Patterns of variability of sea surface chlorophyll in the Mozambique Channel: A quantitative approach Emilie Tew-Kai, Francis Marsac PII: DOI: Reference: S0924-7963(08)00315-1 doi: 10.1016/j.jmarsys.2008.11.007 MARSYS 1745 To appear in: Journal of Marine Systems Received date: Revised date: Accepted date: 18 June 2008 4 November 2008 11 November 2008 Please cite this article as: Tew-Kai, Emilie, Marsac, Francis, Patterns of variability of sea surface chlorophyll in the Mozambique Channel: A quantitative approach, Journal of Marine Systems (2008), doi: 10.1016/j.jmarsys.2008.11.007 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. ACCEPTED MANUSCRIPT Patterns of variability of sea surface chlorophyll in the Mozambique IP T Channel: a quantitative approach UR 109 THETIS, Centre de Recherche Halieutique, Avenue Jean Monnet - BP 171, Sète Cedex CE PT 34203 AC 1 IRD ED MA NU SC R Emilie Tew-Kai1 and Francis Marsac1 * Email : emilie.tewkai@ird.fr 1 ACCEPTED MANUSCRIPT Abstract We analyse the coupling between sea surface chlorophyll concentration (CC) and IP T the physical environment in the Mozambique Channel (MZC) using statistical models. Seasonal and interannual patterns are studied along with the role of mesoscale dynamics on SC R enhancement and concentration processes for phytoplankton. We use SeaWifs data for CC and two other remotely sensed data sets, TMMI for sea surface temperature (SST) and merged altimetry products for sea level anomaly and geostrophic current. Empirical NU Orthogonal Functions (EOF) on SSC and SST show strong seasonality and partition the MA MZC into three distinct sub-areas. The chlorophyll variability is mostly driven by seasonality, but more in the North (10°S-16°S) and South (24°S-30°S), and explains respectively 64% and 82% of the CC variance. In the Central part (16°S-24°S), the ED seasonal signal has less influence (60% variance). There, complex EOFs on Sea Level PT Anomaly (SLA) highlight the role of mesoscale activity (i.e. eddies and filament structures) in the spatial distribution of chlorophyll. Five mesoscale descriptors (shear, stretch, CE vorticity, deformation and eddy kinetic energy) are derived from the altimetry data to AC quantify the eddies-related physical patterns in the central region of the MZC. We use generalized Additive Models to explain the effect of those features on phytoplankton enhancement. The best model fit (r²=0.73) includes shear, stretch, vorticity and the latitudelongitude interaction as eddies are well structured in space. Cyclonic eddies associated with negative vorticity are conductive to phytoplankton enhancement by the effect of upwelling in the core notably during the spin-up phase. The interaction between eddies generate strong frontal mixing favourable to the production and aggregation of organic matter. The mesoscale activity is also affected by interannual variability with consequences on CC. We highlight a substantial reduction of the SLA pattern in 2000-2001 when the SOI positive phase is peaking (Nina-type pattern). The strong relationship between mesoscale eddies 2 ACCEPTED MANUSCRIPT and SOI suggests that primary productivity in the MZC is also under the influence of IP T distant forcing at a basin scale. Key words: Mozambique Channel, Sea Surface chlorophyll, mesoscale, seasonal AC CE PT ED MA NU SC R variability, climate forcing, quantitative approach. 3 ACCEPTED MANUSCRIPT 1. Introduction IP T The pelagic ecosystem, as reflected by the distribution of the living organisms, is heterogeneous at various space- and time-scales (Fennel 2001, Hyrenbach et al, SC R 2007; Cotté et al, 2007; Gaspar et al, 2006). At the first levels of the food web, biological patterns are strongly coupled to physical processes (Denman et al., 1977; Steele, 1989). Since the early 80s, the satellite imagery has been intensively used to NU characterize spatial patterns in the global ocean. After 1997, as the chlorophyll content MA at the sea surface became measurable routinely from the space (SeaWiFS), direct studies of the linkages between climate and biology (Yoder et al 2003, Uz and Yoder 2004, Murtugudde et al 1999, 2004) and their consequences on marine ecosystem ED dynamics (Brentnall et al 2003) could be developed. PT Phytoplankton concentration is bound to seasonal variation in temperate areas (Levy et al, 1999) and to a lesser extent in the tropical areas. But the occurrence of CE plankton patches and local blooms can occur independently of the season at mesoscale. AC Martin et al (2003) highlighted the complexity of interactions between endogenous (physiology) and external (environment) factors to maintain the local algal production. The KISS theory developed by Okubo stipulates that two antagonist processes influence the development of phytoplankton patches. Physiological processes lead the growth of algal populations, then dispersion processes decrease phytoplankton biomass from the patches. Features such as fronts, upwelling, river plumes and eddies play a key role in the dynamics and functioning of regional ecosystems (Falkowski et al 1991, Strass 1992, Flierl and Davis 1993, Dadou et al 1996, Spall and Richards 2000, Martin et al 2001). The influence of mesoscale eddies on nutrient supply in the euphotic zone, new production (McGillicuddy and Robinson 1997; McGillicuddy et al 1998) and 4 ACCEPTED MANUSCRIPT development of zooplankton communities (Huskin et al 2001, Lee and Park 2002, Kang et al 2004) has been well studied. Lima et al (2002) underline the role of the turbulent IP T eddy field, eddies’ edge and eddy-eddy interaction in boosting the primary production. In the Indian Ocean, the Mozambique Channel (MZC) is a region where the SC R mesoscale activity is strongly developed. It can even be considered as a natural laboratory to analyse the coupling between eddies and primary production. Cyclonic and anticyclonic eddies are generated in the narrower part of the MZC, at 17°S, and NU propagate southwards along the coast of Mozambique (Schouten et al. 2003; Quartly MA and Srokosz, 2004, de Ruijter et al., 2002). The MZC consists of three linked systems. In the North, the Rossby forcing dominates and generates eddies at a frequency of seven per year. Then in the Central zone of the MZC, eddies become more energetic ED and the frequency slows down to 5 per year. Finally in the South, eddies merge with PT those travelling from the southern tip of Madagascar and cause a subsequent reduction in the frequency of the mesoscale signal, at 4 per year, before they get trapped in the CE Agulhas Current. The shape of the MZC and its seabed topography mostly drive these AC peculiarities that are likely to affect the distribution of biological enrichment in different manners. So far, phytoplankton enrichment processes have been explored in the region mainly with descriptive approaches (Quartly and Srokosz 2004, Zubkov et al. 2003). The aim of this paper is to develop a quantitative