One Ecosystem :
Research Article
|
Corresponding author: Vladyslav Evstigneev (vald_e@rambler.ru)
Academic editor: Gbenga Akomolafe
Received: 20 Aug 2021 | Accepted: 24 Nov 2021 | Published: 02 Dec 2021
© 2021 Natalya Kyrylenko, Vladyslav Evstigneev
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Kyrylenko N, Evstigneev V (2021) Independent components of spatial-temporal structure of chlorophyll a patterns in the upper layer of the north-western shelf of the Black Sea. One Ecosystem 6: e73269. https://doi.org/10.3897/oneeco.6.e73269
|
|
In the present study, the results of independent component decomposition of satellite-derived chlorophyll a (Chla) patterns for the north-western part of the Black Sea are presented. The study has been carried out on the basis of the DINEOF-reconstructed dataset of 8-day average log-transformed Chla (alChla) patterns for 1997-2016. The alChla patterns were decomposed into six independent components of its spatio-temporal variability in the north-western shelf of the Black Sea. The independent components reflect the spatial distribution of alChla anomalies which are likely to be formed under the influence of sea circulation factors driven by wind. The paper presents the results of the analysis of the intra-annual variability of independent components. The interpretation of the patterns of intra-annual independent components variability is given, taking into account the seasonal variability of the wind factor, the flow of the Danube, the Dnieper and Southern Bug rivers and the fact of modulation of independent components dynamics by seasonal phytoplankton succession.
independent component analysis, chlorophyll a, the Black Sea, satellite-derived patterns
The Black Sea is a semi-enclosed basin with limited water exchange through the Bosphorus Strait with the Mediterranean Sea. Strong salinity and density stratification of the Black Sea waters are due to the influx of fresh waters of the Danube, the Dnieper, the Dniester and the Southern Bug rivers on the one hand and salty Mediterranean waters coming into deep layers of the sea on the other hand. The main pycnocline, formed as a result of this, limits the ventilation of the lower layers of the Black Sea (below 150 m). Thus, only the upper oxygen-containing thin layer (60-200 m) provides the conditions for the existence of biota (
One of the indicators of the marine ecosystem functioning is considered to be the content of the photosynthetic pigment - chlorophyll a (Chla) (
The aim of this study is to identify the main types of spatio-temporal structure of Chla patterns, which are formed under the influence of hydrodynamic factors of the marine environment in the surface layer of the north-western shelf zone of the Black Sea. The main source of information about Chla is satellite-sensing data, presented in different time resolutions. On the one hand, wind-induced circulation adjustment on the shelf is related to the synoptic timescale processes; therefore, satellite data of sub-monthly resolution should be used. On the other hand, the dataset of this temporal resolution may contain lost pixels (or grid cells) due to cloud cover, atmospheric correction algorithm errors etc. The array of 8-day Chla patterns seemed to us to be a reasonable compromise between these two factors determining the quality of the study results.
The paper is organised as follows. The data used in the study, as well as peculiarities of data transformation, are described in Section "Data". Section "Chla patterns decomposition technique" briefly describes a method for independent component analysis to be applied to Chla patterns. Section "Results and Discussion" presents the results of identifying independent spatio-temporal modes of Chla distribution for the north-western area of the Black Sea and joint analysis of the frequency of types and characteristics of wind circulation factors in a specified area.
In order to carry out the study, we combine satellite data and grid data of re-analysis of wind patterns over the north-western part of the Black Sea. The time period to be analysed was limited by the span of available satellite data for this region.
In this work, a dataset of the 8-day average Chla fields for the north-western region of the Black Sea was used. The dataset is presented in a single spatial resolution grid of 4.63 km and encompasses the period 1997-2016. These data have been produced in the framework of the GlobColour project of the European Space Agency (ESA) using the datasets of SeaWiFS, MODIS, MERIS and VIIRS scanners. Chla patterns from different scanners were combined by the Garver-Siegel-Maritorena (GSM) method (
The 8-day Chla arrays, as a rule, contain lost pixels due to a number of factors (cloud cover, errors of the atmospheric correction algorithm etc.). The gaps were statistically reconstructed using empirical orthogonal functions according to the DINEOF method (
The obtained Chla dataset was preprocessed to increase the reliability of the analysis results. The pigment content in the sea surface layer is a substantially positive log-normally distributed random variable, not only in the Black Sea (
\(alChla = \log Chla - \overline{\log Chla}\)
where \(\overline{\log Chla}\) is the average long-term value of the log-transformed Chla value in each pixel of the field. The alChla data are grouped in the form of an observation matrix Xt×m, where m = 1, ..., M is the spatial index (pixel index) and t = 1, ..., T is the time index.
