This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright Author's personal copy Remote Sensing of Environment 115 (2011) 2471–2485 Contents lists available at ScienceDirect Remote Sensing of Environment j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e Remote sensing chlorophyll a of optically complex waters (rias Baixas, NW Spain): Application of a regionally specific chlorophyll a algorithm for MERIS full resolution data during an upwelling cycle Evangelos Spyrakos a, Luis González Vilas a, Jesus M. Torres Palenzuela a,⁎, Eric Desmond Barton b a b Remote Sensing and GIS Laboratory, Department of Applied Physics, Sciences Faculty, University of Vigo, Campus Lagoas Marcosende, 36310, Vigo. Spain Instituto de Investigaciones Marinas, C/Eduardo Cabello, 6, E-36208, Vigo, Spain a r t i c l e i n f o Article history: Received 17 January 2011 Received in revised form 5 May 2011 Accepted 7 May 2011 Available online 12 June 2011 Keywords: Chlorophyll a MERIS Algorithms Upwelling Galician rias a b s t r a c t This study takes advantage of a regionally specific algorithm and the characteristics of Medium Resolution Imaging Spectrometer (MERIS) in order to deliver more accurate, detailed chlorophyll a (chla) maps of optically complex coastal waters during an upwelling cycle. MERIS full resolution chla concentrations and in situ data were obtained on the Galician (NW Spain) shelf and in three adjacent rias (embayments), sites of extensive mussel culture that experience frequent harmful algal events. Regionally focused algorithms (Regional neural network for rias Baixas or NNRB) for the retrieval of chla in the Galician rias optically complex waters were tested in comparison to sea-truth data. The one that showed the best performance was applied to a series of six MERIS (FR) images during a summer upwelling cycle to test its performance. The best performance parameters were given for the NN trained with high-quality data using the most abundant cluster found in the rias after the application of fuzzy c-mean clustering techniques (FCM). July 2008 was characterized by three periods of different meteorological and oceanographic states. The main changes in chla concentration and distribution were clearly captured in the images. After a period of strong upwelling favorable winds a high biomass algal event was recorded in the study area. However, MERIS missed the high chlorophyll upwelled water that was detected below surface in the ria de Vigo by the chla profiles, proving the necessity of in situ observations. Relatively high biomass “patches” were mapped in detail inside the rias. There was a significant variation in the timing and the extent of the maximum chla areas. The maps confirmed that the complex spatial structure of the phytoplankton distribution in the rias Baixas is affected by the surface currents and winds on the adjacent continental shelf. This study showed that a regionally specific algorithm for an ocean color sensor with the characteristics of MERIS in combination with in situ data can be of great help in chla monitoring, detection and study of high biomass algal events in an area affected by coastal upwelling such as the rias Baixas. Published by Elsevier Inc. 1. Introduction Although remote sensing tools can be used with a relatively high precision at global scale for the calculation of chlorophyll a (chla), they are not always totally accurate in local areas (Cota et al., 2004; Ruddick et al., 2008) and highly dynamic systems such upwelling regimes. Eastern boundary upwelling systems cover a small percentage of the ocean surface, but account more than 20% of the global fish catch. In these high productive systems harmful algal events due to toxic phytoplankton species and/or high-biomass blooms pose an increasing threat for aquaculture and fishing industries, ecosystem health and diversity and have possible implications for human health and activities (Trainer et al., 2010). Harmful algal events in eastern boundary ⁎ Corresponding author. Tel.: + 34 986 812 631; fax: + 34 986 812 556. E-mail address: jesu@uvigo.es (J.M. Torres Palenzuela). 0034-4257/$ – see front matter. Published by Elsevier Inc. doi:10.1016/j.rse.2011.05.008 upwelling systems have been closely associated with the wind properties (Bates et al., 1998; Pitcher et al., 1998) and have become a focal point of numerous studies (e.g. Fawcett et al., 2007; Kudela et al., 2005; Pitcher & Nelson, 2006). For example, in their review on harmful algal events in upwelling systems, Pitcher et al. (2010) noticed that variations in wind-stress fluctuations and buoyancy inputs in upwelling systems are controlling factors of the bloom timing and pointed out the role of inner-shelf dynamics on the spatial distribution of the bloom. Upwelling waters are characterized by considerable variability in the vertical distribution of phytoplankton (Brown & Hutchings, 1987) and in their optical properties (Morel & Prieur, 1977). Optically active water constituents such as SPM which are brought into the surface because of the strong mixing and vertical advection that takes place during upwelling events may vary independently of the surface chla, as they do in typically shallow estuarine case II waters. In typical case II waters the traditional satellite-derived chla models (empirical: Aiken et al., 1995; Brown et al., 2008; Evans & Gordon, 1994; Author's personal copy 2472 E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 McClain et al., 2004; Muller-Karger et al., 1990; O'Reily et al., 2000 and semi-analytical: Carder & Steward, 1985; Carder et al., 1999) based on the ratio between the radiance of blue and green light reflected by the surface waters cannot be used for an accurate retrieval of chla (Gitelson et al., 2007; Gons, 1999; Morel & Prieur, 1977). However, chla algorithms that use green to red and near infrared band ratios have shown good performance in inland and coastal waters (Gilerson et al., 2010). In the effort for more accurate retrieval of water constituents in optically complex waters, neural network (NN) techniques can play an important role, since they seem ideal for multivariate, complex and non-linear data modeling (Thiria et al., 1993). Dransfeld et al. (2004) emphasized the role of NNs in the retrieval of water constituents especially in Case 2 waters. In recent decades the application of neural network (NN) techniques for the estimation of selected water quality parameters from ocean-color has increased (Atkinson & Tatnall, 1997; Dzwonkowski & Yan, 2005; Keiner & Yan, 1998; Shahraiyni et al., 2009; Zhang et al., 2003). NN based algorithms are currently used as standard products for the estimation of chla, suspended particulate matter (SPM) and yellow substances by the European Space Agency (ESA) for Medium Resolution Imaging Spectrometer (MERIS) data (Doerffer & Schiller, 2007, 2008). The Galician rias are V-like embayments along the northwest part of the Iberian Peninsula formed by sunken river valleys flooded by the sea, whose ecosystems are strongly influenced by oceanic conditions on the adjacent continental shelf. Interest in developing an accurate estimation of chla in these rias is considerable, mainly because of the economic and social importance of the extensive culture of mussels (Rodríguez Rodríguez et al., 2011), and the frequent occurrence of harmful algal events (GEOHAB, 2005). Although MERIS is an ocean color sensor with characteristics considered suitable for chla monitoring and detection of HABs in coastal areas (Doerffer et al., 1999), to our knowledge the studies using MERIS data in the Galician rias are limited to those of Torres-Palenzuela et al. (2005a, b), Spyrakos et al. (2010) and González Vilas et al. (2011). The latter authors developed a chla algorithm based on NNs and classification techniques from MERIS full resolution data for rias Baixas coastal waters. Previous ocean color studies by satellite sensors (CZCS, SeaWiFS, MODIS) during active upwelling in the Iberian system (Bode et al., 2003; Joint et al., 2002; McClain et al., 1986; Oliveira et al., 2009a, 2009b; Peliz & Fiuza, 1999; Ribeiro et al., 2005) played an important role in the identification of chla patterns and study of harmful algal blooms and primary production but were restricted to the ocean shelf because of insufficient spatial resolution. Another problem that affected many of these previous satellite remote sensing studies in the area was the failure of the algorithms used to provide reliable chla data during upwelling favorable conditions especially in the areas closest to the coast. In the present paper a set of neural network-based chla algorithms previously developed for the Galician rias waters (within the rias and for coastal waters on the continental shelf) are applied for the first time in a short series of MERIS (FR) images delivered during an upwelling cycle in order to obtain maps of chla. This study tests the potential of the algorithms to map the spatial extent of possible algal blooms caused by coastal upwelling. Also, the temporal and spatial distributions of the chla patterns, captured in the MERIS images using the local adapted algorithm, are discussed in relation to the meteorological and oceanographic conditions in the area. Finally, the performance of the neural network-based chla algorithm is compared to in situ measurements. 