analysis of the coupling between physical features and surface chlorophyll concentration at various scales (seasonal and interannual) with emphasis on the role of mesoscale eddies as drivers of the biological enrichment. In a first phase, we shall characterize the spatio-temporal variability of sea surface chlorophyll content (CC), Sea Surface Temperature (SST) and Sea Level Anomaly (SLA) in the whole Mozambique Channel, using empirical 5 ACCEPTED MANUSCRIPT orthogonal functions (EOF). In a second phase, we shall focus on the influence of 2. Data and methods SC R 2.1. Chlorophyll concentration from ocean colour data IP T mesoscale eddies on CC, using generalized additive models. The SeaWiFS global area coverage can be obtained as 8-day composites with a spatial resolution of 9 km (level 3 product). We extracted original data for the study NU area (10°S-30S / 30°E-50°E) and reflectance values were converted into chlorophyll MA concentration (Chl-a, mg.m-3) by the algorithm developed by O’Reilly et al. (1998). The study covers a period of 88 months (September 1997 to December 2004) representing an uninterrupted sequence of 335 weekly images available through the ED Goddard DAAC (http://daac.gsfc.nasa.gov). In the region of interest, the chlorophyll PT concentration (CC) exhibits large differences between coastal regions and the high sea. CC does not exceed 1 mg.m-3 in the deep sea of the MZC when concentrations in the CE coastal areas can be greater by one order of magnitude. As this study focuses on the AC processes in the deep sea, we have masked the continental shelf areas where the depth is less than 200 m. Bio-optical variability of the ocean follows a log normal distribution (Cambell 1995). Chlorophyll is a bio-optical particle that varies spatially by over three orders of magnitude. Hence, ocean colour data were log-transformed prior to any data treatment. The EOF method used in this paper requires no missing value in the data set, whereas pixels can be masked by clouds in visible remotely-sensed data. Spatial gaps were filled-in with a two-step interpolation procedure (kriging in space, then principal component analysis in time for each pixel, i.e. Eigenvector Filtering method). 2.2. Sea surface temperature data 6 ACCEPTED MANUSCRIPT We used the TRMM Microwave Imager (TMMI) data produced by the Remote Sensing Systems sponsored by the NASA Earth Science REASON DISCOVER IP T Project. The data are available weekly on a 0.25° grid. As TMI is a microwave product, the spatial resolution is much less than visible sensors (such as AVHRR) but in return, SC R the clouds have no effect on the measurement. 2.3. Mesoscale descriptors NU Sea level anomaly (SLA) and the zonal (U) and meridional (V) components of the MA geostrophic current are produced by the Ssalto/Duacs multimission project and distributed by Aviso (http://www.aviso.oceanobs.com/) with support from CNES. Merged SLA data have a resolution of 0.25 degree and merged U and V have a ED resolution of 0.33 degree. Those products are a composite of the three sensors from PT TOPEX POSEIDON, JASON and ERS-2 satellites. We used weekly time steps for 1998-2004 and we derived five parameters characterizing the dynamical velocity field. CE Vorticity is a reliable descriptor of the rotation of the water mass (Eq.1), shear (Eq.2), AC stretch (Eq.3) and deformation rate (Eq.4) are components of filament quantity (fig.2.) and EKE (Eddy Kinetic Energy) (Eq.5) is an indicator of the intensity of the movement of the water mass: Vorticity: Z Stretch:ın: V n Shear ıs: V s wv wu  (Eq.1), wx wy wu wv  (Eq.2), wx wy wv wu  (Eq.3), wx wy Deformation=ın2+ıs2 (Eq.4), Eddy Kinetic Energy: EKE (U ²  V ²) (Eq.5). 2 7 ACCEPTED MANUSCRIPT 2.4. Patterns detection The EOF analysis is used to summarize the dominant space and time patterns IP T from the series of satellite images (Nezlin and McWilliams, 2003). The multivariate gridded data used consist of time series (t) over a spatial grid (latitude l and longitude SC R L). The field F (l x L, t) is such than each rows corresponds to a time series and each column to one map. In order to save computational time and achieve a more stable and robust decomposition, we used the Singular Value Decomposition method (Venegas, NU 2001) instead of the classical covariance matrix approach. F (n, m) is composed of location and m, the time coordinate. U .S .V T ED F MA centred data (overall mean subtracted from observations) with n representing spatial U (n, m) is the left singular vectors corresponding to the eigenvectors in a classical A PT EOF (spatial mode) and V is the right singular vector, with S * V T , where A (n, n) is the temporal amplitude in a classical EOF. S k , where O k is the kth eigenvalue in a classical EOF. 2 AC that: O k CE The singular value S is proportional to the eigenvalue obtained by a classical EOF such We also used Complex EOF (or CSVD) to allow an efficient detection of propagating features, such as mesoscale eddies. This method also completes the initial SVD approach as it emphasizes signature of low amplitude modes (see Susanto et al., 1998 for a full description). In terms of computational procedure, the difference lies in the initial matrix F (n, m) which is transformed into a complex signal using the Hilbert transformation. Then U, V and S are determined as in the equation (2). The moving features of the original field are described by four parameters in the CSDV: the spatial amplitude, the spatial phase, the temporal amplitude and the phase. In this paper, we shall only represent the spatial and temporal amplitudes. The combination between the 8 ACCEPTED MANUSCRIPT spatial and temporal amplitudes, i.e. variability of a phenomenon is obtained by IP T multiplying the spatial and temporal modes (Baldacci et al., 2001). 2.5. Regression models SC R The exploration of functional linkages between sea surface chlorophyll and physical fields was undertaken with Generalized Additive Models (GAMs) (McCullagh and Nelder, 1989; Chambers and Hastie, 1992, Hastie & Tibshirani, 1990). NU GAMs are nonparametric generalizations of multiple linear regression techniques. This MA method allows in particular not to be in a rigid parametric environment, expressing the dependence enter the variable to be explained and explanatory variable. The pros of GAMs are they enable multivariate analysis with no a priori assumptions of linearity. ED Linear and additive predictors are related to the mean of Y by a link function g, which PT is monotone and non differentiable. The general formulation of the GAM is given by (Eq.6): P CE p E[Y ] D  ¦ f j (X j ) g (P ) (Eq.6), where Yi is the dependent variable for j 1 AC observation i, Xj, are covariates and fj correspond to the unknown (non-parametric) functions estimated by smoothed operators. We took the cubic regression spline with shrinkage as smoothed method. The criterion used to select the most appropriate model is the generalized cross validation (GCV), analogous to the Akaike information criterion in GLM), which is defined as: GCV nD , where n is the number of samples, D the deviance and df the (n  df ) 2 effective degrees of freedom of the model. 