Spatio-Temporal Variability of Chlorophyll-A and Environmental Variables in the Panama Bight

: The analysis of synoptic satellite data of total chlorophyll-a (Chl-a) and the environmental drivers that inﬂuence nutrient and light availability for phytoplankton growth allows us to understand the spatio-temporal variability of phytoplankton biomass. In the Panama Bight Tropical region (PB; 1–9 ◦ N, 79–84 ◦ W), the spatial distribution of Chl-a is mostly related to the seasonal wind patterns and the intensity of localized upwelling centers. However, the association between the Chl-a and di ﬀ erent physical variables and nutrient availability is still not fully assessed. In this study, we evaluate the relationship between the Chl-a and multiple physical (wind, Ekman pumping, geostrophic circulation, mixed layer depth, sea level anomalies, river discharges, sea surface temperature, and photosynthetically available radiation) and chemical (nutrients) drivers in order to explain the spatio-temporal Chl-a variability in the PB. We used satellite data of Chl-a and physical variables, and a re-analysis of a biogeochemical product for nutrients (2002–2016). Our results show that at the regional scale, the Chl-a varies seasonally in response to the wind forcing and sea surface temperature. However, in the coastal areas (mainly Gulf of Panama and o ﬀ central-southern Colombia), the maximum non-seasonal Chl-a values are found in association with the availability of nutrients by river discharges, localized upwelling centers and the geostrophic circulation ﬁeld. From this study, we infer that the interplay among these physical-chemical drivers is crucial for supporting the phytoplankton growth and the high biodiversity of the PB region. at http: measurements ccmp . Surface geostrophic velocity ﬁelds and sea level anomalies were generated by DUACS and distributed by the Copernicus Marine and Environment Monitoring (CMEMS; http: marine.copernicus.eu The mixed layer depth was obtained from the NOAA Monthly Isopycnal and Mixed-layer Ocean Climatology (MIMOC) product, available at http: // www.pmel.noaa.gov / mimoc / . The Sea (SST) Active Archive Center (PODAAC), available at https: podaac.jpl.nasa.gov Multi-scale_Ultra-high_Resolution_MUR-SST. Photosynthetically available radiation, remote sensing reﬂectance at 645 and normalized ﬂuorescence line height data were obtained from the MODIS-Aqua mission distributed by NASA, online available at https: oceandata.sci.gsfc.nasa.gov / MODIS-Aqua. Monthly data for nutrient content were obtained from the Ocean Biogeochemistry Non Assimilative Hindcast (Pisces) product freely distributed by CMEMS. We also grateful to the Empresa Transmisi n El é ctrica (ETESA) of Panama and Instituto de Hidrolog a, of for the data.


Introduction
Phytoplankton are the major primary producers in the ocean, being the base of the marine food web through the photosynthetic fixation of carbon and an essential component of the marine biogeochemical cycles [1,2]. Phytoplankton growth is mainly influenced by light and nutrient availability in the upper ocean. These conditions are controlled by several environmental variables and physical processes influencing the composition and abundance of phytoplankton, which determines the functioning of marine ecosystems [3][4][5]. Several studies have considered the link between the physical processes influencing changes in the phenological cycles of phytoplankton (based on chlorophyll-a (Chl-a) concentration as a proxy of phytoplankton biomass) and community structure [6][7][8][9][10].
In low-latitude regions (tropics), a long period of phytoplankton growth (~15-20 weeks) and a low amplitude in the annual cycle of Chl-a (<0.5 mg m −3 ) are mostly controlled by the input of nutrients to the surface layer [5]. By contrast, short growing periods (<10 weeks) with higher Chl-a values (>7 mg m −3 ) have been observed at high latitudes, as a result of light availability [5]. In the tropics, the Eastern Tropical Pacific (ETP) occupies 9% of the global ocean area and contributes 10% of the world's oceanic productivity [11]. However, open ocean areas in the ETP are mainly oligotrophic with low Chl-a concentration (~0.3 mg m −3 ) and only few localized regions present high Chl-a values (e.g., 1.05 mg m −3 in the Gulf of Papagayo,~0.98 mg m −3 in Tehuantepec and~1.06 mg m −3 in Panama). In general, these high Chl-a values have been found during the boreal winter (December to April, dry season in the ETP) in association with intense wind jets as a result of the intensification of the trade winds crossing Central America [12,13]. These jets can produce an intense wind stress curl, promoting sea surface divergence and upwelling, favoring the input of nutrients towards the upper layer and phytoplankton growth [11,14].

Surface Satellite Chlorophyll-a
Daily images of Chl-a were downloaded from the Ocean Colour Climate Change Initiative (OC-CCI; version 3.1) for the period 2002-2016. The OC-CCI is a merged Level 3 product with a spatial resolution of 4 km, available at http://www.oceancolour.org/. The OC-CCI Chl-a dataset is retrieved by the combining of the observational data from the MERIS (MEdium spectral Resolution Imaging Spectrometer) sensor of the European Space Agency, the SeaWiFS (Sea-viewing Wide-Fieldof-view Sensor) and MODIS-Aqua (Moderate-resolution Imaging Spectroradiometer-Aqua) sensors from the National Aeronautics and Space Administration (NASA-USA), and the VIIRS (Visible and Infrared Imaging Radiometer Suite) from the National Oceanic and Atmospheric Administration (NOAA-USA). The remote sensing reflectance data derived from the sensors were merged by band-shifting and bias-correcting the MERIS, MODIS and VIIRS data to match with the SeaWiFS data. Then, a selected in-water algorithm, based on water type, was applied to this merged dataset in order to generate the daily images of Chl-a concentration, including per-pixel uncertainty estimates. Due to the complexity of the study region, version 3.1 of the OC-CCI product was selected as it improves the performance of the ocean color data in coastal Case-2 waters compared to earlier versions that mostly focus on open ocean waters [44]. In addition, monthly composites of normalized fluorescence line height (nFLH) for phytoplankton fluorescence activity, at processing level 3 and spatial resolution of 4 km, were downloaded from the MODIS-Aqua mission (http://oceancolor.gsfc.nasa.gov/), for the period 2002-2016 (Appendix A).