2. Methods and data 2.1. Description of the study area The rias Baixas constitute the southern part of the Galician rias (Fig. 1). They are formed by four large coastal embayments, from north to south: Muros y Noya, Arousa, Pontevedra and Vigo, all oriented in a SW–NE direction, and characterized by strong tides. Surface area covers approximately 600 km 2 and water depths range from 5 to 60 m. This study focuses on three rias (Arousa, Pontevedra and Vigo), each connected to the open sea through two entrances, to the north and south of the islands located at the external part of each ria. The ria de Vigo is the longest of the rias whereas the ria de Arousa is the widest. Rias vary in width from 1–3 km in their inner part to 8–12 km in their external part (Vilas et al., 2005). The main freshwater inputs in the rias are by rivers that located in innermost part of the rias. In these highly primary productive upwelling estuarine systems (Fraga, 1981; Spyrakos et al., in press; Torres & Barton, 2007) transient increases of phytoplankton abundance, referred to as blooms, are a frequent phenomenon occurring mainly between early spring and late fall (Figueiras & Ríos, 1993; Fraga et al., 1988; Varela, 1992). Sporadically, some phytoplankton blooms in the Galician rias are perceived as harmful with direct and indirect impacts to the mussel production that constitute an important economic activity in the area. Harmful algal events in the Galician rias are a well documented phenomenon. Several studies since the 1950s referred to the harmful algal events and in general to phytoplankton ecology on the Galician rias particularizing favorable conditions to the development of HABs, their origin, dynamic, distribution and toxicity levels (Figueiras et al., 1994; GEOHAB, 2005; Margalef, 1956; Tilstone et al., 1994), seasonal taxonomic and chemical composition of phytoplankton and picophytoplankton “patchiness” (Figueiras & Niell, 1987; Nogueira et al., 1997; Tilstone et al., 2003). More specifically, spring and summer upwelling events in the area have been associated with the dominance of diatoms including the potentially toxic, chain-forming diatom Pseudo-nitzschia spp. (Figueiras & Ríos, 1993). Harmful events due to the Paralytic shellfish toxin (PST) producer Gymnodinium catenatum have been occasionally (1976: Estrada et al., 1984; 1985: Laiño, 1991; 2005: Bravo et al., 2010b) or/and annually (1985–1995: Pazos et al., 2006) recorded in the Galician rias. It is generally considered (Fraga et al., 1990; Figueiras et al., 1996) that advection of warmer waters from the shelf into the rias at the end of the upwelling season coincides with the highest abundances of G. catenatum. Pazos et al. (2006) observed a northward progression of G. catenatum along the Iberian Peninsula starting on the Portuguese coast, suggesting this could provide an early notice of G. catenatum events in the Galician rias. 2.2. Sampling regime Two samplings were conducted in 2008 in the ria de Vigo. Twelve fixed stations were visited on cloud-free days (July 9 and 22). The sampling transect was extended from the open sea towards the inner part of the ria. Satellite data from MERIS (FR) were available for the same days. The depth of the sites, where samples were collected, ranged from 5 m inside the ria to 100 m outside. Triplicate water samples from surface to 3 m were collected at each station (Fig. 1B) from a sampler (3524 cm3) for the determination of chla and SPM. 2.3. In situ measurements In situ chla fluorescence profile was monitored by a Turner designs CYCLOPS-7 submersible fluorometer. Profiles of water temperature were provided by a portable meter (HI 9829, Hanna instruments). The depth of the euphotic zone was established with a Secchi disk. For the High Performance Liquid Chromatography (HPLC) chla determination, water samples (100–200 mL) were filtered through a 9 mm diameter Whatman GF/F filters and stored at − 80 °C for 2 weeks, and 95% methanol was used as extraction solvent for the pigments. In this study only chla concentration data are presented, calculated as the sum of chlorophyllide a, chlorophyll a epimer, chlorophyll a allomer and divinyl chlorophyll a. An HPLC method using a reversed phase C8 was applied for the separation of the pigments. Details of pigment extraction and separation are provided in Zapata et al. (2000). Author's personal copy E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 2473 Fig. 1. A) Galician coast and bathymetry of the area. From north to south the rias Baixas: Muros y Noya, Arousa, Pontevedra and Vigo. The location of the Seawatch buoy station off Cabo Silleiro is shown by a black rectangle. B) Map of ria de Vigo showing the locations of the sampling stations. The MERIS FR pixel size is presented in relation to the size of the ria. Author's personal copy 2474 E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 Table 1 MERIS imagery showing the acquisition time (UTC) and mean view zenith angle from west. Sea-truthing mean values (± S.D.) of chlorophyll a (chla), suspended particulate matter (SPM), percentage of inorganic contribution to SPM and Secchi disk depth (Zsd) for ria de Vigo (12 stations) during the two samplings. Chla (mg m− 3) SPM (mg L− 1) Inorganic matter (%) Zsd (m) 0.74 ± 0.80 1.85 ± 0.30 48.17 ± 6.21 7.65 ± 3.30 0.97 ± 0.82 2.01 ± 0.65 48.20 ± 7.70 4.45 ± 1.66 MERIS FR July 03 2008 July 09 2008 July 16 2008 July 19 2008 July 22 2008 July 29 2008 Acquisition time (UTC) View zenith angle (°) 10:59 13.5 11:10 13.0 10:50 20.7 10:56 15.3 11:02 11.7 10:42 20.7 SPM was evaluated in terms of SPM concentration and percent weight of organic matter (%OM). Pre-combusted (450 °C for 24 h), pre-washed in 500 mL of MilliQ, 47 mm Whatman GF/F filters were used. These filters were then dried at 65 °C to a constant weight. Particles were collected by filtering a standard volume (1000 mL) of seawater samples and then rinsed with 50 mL MilliQ in order to remove salts and dissolved organic matter. For the determination of SPM the filters were dried at 65 °C till no weight changes were observed. The filters were then re-combusted at 450 °C for 5 h in order to obtain the inorganic suspended matter (ISM). The percent weight of organic matter (%OM) was determined by subtracting the ISM from the SPM. All the filters were weighted on a Precisa 262 SMA-FR microbalance (10− 5 g precision). 2.4. Oceanographic and meteorological data Oceanographic and meteorological data off the rias Baixas were provided by the Spanish Port System (www.puertos.es). More specifically, wind speed (W) and direction, current data and water temperature were observed at a Seawatch buoy station located off Cape Silleiro (42° 7.8′N, 9° 23.4′W) (Fig. 1). This meteorological station was selected as fairly representative of the study area (Herrera et al., 2005). Daily upwelling index (IW) was estimated from wind by Bakun's (1973) method: IW = −τy = ðρW ·f Þ = −1000·ρa ·CD ·WWy = ðρW ·f Þ 3 m = ðs·kmÞ ð1Þ where τy is the alongshore component of wind stress N m− 2, ρW is the density of seawater (1025 kg m − 3), f is the Coriolis parameter (9.9 · 10− 5 s − 1 at 42° latitude), ρa is the density of air (1.2 kg m − 3 at 15 °C), CD is an empirical dimensionless drag coefficient (1.4· 10− 3 according to Hidy, 1972) and W and Wy are the average daily modulus and northward component of the wind. Moderate Resolution Imaging Spectroradiometer (MODIS-Aqua) sea surface temperature (SST) daily level 2 data for July 2008 were downloaded from the website of the National Aeronautics and Space Administration Goddard Space Flight Center (NASA-GSFC) (http:// oceancolor.gsfc.nasa.gov/). The 1 × 1 km resolution MODIS data were processed using MATLAB software to derive projected SST maps of the study area. The study area for the SST maps was expanded to 42–43° N and 9.3–8.3° W. The MODIS imagery contains 6 images for the dates that MERIS data were available. 