9 ACCEPTED MANUSCRIPT 3. Patterns of variability of temperature, chlorophyll and altimetry field IP T The pattern revealed by the first EOF mode of the sea surface temperature (SST) field holds 89.6% of the variance. The signal is essentially seasonal as reflected by the SC R time series in Fig 3b. There are clear differences from north to south of the MZC, although the SST varies in phase in the whole region as all spatial loadings have the same sign (Fig. 3a). North of 17°S (red area), the seasonal amplitude is around 4°C NU whereas a greater magnitude (5.7°C) characterizes the south west of Madagascar MA (purple area). Overall, the seasonal amplitude gradually increases from North to South of the channel. South east of Madagascar, a core area with relatively weaker seasonal amplitude is found in the lee of the southern limb of the East Madagascar current ED flowing poleward along the eastern shelf of Madagascar. The second EOF mode (not PT shown) holds 1.9% of the variance; it mostly reflects the interannual variability that is very minor here compared to the seasonal variability. CE The first EOF of the sea surface chlorophyll field explains 31.9% of the AC variance. The time sequence is driven by the seasons with a larger inter-annual variability than for SST (Fig. 4b). The highest levels of CC occurred during the southwest monsoons of 2002 and 2003. Except for areas stretching along the western shelf of Madagascar, all spatial loadings of CC are positive and indicate a phased seasonal variability in all locations (Fig 4a). The CC peaks in August-September that is the austral winter. Unlike the SST field showing a North-South gradient, CC exhibits an East-West gradient north of the latitude 24°S, with increasing CC from Madagascar to the African mainland. South of 24°S, the enrichment in phytoplankton stretches out from the South of Madagascar, where the signature of the coastal upwelling is very clear. Actually, the coastal part of the upwelling is not shown because we applied a 10 ACCEPTED MANUSCRIPT mask on the shelf areas (delineated by the isobaths 200 m) to exclude high and local values from the regional analysis. IP T The SLA field in the MZC points out very well-marked mesoscale features in the central and southern portions of the channel. SLA was processed with complex SC R EOFs (CEOF) and the first mode holds 10.4% of the variance. This mode is similar to the second mode of the classical EOF that contains only 5.4% of the variance (not shown). Compared to a classical EOF, the use of the Hilbert transformation gives a NU much better precision in delineating these moving features on a statistical basis. The MA spatial pattern is particularly clear in the central part of the MZC (Fig. 5a). It displays a corridor of 3 core eddies of opposite sign from one to another, 100 x 200 km in size, in a south-westerly direction. In the southern portion of the channel, alternating eddies of ED smaller magnitude can be discernable across the region, from the Southern tip of PT Madagascar to the African coast along 28°S. The time component of the CEOF 1 AC 2001 (Fig. 5b). CE highlights a substantial reduction of the mesoscale activity in 2000 and first semester of 4. Physical forcing on chlorophyll pattern The East-west gradient in CC mentioned earlier suggests that other drivers than the sole seasonal cycle, are involved in the spatial distribution of CC. We explored the amount of variance contained in the CC field. Three distinct areas were examined on the basis of information revealed by the spatial pattern of the first EOFs: North between latitudes 10°S-16°S, Central for 16°S-24°S and South for 24°S-30°S. We decomposed the CC field into monthly time series and seasonal time series and quantified the variability due to the seasonal signal in the whole channel and in each of the three 11 ACCEPTED MANUSCRIPT areas. Overall, the CC variability in the MZC is driven by a clear seasonal signal (68.9% of the total variance). But the Central area appears less related to the seasonal IP T cycle (60.1% of the total variance) than are the North and South areas (64 and 82% of the total variance explained by seasonality respectively). Indeed, eddies are well- SC R developed in the central area, and this simple test confirms that propagating eddies introduce a prominent non-seasonal mode into the underlying seasonal processes. An interannual mode is also evidenced in the CEOF 1 time component of the NU SLA with depressed amplitude in 2000 and 2001 (Figure 5b). Varying amplitudes are MA also noted in the EOF 1 of the CC (Figure 4b). In order to compare the interannual trends of both factors, we applied a cumulative deviation on the monthly anomalies. For CC, subtracting the seasonal component from the original series produces anomalies. ED The analysis is restricted to the central area of the MZC (Figure 6a). Both time series PT are highly correlated with reversed patterns (r=-0.87). We find two phases along the 7 years of the study period. During the first phase (1998-2000), eddies are predominantly CE anticyclonic and chlorophyll is below normal. During the second phase (mid-2001- AC 2004), cyclonic eddies are dominant and chlorophyll is enhanced. Unlike the SLA signal with comparable amplitudes in both phases, CC displays much higher amplitudes during the positive phase. In order to relate this local response to basin-scale forcing, we cumulated the Southern Oscillation Index (SOI) over time (Figure 6b). We found that the transition between two phases (mid 2000 to mid-2001) corresponds to the peak of the positive SOI phase (denoting Niña-type patterns). 5. Coupling mesoscale structures with chlorophyll distribution 5.1. Descriptive approach 12 ACCEPTED MANUSCRIPT We selected a SeaWiFS scene with minimal cloud cover (26-Feb to 5-March 1998) and the corresponding SLA scene (25-Feb 1998) to illustrate the structuring IP T effect of mesoscale features on chlorophyll concentration. The cloud cover (23.6% of the pixels) is mostly restricted to the northern part of the MZC as shown on Fig. 7a. The SC R high chlorophyll concentrations of the shelf areas along Africa and West Madagascar (> 1 mg.m-3) highlight the sharp contrast with the less productive waters of the deep sea (average 0.15 mg.m-3). The use of a 200-m depth mask is due to exclude these coastal NU systems from the analysis. Both coastal mask and chlorophyll-interpolated values under MA clouds are shown in Fig 7b. The CC field exhibits well-marked spatial features, with a succession of high (E3-E5) and low (E1-E2-E4) chlorophyll areas that are located mostly on the western part of the MZC. The CC and SLA fields (Fig. 7c) show a good ED coherence between low-chlorophyll concentrations and anti-cyclonic eddies on one PT hand, high-chlorophyll concentrations and cyclonic eddies on the other hand. Filament features are created at the periphery of anti-cyclonic eddies and seaward transport of CE shelf production of the Sofala Bank (19°S-22°S) along coast of Mozambique is also AC suggested. We shall focus the statistical analysis of the spatial coupling in the central MZC (16°S-24°S / 36°E-44°E) where responses in SLA and CC are well shown. 