Physical Environmental Drivers
Daily surface wind data, with a spatial resolution of 0.25° (~25 km), were acquired from the product V2 CCMP L3.0 (Cross-Calibrated, Multi-Platform Ocean Surface Wind Velocity; http://www.remss.com/measurements/ccmp/). This product integrates measurements from the Version-7 RSS radiometer wind speeds, QuikSCAT and ASCAT scatterometer wind vectors, moored buoy wind data and ERA-Interim wind fields.
The wind stress (τ) was computed from the monthly wind fields as follows:

Surface Satellite Chlorophyll-a
Daily images of Chl-a were downloaded from the Ocean Colour Climate Change Initiative (OC-CCI; version 3.1) for the period 2002-2016. The OC-CCI is a merged Level 3 product with a spatial resolution of 4 km, available at http://www.oceancolour.org/. The OC-CCI Chl-a dataset is retrieved by the combining of the observational data from the MERIS (MEdium spectral Resolution Imaging Spectrometer) sensor of the European Space Agency, the SeaWiFS (Sea-viewing Wide-Field-of-view Sensor) and MODIS-Aqua (Moderate-resolution Imaging Spectroradiometer-Aqua) sensors from the National Aeronautics and Space Administration (NASA-USA), and the VIIRS (Visible and Infrared Imaging Radiometer Suite) from the National Oceanic and Atmospheric Administration (NOAA-USA). The remote sensing reflectance data derived from the sensors were merged by band-shifting and bias-correcting the MERIS, MODIS and VIIRS data to match with the SeaWiFS data. Then, a selected in-water algorithm, based on water type, was applied to this merged dataset in order to generate the daily images of Chl-a concentration, including per-pixel uncertainty estimates. Due to the complexity of the study region, version 3.1 of the OC-CCI product was selected as it improves the performance of the ocean color data in coastal Case-2 waters compared to earlier versions that mostly focus on open ocean waters [44]. In addition, monthly composites of normalized fluorescence line height (nFLH) for phytoplankton fluorescence activity, at processing level 3 and spatial resolution of 4 km, were downloaded from the MODIS-Aqua mission (http://oceancolor.gsfc.nasa.gov/), for the period 2002-2016 (Appendix A).

of 26
The wind stress (τ) was computed from the monthly wind fields as follows: where C d = 0.0015 is the drag coefficient, ρ a = 1.2 kg m −3 is the mean air density, and V 10 is the wind speed at 10 m above the sea surface. The wind stress curl was estimated using the zonal and meridional components of the wind stress: This procedure was performed for each grid point by applying the centered finite difference algorithm, and the Ekman pumping velocity was estimated according to [36,45] as follows: where τ x is the zonal wind stress component, β = (2Ωcosϕ/R) is the latitudinal variability of the Coriolis parameter (f = 2Ωsenϕ), R = 6,371,000 m is the radius of the earth and ρ w = 1025 kg m −3 is the mean water density. Positive (negative) values of Ekman pumping indicate upward (downward) water motion. Daily surface geostrophic velocity fields and sea level anomaly (SLA) data were obtained from the Copernicus Marine and Environment Monitoring Service (CMEMS; http://marine.copernicus.eu/) with a spatial resolution of 0.25 • (~25 km). The mixed layer depth (MLD) was estimated following the density criteria in the algorithm of [46], which is available from the NOAA Monthly Isopycnal and Mixed-layer Ocean Climatology (MIMOC) [47,48] with a spatial resolution of 0.5 • (~50 km). The MIMOC product is based mostly on Argo data, supplemented by shipboard and Ice-Tethered Profiler conductivity-temperature-depth (CTD) data, and data from the World Ocean Database (http://www.pmel.noaa.gov/mimoc/).
Daily sea surface temperature (SST) and photosynthetically available radiation (PAR) data were obtained from the Multi-scale Ultra-high Resolution Sea Surface Temperature product (MUR-SST; https://podaac.jpl.nasa.gov/Multi-scale_Ultra-high_Resolution_MUR-SST, with a 1 km spatial resolution) and from the MODIS-Aqua mission (http://oceancolor.gsfc.nasa.gov/, with a 9 km spatial resolution), respectively. Daily images were averaged in monthly composites in order to aggregate all datasets into monthly intervals.

River Discharges
Monthly discharge data from the main rivers off the Panama and Colombia coasts were obtained from ETESA (Empresa Transmisión Eléctrica, Panama) and IDEAM (Instituto de Hidrología, Meteorología y Estudios Ambientales, Colombia) [49,50]. In addition, monthly composites of remote sensing reflectance at 645 nm (Rrs645), as a proxy of turbid plumes from river discharges, were obtained from the MODIS-Aqua mission (http://oceancolor.gsfc.nasa.gov/) for the period 2002-2016, at processing level 3 and a 4 km spatial resolution (Appendix A).

Nutrient Content
Monthly data of surface (~0.5 m depth) nutrient (silicate, nitrate and phosphate) and iron concentrations for the period 2002-2016 were obtained from the Ocean Biogeochemistry Non Assimilative Hindcast (Pisces) product from the Copernicus Marine and Environment Monitoring Service (CMEMS; http://marine.copernicus.eu/) with a spatial resolution of 0.25 • (~25 km).