2.5. MERIS data and MERIS chla algorithms for rias Baixas 2.5.1. Regional neural network for rias Baixas (NNRB) Developed by González Vilas et al. (2011), this set of algorithms represents feed-forward NNs trained by supervised learning using iterative back-propagation of error for the retrieval of chla from MERIS FR data. These algorithms approximate sets of different classes of waterleaving radiance reflectances data, determined after the application of Fuzzy c-means clustering techniques (FCM), to a set of appropriate chla concentrations. Input variables are 11 MERIS water leaving radiance reflectance and 3 geometry values. It was found that MERIS data can be classified in 3 clusters (#1, #2 and #3) but only one could be used for the 3 different NNs that were developed for the retrieval. The method performs well in the estimation of chla from MERIS (FR) data in the optically complex waters of the rias Baixas and detects accurately the peaks of chla. NNRB is based on in situ chla data collected from the rias Baixas during a long period survey (2002–2008), covering the temporal variability of chla in all the part of the rias. In contrast with the Doerffer and Schiller algorithm, this algorithm does not use simulated data. The result is a narrower range (0.03–7.73 mg m − 3), but this is considered as sufficient for the study area. 2.5.2. Application of chla algorithms to MERIS imagery The MERIS satellite imagery used in this study contains 6 fullresolution level-1b images derived from the area in July 2008. MERIS overpasses were within 2 h of the time that samples and data were collected in situ. Beam 4.2 (Brockmann Consult and contributors, Germany) software was used for the analysis of the imagery. The BEAM-4.6′s smile correction was applied to the original level-1b data. For the atmospheric correction the ocean color data were processed with a NN-based algorithm implemented alongside the MERIS Case-2-Regional Processor (C2R) which was developed by Doerffer and Schiller (2008). This NN algorithm for dedicated atmospheric correction over turbid Case 2 waters is based on radiative transfer simulations. The performance test of the atmospheric correction showed increasing uncertainty with decreasing values of water leaving radiance reflectances (Doerffer & Schiller, 2008). The flags for coastline, land, clouds and invalid reflectance were raised using the Beam software. Ocean color data derived from areas significantly affected by sun glint (beyond a solar zenith angle limit of 60°) were characterized invalid and removed from the analysis. The FCM algorithm that was proposed by González Vilas et al. (2011) was applied to the level-2 reflectance data in order to identify the different clusters. Classification images were then obtained for the available MERIS images using the same FCM algorithm. The pixels in these images were assigned to the cluster with the highest value in its corresponding membership function and the percentage of pixels belonging to each cluster was computed. The performance of the available algorithms was then tested and the NN with the best performance measures was applied to the MERIS leaving radiance reflectance values in order to deliver the chla maps for the study area. Chla data points delivered from cloud-free scenes and areas that were not flagged for coastline and invalid reflectance were considered to be valid match-up data and were used for the performance testing of the chla algorithms. Water-leaving radiance reflectances and chla concentrations were computed as mean values of the pixel corresponding to the sampling station location and the 8 surrounding pixels. These 9 pixels cover approximately 0.8 km 2 of surface area and it was considered that this averaging was able to reduce MERIS instrument noise. Although MERIS spatial resolution of chla is considered suitable to study coastal areas there are cases (e.g. Kutser, 2004) where variability occurs even within one MERIS pixel. Chla spatial variability in the Galician rias has not been studied in sufficient detail to determine sub-pixel variability, but available information Author's personal copy E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 2475 Fig. 2. Plots of chorophyll a fluorescence (mg m− 3) vertical profiles for the upper 10 m of the water column in ria de Vigo on A) July 09 2008 and B) July 22 2008. from previous campaigns in the ria de Vigo using MiniBAT undulating vehicle and NERC-CASI instrument (http://www.iim.csic.es/~barton/ cria) and previous studies on the spatial distribution of algae (dinoflagellates and diatoms) blooms from station samples in the area (Figueiras & Ríos, 1993; Bravo et al., 2010b) show variability well represented by MERIS. For each sampling point, the number of pixels included in the median computation was also extracted as a quality flag, ranging from 9 (highest quality) to 1 (lowest quality). Low quality values indicate that the sampling station is located in the proximity of the coast or cloudy or foggy areas, so that the reflectance values could be affected. The imagery was then remapped using the standard Mercator projection with a fixed grid of 890 by 890 pixels. Each chla image ranges from 42° 04′ N to 42° 40′ N latitude and from 8° 32′W to 9° 32′ W longitude, which covers approximately 3.1 × 10 3 km 2. 2.6. Performance measures The following statistical measurements were used to evaluate the performance of the chla models. For the measured chla concentration (Chlx) and the chla retrieved by the models (Chlx̂) the difference PEi = Chlxi −Chl x̂i ð2Þ was noted and was used to obtain the root mean square error (RMS error) and the relative RMS error, which are defined as: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ∑Ni= 1 PEi2 RMS error = N ð3Þ Author's personal copy 2476 E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 Fig. 3. Regression analysis between Total Suspended Material (TSM, also termed Suspended Particulate Matter) and chla concentrations. (y = 1.76 + 0.13x, R2 = 0.1 and sample size N = 41). sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi RMSE Rel:RMS error = 1 N 100 ∑ i Chlxi N ð4Þ In addition, the coefficient of determination (R 2) was computed as a measurement of the correlation between the Chlx and the Chlx̂. RMSE and relative RMSE were used in this work as measurements of absolute error (in mg m − 3) and relative error respectively. 3. Results and discussion 3.1. In situ data Sea-truthing mean values of HPLC chla, SPM, percentage of inorganic matter and Secchi disk depth for the two samplings are given in Table 1, which also summarizes information about the available MERIS imagery. Water temperatures near surface ranged from 16.90 to 19.54 °C and from 16.58 to 18.91 °C, respectively, for the two samplings. Temperature at 10 m depth during the first campaign was between 15.33 and 17.70 °C, whereas temperatures dropped to 13.50–14.47 °C at the sampling stations on July 22. Water temperature decreased from the outer part towards the inner part of the ria. Values of Secchi disk depth between 2 and 12 m were measured in the ria de Vigo, generally less than half the water column depth. Chla concentration in the surface water samples did not show a wide variation. Chla levels were relatively low in comparison with the temporal pattern proposed for the rias Baixas by Nogueira et al. (1997) where chla concentrations close to 5 mg m − 3 are described as typical during the summer period. Mean chla determined by HPLC varied from 0.03 at station 12 to 2.65 mg m − 3 in the inner part during the first sampling. On July 22 the highest chla concentration (2.72 mg m − 3) was recorded in the innermost ria station. Although the range in the surface chla concentration was similar in both samplings, differences were observed in the chla profiles. In the sampling conducted on July 9, small differences were observed in the chla concentration profiles in the first 10 m of the water column in all stations except the three at the inner part of the ria where a chla maximum was recorded at 4 m depth (Fig. 2A). On the other hand, a vertical gradient of chla was detected in almost all sampling stations during the second sampling (Fig. 2B), with the highest values of chla (up to 16 mg L − 1) found close to 10 m. The same vertical distribution pattern during the month of July in ria de