5.2. Statistical approach We investigated the relationships between physical descriptors and the response of chlorophyll concentration with generalized additive models. The work was initially conducted on the SeaWiFS and altimetry scenes presented in the previous section, and also on the SST scenes. In order to prepare coherent data tables containing all pixel values at the same spatial resolution, we had to downscale the initially 9-km grid of CC and the 25-km grid of SST and SLA to the 33-km grid of the U and V components of 13 ACCEPTED MANUSCRIPT the geostrophic current. The eleven variables considered in the analysis are listed in Table 1. Among those, we had to consider latitude and longitude due to the spatially IP T structured nature of the coupling. It was necessary to study the correlations between variables in order to select the SC R most appropriate set of independent variables to incorporate in the models. The correlation matrix (Spearman coefficients) points out high degree of colinearity between SST and latitude-longitude, SLA, vorticity and Chl concentration (Table 2). NU Model selection was based on the minimisation of the GCV criteria. The selected MA model is detailed in Table 3. All smoothed terms are highly significant (p<0.001). The interaction between latitude and longitude and the vorticity are the variables that mostly contribute to the deviance of the model. The selected GAM explains 75.1% of the total ED deviance. The PT Graphic representations of the GAM regression are given in Fig. 8. strongest positive responses of chlorophyll concentration are located along the western CE shelf of the narrow section of the Channel (16°S-17°S) and off Sofala Bank (19°S- AC 21°S) (Fig. 8a). The latitude-integrated response (Fig. 8b) shows an overall decreasing trend from North to South, after a maximum value at 17°S. The longitude-integrated response (Fig. 8c) is a U-shaped curve, where peak values of chlorophyll concentration are located at the slope of the two bordering continental shelves. Among the structural descriptors, the stretch (Fig. 8d) has a monotonous positive effect on CC and the shear (Fig. 8e) has a similar effect in its range of positive values. The chlorophyll concentrations decrease almost linearly with vorticity (Fig. 8f) reflecting the enhanced (depressed) primary productivity in the cyclonic (anti-cyclonic) eddies. In order to test and strengthen the results produced by this single snapshot, we computed eight other GAMs shared out among summer (February) and winter 14 ACCEPTED MANUSCRIPT (August). We only introduced mesoscale descriptors in the model, excluding latitude and longitude. We selected scenes with minimal cloud cover in order to maximize the IP T number of chlorophyll information in the analysis. The same set of mesoscale descriptors was used to compare the models. Results are presented in Table 4. The SC R percent of deviance explained by mesoscale activity is satisfactory, ranging from 13.7 to 43.7% of the total deviance, and no specific trend is noted from summer to winter. In all models, vorticity is highly significant. On the other hand, descriptors of eddy NU boundary are not significant everywhere but at least one of those is found significant in MA each model. The shape of the relationships (not shown) is similar to those found in the first example (February 1998). Overall, these results emphasize the major role played by cyclonic eddies on phytoplankton enhancement. Therefore, any change in the ED balance between cyclonic and anticyclonic eddies in the MZC may result in significant PT impacts on the dynamics of consumers. Moreover, dynamical gradients created at the AC 6. Discussion CE eddy boundary have statistically a positive influence on phytoplankton enhancement. 6.1. Influence of seasonality on phytoplankton enhancement The seasonal cycles of SST and CC in the MZC are well summarized by the EOF analysis. The seasonal signal dominates in the northern (10°S-16°S) and southern (24°S-30°S) regions, whereas the variability in the central region (16°S-24°S) is driven by the mesoscale dynamics. In the northern region of the MZC, the spatial patterns of SST and CC depict the large anti-cyclonic gyre described by several authors (Piton and Poulain 1974, Donguy and Piton 1969, 1991, Lujeharms 2004). The northern branch of the East Madagascar current and the South Equatorial current pass the north tip of 15 ACCEPTED MANUSCRIPT Madagascar and move onwards to the African coast where the current splits in two branches. The southern branch then feeds the western limb of the north basin gyre, IP T flowing poleward. Particularly noteworthy are the cooler SST and higher CC shown by the EOF spatial pattern all along the Mozambican coast to 15°S. This feature would SC R suggest the existence of an upwelling during the austral summer. At this season, the mean winds are from a south-easterly direction (Hastenrath and Lamb 1979), thus not favourable to cause an offshore Ekman transport. Hence, the current along the NU continental slope would drive the presumed upwelling. The SST in the southern and MA eastern limbs of the gyre is rather homogeneous whereas the CC is more spatiallystructured. Indeed, there is no any enrichment process occurring in the course of the current; hence, the phytoplankton crop vanishes gradually from west to east, which we ED interpret as the effect of grazing by the consumers. Another noteworthy feature is the PT dramatic increase of chlorophyll concentration along the west shelf of Madagascar during the austral summer. This enhancement is not due to an upwelling (winds are CE from the north-west and the SST is high) but to river runoff. Indeed in the north AC Madagascar, the rivers are often flooded by high precipitations in relation with the intertropical convergence zone and passing of cyclones during the warm season (Nassor and Jury, 1998). The spatial pattern of the SLA does not highlight any striking mesoscale activity in the north MZC. South of 24°S, the SST exhibits larger seasonal amplitude than in the northern region of the MZC. The chlorophyll is also found in higher concentration. The primary productivity is enhanced by three main processes: wind-induced turbulence over most of the area during the austral winter, upwelling south-east of Madagascar off Fort Dauphin (Lutjeharms and Machu, 2000; DiMarco et al., 2000, Machu et al 2002), and presumably, a cyclonic lee-eddy located in the