Data Processing and Statistical Analysis
The mean annual cycles of Chl-a, nutrient content and physical variables in the study region were characterized through monthly mean climatology. In the case of total Chl-a, a standard Empirical Orthogonal Function (EOF) analysis was performed in order to evaluate the principal modes of the Chl-a variability in the PB. For this, the spatial and temporal patterns of Chl-a were determined using the singular value decomposition (SVD) technique, which decomposes the space-time dataset into a linear combination of orthogonal standing oscillations (EOF modes) [51]. The SVD method estimates the temporal amplitudes of the spatial eigenvectors and their associated eigenvalues, allowing us to quantify the total variance of the Chl-a concentration in both orthogonal and independent modes. These are the modes or structures containing the highest percentage of the Chl-a variability that enable the assessment of (i) the most contrasting areas of the spatial Chl-a variability, and (ii) the main contrasting months in terms of the temporal annual Chl-a variability in the PB. This approach has been successfully used to interpret the space-time features of Chl-a data [52,53].
To assess the regional association between the Chl-a and its predicting physical variables for the most contrasting months, a multivariate analysis was performed using the Primer-E7 Software through a distance-based linear model (DISTLM) [54]. This analysis consists of a permutational test for the multivariate null hypothesis of no relationship between the Chl-a and the physical variables as predictors, on the basis of a resemblance measure (dissimilarity matrix D), using permutations of the samples to obtain a p-value. Therefore, it is possible to determine the variation in the data (described by D) that is explained by each physical variable. To find the best explanatory model from all the possible combinations of the predictor variables, this analysis followed a step-wise procedure selection and Akaike's information criterion (AIC) [55]. The smallest achieved value of AIC indicates the best model, i.e., the best combination of the environmental predictor variables that explains the largest amount of the variation in the response variable (Chl-a). Additionally, further marginal tests explain the percentage variance of the Chl-a when each environmental variable is considered individually through a Pseudo-F value, which is a direct multivariate analogue of the Fisher's F ratio used for standard regressions. The higher the Pseudo-F value, the higher the probability of the null hypothesis being false, i.e., a higher Pseudo-F value indicates a high probability of a relationship between the Chl-a and the physical variable as a predictor.
Then, from the resemblance dissimilarity matrix, a non-metric multidimensional scaling ordination (nMDS) was performed to assess the spatial differences of the Chl-a values across the PB by dividing the region into several subdomains (coastal and oceanic), which were used as the factors [56]. A multidimensional scaling analysis is an ordination method used to reduce the dimensionality of the data, allowing the detection of the most prominent observable patterns and structures. In particular, the non-metric multidimensional scaling focuses on preserving the rank order of the inter-point dissimilarities allowed within the constraints of a small number of dimensions (two or three). The procedure is based on the minimization of a monotonic function of the dissimilarities that is called "stress". In consequence, the suitability of the resulting plot is determined by how well the inter-point distances in the reduced dimension correspond to the rank orders of the underlying dissimilarities [57,58], allowing us to visualize how different the subdomains are in terms of their Chl-a concentration values. Finally, linear Pearson correlations between Chl-a and the physical and chemical variables in each subdomain were calculated, in order to analyze the prominent factors promoting the low or high Chl-a values. For this, all the variables were re-gridded to the lowest spatial resolution (~25 km), excluding the MLD fields from the analysis due to their coarser resolution (~50 km).

Spatio-Temporal Variability of the Annual Cycle of Chlorophyll-a
The climatological annual cycle of Chl-a revealed a seasonal signal in the oceanic zone, with the highest Chl-a concentration values from January to May (~0.5-1.5 mg m −3 ) and the lowest, from June to December (~0.25 mg m −3 ; Figure 2). In addition, the maximum Chl-a concentrations in the Gulf of Panama generated a Chl-a plume (>1 mg m −3 ), which extends southwestward into the oceanic area during January-April. By contrast, the maximum Chl-a concentrations (>2 mg m −3 ) in the PB were observed in the coastal areas of the Gulf of Panama and Gulf of Chiriqui and the coastal area off central-southern Colombia, whereas its northern coastal area showed Chl-a values similar to those of the oceanic zone (~0.5 mg m −3 ). The Empirical Orthogonal Functions (EOFs) performed on the monthly total Chl-a values revealed a dominant signal of seasonal variation represented by the first and second modes, explaining 56% of the total variance in the annual cycle of Chl-a in the region ( Figure 3). The first mode explains 38% of the Chl-a annual variability (Figure 3a,c). The annual harmonic applied to the first mode showed a marked seasonality, characterized by maximum (minimum) Chl-a values in March (September, Figure 3a), together with a maximum amplitude of the annual cycle in the Gulf of Panama and the coastal area off southern Colombia, decreasing towards the oceanic zone ( Figure  3c). The second mode explained 18% of the Chl-a variability; however, its spatial structure showed a smaller amplitude of the annual cycle in the region (Figure 3d). Therefore, in terms of the temporal variability, we determined March and September as the most contrasting months for explaining the Chl-a variation in association with the environmental drivers. In terms of the spatial variability of Chl-a, we expected different behavior in the coastal areas off the Gulf of Panama and southern Colombia in comparison with the other coastal and oceanic zones in the PB region. The Empirical Orthogonal Functions (EOFs) performed on the monthly total Chl-a values revealed a dominant signal of seasonal variation represented by the first and second modes, explaining 56% of the total variance in the annual cycle of Chl-a in the region ( Figure 3). The first mode explains 38% of the Chl-a annual variability (Figure 3a,c). The annual harmonic applied to the first mode showed a marked seasonality, characterized by maximum (minimum) Chl-a values in March (September, Figure 3a), together with a maximum amplitude of the annual cycle in the Gulf of Panama and the coastal area off southern Colombia, decreasing towards the oceanic zone ( Figure 3c). The second mode explained 18% of the Chl-a variability; however, its spatial structure showed a smaller amplitude of the annual cycle in the region ( Figure 3d). Therefore, in terms of the temporal variability, we determined March and September as the most contrasting months for explaining the Chl-a variation in association with the environmental drivers. In terms of the spatial variability of Chl-a, we expected different behavior in the coastal areas off the Gulf of Panama and southern Colombia in comparison with the other coastal and oceanic zones in the PB region.

