quite accurate to resolve adequately vertical stratification. Newer float technology that uses IRIDIUM communications yields much higher vertical resolution, usually 2 m throughout the water column. The possible consequences of inaccuracy in the vertical resolution is addressed in the present study and taken into account in the SEON and SESON.

Appendix 4: Mathematical expressions for observation operators

In the following the exact mathematical expressions used for the observation operator matrix H introduced in Section 2.4 are summarized. The measurement operator $H_{{\theta _{s}}/{\zeta }}$ for simultaneous SST and SLA measurements is given as

$$\begin{array}{@{}rcl@{}} H_{\theta_{s}/\zeta}=\left( \begin{array}{lll}I_{m} & Z_{m,m} & Z_{m,N-2m}\\ Z_{m,m} & I_{m} & Z_{m,N-2m}\end{array} \right)\;\;\; , \end{array} $$

(12)

where I _m denotes the m×m identity matrix and Z _{m, n} is the zero matrix of dimension m×n. We then have

$$ H_{{\theta_{s}}/{\zeta}} \; x +\epsilon_{{\theta_{s}}/{\zeta}} = \left( \zeta^{o}, {\theta_{s}^{o}} \right)^{T}\,, $$

(13)

with observed (index ^o) surface temperatures ${\theta _{s}^{o}}$, sea level anomaly ζ ^o, and respective observation errors $\epsilon _{{\theta _{s}}/{\zeta }}$. For the vertical profile measurements of salinity and temperature provided by Argo floats, the observation operator H _s/θ is given by

$$ H_{s/\theta} = \left( \begin{array}{lll} Z_{m,m} & I_{m} & Z_{m,N-2m}\\ Z_{p,2 m} & Z_{p,r} & Q \\ Z_{p,2 m} & Q & Z_{p,r} \end{array} \right) \;\;\; . $$

(14)

Here r denotes the number of bins in the three-dimensional grid below the surface and p is the number of bins below the surface in which salinity and temperature measurements are taken. The matrix Q of dimension p×r depends on the specific trajectory of the floats and is given by

$$\begin{array}{@{}rcl@{}} Q_{i,j} \,=\, \left\{\begin{array}{ll} 1& \text{if measurement number} \;i \; \text{is taken at bin number}\; j\\ 0& \text{else} \end{array}\right.. \end{array} $$

(15)

The index j refers to the bin number below the surface with ordering as given in the definition of the state vector (Eq. 1). We then have

$$ H_{s/\theta} \; x + \epsilon_{s/\theta} = \left( \overline{\theta}^{o}_{s} ,s^{o} ,\theta^{o} \right)^{T} , $$

(16)

where $\overline {\theta }^{o}_{s}$ is the surface temperature provided by the profiling instrument, which may differ from the remote sensing measurement ${\theta _{s}^{o}}$, while s ^o and θ ^o are the salinity and temperature measurements below the surface. The observation errors of the Argo floats is represented by 𝜖 _s/θ .

The observation operator for the combined SST, SLA and Argo measurements can then be written as

$$\begin{array}{@{}rcl@{}} H_{{\theta_{s}}/{\zeta},s/\theta}= \left( \begin{array}{l} H_{{\theta_{s}}/{\zeta}}\\ H_{s/\theta} \end{array}\right) \;\; . \end{array} $$

(17)

Appendix 5: Horizontal patterns of the seasonal variability of errors in the SEON

The results from SEON analyzed in the main part of this paper were annual mean. As it has been shown averaging over long periods appeared usefull to derive first order information but gives only an insufficient information of the temporal and spatial dynamics. In Fig. 15 additional seasonal averages for the presented SEONs are given to demonstate that the individual estimates for shorter times could largely differ from the annual mean ones. It is clearly seen from Fig. 15 that the differences exist not only between the individual SEONs but also during the individual seasons.

Fig. 15

Seasonal mean relative reconstruction errors vertically averaged over the upper 70 m (top) and meridionally averaged for all latitudes (bottom) for temperature (left-hand side) and salinity (right-hand side) fields. Estimates for the observation networks (from top to bottom) AT-1, F _a -1, ATF _a -2 and ATF _a -1 (see Table 1 for the nomenclature) are shown