Delagoa Bight off Maputo (Lutjeharms 16 ACCEPTED MANUSCRIPT 2006). The south-east Madagascar upwelling is mostly located on the shelf, thus it cannot be seen on the SST spatial pattern because of the relatively low spatial IP T resolution of the data in comparison to the width of the shelf (<40 km). However, the CC enrichment caused by the upwelling is discernable in the CC spatial pattern at the SC R southern bound of the shelf, where the coastal waters are advected. The core of relatively warmer waters (appearing in red in Fig. 3a), southeast of Madagascar, is located in the lee of the southern branch of the East Madagascar current. This poleward NU flowing current can be identifiable in the SST spatial pattern (yellow corridor along the MA eastern shelf of Madagascar). The origin of the SST core is uncertain but we suggest that it might be linked to the retroflection of the East Madagascar current (Lutjeharms 1988, De Ruijter et al. 2004). Along the western shelf of the South MZC, the Delagoa ED Bight is a large offset of the coastal shelf at 26°S. The southward passing eddies cause PT here a cyclonic cell and presumably, nutrient inflow in the surface waters enhancing the primary productivity. Such enhancement has been shown only intermittently (Quartly CE and Srokosz 2003). The CC spatial pattern that we present brings evidence of seasonal AC recurrent chlorophyll enrichment in this region. South of Delagoa Bight, a tongue of cool water along the shelf during the austral winter is seen in the SST spatial pattern with elevated chlorophyll concentrations associated to this cooling. The cold waters originate from the upwelling formed in the Natal Bight, between 29°S-30°S, as evidenced by Lutjeharms et al (1988, 2000) and Meyer et al (2002). The SLA spatial pattern in the southern MZC highlights a pathway of alternating cyclonic and anticyclonic eddies along 28°S. Those were described as paired vortices along 28°S formed at the south Madagascar retroflection and travelling westward to ultimately feed the Agulhas Current (de Ruijter et al 2003). 17 ACCEPTED MANUSCRIPT In the central region of the MZC the longitudinal gradients in SST and CC are caused by the passing of eddies in the western region. The SST spatial pattern reflects IP T advection of north Mozambique surface water along the Mozambique shelf as the resultant water transport is to the South (de Ruijter et al 2002). CC is higher in the SC R western “corridor” than in the eastern part of the channel. 6.2. Influence of mesoscale eddies on phytoplankton enhancement NU The GAM carried out in the central part of the channel for the week of February MA 26, 1998 demonstrated that physical quantities characterizing the core and boundary of mesoscale eddies explain the spatial distribution of phytoplankton in a significant way. Cyclonic eddies associated with negative vorticity are conductive to phytoplankton ED enhancement, by the effect of divergence at the surface and upwelling in the core of the PT eddy. Indeed, such a result was expected. But we demonstrate the role of dynamical gradients –strong positive shear and stretch- in the concentration process (elevated CC). CE Vortices alone cannot generate phytoplankton patches because of dispersing processes AC occurring in the core. The formation of persistent phytoplankton patches is the results of combined effects of vortex dynamics and filament structures generated at meso- and sub-mesoscale (Levy et al 2001) Spatial variables have also a clear impact on chlorophyll distribution as depicted by the interaction between latitude and longitude. Coastal and shelf areas are known to be more productive than offshore deep water because of land-originated nutrient inputs and tidal mixing. Although large shelf areas are present in the Eastern side of the MZC (Madagascar), the primary production is substantially higher on the Western side. Waters along the African landmass are enriched by two major sources: run-offs of the 4th largest river in Africa, the Zambezi, and mesoscale eddies passing in the west part of 18 ACCEPTED MANUSCRIPT the channel. Chlorophyll-enriched waters from the shelves are advected offshore by the “succion” effect of those eddies. These observations, and notably the distance to the IP T coast, are reflected by the longitude in the GAM analysis. Similarly, there is a chlorophyll response with latitude: CC declines in a southerly direction, along the SC R trajectory of the vortices. We therefore attempt to develop a sketch of the processes controlling phytoplankton patchiness in relation with mesoscale eddies in the MZC (Fig. 9). NU The central portion of the MZC (16°S-24°S) can be subdivided in three sub- MA systems. The first one occupies the northern part, more specifically the narrows of the channel. There, at the narrows of the Channel, eddies are at early life stages. The spinup process of cyclonic eddies is associated with nutrient inflow in the core in the ED vertical plane (McGillicuddy and Robinson, 1997; McGillicuddy et al, 1998) that PT boosts the primary production. The phytoplankton growth takes place locally: core water is being isolated from the surroundings and vertical mixing is reduced (Fennel, CE 2001; Onken, 1990). The second sub-system stretches out in the median part: it is AC characterized by eddies becoming mature as they move along the west coast. There, eddies may control the biological processes. The interaction between anticyclonic and cyclonic eddies generates high dynamical and complex barriers consisting of multiple fronts at different scales favourable to phytoplankton enhancement. Such features are known to play a major role in the production and distribution of chlorophyll surface locally (Levy et al, 2001; Martin et al, 2002; Lima et al, 2002) by allowing the exchange of organic matter between the core of the vortex and the periphery. Furthermore, theoretical studies show that the presence of a vortex in a turbulent field allows heterogeneous input of resources, and the coexistence of competitive species of phytoplankton (Bracco et al, 2000;Martin, 2003 - for example). These phenomena 19 ACCEPTED MANUSCRIPT trigger the maturation of the biological compartments from the coast to the deep sea at the edge of eddies. In some cases, the enrichment process propagates along the food IP T chain and lead to favourable foraging areas for top predators, as demonstrated for great frigatebirds, boobies and sooty terns (Weimerkirch et al, 2004, 2005; Jaquemet et al SC R 2005). The third sub-system is located in the South part. There, eddies enter a spin-off process, with decreasing energy and phytoplankton growth is much reduced. MA NU 6.3. Interannual variability of mesoscale activity and surface chlorophyll concentration Cumulative deviation has pointed out an alternation between anticyclonic (positive) and cyclonic (negative) phases from 1998 to 2004. A below normal phytoplankton ED production is