Pontevedra is described in Varela et al. (2008) and is imputed to the presence of upwelled waters. This depth-varying chla distribution may affect the remote sensing reflectance spectra especially at stations where light penetrates deeper into the water column. Kutser et al., 2008 showed that different vertical distributions of the same total cyanobacterial biomass may strongly impact the remote sensing signal. SPM concentrations varied from 1.17 to 3.15 mg L − 1 in the ria de Vigo and showed decreasing values with distance from St. 1, which is located in the inner, narrow part of the ria and closer to the main freshwater inputs. In this part of the ria sediment resuspension and continental runoff are probably higher having as a result high concentrations of SPM. The results of the Chla and SPM analyses of this study combined with available unpublished data from the same sampling stations using the same methodology showed that these two variables vary independently (Fig. 3, determination coefficient of linear relationship R 2 = 0.1). This confirms the initial assumption that rias Baixas waters can be categorized as Case 2 (Morel & Prieur, 1977). This classification is not always a simple distinction of coastal and oceanic waters as Morel and Maritorena (2001) describe in a later study using a new dataset of optical properties, indicating the need for models more restricted in geographical and seasonal terms. 3.2. Classification results and comparison of MERIS chla algorithms with in situ data Classification images, showing the cluster value for each sea pixel, were obtained for the MERIS images involved in the analysis (Fig. 4). In theory, the membership grades for each cluster would allow us to blend the chla concentration, obtained from the different neural networks developed for each cluster, into a given pixel, so that chla maps with soft transitions would be created (Moore et al., 2009). In practice, the NNRB model was only developed for Cluster#1 (González Vilas et al., 2011), so that these classification images presented here were only useful for detecting the zones where Cluster#1 is the dominant cluster and therefore the areas where NNRB can be best applied to obtain more reliable results. Fig. 4 shows that Cluster#1 is dominant in almost all the images in the rias Baixas and the adjacent area. Table 2 shows the percentage of pixels belonging to each cluster for each image over the rias Baixas. Cluster#1 includes the majority of the pixels in ria de Vigo, with more than 72% of pixels in the six images. On average, 71% and 65% of the pixels over ria de Pontevedra and ria de Arousa belong to this cluster. Cluster#2 is the predominant one in the ria de Pontevedra and ria de Arousa in the image delivered on July 29. Cluster#3 is the least abundant in most of the images, with less than 3.25% of pixels in all of them. However, the presence of Cluster#2 and Cluster#3 does not prevent the continuous chlorophyll mapping over large areas in the rias, because of the predominance of Cluster#1. The image acquired on the July 29 is more problematic (referring to the high percentage of pixels belonging to Cluster#2), although the mapping of a large part of ria de Vigo and small parts of ria de Pontevedra and ria de Arousa was possible. Insofar as the two cloud-free field campaigns were designed specifically to collect samples within a time period of 2 h from the MERIS overpasses time, 24 valid data were available to test performance of the MERIS chla algorithms. The performance Fig. 4. Classification of MERIS images derived from the study area. The 3 classes identified using the FCM are shown. Author's personal copy E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 2477 Author's personal copy 2478 E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 Table 2 Percentage of pixels belonging to each cluster over the study area (rias Baixas), obtained from classification images derived from the MERIS images used in this study. Date ria Cluster#1 Cluster#2 Cluster#3 July 03 2008 Vigo Pontevedra Arousa Vigo Pontevedra Arousa Vigo Pontevedra Arousa Vigo Pontevedra Arousa Vigo Pontevedra Arousa Vigo Pontevedra Arousa 74.22 62.50 76.33 95.40 96.14 94.24 84.52 70.48 49.56 82.17 83.68 75.68 87.41 90.06 82.92 72.43 24.17 11.67 23.95 37.35 22.55 4.04 3.57 3.01 15.48 28.78 47.19 17.66 16.18 22.48 11.88 9.79 15.86 27.57 75.40 86.60 1.83 0.15 1.13 0.55 0.30 2.75 0.00 0.73 3.25 0.17 0.15 1.84 0.71 0.15 1.22 0.00 0.43 1.73 July 09 2008 July 16 2008 July 19 2008 July 22 2008 July 29 2008 Table 3 Performance parameters for the Chla neural networks tested in this study. Chla model Data base R2 RMS error (mg m− 3) Relative RMS error % NNRB#1 NNRB#2 NNRB#3 Whole Cluster#1 Cluster#1 High quality 0.17 0.24 0.70 0.74 0.69 0.46 93 85 65 Table 4 Dominant atmospheric and oceanographic conditions off the rias Baixas categorized as three different states during the upwelling cycle in summer 2008. Period Date 1 2 3 Dominant atmospheric and oceanographic conditions off rias Baixas July 1–10 Winds blowing mainly in south direction (Iw = − 108) after a period of favorable upwelling winds, mostly northward surface flow July 11–21 Strong north winds (Iw = 900), surface flow towards southwest July 22–31 Mainly south blowing winds (Iw = − 230), southward surface flow parameters for the match-up data of MERIS chla retrieved by the three NNRB algorithms (NNRB#1, NNRB#2 and NNRB#3) are shown in Table 3. The available dataset did not show a wide range of environmental variation, with chla ranging from 0.03 to 2.72 mg m − 3. NNRB for the Cluster#1, high quality data produced well correlated results (R 2 = 0.70). Bottom effects on the reflectance or the presence of macroalgae and adjacency effects might be the factors responsible for the difference in the performance parameters that were observed between NNRB#1/NNRB#2 and NNRB#3. The good performance of NNRB#3 is not surprising considering that NNRB#3 can clearly follow the cycle of chlorophyll recorded in the rias Baixas: concentrations lower than 1 mg m − 3 during the winter months, up to 8 mg m − 3 during the spring and autumn maxima and close to 5 mg m − 3 during the summer. Moreover, the NNRB algorithm is trained with MERIS and in situ chla data during upwelling events. NNRB#3 seems to be robust and ideal for the rias Baixas coastal waters where it can be used for a more accurate mapping of chla in order to improve the understanding of the spatial and temporal distributions. 3.3. Upwelling cycle Different meteorological and oceanographic periods were identified and categorized as three different states in the area during July of 2008 (Table 4, Fig. 5). The states lasted from 9 to 11days which is typical in an upwelling cycle in the area (Nogueira et al., 1997). 3.3.1. State 1 (July 1–11) This 10 d period state comes after a strong upwelling that occurred in the area at the end of June (Fig. 5A) and it is characterized mainly by weak winds of variable direction which are typical of upwelling relaxation in the area (deCastro et al., 2004). An exception of strong downwelling-favorable wind from the south was recorded on July 4. Surface flow off the rias had a northward direction with a speed ranging between 0.5 and 7.5 cm s − 1 (Fig. 5B). On July 3, SST ranged between 16 and17 °C inside the rias, but an area with temperature higher than 17 °C was observed outside the rias. In the next SST image (July 9) an increase in temperature was recorded in the rias Baixas and the adjacent area (Fig. 6). The temperature increase was confirmed by the Seawatch data (Fig. 5C). The daily mean water temperature off the rias Baixas increased from 16.4 °C, in the first days of July up to 18 °C in a period of 10 days after the upwelling. Two MERIS (FR) images (Fig. 7) from the study area were available during State 1, one on July 3 and the other on July 9. In both, several high chla“patches” were mapped inside and in the outer parts of the rias. In the area off the external coast of the rias the chla concentration in the images remained at levels close to 0 mg m − 3. This pattern of the phytoplankton biomass principally confined in the rias while in neighboring shelf area the chla levels remained very low, seems to be generated by the northward flow of surface waters outside the ria. The development of northward currents in the relaxation following intense north winds, responsible for the upwelling recorded at the end of June, may introduce water of high chla to the three rias from the ocean area outside them in the first days of July. The