Wind Field and Ekman Pumping
In March, the north trade winds dominated, being intensified in Panama and generating the Panama Jet with velocities of ~5 m s −1 (Figure 4a). The jet crosses the PB in a southwest direction down to ~3°N, where it splits zonally and decreases its intensity (~2 m s −1 ). In the area between 1 and 3°N, winds from the north collided with winds from the south, deflecting east and west with lower speeds (Figure 4a). Associated with this wind pattern, the Ekman pumping showed a dipole pattern with upwelling-favorable vertical velocities on the east side (~78-80°W) and downwelling conditions on the west side (~80-84°W) (Figure 4b). The maximum Ekman pumping values (± 2 m day −1 ) were observed in the northern coastal zone of the PB (see Figure 4b, red contours). In September, the south trade winds dominated the wind field, blowing in the northeast direction and reaching higher intensities (~7 m s −1 ) than those observed in March (

Wind Field and Ekman Pumping
In March, the north trade winds dominated, being intensified in Panama and generating the Panama Jet with velocities of~5 m s −1 (Figure 4a). The jet crosses the PB in a southwest direction down to~3 • N, where it splits zonally and decreases its intensity (~2 m s −1 ). In the area between 1 and 3 • N, winds from the north collided with winds from the south, deflecting east and west with lower speeds (

Geostrophic Circulation Field and Sea Level Anomalies
The Panama Jet Surface Current (PJSC) crossed the PB region in March, flowing from the Gulf of Chiriqui to the south around ~82°W (Figure 5a). On its west side, a portion of an anticyclonic gyre was observed with positive SLA values of ~5 cm (~5-7°N, 82-84°W). There were two cyclonic gyres to the east of the PJSC, one located in the Gulf of Panama (~7-9°N, 78-80°W) and the other, off the Colombian coast (~3-7°N, 78-80°W), both with negative SLA values of about −10 to −15 cm. In addition, a coastal current flowing northward along the southern Colombian margin was formed and reached the southern flank of the largest oceanic cyclonic gyre ( Figure 5a). By contrast, the surface circulation field reversed in September, and an anticyclonic gyre was observed off the Colombian coast (~3-7°N, 78-80°W) with positive SLA values (~10 to 15 cm). Moreover, the cyclonic gyre in the Gulf of Panama and the anticyclonic gyre in the northwest area of the PB were no longer observed. Instead, these features were replaced by a lower intensity anticyclonic gyre centered around 4°N, 82°W, a cyclonic gyre in the northwest part of the domain (~7°N, 84°W), and a westward surface current in the southern area of the PB with a maximum speed of ~25 cm s −1 (Figure 5b). circulation field reversed in September, and an anticyclonic gyre was observed off the Colombian coast (~3-7°N, 78-80°W) with positive SLA values (~10 to 15 cm). Moreover, the cyclonic gyre in the Gulf of Panama and the anticyclonic gyre in the northwest area of the PB were no longer observed. Instead, these features were replaced by a lower intensity anticyclonic gyre centered around 4°N, 82°W, a cyclonic gyre in the northwest part of the domain (~7°N, 84°W), and a westward surface current in the southern area of the PB with a maximum speed of ~25 cm s −1 (Figure 5b). Red (blue) shading represents the regions with sea level anomalies above (below) the regional average.

Mixed Layer Depth
The observed spatial variability of the MLD in the PB is shown in Figure 6. In March, the MLD was found at depths between 18 and 20 m throughout the region ( Figure 6a) and was estimated at similar depths in the northern part of the domain in September (~5-8 • N). However, it deepened considerably (~32 m) in the southwest corner of the PB (Figure 6b).  Red (blue) shading represents the regions with sea level anomalies above (below) the regional average.

Mixed Layer Depth
The observed spatial variability of the MLD in the PB is shown in Figure 6. In March, the MLD was found at depths between 18 and 20 m throughout the region ( Figure 6a) and was estimated at similar depths in the northern part of the domain in September (~5-8°N). However, it deepened considerably (~32 m) in the southwest corner of the PB (Figure 6b).    Colombia (Figure 8b). Thus, there was a change of direction in the dominant regional SST gradient from a zonal to a meridional gradient in March and September, respectively.

Spatial Distribution of Nutrient Content in March and September
The surface nutrient (nitrate, phosphate and silicate) and iron concentrations in March and September are shown in Figure 9. The observed nitrate/phosphate (N:P) and nitrate/silicate (N:Si) ratios are also presented for a better understanding of the nutrient distribution in the region ( Figure  10). Nitrate and phosphate had similar spatial patterns, with values ranging between ~2-6 mmol m −3 and ~0.5-1.5 mmol m −3 in the northeastern part of the PB (Figures 9a,b). Silicate and iron concentrations ranged between ~5-20 mmol m −3 and ~0.3-1 mmol m −3 in the Gulf of Panama, off the Azuero peninsula, and in the oceanic region off central Colombia (Figures 9c,d). In September, nitrate and phosphate concentrations were elevated (~3 mmol m −3 and ~1 mmol m −3 , respectively) over the southern oceanic area of the PB (Figures 9e,f), while the silicate concentration reached ~5-10 mmol m -3 in the coastal zones. The iron content was mostly depleted in the PB (Figure 9h). However, localized maxima of silicate and iron (~20 mmol m −3 and ~0.4 mmol m −3 , respectively) were associated with the San Juan river discharge (Figures 9g,h, see also Appendix A). The N:P and N:Si ratios indicated deficits in nitrate (Figure 10a) and silicate (Figure 10b) across the PB, with contrasting spatial distributions in March versus September (Figures 10c,d). Whereas the deficits of nitrate and silicate were confined mostly to the northeast area in March, these were shifted to the southwestern region in September (Figure 10).