associated with the positive SLA phase (1998-2000) whilst production is PT enhanced during the negative phase (2002-2004). A transition occurs from mid-2000 to mid-2001. Complex EOF computed on the SLA data points out interannual variability CE of the mesoscale activity in the MZC, with minimal activity recorded in 2000 and 2001. AC The influence of the ENSO/Dipole mode (IOD, Saji et al 1999, Webster et al 1999) might be suggested to explain this reduced mesoscale intensity around Madagascar. Palastanga et al (2006) have shown a dramatic reduction of the flow at the narrows of the MZC in 1997 and 2000-2001. They relate the less intense flow observed in 1997 to a negative IOD event, but they do not elaborate on the 2000-2001 event that was not characterized by a prominent negative IOD. Rather, we observe a shift from a moderate positive IOD (February-2000 to March-2001) to a moderate negative IOD (May to October 2001). However, the oscillation of the flow shown by Palastanga et al (2006) corresponds with the oscillation of the cumulated SOI, with lower mesoscale activity (i.e. weaker southward flow at the narrows of the Channel) during the SOI positive 20 ACCEPTED MANUSCRIPT peak, as shown in this study. These observations are a confirmation that climate forcing is an important driver of the fluctuations in the shedding rate and intensity of eddies in IP T the central MZC. SC R Conclusions Patterns of variability of Sea Surface Chlorophyll in the Mozambique Channel have been assessed. The influence of ocean processes on the phytoplankton production at different NU spatial and temporal scales has been quantified. The Channel can be decomposed in three sub MA areas. In the north and south seasonality is the dominant physical processes controlling CC. The central part differs from the two others by the passing of mesoscale eddies. In this place mesoscale activity seems to be a mainly factor of CC variability, and especially cyclonic ED vortex and edge of eddies. Cyclonic vortex allows the upwelling of nutrient and the PT enhancement of phytoplankton in the core. Interactions between eddies generate strong dynamical barriers at meso and sub mesoscale favourable to the phytoplankton. Interannual CE variability of mesoscale eddies is negatively correlated to the interannual cycle of CC, AC confirming the importance of cyclonic activity in the central part of the channel. In addition, interannual variability of Sea Level anomaly may be related to climate forcing (IOD/SOI). This implies that the productivity of the MZC regional ecosystem is also under the influence of remote forcing at a basin scale. This study shows the need to work at meso and particularly at sub-mesoscale to better understand the interactions between physical and ecological processes. Finally the better knowing of these processes appears essential to better understand the consequences of climate change on mesoscale ocean dynamics and first trophic levels. 21 ACCEPTED MANUSCRIPT Acknowledgements IP T The study was funded by the Institut de Recherche pour le Développement (IRD). The authors wish to thank Hervé Demarcq (IRD-ECO-UP) for the pre-processing of AC CE PT ED MA NU SC R SeaWiFS data. 22 ACCEPTED MANUSCRIPT References IP T Baldacci, A., Corsini.G, Grasso.R, Manzella.G, Allen.J.T, Cipollini.P, Guymer.T.H, Snaith.H.M. 2001. A Study of the Alboran sea mesoscale system by means of empirical SC R orthogonal function decomposition of satellite data. Journal of Marine Systems, 29, 293–31 NU Bracco, A., Provenzale, A., Scheuring, I. 2000. Mesoscale vortices and the paradox of MA the plankton. Proceedings of the Royal Society, B, 267, 1795-1800 Brentnall, S.J., Richards, K.J., Brindley, J., Murphy, E. 2003. Plankton patchiness and PT ED its effect on larger-scale productivity. Journal of Plankton Research, 25, (2), 121-140 Campbell, J.W. 1995.The lognormal distribution as a model for bio optical variability AC CE in the sea. Journal of Geophysical Research, 100, C7, 237-254 Chambers, J.M., Hastie, T.J., 1992. Statistical Models in S. Wadsworth and Brooks/Cole, Pacific Grove, USA. 608 pp. Cotté C., Park Y.H., Guinet C., Bost C.A. 2007. Movements of foraging king penguins through marine mesoscale eddies. Proceedings of the Royal Society B, 274, 2385-2391 Dadou, I., Garçon, V., Anderson, V., Flier, G., Davis, C. 1996. Impact of the North Equatorial Current meandering on a pelagic ecosystem: A modelling approach. Journal of marine Research, 54, 311-342 23 ACCEPTED MANUSCRIPT Denman, K.L., Okubo, A., Platt, T. 1977.The chlorophyll fluctuation spectrum in the IP T sea. Limnology and Oceanography. 22, 1033–1038. SC R De Ruijter, W., Ridderinkhof, H., Lutjeharms, J.R., Schouten, M., Veth, C. 2002. Observations of the flow in the Mozambique Channel. Geophysical Research Letters, NU 29(10. 1502, doi :10.1029/2001GL013714 MA De Ruijter, W., van Aken, H., Beier, E., Lutjeharms, J., Matano, R., Schouten, M., 2004. Eddies and dipoles around South Madagascar: formation, pathways and large- ED scale impact. Deep-Sea Research I, 51, 383-400 PT Di Marco, S. F., Chapman, P., Nowlin, W. D. 2000. Satellite observations of upwelling on the continental shelf south of Madagascar. Geophysal Research Letter, 27, 3965- AC CE 3968. Donguy, J.-R., Piton, B.,1969. A perçu des conditions hydrologiques de la partie nord du canal de Mozambique. Cahiers ORSTOM Océanographie, 7, 3-26. Donguy, J.-R., Piton, B. 1991. The Mozambique Channel revisited. Oceanologica Acta, 14, 549-558. Gaspar, P., Georges, J.Y., Fossette, S., Lenoble, A., Ferraroli, S. Le Maho, Y., 2006. Marine animal behaviour: neglecting ocean currents can lead us up the wrong track. Proceedings of the Royal Society .B 273, 2697–2702,doi:10.1098/rspb.2006.3623 24 ACCEPTED MANUSCRIPT Hastenrath, S., Lamb, P.J. 1979. Climatic Atlas of the Indian Ocean. Part I: Surface IP T Climate and Atmospheric Circulation. University of Wisconsin Press, Madison, WI, 19 SC R pp., 97 charts. Huskin, I., Anadon, R., Medina, G. et al. 2001. Mesozooplankton distribution and NU copepod grazing in the Subtropical Atlantic near the Azores: Influence of mesoscale MA structures. Journal of Plankton Research, 23, 671–691. Falkowski, P. G., Ziemann, D., Kolber, Z. et al. 1991. Role of eddy pumping in ED enhancing primary production in the ocean. Nature, 352, 55–58. PT Fennel, K., 2001. The generation of phytoplankton patchiness by mesoscale current CE patterns. Ocean Dynamics, 52, 58-70 AC Flier, G., Davis, C. 1993. Biological effects of Gulf Stream meandering. Journal of Marine Research, 51, 529-560 Hastie, T.J., Tibshirani,,R.J. 1990. Generalized Additive Models. Edited by Chapman and Hall, New York. Hyrenbach, K.D., Veit, R.R. Weimerskirch, H., Metzl, N., Hunt, Jr., G.L. 2007. Community structure across a large-scale ocean productivity gradient: Marine bird assemblages of the Southern Indian. Deep Sea Research I, 54, 1129-1145. 