continuing mainly north-westward directed transport over several days may have been responsible for the chla distribution observed on July 9, where chla concentration was significantly higher in the ria de Arousa than in the other two rias in the south. Different patterns of the higher chla areas in the rias were mapped during State 1. On July 3, areas of high chla concentrations were mapped in the ria de Arousa (3–4.5 mg m − 3) and close to the mouths of the three rias (higher than 2 mg m − 3). On July 9, chla concentrations greater than 2 mg m − 3 were mapped mainly in the middle and inner parts of the ria de Vigo and in the outer part of the ria de Arousa. In the image obtained on July 3, areas of high chla concentrations were observed in the outer part of the three rias, whereas chla decreased towards the inner part of the rias. Varela et al. (2008) reported that this gradient is common in the ria de Pontevedra during the upwelling period when the meteorological forcing is the main factor responsible for the circulation of the ria. Six days after the first available image, the gradient of chla in the rias described above was observed only in the ria de Arousa. On the contrary, Vigo and Pontevedra were characterized by a chla gradient where concentration increased toward inshore. In the ria de Pontevedra areas of higher chla were recorded at the innermost part and close to the northern mouth of the ria. In the rest of the ria de Pontevedra chla concentration was close to 0 mg m − 3. In the inner part of ria de Vigo, Fig. 5. A) Daily upwelling index off the rias Baixas. The Iw in m− 3 s− 1 100 m− 1 represents the offshore Ekman flux in the surface layer. Arrows indicate the days where MERIS FR images were available, numbers define the different states identified during the studied period, and red dots denote the days of sampling. B) Surface currents (cm s− 1) recorded by the Seawatch buoy station located off Cape Silleiro (42° 7.8′N, 9° 23.4′W). Data were daily averaged from 0 a.m. on July 1 2008 to 12 p.m. on July 31 2008. Symbols shown in Fig. 2. C) Daily average of sea surface temperature for the month of July 2008 off the rias Baixas. All data are means ± 1 S.D. Author's personal copy E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 2479 Author's personal copy 2480 E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 Author's personal copy E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 2481 Fig. 7. Chla maps for MERIS FR data derived during the upwelling cycle of July 2008 in the study area. White color indicates masking of land and clouds. MERIS chla varied between 2 and 3 mg m− 3. MERIS data delivered from areas like the most interior, shallow part of the rias normally considered as suspicious because the high abundance of macroalgae increases the chla signal (Gons, 1999) were here characterized as reliable, since they were confirmed by in situ data. Moreover, water transparency in the 20 m station (St. 1) as determined by the Sechhi disk measurements during the first campaign was 2 m, decreasing the effect of the bathymetry. This part of the rias can be firmly considered as estuary and during nutrient enrichment from river flows, high concentrations of chla have been recorded (Evans & Prego, 2003). In this case the observed relatively high concentrations of chla at the inner part of the two southern rias may be the result of the mixture of estuarine water with Eastern North Atlantic Central Water (ENACW) combined with high residence times. The different offshore–inshore gradient of ria de Arousa seems to be formed by material transferred to the north from the rias de Vigo and Arousa due to the northward surface currents. Differences in topography and local winds should also be considered as possible factors for the observed differences. Ria de Arousa is considered to be the most productive of the rias Baixas (Bode & Varela, 1998). In the classification of Vidal-Romaní (1984)ria de Arousa is categorized as open bay, while ria de Pontevedra and ria de Vigo as fjord-like. Though fjords have deep quiescent interiors, only intermittently renewed, and a shallow sill at the entrance, while the rias are shallow and have a 2 layer circulation that reverses between up and downwelling. 3.3.2. State 2 (July 12–21) State 2 was characterized by 9 d of sustained upwelling favorable winds (Fig. 5A) and southwest currents of up to 6 cm s − 1 (Fig. 5B). SST maps showed that temperature ranged between 16 and 17 °C in the coastal area outside the rias (Fig. 6). Temperature recorded by the Seawatch buoy decreased more than 1 °C during the upwelling (Fig. 5C). The two chla maps (Fig. 7) for this state trace the primary results of the upwelling favorable winds. On July 16 map areas with the highest chla concentrations were recorded in the middle part of ria de Pontevedra, at the mouth of ria de Arousa and through the entire ria de Fig. 6. MODIS-derived Sea Surface Temperature maps for rias Baixas and adjasted coastal waters during the upwelling cycle of July 2008. White patches represent clouds. Author's personal copy 2482 E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 Fig. 8. RGB MERIS (l2) FR composite image acquired on July 22 2008 over the study area. Land was masked in black. Vigo. The distributions were similar in form in the ria de Vigo and ria de Pontevedra but higher chla (N2.5 mg m − 3) was found in the former. Unpublished data showed the outflow of ria water towards offshore in speeds that reached 4 cm s − 1. This situation of the surface water being advected offshore in the rias, when upwelling favorable wind started to blow off the rias Baixas is typical of the positive estuarine circulation that has been described in the area (Figueiras & Pazos, 1991; Fraga & Margalef, 1979). In this two-layer circulation, the offshore surface Ekman-transport advects the low salinity water out of the ria, while the denser upwelled water flows into the ria along the sea bed. The zone of enhanced surface chla concentration that in the MERIS images extend throughout the ria de Pontevedra and ria de Vigo is probably surface water that is flowing out of the rias due to the positive estuarine circulation generated during the upwelling favorable conditions. The July 19 chla image shows a noticeable increase in chla with concentrations higher than 1 mg m − 3 over the entire continental shelf zone, although chla decreased slightly within the rias. In the study of Ospina-Álvarez et al. (2010) it was found out that during the upwelling favorable conditions that characterized the Northern Galician rias during the period of July 13–22 2008 the ENACW did not enter in the rias. While in that period chla in the Northern Galican rias did not exceed the value of 1 mg m − 3 (Ospina-Álvarez et al., 2010), in the rias Baixas it was generally higher than 1 mg m − 3. 3.3.3. State 3 (July 22–31) As a result of the strong upwelling event a peak of chla with concentrations up to 5 mg m − 3 was mapped on July 22 in the coastal area off Galicia. The high chla concentration was extended from the northern offshore area to the interior of the rias (Fig. 7). A coincident area of relatively low temperature was mapped in the north part of the study area, whereas an area of warmer water was detected at the south. Differences up to 2 °C were obtained between the rias (Fig. 6). This alongshore difference probably reflects the persistence of stronger coastal upwelling in the north after the event of 10–21 July and an earlier onset of relaxation in the south. It is often the case that upwelling is more persistent in the north of the area (Torres & Barton, 2007). Fig. 8 shows the development of the upwelling on the Galician coast. With the abrupt decrease of upwelling-favorable to zero wind on July 22, currents at the Seawatch buoy became briefly northward as expected, but subsequently returned to southward despite the onset of intermittent northward winds. The last MERIS image (July 29) is consistent with strong relaxation: the offshore region has near-zero chlorophyll and a region of moderately high chla is bound to the coast. Within the rias values tend to be low, reflecting downwelling conditions. It seems probable that more flow inshore of the Seawatch buoy was northward and convergent to shore. At the end of July chla in the ria de Vigo showed the lowest concentration of all the images of previous days. The high chla concentrations along the Galician shelf coupled with low SST. MERIS and MODIS images at the start of this state on July 22 show