In this paper, to characterise the external environmental factors and their variability, a daily data array of re-analyses ERA-Interim (at grid nodes of 0.25º x 0.25º) consisting of wind components (U, V), tangential wind stress in zonal (τX) and meridional directions (τY) was involved. In addition, to characterise the wind circulation of water in the surface layer of the Black Sea, the wind stress curl was calculated from the data of τ X , τ Y
\(rot_z\overrightarrow\tau = \frac{\partial\tau_X}{\partial X}-\frac{\partial\tau_Y}{\partial Y}\).
Eight-day fields, corresponding to Chla data, were calculated from the re-analysis data by simple averaging.
Analysis of multivariate time series, as a rule, is carried out to reveal hidden regularities associated with factors impacting the system under study and manifesting themselves as modes both in time and in space (
- the resulting space-time array is a linear mixture of principal components corresponding to factors impacting the system under study;
- the distribution of stochastic variables, associated with principal components, are normally distributed random variables.
In fact, alChla fields have non-Gaussian distribution. Fig.
Consequently, application of standard techniques for decomposition of multivariate Chla dataset is incorrect. In this work, the recently-developed method of decomposition into independent components is used for these purposes. Independent component analysis (ICA) (
The idea of ICA decomposition is to treat observation matrix Xl×m as a mixture of N ≤ M independent “sources” of signals \(S=(s_1,...,s_N)^T\). The requirement for component independence is more stringent than their non-correlation (as one assumes when applying PCA). For independent random variables \((s_1,...,s_N)^T\), the joint probability density function is expressed as a product of univariate functions. In case of non-Gaussian components, uncorrelatedness of si is not obvious. That is why any estimate of deviation of empirical probability density function from the Gaussian one may be considered as a measure of independence in ICA. According to the FastICA algorithm to be applied in this study, negentropy is considered to be a measure of independence (
The independent components to be extracted represent a certain type of evolution of the alChla patterns or the so-called “prototype” of behaviour (
Six independent components of alChla patterns were determined by the ICA method applied to a dataset of 8-day average satellite data on the content of Chla in the surface layer of the north-western shelf of the Black Sea. The choice of the number of types is ambiguous as there is no universal way to determine their optimal number. However, we focused on the results of Chla patterns decomposition into empirical orthogonal functions (EOF). Calculations shows that up to six empirical orthogonal modes account for more than 80% of the field total variance. The variance of the rest of the EOFs has values less than 2% and their loading maps are much distorted and badly distinguished. A comparison of the IC scores with their physical factors temporal variability allowed us to give an interpretation of the possible reasons for their formation.
Fig.
The peak of flood for the rivers of the north-western shelf, in particular the Danube and Dnieper Rivers, falls on April-May. With an increasing volume of fresh river waters carried to the shelf, the influx of nutrients assimilated by the phytoplankton within 1-2 months increases. The spring-summer peak of Chla is well-recognised by many studies (
Despite all the variety of factors affecting Chla during this period, we will focus on wind circulation over the area under study. Maximum correlation between 8-day mean values Chla/wind stress curl and Chla/vertical component of the wind stress was -0.38 and 0.32 for April and May, respectively. Significant correlation coefficients indicate the indirect influence of wind by forming the corresponding sea-surface drift currents on the shelf (
In the autumn-winter period, there is a mismatch between the course of IC1 and the run-off volume. In our opinion, this is due to the dominant influence of the wind factor. Fig.
IC2 describes about 13.5% of the total variance of alChla log fields. The analysis of the base temporal function of IC2 shows an obvious intra-annual variability - the presence of two peaks in winter (February-March) and in autumn (September-November). On the loading map, these peaks correspond to two unrelated areas of localisation of positive values of the IC2 decomposition coefficients. The first narrow area stretched along the coastline near the mouths of the main rivers (Fig.
In order to determine hydrodynamical processes that form this type of aj distribution, it is probably worth noting the wind factor under the influence of which sea surface currents develop both in the coastal region and on the slope of the north-western shelf of the Black Sea. This assumption is confirmed by correlation estimates (see alsoFig.
For the zonal wind component, the minimum negative correlations (r = -0.47 ... -0.32) were obtained for January to April. In summer, IC2 significantly correlated with the zonal component of tangential wind stress vector (r = -0.35 for July). For the meridional component of the wind velocity and wind stress, negative correlation values were obtained for the winter-spring months (r = -0.31 ... -0.5). Under the influence of northern winds, the input of sea waters from the shelf zone to the deep sea is predominant. This leads to the formation of negative anomalies in the content of Chla in the shelf zone, while in the deep-sea region, a winter maximum of pigment content can be observed (
During autumn-winter, convection surface water mixes with the bottom water, resulting in a gradual cooling. Ongoing convection until late autumn causes surface water to be completely transformed into bottom water over the entire sea area under study. In the estuaries, river water continues to flow into the sea providing an uninterrupted input of nutrients, mixing with and transforming into bottom water within the estuaries or in the immediate vicinity. It is worth noting that, according to the expedition data, a vertical distribution of nutrients has specific regional features (
The structure of the IC2 pattern corresponding to the autumn peak (September - November) (Fig.