Spatial Distribution of Nutrient Content in March and September
The surface nutrient (nitrate, phosphate and silicate) and iron concentrations in March and September are shown in Figure 9. The observed nitrate/phosphate (N:P) and nitrate/silicate (N:Si) ratios are also presented for a better understanding of the nutrient distribution in the region (Figure 10). Nitrate and phosphate had similar spatial patterns, with values ranging between~2-6 mmol m −3 and 0.5-1.5 mmol m −3 in the northeastern part of the PB (Figure 9a,b). Silicate and iron concentrations ranged between~5-20 mmol m −3 and~0.3-1 mmol m −3 in the Gulf of Panama, off the Azuero peninsula, and in the oceanic region off central Colombia (Figure 9c,d). In September, nitrate and phosphate concentrations were elevated (~3 mmol m −3 and~1 mmol m −3 , respectively) over the southern oceanic area of the PB (Figure 9e,f), while the silicate concentration reached~5-10 mmol m −3 in the coastal zones. The iron content was mostly depleted in the PB (Figure 9h). However, localized maxima of silicate and iron (~20 mmol m −3 and~0.4 mmol m −3 , respectively) were associated with the San Juan river discharge (Figure 9g,h, see also Appendix A). The N:P and N:Si ratios indicated deficits in nitrate ( Figure 10a) and silicate (Figure 10b) across the PB, with contrasting spatial distributions in March versus September (Figure 10c,d). Whereas the deficits of nitrate and silicate were confined mostly to the northeast area in March, these were shifted to the southwestern region in September ( Figure 10).
Azuero peninsula, and in the oceanic region off central Colombia (Figures 9c,d). In September, nitrate and phosphate concentrations were elevated (~3 mmol m −3 and ~1 mmol m −3 , respectively) over the southern oceanic area of the PB (Figures 9e,f), while the silicate concentration reached ~5-10 mmol m -3 in the coastal zones. The iron content was mostly depleted in the PB (Figure 9h). However, localized maxima of silicate and iron (~20 mmol m −3 and ~0.4 mmol m −3 , respectively) were associated with the San Juan river discharge (Figures 9g,h, see also Appendix A). The N:P and N:Si ratios indicated deficits in nitrate ( Figure 10a) and silicate (Figure 10b) across the PB, with contrasting spatial distributions in March versus September (Figures 10c,d). Whereas the deficits of nitrate and silicate were confined mostly to the northeast area in March, these were shifted to the southwestern region in September ( Figure 10).

Association Between Chlorophyll-a and the Environmental Drivers
The climatology of the annual cycle of Chl-a revealed a seasonal pattern variation, with the highest Chl-a values observed from January to May and the lowest, from June to December, as previously reported by [30]. The PB region is characterized by having two climatic seasons, the dry (December to April) and rainy (May to November) seasons, both forced by the trade winds and the

Association between Chlorophyll-a and the Environmental Drivers
The climatology of the annual cycle of Chl-a revealed a seasonal pattern variation, with the highest Chl-a values observed from January to May and the lowest, from June to December, as previously reported by [30]. The PB region is characterized by having two climatic seasons, the dry (December to April) and rainy (May to November) seasons, both forced by the trade winds and the meridional migration of the ITCZ [13,36].
During the dry season, and specifically in March, the ITCZ reaches its southernmost position in the PB together with the intensification of the northeastern trade winds as they cross the Isthmus of Panama (intensification of the Panama Jet; Figure 4a), being the primary driver of the climatic seasonality in the region [11,32,34]. The wind pattern around the Panama Jet axis generates a dipole in the Ekman pumping (Figure 4b), together with a surface circulation characterized by a cyclonic gyre on the east of the Panama Jet, an anticyclonic gyre on the west, and the PJSC flowing southward along the PB (Figure 5a) [34][35][36]. The intense wind jet promotes the high availability of light by displacing the clouds across the oceanic zone ( Figure 7a) [31,33], together with a cooler water tongue from the Gulf of Panama towards the south (Figure 8a) coinciding with the intensification of the upward vertical water velocities, and the maximum rise of the thermocline in the northern part of the PB [32,34]. These atmospheric and oceanographic features can promote both light availability (PAR; Figure 7a) and the input of nutrients into the upper layer, with a spatial structure as the plume of high Chl-a in March (Figures 2 and 9a-d). Despite this, the low Chl-a values found in the western part of the domain can be associated with downwelling-favorable Ekman pumping velocities (Figure 4b) and the anticyclonic gyre (Figure 5a), which could locally produce a downward isopycnal displacement, modulating the phytoplankton growth rates in association with variations in the availability of nutrients and/or light with depth [59].
By contrast, the ITCZ was located in the northern part of the PB during the rainy season. This results in a shift of the wind pattern and an intensification of the southwestern trade winds as they cross the region flowing towards the continent (the Choco Jet; Figure 4c) [60]. The shift in the wind pattern leads to an Ekman pumping favorable to downwelling in the eastern half of the domain (Figure 4d), an anticyclonic oceanic gyre (also favorable to downwelling; Figure 5b), a deep mixed layer (Figure 6b), and a low nutrient content (Figure 9e-h)-altogether promoting the observed regional low Chl-a values in September ( Figure 2). In addition, the deeper mixed layer over the west side of the domain (Figure 6b) could explain the low Chl-a concentrations despite having upwelling-favorable Ekman pumping velocities ( Figure 4d) and a high nutrient content (Figure 9e,f). Regarding SST and PAR, the prevailing wind field enhances the north-south gradient of SST (Figure 8b) [34,60], and the greater cloud coverage decreases the light availability (Figure 7b) for phytoplankton cells to grow across the oceanic region.
Among these relationships between the environmental variables and Chl-a, the DISTLM analysis highlights different key physical drivers affecting the increase or decrease in phytoplankton biomass on a regional scale (Table 1). In March, the high Chl-a values were mainly explained by the SST (60.82%), whereas the low Chl-a values were better correlated with the wind field (32.05%) and light availability (22.24%) in September. Table 1. Chlorophyll-a variance explained by the environmental variables in the PB for March and September. The marginal tests shows the relative contribution of each variable tested individually: SS denotes the total sum of squares of the complete multivariate data, the Pseudo-F value indicates the probability of a relationship between the Chl-a and the physical variable as a predictor, P is the p-value, and the % explained is the proportion of the variation explained by each variable. The conditional tests show the best combination of the environmental predictor variables that explain the largest amount of the variation in the response variable (Chl-a) based on the smallest value of Akaike's information criterion (AIC). R 2 is the proportion of the explained variation for the model, RSS is the SS residual and No. Vars refers to the number of predictor variables used by the best solution model. Changes in the abundance and composition of the phytoplankton community have been reported to coincide with changes in the SST along the PB, mostly related to upwelling areas and/or El Niño Southern Oscillation (ENSO) events. In neutral years or with weak to moderate ENSO conditions, the dominance of diatoms has been reported during both climatic seasons, with higher diatom abundance and higher Chl-a values during the dry season [61][62][63]. From January to March, the dominance of diatoms is mostly associated with the main oceanic and coastal upwelling areas (central oceanic region and coastal areas off southern Colombia), displaying a negative correlation with the SST; i.e., diatom abundance is high in upwelling areas with low SST values [62]. Instead, in years with the strong warm phase of ENSO (El Niño conditions), a high abundance of diatoms (e.g., Chaetoceros taciniosos, Skeletonema costatum and Dytylum brightwellii) was detected from March to April in areas with low SST values (25-26 • C), corresponding to localized upwelling zones. By contrast, under the same conditions (El Niño), an atypical dominance of dinoflagellates (e.g., Ceratium deflexum and Ceratium furca var furca) was observed from September to October, concentrated in the oceanic areas with high SST values (>27 • C) [64]. These reports indicate a relationship between the phytoplankton biomass and the community composition in association with the SST; however, the temperature was never a variable that was independent from other oceanographic processes (mainly upwelling) or ENSO events. This is also indicated in the DISTLM analysis through Akaike's criterion, which shows that in total four predictors or physical variables (SST, SLA, geostrophic circulation, and wind) should be considered to better explain the high Chl-a concentrations in March (Table 1).