25 ACCEPTED MANUSCRIPT Jaquemet, S., le Corre, M., Marsac, F., Potier, M., Weimerskirch, H. 2005. Foraging habitats of the seabird community of Europa island (Mozambique Channel). Mar. Biol. IP T 147: 573-582. SC R Kang, J., Kim,W., Chang, K., Noh, J. 2004. Distribution of plankton related to the mesoscale physical structure within the surface mixed layer in the southwestern East NU Sea, Korea. Journal of Plankton Research, 26, 12, 1515-1528. MA Lee, C.R. and Park, C., 2002. Coastal upwelling of the East Sea and zooplankton distribution. Proceedings of the Spring Meeting of the Korean Society of ED Oceanography, 9–11, May 2002, Kwangju, Korea (in Korean). PT Levy, M., Klein, P., Treguier, A. 2001. Impact of sub-mesoscale physics on production and subduction of phytoplankton in an oligotrophic regime. Journal of Marine AC CE Research, 59, 535–565. Lima, I., Olson, D., Doney, S., 2002. Biological response to frontal dynamics and mesoscale variability in oligotrophic environments: Biological production and community structure. Journal of geophysical research 107, C8, 10. Lutjeharms, J.R.E., 1988. Remote sensing corroboration of retroflection of the East Madagascar Current. Deep-Sea Research, 35, 2045-2050 Lutjeharms, J. R. E., Cooper, J., Roberts, M. 2000. Upwelling at the inshore edge of the Agulhas Current. Continental Shelf Research, 20,737–761 26 ACCEPTED MANUSCRIPT Lutjeharms, J.R.E., 2004. The coastal oceans of south-Eastern Africa. The global IP T coastal interdisciplinary regional Studies and syntheses, vol 20. edited by Harward SC R University Press. Lutjeharms, J. R. E., 2006. The Agulhas Current. Edited by Springer-Verlag, Berlin, NU 329 pp MA Machu, E., Lutjeharms, J. R. E., Webb, A. M., Van Aken, H. M. 2002. First hydrographic evidence of the south-east Madagascar upwelling cell. Geophysical ED Research Letter, 29, doi: 10.1029/2002GL015381. PT Martin, A., Richards, K., Law, C., Liddicoat, M. 2001. Horizontal dispersion within an CE anticyclonic mesoscale eddy. Deep-Sea Research II, 48, 739-755 AC Martin, A., Richard, K., Bracco, A. 2002. Patchy productivity in the open ocean. Global Biogeochemical cycles, 16. Martin, A. 2003. Phytoplankton patchiness: the role of lateral stirring and mixing. Progress in Oceanography, 57, 125–174 McCullagh, P., Nelder, J.A., 1989. Generalized linear models. Chapman & Hall, London 27 ACCEPTED MANUSCRIPT McGillicuddy, D.J., Robinson, A.R., Siegel, D.A., Jannasch, H.W., Johnson, R., Dickey, T.D., McNeil, J., Michaels, A.F., Knap, A.H. 1998. Influence of mesoscale IP T eddies on new production in the Sargasso Sea. Nature, 394, 263–266 SC R McGillicuddy, D., J., and Robinson, A., R. 1997. Eddy induced nutrient supply and new production in the Sargasso Sea. Deep-Sea Research I, 44, 1427–1449. NU Meyer, A. A., Lutjeharms, J. R. E., de Villiers, S. 2002. The nutrient characteristics of Murtugudde, R. G., Wang, L. MA the Natal Bight, South Africa. Journal of Marine Systems, 35, 11–37 Hackert, E., Beauchamp, J., Busalacchi, A. 2004. ED Remote sensing of the Indo-Pacific region: ocean colour, sea level, winds and sea PT surface temperatures. Int. J. remote Sensing, 25 (7-8), 1423-1435. CE Murtugudde, R. G., Signorini, S. R., Christian, J. R., Busalacchi, A., Mc Lain, C. R, Picaut, J. 1999. Ocean color variability of the tropical Indo-Pacific basin observed by 18366 AC SeaWiFS during 1997–1998. Journal of Geophysical Research, 104 (C8), 18351– Nassor, A., Jury, M.R. 1997. Intra-Seasonal Climate Variability of Madagascar. Part 2: Evolution of Flood Events. Meteorology and Atmospheric Physics, 64, 243-254 Nezlin, N.P., Mc Williams, J.C. 2003. Satellite data, Empirical Orthogonal Functions, and the 1997-1998 El Nino off California. Remote Sensing of Environment, 84, 234254. 28 ACCEPTED MANUSCRIPT Onken, R. 1990. The creation of reversed baroclinity and subsurface jets in oceanic IP T eddies. Journal of Physical Oceanography, 20, 786–791 O’ Reilly, J.E., Maritorena, S., Mitchell, B.G., Siegel, D.A., Carder, K.L., Garver, S.A., SC R Kahru, M., McClain, C. 1998. Ocean color chlorophyll algorithm for SeaWiFS. Journal of Geophysical Research, 103, 24937-24953 NU Palastanga,V., van Leeuwen,P.J., de Ruijter,W.P.M. 2006. A link between low- MA frequency mesoscale eddy variability around Madagascar and the large-scale Indian Ocean variability. Journal of geophysical research, 111, C9, C09029. ED Piton, B., and Poulain, J. F. 1974. Résultats de mesures de courants superficiels au PT G.E.K. effectuées avec N.O. Vauban dans le sud-ouest de l’océan Indien (1973-1974). Office de la Recherche Scientifique et Technique Outre-mer, Documents Scientifique AC CE de la Mission de Nosy-Bé, 47, 14 pp. Quartly, G. and Srokosz, M.A. 2003. Visible Record of Eddies in the Southern Mozambique Channel. Geoscience and Remote Sensing Symposium Quartly, G.D., Srokosz, M.A. 2004. Eddies in the southern Mozambique Channel. Deep Sea Research II, 51 (1-3), 69-83. Saji, N.H., Goswami, B.N.,Vinayachandran, P.N., Yamagata, T. 1999. A dipole mode in the tropical Indian Ocean. Nature, 401, 360-363 29 ACCEPTED MANUSCRIPT Schouten, M.W., de Ruijter,W.P.M., van Leeuwen, P.J., Ridderinkhof, H. 2003. Eddies IP T and variability in the Mozambique Channel. Deep-Sea Research II, 50, 1987-2003 Spall, S., Richards, K., 2000. A numerical model of mesoscale frontal instabilities and SC R plankton dynamics: I. Model formulation and initial experiments. Deep-Sea Research I, 47, 1261-130 NU Steele, J.H. 1989. The ocean ‘landscape’. Landscape Ecology, 3, 185–192. MA Strass,V.H. 1992. Chlorophyll patchiness caused by mesoscale upwelling at fronts. Deep Sea Research Part A-Oceanographic research paper, 39(1A), 75-96 ED Susento, R., Zheng, Q., Yan, X. 1998. Complex Singular Value Decomposition PT Analysis of Equatorial Waves in the Pacific Observed by TOPEX/Poseidon Altimeter CE .Journal of atmospheric and oceanic technology,15 , 764-774 AC Uz, B., Yoder, J. 2004. High frequency and mesoscale variability in SeaWiFS chlorophyll imagery and its relation to other remotely sensed oceanographic variables. Deep-Sea Research II, 51, 1001–1017 Venegas, S., A., 2001. Danish Center for Earth System Science Report Yoder, J.A., Kennelly, M.A. 2003. Seasonal and ENSO variability in global ocean phytoplankton chlorophyll derived from 4 years of SeaWIFS measurements. Global biogeochemical cycles, 17(4), 1112, doi: 10.1029/2002GB0019. 