clearly the presence of a cold, chlorophyll-rich area resulting from the previous 10 days of upwelling. Although high chla water was recorded below the surface in the in situ profiles (Fig. 2B) during the second sampling in the ria de Vigo, MERIS data recorded the low surface values present. Figueiras and Pazos (1991) noted the presence of nutrient-rich water during a summer upwelling event in the rias Baixas that did not reach the surface. As soon as upwelling ceases, the 2-layer circulation reverses and surface waters flow inwards and sink to the lower layer carrying with them the higher surface concentrations of Author's personal copy E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 chla. The possible non uniformity of the Inherent Optical Properties (IOP) in the water profiles (Stramska & Stramski, 2005) and the development and validation of the water constituent algorithms based on water samples from certain depths (e.g. González Vilas et al., 2011; O'Reily et al., 2000) affirms the necessity of the in situ data. Although this high biomass area was not sampled directly, in situ data from the ria de Vigo revealed relatively high concentrations of diatoms (mainly Chaetoceros spp.) and small flagellates (personal observation). The potentially domoic acid producing Pseudo-nitzschia spp. was also present in the ria de Vigo but in relatively low concentrations. This phytoplankton composition seems to be typical in the rias Baixas during the summer according to the annual cycle of phytoplankton abundance proposed in 1987 by Figueiras and Niell. Moreover, Frangópulos et al. (2011) mentioned the presence of the red-tide dinoflagellate Noctiluca scintillans in high abundances in ria de Vigo during summer 2008. 4. Summary and conclusions Three different states of meteorological and oceanographic periods were identified in the area during the July of 2008. Surface currents and winds off the rias Baixas affected the distribution of chla in the rias Baixas. At the beginning of July (State 1) the variable and weak wind and the resulting northward surface currents limited the high chla concentrations to the rias so that only low chla values were found in the offshore area. Differences in the topography of the rias, effects of local winds and transport by currents between the rias seem to be the main factors for the observed differences in the gradients of chla along the rias between ria de Arousa and the two southern rias (Vigo and Pontevedra). MERIS images obtained during State 2 showed the first response of chla distribution due to the strong favorable winds that were blowing in the area. With the development of strong upwelling the circulation in the rias is reinforced in the estuarine sense so that chla increases rapidly there. After a period of six days of continued upwelling, chla concentrations higher than 1 mg m − 3 were observed in all the area mapped according to MERIS data. State 3 commences with the appearance of a high biomass algal event coincident with the area of low SST as the culmination of the preceding, extended upwelling. The weak northward winds that characterized this state permitted downwelling that transferred chla rich water toward the rias. The upwelled water was recorded in the chla profiles below the surface in the ria de Vigo but was missed by the MERIS. The continuing downwelling circulation resulted in decay of the bloom and subduction of surface waters in the rias compatible with the decrease of chla observed in the last MERIS image. Although MERIS has a repeated interval of 3 days, cloud cover prevented acquisition of all possible images. Nevertheless the 6 images obtained in July 2008 captured the main changes in chla concentration and distribution during the three periods of different meteorological and oceanographic states. The application of FCM revealed the areas where the NNRB for chla retrieval can be best applied to obtain the most reliable results. Even if NNRB#1 and #2 did not give reliable results, NNRB#3 showed good performance indices and seems suitable for chla determination in the area. The chla concentrations observed in this study fall into the scope of the NNRB algorithms proposed by González Vilas et al. (2011). According to the data recorded in situ, our models perform better than the C2R algorithm proving the necessity for regionally specific models. The present study allows more detailed examination of the chla distribution and detection of high biomass “patches” in the Galician rias and the adjacent area during a summer upwelling cycle due to the finer spatial resolution and precise atmospheric correction offered by MERIS. The application of an algorithm specially developed for the study area provides more accurate mapping of chla, which has, for the first time to our knowledge, provided surface chla mapping of the interior of the rias Baixas. 2483 There was a significant variation in the timing and the extent of the chla peak areas. The maps confirmed that the complex spatial structure of the phytoplankton distribution in the rias Baixas is affected by the surface currents and winds on the adjacent continental shelf. Field studies have limited spatial coverage and temporal frequency. Some of these areas of high chla and the dynamic changes in chla distribution that are apparent in satellite images can be missed by in situ monitoring. High chla levels in the rias due to the increase in the concentration of harmful phytoplankton species have been recorded in the past especially in summer (GEOHAB, 2005). Moreover, some potentially toxic species such as Pseudo-nitzschia spp., which form blooms in the study area and in other upwelling systems, can be found in high concentrations within these high biomass phytoplankton “patches”. The approach followed in this study can be particularly useful in the case of blooms of Gymnodynium catenatum. The progression of G. catenatum blooms from the Portuguese coast to the Galician rias can be tracked by MERIS and current data, providing useful advanced information, of importance to the local mussel industry. It is worth mentioning that a harmful algal event due to G. catenatum caused the closure of the shellfishery in the Galician rias from October 2005 till February 2006 (Caballero Miguez et al., 2009). An example of a localized feature is that constantly high surface chla was observed in the Bay of Baiona, located in the southern mouth of ria de Vigo (Fig. 1A). This bay is characterized as a zone where harmful algal events due to species like Alexandrium minutum are a frequent and recurrent phenomenon (Bravo et al., 2010a). It is worth noting that for the area seaward of the rias all the algorithms used in this study came up with very similar values and patterns for chla. Overall, this study showed that the synergy of two space borne sensors (MERIS, MODIS) in combination with in situ data can be of great help in the monitoring, detection and study of high biomass algal events in an coastal upwelling areas. Acknowledgments MERIS data were obtained through ESA/ENVISAT project AO-623. We are very grateful to A. Acuña and D. Perez Estevez for their helpful assistance during the field work. This work was partially funded by the European Commission's Marie Curie Actions (project 20501 ECOsystem approach to Sustainable Management of the Marine Environment and its living Resources [ECOSUMMER]) through a grant supported ES. Part of this work has been financed by European Commission's EUFAR/RIAWATER project and by the Xunta de Galicia project PDIGIT05RMA 40201PR. We would like to thank the two anonymous reviewers for their very thorough review and detailed suggestions. References Aiken, J., Moore, G. F., Trees, C. C., Hooker, S. B., & Clark, D. K. (1995). The SeaWiFS CZCStype pigment algorithm. In S. B. Hooker, & E. R. Firestone (Eds.), NASA technical memorandum, 104566. SeaWiFS Technical Report Series, vol. 29. (pp. 1–32). Atkinson, P. M., & Tatnall, A. R. L. (1997). Introduction Neural networks in remote sensing. International Journal of Remote Sensing, 18, 699–709. Bakun, A. (1973). Coastal upwelling indices, west coast of North America 1946–71. NOAA Technical Report NMFS-SSRF 671 (pp. 1–103). Seattle: US Dept. Commerce. Bates, S. S., Garrison, D. L., & Horner, R. A. (1998). Bloom dynamics and physiology of domoic-acid producing Pseudo-nitzschia species. In D. M. Anderson, A. D. Cembella, & G. M. Hallegraeff (Eds.), Physiological ecology of harmful algal blooms (pp. 267–292). Heidelberg, Germany: Springer-Vergal. Bode, A., Carrera, P., & Lens, S. (2003). The