March
Water temperature variations may also have an impact on the ecosystem's functioning through changes in the metabolic rates and phytoplankton cell size. Although this relationship is difficult to test, even under controlled experimental conditions, an inverse relationship between temperature and final body size has been observed, i.e., a colder temperature induces a larger cell size [65][66][67]. In March, when the DISTLM analysis indicates an important relationship between Chl-a and SST, the reported dominance of large phytoplanktonic cells in the PB [61][62][63] also suggests a higher requirement for nutrients, due to which the effect of the nutrient supply could be stronger than the temperature effect upon the phytoplankton metabolic rates. In that case, the nutrient availability can have a more significant role than temperature in controlling the efficiency of photosynthetic carbon conversion into new biomass, especially in nitrogen-limited regions [68], as is the PB (Figures 9 and 10). However, we could not test these assumptions, and they should be explored in future work, preferably through determining biomass-specific carbon fixation rates under different temperatures (as the independent variable) and nutrient supply conditions. Moreover, considering that (i) variations in the tropical phytoplankton community structure are strongly linked to ENSO events, and (ii) the use of Chl-a data alone causes bias in the estimation of phytoplankton biomass, which can be misleading when the relationships between physical drivers and phytoplankton growth are evaluated [24,69].
In that sense, phytoplankton cells need to produce an amount of Chl-a in response to the nutrient content and light availability to optimize photosynthesis, but this does not necessarily imply a linear covariation between Chl-a biomass and phytoplankton carbon. For example, under stable light conditions, there is no need for the cells to produce large amounts of the energetically expensive Chl-a pigment, so the phytoplankton exploit the availability of nutrients to grow, despite the decrease in the Chl-a values. Instead, under decreasing light availability, the phytoplankton cells allocate a greater amount of the energy from the nutrients into the production of Chl-a, even though this limits their growth rate. Consequently, the use of Chl-a data alone does not necessarily take into account the physiological adjustments that phytoplankton undergo in response to changes in the environment, such as in light and nutrient conditions [70][71][72]. Future studies are encouraged, combining in situ experiments with satellite-based measurements of Chl-a and/or size-fractionated Chl-a, as well as the use of Chl-a biomass and phytoplankton carbon on both seasonal and interannual scales.
On the other hand, the association among the observed low regional Chl-a values, the wind field and the low light availability in September (Table 1) are in agreement with previous studies describing a decrement in Chl-a during the rainy season, mainly due to the weakening of the upwelling [30,32,34]. Periods with increased cloud coverage and low PAR (Figure 7b) limit the growth rates of phytoplankton cells due to the light limitation, which in the Eastern Pacific has been reported to be co-limited by macronutrients and iron [73,74], as is the case for the PB (Figure 9e-h). In addition, the DISTLM analysis also shows other four predictors to be considered to better explain the observed low Chl-a values in September, such as the Ekman pumping, SLA, geostrophic circulation and SST. In this sense, the reduction in cell size under increased temperatures could also be an adaptive response by a change in the species composition in association with resource availability [66,75]. Furthermore, the ratio of the supply to potential consumption of limiting nutrients is reduced when the temperature increases; therefore, a reduced cell size can compensate for the limitation of resources, since their uptake is more efficient by smaller cells [66,67,76]. Beside these regional relationships, we further explore the spatial Chl-a variability by dividing the PB into several subdomains in order to better understand the variability of Chl-a in the oceanic and coastal areas.

Characterization of Chlorophyll-a by Subdomains in Association with Physical Drivers and Nutrient Availability
According to bathymetry features, number of rivers and Chl-a concentration range, the PB region was divided into subdomains, seven in March and five in September ( Figure 11). The coastal subdomains (1, 2, 3 and 4) remain the same for both periods, whereas the oceanic subdomains (5, 6 and 7) in March became only one (5) in September due to the homogeneity of low Chl-a values across the oceanic region.
nutrients is reduced when the temperature increases; therefore, a reduced cell size can compensate for the limitation of resources, since their uptake is more efficient by smaller cells [66,67,76]. Beside these regional relationships, we further explore the spatial Chl-a variability by dividing the PB into several subdomains in order to better understand the variability of Chl-a in the oceanic and coastal areas.