30 ACCEPTED MANUSCRIPT Webster, P. W., Moore, A. M. Loschnigg, J. P, Lebben, R. R. 1999. Coupled ocean- IP T atmosphere dynamics in the Indian Ocean during 1997–98. Nature, 401, 356–360. Weimerskirch, H., Le Corre, M., Jaquemet, S., Potier, M., Marsac, F. 2004. Foraging SC R strategy of a top predator in tropical waters : great frigatebirds in the Mozambique Channel. Mar. Ecol. Prog. Ser. 275:287-308. NU Weimerskirch, H., Le Corre, M., Jaquemet, S., Marsac, F. 2005. Foraging strategy of a MA tropical seabird, the red-footed booby, in a dynamic marine environment. Mar. Ecol. Prog. Ser. 288: 251-261. ED Zubkov, M., Quartly, G. 2003. Ultraplankton distribution in surface waters of the AC CE Ecology, 33, 155-161 PT Mozambique Channel – flow cytometry and satellite imagery. Aquatic Microbial 31 ACCEPTED MANUSCRIPT Legends to Figures bathymetry. IP T Figure 1 - Major circulatory features and bathymetry in the Mozambique Channel with The main current and the mesoscale features are schematically shown. SC R Hatched areas denote upwelling. In the north of the channel, the coastal current shown is fed by the South Equatorial Current (SEC) and later depicts a large anticyclonic cell in the Comoro basin. The white area with black points represents the lee eddy off Angoche. In the NU west, along Mozambique coasts, mesoscale eddies (MCE) move in a southwesterly MA direction. In the east coast of Madagascar, the feature shown is the East Madagascar Current (EMC) and in the south, the south east Madagascar dipolar eddies (SEME) moving westward and little north ward. The mesoscale eddies from the Mozambique channel and ED the dipolar structures from the south of Madagascar reach the Agulhas Current (AC). PT (adapted from Lutjeharms, 2003 and Schouten et al, 2003). CE Figure 2 - Sketch representing the different mesoscale descriptors used. The Sla and vorticity AC are specific to the core of the eddy. The physical quantities shear, stretch, Eddy Kinetic Energy (EKE), the SST gradient and the deformation characterize the filaments and the edge of vortices. Light grey represents a cyclonic eddy (clockwise) and a dark grey an anticyclonic eddy (anti-clockwise) in the southern hemisphere. Figure 3 - Spatial pattern (a) and temporal amplitudes (b) of the first EOF computed on SST data. Percentage of variance explained is 89.6%. Figure 4 - Spatial pattern (a) and temporal amplitudes (b) of the first EOF computed on Sea Surface Chlorophyll data. Percentage of variance explained is respectively 31.9%. 32 ACCEPTED MANUSCRIPT Figure 5 - Spatial pattern (a) and temporal amplitudes (b) of the complex-EOF computed on IP T Sea Level Anomaly data. Percentage of variance explained is 10.4%. SC R Figure 6 – a) Time series of cumulative deviation of Sea Level Anomalies (solid line) and chlorophyll concentration anomalies (mg.m-3) (dotted line) in the central part of the Mozambique Channel from January 1998 to December 2004. b) Cumulative deviation of the NU Southern Oscillation Index. MA Figure 7 - a) Original map of chlorophyll concentration (log(Chla) mg.m-3) for the week 2/26/1998 with clouds (in white). b) Map of chlorophyll concentration (log (Chla) mg.m-3) ED with estimation of missing values by interpolation, and the 200-m depth mask. c) Map of Sea PT Level Anomaly (cm) for the same date; E indicates Eddy structures (E1,E2,E4 anticyclonic CE eddies and E3,E5 cyclonic eddies). AC Figure 8 - Partial residuals of the nonlinear terms estimated by means of generalized additive models (GAM) with smoothing splines. The appropriate smoothness for each applicable model term was selected using generalized cross validation (GCV). a) The map displays to the non linear relation between the longitude (long) - latitude (lat) interaction and the sea surface chlorophyll. Mean smooth terms are summarised on (b) the latitude and (c) the longitude. Effect of the stretch (d), Shear (e) and vorticity (f) on SSC. Figure 9 - Sketch of the different phases of eddies in the Mozambique Channel and their influence on the phytoplankton. 1) In the narrow part of the channel, eddies are being spinned 33 ACCEPTED MANUSCRIPT up, and highly energetic. The cyclonic eddies can then generate primary production in the centre by upwelling. 2) Vortices moving along the African coast, reaching a mature stage. IP T Anticyclonic eddies export offshore matter from the coast. The succession of cyclonic and anticyclonic eddies generates strong boundary gradients. These eddy-eddy interactions lead to SC R concentrate biomass at eddies’ periphery. Physical processes are conducive to the maturation of the system. 3) Finally, in the south, eddies are being spinned off and join the Agulhas AC CE PT ED MA NU current. 34 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 1 35 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 2 36 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 3 37 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 4 38 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 5 39 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 6 40 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 7 41 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 8 42 ACCEPTED MANUSCRIPT AC CE PT ED MA NU SC R IP T Fig 9 43 ACCEPTED MANUSCRIPT Table 1. List of the environmental variables used as input for the models. MA ED PT CE AC 44 IP T Units °C cm cm2.s-2 s-1 s-1 s-1 s-1 degree degree SC R Symbol SST SST gradient SLA EKE def SS SN vort Lat long NU Explicative variables Temperature Gradient of temperature Sea Level anomaly Eddy kinetic Energy deformation Shear Stretch Vorticity Latitude Longitude ACCEPTED MANUSCRIPT Table 2. Correlation analysis between the 11 explanatory variables and the sea surface chlorophyll (sw). 1.00 0.30 0.13 0.76 -0.02 0.18 0.14 -0.19 -0.08 0.20 1.00 0.17 0.19 -0.62 0.07 0.22 0.27 -0.58 0.08 1.00 0.05 -0.15 0.17 -0.02 0.10 -0.05 0.06 sst sla 1.00 0.01 0.06 0.14 -0.02 -0.04 0.02 1.00 -0.05 -0.28 -0.37 0.84 0.14 eke ss sn vort def IP T gradsst SC R sw NU lat CE PT ED MA long 1.00 0.41 0.07 0.03 0.61 -0.04 -0.30 0.14 -0.07 0.00 -0.23 AC long lat sw gradsst sst sla eke ss sn vort def 45 1.00 -0.04 1.00 0.02 0.00 1.00 -0.07 -0.24 -0.27 0.14 -0.11 0.06 1.00 0.20 1.00 ACCEPTED MANUSCRIPT Parametric coefficients: Estimate -2.09 Approximate significance of smooth terms: SC R t value -149.00 IP T Table 3. Coefficients of the selected GAM. Smooth terms are represented using penalized regression spline with smoothing parameters selected by GCV (generalized cross validation).Parametric coefficients represent the linear terms the model, edf are the estimated degrees of freedom of the smooth terms using cubic regression spline with shrinkage. Pr(>|t|) <2e-16 AC CE PT ED MA NU Edf F p-value deviance 13.25 s(ss) 5.10 36.10 1.700E-04 29.03 s(sn) 0.94 4.00 3.230E-04 40.95 s(vort) 8.89 12.50 < 2e-16 48.05 te(lat:long) 4.98 8.70 < 2e-16 R-sq.(adj) = 0.731 Deviance explained = 75.1% deviance null =53.12 GCV score = 0.04216 46 ACCEPTED MANUSCRIPT FEB2004 14.8 24.3 *** *** ** ** * AUG2000 7 18.3 ** ** ** ** MA ED PT CE AC 47 winter AUG2001 AUG2002 7.7 14.16 14.5 13.9 *** *** * * SC R summer FEB2001 FEB2002 23.4 18.15 13.7 43.7 *** ** ** *** * * ** ** ** *** NU % of clouds % od devience gradsla ss vort gradsst sn def FEB2000 27.35 31.9 *** *** *** * ** IP T Table 4. Results of 8 GAMs, computed on 8 SeaWIFs scenes in the Central part of the Mozambique Channel during winter and summer. Least cloudy scenes were selected for this analysis. p<0.001;**p<0.01;*p<0.05. AUG2004 6.6 34 *** * *** *** *