pelagic foodweb in the upwelling ecosystem of Galicia (NW Spain) during spring: natural abundance of stable carbon and nitrogen isotopes. ICES Journal of Marine Sciences, 60, 11–22. Bode, A., & Varela, M. (1998). Primary production and phytoplankton in three Galician Rias Altas (NW Spain): seasonal and spatial variability. Scientia Marina, 62(4), 319–330. Bravo, I., Figueroa, R. I., Garcés, E., Fraga, S., & Massanet, A. (2010a). The intricacies of dinoflagellate pellicle cysts: the example of Alexandrium minutum cysts from a bloom-recurrent area (Bay of Baiona, NW Spain). Deep Sea Research Part II: Topical Studies in Oceanography, 57, 166–174. Author's personal copy 2484 E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 Bravo, I., Fraga, S., Figueroa, R. I., Pazos, Y., Massanet, A., & Ramilo, I. (2010b). Bloom dynamics and life cycle strategies of two toxic dinoflagellates in a coastal upwelling system (NW Iberian Peninsula). Deep-Sea Research Part II: Topical Studies in Oceanography, 57, 222–234. Brown, C. A., Huot, Y., Werdell, P. J., Gentili, B., & Claustre, H. (2008). The origin and global distribution of second order variability in satellite ocean color and its applications to algorithm development. Remote Sensing of Environment, 112, 4186–4203. Brown, C., & Hutchings, L. (1987). The development and decline of phytoplankton blooms in the Benguela upwelling system. 1. Drogue movements, hydrography and bloom development. South African Journal of Marine Science, 5, 171–193. Caballero Miguez, G., Garza Gil, M. D., & Varela Lafuente, M. M. (2009). The institutional foundations of economic performance of mussel production: the Spanish case of the Galician floating raft culture. Marine Policy, 33, 288–296. Carder, K. L., Chen, F. R., Lee, Z. P., Hawes, S. K., & Kamykowski, D. (1999). Semianalytic moderate-resolution imaging spectrometer algorithm for chlorophyll a concentration and absorption with bio-optical domains based on nitrate-depletion temperatures. Journal of Geophysical Research, 104, 5403–5421. Carder, K. L., & Steward, R. G. (1985). A remote-sensing reflectance model of a red-tide dinoflagellate off west Florida. Limnology and Oceanography, 30, 286–298. Cota, G. F., Wang, J., & Comiso, J. C. (2004). Transformation of global satellite chlorophyll retrievals with a regionally tuned algorithm. Remote Sensing of Environment, 90, 373–377. deCastro, M., Gómez-Gesteira, M., Alvarez, I., & Prego, R. (2004). Negative estuarine circulation in the Ria of Pontevedra (NW Spain). Estuarine, Coastal and Shelf Science, 60, 301–312. Doerffer, R., & Schiller, H. (2007). The MERIS Case 2 water algorithm. International Journal of Remote Sensing, 28, 517–535. Doerffer, R., & Schiller, H. (2008). MERIS regional coastal and lake case 2 water project— Atmoshperic correction ATBD. GKSS Research Center 21502 Geestacht. Germany Version 1.0, 18 May 2008. Doerffer, R., Sorensen, K., & Aiken, J. (1999). MERIS potential for coastal zone application. International Journal of Remote Sensing, 20(9), 1809–1818. Dransfeld, S., Tatnall, A. R., Robinson, I. S., & Mobley, C. D. (2004). A comparison of Multi-layer multilinear regression algorithms for the inversion of synthetic ocean colour spectra. International Journal of Remote Sensing, 25(21), 4829–4834. Dzwonkowski, B., & Yan, X. -H. (2005). Development and application of a neural network based ocean colour algorithm in coastal waters. International Journal of Remote Sensing, 26, 1175–1200. Estrada, M., Sanchez, F. J., & Fraga S. (1984). Gymnodinium catenatum (Graham) en las rias gallegas (NO de Espana). Investigación Pesquera, 48, 31–40. Evans, R. H., & Gordon, H. R. (1994). Coastal zone color scanner system calibration: a retrospective examination. Journal of Geophysical Research, 99, 7293–7307. Evans, G., & Prego, R. (2003). Rias, estuaries and incised valleys: is a ria an estuary? Marine Geology, 196, 171–175. Fawcett, A., Pitcher, G. C., Bernard, S., Cembella, A. D., & Kudela, R. M. (2007). Contrasting wind patterns and toxigenic phytoplankton in the southern Benguela upwelling system. Marine Ecology Progress Series, 348, 19–31. Figueiras, F. G., Gómez, E., Nogueira, E., & Villarino, M. L. (1996). Selection of Gymnodinium catenatum under downwelling conditions in the Ría de Vigo. In T. Yasumoto, Y. Oshima, & Y. Fukuyo (Eds.), Harmful and toxic algal bloom (pp. 215–218). Paris, France: IOC/ Unesco. Figueiras, F. G., Jones, K. J., Mosquera, A. M., Alvarez-Salgado, X. A., Edwards, E., & MacDougall, N. (1994). Red tide assemblage formation in an estuarine upwelling ecosystem: Ria de Vigo. Journal of Plankton Research, 16, 857–878. Figueiras, F. G., & Niell, F. X. (1987). Composición del fitoplankton de la ría de Pontevedra (NO de España). Investigacion Pesquera, 51, 371–409. Figueiras, F., & Pazos, Y. (1991). Microplankton assemblanges in three Rías Baixas (Vigo, Arosa and Muros, Spain) with a subsurface chlorophyll maximum: their relationships to hydrography. Marine Ecology Progress Series, 76, 219–233. Figueiras, F. G., & Ríos, A. F. (1993). Phytoplankton succession, red tides and the hydrographic regime in the Rías Bajas of Galicia. In T. J. Smayda, & Y. Shimizu (Eds.), Toxic Blooms in the Sea (pp. 239–244). New York: Elsevier. Fraga, F. (1981). Upwelling off the Galician coast North West Spain. In E. Suess, & J. Thiede (Eds.), Coastal upwelling (pp. 176–182). New York: Plenum Publishing Corp. Fraga, S., Anderson, D. M., Bravo, I., Reguera, B., Steidinger, K. A., & Yentsch, C. M. (1988). Influence of upwelling relaxation on dinoflagellates and shellfish toxicity in Ria de Vigo, Spain. Estuarine Coastal Shelf Science, 27, 349–361. Fraga, F., & Margalef, R. (1979). Las Rias Gallegas. In Estudio y explotacion del mar en Galicia (pp. 101–122), University of, Santiago (Ed.). Fraga, S., Reguera, B., & Bravo, I. (1990). Gymnodinium catenatum bloom formation in the Spanish rías. In E. Graneli, B. Sundstrom, L. Edler, & D. M. Anderson (Eds.), Toxic marine phytoplankton (pp. 149–154). New York: Elsevier. Frangópulos, M., Spyrakos, E., & Guisande, C. (2011). Ingestion and clearance rates of the red Noctiluca scintillans fed on the toxic dinoflagellate Alexandrium minutum (Halim). Harmful Algae, 10, 304–309. GEOHAB (2005). Global Ecology and Oceanography of Harmful Algal Blooms (GEOHAB). In G. Pitcher, T. Moita, V. Trainer, R. Kudela, F. G. Figueiras, & T. Probyn (Eds.), GEOHAB Core Research Project: HABs in Upwelling Systems (pp. 82). Paris and Baltimore: IOC and SCOR. Gilerson, A. A., Gitelson, A. A., Zhou, J., Gurlin, D., Moses, W. J., Ioannou, I., et al. (2010). Algorithms for remote estimation of chlorophyll-a in coastal and inland waters using red and near-infrared bands. Optical Express, 18, 24109–24125. Gitelson, A. A., Schalles, J. F., & Hladik, C. M. (2007). Remote chlorophyll-a retrieval in turbid, productive estuaries: Chesapeake Bay case study. Remote Sensing of Environment, 109, 464–472. Gons, H. (1999). Optical teledetection of chlorophyll a in turbid inland waters. Environmental Science and Technology, 33, 1127–1132. González Vilas, L., Spyrakos, E., & Torres Palenzuela, J. M. (2011). Neural network estimation of chlorophyll a from MERIS full resolution data for the coastal waters of Galician rias (NW Spain). Remote Sensing of Environment, 115, 524–535. Herrera, J. L., Piedracoba, S., Varela, R. A., & Roson, G. (2005). Spatial analysis of the wind field on the western coast of Galicia (NW Spain) from in situ measurements. Continental Shelf Research, 25, 1728–1748. Hidy, G. M. (1972). A view of recent air–sea interaction research, Bulletin of American Meteorological Society, 53, 1083-1102. Joint, I., Groom, S., Wollast, R., Chou, L., Tilstone, G., Figueiras, F., et al. (2002). The response of phytoplankton production to periodic upwelling and relaxation events at the Iberian shelf break: estimates by 14C method and by satellite remote sensing. Journal of Marine Systems, 32, 219–238. Keiner, L. E., & Yan, X. H. (1998). A neural network model for estimating sea surface chlorophyll and sediments from thematic mapper imagery. Remote Sensing of Environment, 66, 153–165. Kudela, R., Pitcher, G., Probyn, T., Figueiras, F., Moita, T., & Trainer, V. (2005). Harmful algal blooms in coastal upwelling systems. Oceanography, 18, 184–197. Kutser, T. (2004). Quantitative detection of chlorophyll in cyanobacterial blooms by satellite remote sensing. Limnology and Oceanography, 49, 2179–2189. Kutser, T., Metsamaa, L., & Dekker, A. G. (2008). Influence of the vertical distribution