Characterization of Chlorophyll-a by Subdomains in Association with Physical Drivers and Nutrient Availability
According to bathymetry features, number of rivers and Chl-a concentration range, the PB region was divided into subdomains, seven in March and five in September ( Figure 11). The coastal subdomains (1, 2, 3 and 4) remain the same for both periods, whereas the oceanic subdomains (5, 6 and 7) in March became only one (5) in September due to the homogeneity of low Chl-a values across the oceanic region. In general, higher Chl-a values were observed in the coastal subdomains for both periods, mainly in Subdomains 1 (Gulf of Panama) and 2 (coastal area off central-southern Colombia; Figure  11), also in agreement with the spatial differences of the Chl-a values observed in the nMDS analysis ( Figure 12). Note that Subdomain 4 was not included in this analysis or in the calculation of the linear Pearson correlations between Chl-a and the physical/chemical variables (Table 2) due to the restricted spatial coverage of this subdomain, which implies low data availability. Nonetheless, we analyzed the Chl-a variability in Subdomain 4 based on the local oceanographic processes, bathymetry and river discharges.
The PB is located in the second rainiest region of the world with maximum and minimum rainfall values in the Colombian Pacific (CP) occurring in August and January-March, whereas in Panama, these extremes occur in October and February-March, respectively [38,39]. In addition, maximum river discharges into the Gulf of Panama, the Azuero Peninsula, and the Gulf of Chiriqui occur from September to November [77]. By contrast, low river discharges are reported in northern Colombia (north of Point Arusi; Figure 1), and the highest, in the central-southern coast, with maximum values from October to November, except for the Patia and Mira rivers, with maximum discharge rates restricted to the period of March-May [49,50,78] not shown. High rainfall rates and river discharges promote the input of not only nutrients to the continental shelf but also of sediments In general, higher Chl-a values were observed in the coastal subdomains for both periods, mainly in Subdomains 1 (Gulf of Panama) and 2 (coastal area off central-southern Colombia; Figure 11), also in agreement with the spatial differences of the Chl-a values observed in the nMDS analysis ( Figure 12). Note that Subdomain 4 was not included in this analysis or in the calculation of the linear Pearson correlations between Chl-a and the physical/chemical variables (Table 2) due to the restricted spatial coverage of this subdomain, which implies low data availability. Nonetheless, we analyzed the Chl-a variability in Subdomain 4 based on the local oceanographic processes, bathymetry and river discharges.
Remote Sens. 2020, 12, x FOR PEER REVIEW 16 of 25 (e.g., > 30.13 × 10 6 tons of sediments per year in the Colombian coast) [79]. This could lead to a misinterpretation of the satellite Chl-a signal in the coastal areas due to the masking of Chl-a in high turbidity regions. The combined use of remote sensing reflectance at 645 nm (Rrs645) as a proxy for turbid plumes with high loads of sediments [80,81] and the normalized fluorescence line height (nFLH) for phytoplankton fluorescence activity [82] helped us to confirm that high phytoplankton biomass is found in the coastal areas of the Gulf of Panama and off central-southern Colombia year-round (Appendix A).     The PB is located in the second rainiest region of the world with maximum and minimum rainfall values in the Colombian Pacific (CP) occurring in August and January-March, whereas in Panama, these extremes occur in October and February-March, respectively [38,39]. In addition, maximum river discharges into the Gulf of Panama, the Azuero Peninsula, and the Gulf of Chiriqui occur from September to November [77]. By contrast, low river discharges are reported in northern Colombia (north of Point Arusi; Figure 1), and the highest, in the central-southern coast, with maximum values from October to November, except for the Patia and Mira rivers, with maximum discharge rates restricted to the period of March-May [49,50,78] not shown. High rainfall rates and river discharges promote the input of not only nutrients to the continental shelf but also of sediments (e.g., > 30.13 × 10 6 tons of sediments per year in the Colombian coast) [79]. This could lead to a misinterpretation of the satellite Chl-a signal in the coastal areas due to the masking of Chl-a in high turbidity regions. The combined use of remote sensing reflectance at 645 nm (Rrs645) as a proxy for turbid plumes with high loads of sediments [80,81] and the normalized fluorescence line height (nFLH) for phytoplankton fluorescence activity [82] helped us to confirm that high phytoplankton biomass is found in the coastal areas of the Gulf of Panama and off central-southern Colombia year-round (Appendix A).
In March, the localized input of nutrients by rivers into the coastal subdomains (1 to 3) could have promoted the observed high Chl-a values, which together with the widest continental shelf could imply a longer residence time of nutrients to be used by phytoplankton in those areas (Figures 1 and  9a-d). Moreover, other factors were favoring the increase in the phytoplankton signal in the coastal subdomains (see the linear correlations in Table 2). In the Gulf of Panama (Subdomain 1), the high Chl-a values were correlated with the Ekman pumping and the cyclonic gyre (Figures 4b and 5a), both favorable to the upwelling of nutrients towards the upper layer, and were also positively correlated with the availability of iron (Figure 9d). In the central-southern region off Colombia (subdomain 2), the configuration of the geostrophic circulation field with the Coastal Current flowing northward (Figure 5a) probably kept the primary production restricted to the coast, together with localized upwelling centers previously reported by [30]. In the Gulf of Chiriqui (Subdomain 3), lower Chl-a values were found in comparison with those in the other coastal subdomains (1-2; Figure 12a). However, the Chl-a concentrations here appeared to be strongly associated with the availability of all nutrients (0.87 < r < 0.91; Table 2), possibly related to a localized input of nutrients by river discharges (Figure 1 and Appendix A) and/or to an advection of nutrients and Chl-a from the Gulf of Panama (closest to the Azuero Peninsula). The input of nutrients by upwelling in this area has been discarded as a dominant Chl-a driver due to a lower wind intensity during March (Figure 4a) [37]. Lastly, in the northern region off Colombia (Subdomain 4), the Chl-a signal seems to be mostly related to an advection of nutrients or Chl-a from the Gulf of Panama and from the eastern edge of the largest cyclonic gyre in the PB (Figure 5a), because other factors such as the narrow continental shelf and the absence of large rivers (Figure 1 and Appendix A) may have hindered the increase in phytoplankton biomass.
Regarding the oceanic subdomains in March, Chl-a decreased from Subdomain 5 to 7 (Figure 12a). In Subdomain 5, the Chl-a values were highly correlated to the upwelling-favorable wind field, also bringing colder deep waters and nutrients to the upper layer (Figure 4a,b, Figure 8a, and Figure 9a,b). The positive correlation values (0.58 < r < 0.64; Table 2) between Chl-a and the availability of nitrate and phosphate may indicate favorable conditions for phytoplankton growth. In Subdomain 6, Chl-a was mainly associated with the Ekman pumping favorable to upwelling although with lower intensity than in Subdomain 5 (Figure 4a), and relatively low PAR values as indicated by its negative correlation. By contrast, the lowest Chl-a values in Subdomain 7 were associated with multiple factors ( Table 2), mainly the wind conditions favorable to downwelling (Figure 4a,b) and the lowest nutrient content (Figure 9a-d).
In September, lower Chl-a values were observed in all the subdomains compared to those in March, except in Subdomain 3, where the Chl-a values remained similar for both periods ( Figure 12 and Table 2). In terms of the spatial variability of Chl-a by subdomain in September, Subdomains 1 and 2 show higher Chl-a values than those of Subdomains 3 to 5 (Figures 2 and 12 and Table 2). In the Gulf of Panama (Subdomain 1), the high Chl-a values were strongly restricted to the coast and potentially associated with a localized input of nutrients by rivers (Figure 1 and Appendix A), which had a maximum discharge (~4500 m 3 s −1 ) from September to November [49,50]. In addition, the low nutrient content observed in this area (Figure 9e-h) could imply a faster uptake of nutrients by phytoplankton. Nonetheless, there was also a wind pattern favorable to downwelling (Figure 4c,d and Table 2), which could have contributed to the lower nutrient content and the decrease in Chl-a in the offshore area. Likewise, off central-southern Colombia (Subdomain 2), the high Chl-a values were restricted to the coast in association with high river discharges during the rainy season and the coastal signal of Ekman pumping favorable to upwelling (Figures 1 and 4d and Appendix A). However, the negative correlation between Chl-a and the wind magnitude (r= −0.79, Table 2) could indicate an adverse effect on phytoplankton due to higher turbulence, not favorable for phytoplankton growth [83]. In terms of subdomains with low Chl-a in September, the absence of upwelling and elevated river discharges could limit the input of nutrients for phytoplankton growth off northern Colombia (Subdomain 4). Across the oceanic region (Subdomain 5), the low Chl-a values were associated with low PAR (Figure 7b), a deficit of silicate and a higher magnitude of wind (Figures 4c  and 10d and Table 2), despite the Ekman pumping pattern favorable to upwelling (Figure 4d). In the case of the Gulf of Chiriqui (Subdomain 3), the localized intense downwelling conditions (Figure 4d) could generate the decrease in Chl-a, despite the maximum river outflows reported from September to November [49,50,77].
In all subdomains and for both study periods (March and September), we highlight the physical and chemical variables that may promote the high and low Chl-a values across the PB. However, further detailed analyses are needed in order to more effectively test the sources of the nutrients (e.g., by upwelling or rivers) in the areas of high Chl-a values, as well as more frequent biological sampling (e.g., time series at fixed stations) across the region to obtain a higher spatio-temporal resolution of the phytoplankton's responses to different oceanographic conditions. In addition, measurements focused on the phytoplankton community are needed to obtain a better picture of (i) how phytoplankton changes in association with the physical and chemical drivers, and (ii) how they adapt to the high spatio-temporal environmental variability to maintain this highly productive region, which is also characterized by high biodiversity and endemism [84,85].