Intra-annual variability of IC2 is characterised by the prevalence of its negative values in summer. Thermohaline stratification of the sea water takes place in summer significantly complicating the upward transport of nutrients into the upper layer above the thermocline. Lack of nutrients leads to phytoplankton diversity and biomass decline in the upper layer (
The IC 3-5 components describe about 23% (9.3%, 8.2% and 5.4%, respectively) of the total variance of alChla patterns. All three ICs are depicted in Figs
Maximum absolute values of the correlation coefficient between wind characteristics and independent component ICs. Coefficients are significant at 5% level.
Covariates |
IC3 |
IC4 |
IC5 |
U |
-0.38 (Apr) -0.41 (Jul) |
||
V |
0.32 (Jul) | ||
τX |
-0.31 (Jul) -0.3 (Nov) |
0.31 (Sep) | |
τY |
-0.32 (Apr) 0.33 (Jun) -0.32 (Oct) |
||
rotzτ |
-0.37 (Jul) |
0.39 (Apr) -0.42 (Jul) |
The vorticity of the wind field in the spring months is insignificant on average, whereas the meridional wind component reaches its maximum. In particular, under the influence of the meridional component of the wind, the Danube waters can spread to the shelf north of Snake Island up to the Dniester Estuary (see Fig.
In summer, the wind curl takes negative values which favours anticyclonic circulation in the north-western shelf zone. Thus, loading maps in the summer period reflect different variants of evolution of alChla patterns under the influence of the anticyclonic vorticity of the wind field (
It should be noted that mesoscale variability in the north-western shelf plays an important role in the structure of "prototypes" IC3, 4 and 5. As a result of interaction of the Black Sea Rim Current and Sevastopol Anticyclone which form a system of currents on the outer boundary of the shelf, the open sea waters expand to the north-western shelf zone. In addition, it was found out that, with considerable variability in surface currents under the influence of the rivers’ freshwater inflow from the Dnieper-Bug Liman, an anticyclonic eddy was also formed near Odessa. Such a process development pattern is quite likely. For example, in (
The IC6 component describes 6.7% of variance of alChla patterns. Despite the insignificant contribution, IC6 cannot be ignored since it reflects the dynamics of changes in the Chla pattern on the north-western shelf in the winter-spring period.
Large positive values of the wind curl indicate an increase in cyclonic circulation in the region. This type of circulation in the cold half-year determines the intensification of the local anticyclonic eddies. In the western part of the shelf, the Danube anticyclonic eddies are formed in early spring, which contribute to the spread of river waters. The region of its formation coincides with the western area of maximum matching of loadings aj on the map (see Fig.
In summer, the IC6 loading map has an opposite distribution of maxima and minima (negative IC6 values in Fig.
This study has been carried out on the basis of reconstructed arrays of Chla fields of 8-day averaging (using the DINEOF method) during the period of 1997-2016. The Chla fields were decomposed into six independent components of spatio-temporal organisation of Chla (more precisely, the logarithm of the pigment concentration alChla) in the surface layer in the shelf zone of the northwest of the Black Sea. The types reflect the spatial distribution of alChla anomalies which are likely to be formed under the influence of sea circulation factors driven by wind. However, this conclusion should be confirmed by a detailed analysis of the consistency of alChla types and current fields on the north-western shelf of the Black Sea. This work is currently under way in our laboratory and the results will be reported soon.
The paper presents the results of the analysis of the intra-annual variability of the temporal decomposition coefficients IC. The interpretation of the patterns of intra-annual IC variability is given, taking into account the seasonal variability of the wind factor, the flow of the Danube, the Dnieper and Southern Bug rivers and the fact of modulation of IC dynamics by natural seasonal phytoplankton succession. It has been found that component IC1 mainly reflects the influence of the river flow, while the effect of the wind factor is more evident in the autumn-winter period. The component IC2 significantly reflects the effect of wind circulation in the surface layer of the sea, taking into account the succession of phytoplankton. The components of IC3, 4 and 5 are correlated with the action of such a factor as the meridional component of wind velocity. The independent component of IC6 reflects the dependence of the occurrence of phytoplankton “bloom” with changing weather conditions, determining the development of coastal wind upwelling in the spring-summer period.
Application of the ICA method allows one to represent the entire variety of Chla pigment fields in the surface layer as a small number of independent components that can be interpreted. The use of these types will make it possible, in the future, to give a qualitative or quantitative assessment of the spatial "restructuring" of the Chla field under the influence of hydrodynamic factors, as well as to reveal the statistical patterns of the formation of the Chla response to this action by analysing type interchangeability or their joint repeatability.
The work was supported by the Ministry of Science and Education of Russian Federation project No. FEFM-2020-0003; and partly funded by Sevastopol State University within the framework of grant no. 42-01-09/90/2020-3.
NK - Investigation, Visualisation, Writing- Original draft preparation. VE - Supervision, Formal analysis, Methodology, Writing- review and editing.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.