of cyanobacteria in the water column on the remote sensing signal. Estuarine Coastal and Shelf Science, 78, 649–654. Laiño, C. C. (1991). Gymnodinium catenatum toxins from mussels (mytilus galloprovincialis). Environmental Technology, 12, 33–40. Margalef, R. (1956). Estructura y dinámica de la “purga de mar” en Ría de Vigo. Investigacion Pesquera, 5, 113–134. McClain, C. R., Chao, S. -Y., Atkinson, L. P., Blanton, J. O., & Castillejo, F. (1986). Winddriven upwelling in the vicinity of Cape Finisterre, Spain. Journal of Geophysical Research, 91, 8470–8486. McClain, C. R., Feldman, G. C., & Hooker, S. B. (2004). An overview of the SeaWiFS project and strategies for producing a climate research quality global ocean biooptical time series. Deep-Sea Research II, 51, 6–42. Moore, T. S., Campbell, J. W., & Dowell, M. D. (2009). A class-based approach to characterizing and mapping the uncertainty of the MODIS ocean chlorophyll product. Remote Sensing of Environment, 113, 2424–2430. Morel, A., & Maritorena, S. (2001). Bio-optical properties of oceanic waters: a reappraisal. Journal of Geophysical Research, 106, 7163–7180. Morel, A., & Prieur, L. (1977). Analysis of variations in ocean color. Limnology and Oceanography, 22, 709–722. Muller-Karger, F. E., McClain, C. R., Sambrotto, R. N., & Ray, G. C. (1990). A comparison of ship and coastal zone color scanner mapped distribution of phytoplankton in the southeastern Bering Sea. Journal of Geophysical Research, 95, 483–499. Nogueira, E., Perez, F. F., & Ríos, A. F. (1997). Seasonal patterns and long-term trends in an estuarine upwelling ecosystem (Ria de Vigo, NW Spain). Estuarine Coastal and Shelf Science, 44, 185–300. Oliveira, P., Moita, T., Silva, A., Monteiro, I., & Sofia Palma, A. (2009a). Summer diatoms and dinoflagellate blooms in Lisbon Bay from 2002 to 2005: pre-conditions inferred from wind and satellite data. Progress in Oceanography, 83, 270–277. Oliveira, P., Nolasco, R., Dubert, J., Moita, T., & Peliz, A. (2009b). Surface temperature, chlorophyll and advection patterns during a summer upwelling event off central Portugal. Continental Shelf Research, 29, 759–774. O'Reily, J. E., Maritorena, S., Siegel, D. A., O'Brien, M. C., Toole, D., Mitcell, B. G., et al. (2000). In S. B. Hooker, & E. R. Firestone (Eds.), Ocean color chlorophyll a algorithms foe SeaWiFS, OC2 and OC4. Version 4. NASA technical memorandum 2000–206892, vol. 11, 49 pp. Ospina-Álvarez, N., Prego, R., Álvarez, I., deCastro, M., Álvarez-Ossorio, M. T., Pazos, Y., et al. (2010). Oceanographic patterns during a summer upwelling–downwelling event in the Northern Galician Rias: comparison with the whole Ria system (NW of Iberian Peninsula). Continental Shelf Research, 30, 1362–1372. Pazos, Y., Moroño, A., Triñanes, J., Doval, M., Montero, P., Vilarinho, M. G., et al. (2006). Early detection and intensive monitoring during an unusual toxic bloom of Gymnodinium catenatum advected into the Galician Rías (NW Spain). Proceedings of 2006 International Conference on Harmful Algae, September 4–8. Copenhagen, Denmark. Peliz, A., & Fiuza, A. (1999). Temporal and spatial variability of CZCS-delivered phytoplankton concentrations off the western Iberian Peninsula. International Journal of Remote Sensing, 20, 1363–1407. Pitcher, G. C., Boyd, A. J., Horstman, D. A., & Mitchell-Innes, B. A. (1998). Subsurface dinoflagellate populations, frontal blooms and the formation of red tide in the southern Benguela upwelling system. Marine Ecology Progress Series, 172, 243–264. Pitcher, G. C., Figueiras, F. G., Hickey, B. M., & Moita, M. T. (2010). The physical oceanography of upwelling systems and the development of harmful algal blooms. Progress in Oceanography, 85, 5–32. Pitcher, G. C., & Nelson, G. (2006). Characteristics of the surface boundary layer important to the development of red tide on the southern Namaqua shelf of the Benguela upwelling system. Limnology and Oceanography, 51, 2660–2674. Ribeiro, A., Peliz, A., & Santos, A. M. P. (2005). A study of the response of chla-biomass to a winter upwelling event off western Iberia using SeaWiFS and in situ data. Journal of Marine Systems, 53, 87–107. Rodríguez Rodríguez, G., Villasante, S., & García-Negro, M. (2011). Are red tides affecting economically the commercialization of the Galician (NW Spain) mussel farming? Marine Policy, 35, 252–257. Ruddick, K., Lacroix, G., Lancelot, C., Nechad, B., Park, Y., Peters, S., et al. (2008). Optical remote sensing of the North Sea. In V. Barale, & M. Gade (Eds.), Remote Sensing of the European Seas (pp. 79–90). New York: Springer-Verlag. Author's personal copy E. Spyrakos et al. / Remote Sensing of Environment 115 (2011) 2471–2485 Shahraiyni, H. T., Shouraki, S. B., Fell, F., Schaale, M., Fischer, J., Tavakoli, A., et al. (2009). Application of an active learning method to the retrieval of pigment from spectral remote sensing reflectance data. International Journal of Remote Sensing, 30, 1045–1065. Spyrakos, E., González Vilas, L., & Torres Palenzuela, J. (2010). Development and application of neural network-based chlorophyll-a algorithms from MERIS fullresolution data for optically complex waters. Proceedings of Remote sensing and photogrammetry society annual conference with Irish earth observation symposium, Septermber 1–3. Cork, Ireland. Spyrakos, E., Santos-Diniz, T. C., Martinez-Iglesias G., Torres-Palenzuela, J. M., & Pierce, G. J. (2011). Spatiotemporal patterns of marine mammal distribution in coastal waters of Galicia, NW Spain. Hydrobiologia, 670, 87–109. Stramska, M., & Stramski, D. (2005). Effects of a nonuniform vertical profile of chlorophyll concentration on remote-sensing reflectance of the ocean. Applied Optics, 44, 1735–1747. Thiria, S., Mejia, C., Badran, F., & Crepon, M. (1993). A neural network approach for modelling nonlinear transfer functions: Application for wind retrieval from spaceborne scatterometer data. Journal of Geophysical Research, 98(12), 22827–22841. Tilstone, G. H., Figueiras, F. G., & Fraga, F. (1994). Upwelling–downwelling sequences in the generation of red tides in a coastal upwelling system. Marine Ecology Progress Series, 112, 241–253. Tilstone, G. H., Figueiras, F. G., Lorenzo, L. M., & Arbones, B. (2003). Phytoplankton composition, phytosynthesis and primary production during different hydrographic conditions at the Northwest Iberian upwelling system. Marine Ecology Progress Series, 252, 89–104. Torres, R., & Barton, E. D. (2007). Onset of the Iberian upwelling along the Galician coast. Continental Shelf Research, 27, 1759–1778. 2485 Torres-Palenzuela, J. M., Vilas-González, L., & Mosqura-Giménez, Á. (2005a). Correlation between MERIS and in-situ data for study of Pseudo-nitzschia spp. toxic blooms in Galician coastal area. Rotterdam: Millpress. Torres-Palenzuela, J. M., Vilas-González, L., & Mosqura-Giménez, Á. (2005b). Detection of Pseudo-nitzschia spp. toxic blooms using MERIS images on the Galician coast (pp. 1915–1919). : European Space Agency (Special Publication). Trainer, V. L., Pitcher, G. C., Reguera, B., & Smayda, T. J. (2010). The distribution and impacts of harmful algal bloom species in eastern boundary upwelling systems. Progress in Oceanography, 85, 33–52. Varela, M. (1992). Upwelling and phytoplankton ecology in Galician (NW Spain) rías and shelf waters. Boletín del Instituto Español de Oceanografía, 8, 57–74. Varela, M., Prego, R., & Pazos, Y. (2008). Spatial and temporal variability of phytoplankton biomass, primary production and community structure in the Pontevedra Ria (NW Iberian Peninsula): oceanographic periods and possible response to environmental changes. Marine Biology, 154, 483–499. Vidal-Romaní, J. R. (1984). A orixe das rias galegas: Estado da censtión (1886–1983). Cuardenos da Área de Ciencias Mariñas, Seminario de Estudos Galegos. Vol. 1. Ediciós do Castro. A Coruña, Spain. Vilas, F., Bernabeu, A., & Méndez, G. (2005). Sediment distribution pattern in the Rías Baixas (NW Spain): main facies and hydrodynamic dependence. Journal of Marine Systems, 54, 261–276. Zapata, M., Rodríguez, F., & Garrido, J. L. (2000). Separation of chlorophylls and carotenoids from marine phytoplankton: A new HPLC method using a reversed phase C8 column and pyridine-containing mobile phases. Marine Ecology Progress Series, 195, 29–45. Zhang, T., Fell, F., Liu, Z. -S., Preusker, R., Fischer, J., & He, M. -X. (2003). Evaluating the performance of artificial neural network techniques for pigment retrieval from ocean color in Case I waters. Journal of Geophysical Research, 108, 3286–3298.