Reliability of Satellite Data
The merged OC-CCI total Chl-a data product has been widely used in global and regional studies, displaying high correlation coefficient values (>0.85) and low root-mean-square errors (<0.30) between the satellite estimates and in situ measurements [86][87][88]. Besides, the fact that this product includes inter-sensor bias-corrected time series data improves the spatial and temporal coverage, maintaining rigorous standards for data quality, error characterization, and per-pixel uncertainty characterization based on its validation [44]. Future work on the study area should include in situ Chl-a data to assess the performance of this product in the PB region.
Regarding wind surface data, the V2 CCMP product is a combination of inter-calibrated wind data from satellite scatterometers (QuikSCAT and ASCAT) and microwave radiometer sensors (SSM/I, TMI and AMSR-E). Both radiometer and scatterometer data are validated against in situ information compiled from moored buoys using a variational analysis method (VAM), with the ERA-Interim reanalysis as a background input of the wind field. In general, the CCMP data are nearly unbiased relative to the assimilated wind speeds, with a root-mean-square fit of~0.5 m s −1 and a root-mean-square vector difference of~0.8 m s −1 from the assimilated QuikSCAT data, and no directional bias. However, a lower accuracy is expected at high wind speeds (>15 m s −1 ) and under strong storm conditions. Beside this, the CCMP winds are useful in descriptions and evaluations of air-sea processes, for driving regional and large-scale ocean models, to provide reliable ocean climatologies, and oceanographic applications [89]. In terms of SLA and geostrophic currents, these products are generated by the DUACS processing system including data from a stable number of altimetry missions in order to ensure the long-term stability of the ocean observation system. The SLA product errors in the coastal areas (<200 km) are estimated at 8.2 cm 2 ; however, higher error values could be observed in high variability coastal regions and in the retrieval of mesoscale oceanic features [90], cases in which additional criteria for data processing should be considered [88]. Furthermore, the results obtained from these products (wind field, SLA and geostrophic currents) are consistent with those previously reported in the study area through the use of other satellite databases and/or in situ data [30,32,34].
Finally, the MUR-SST analysis combines infrared satellite SST data at high (1 km; MODIS) and medium (4 to 8.8 km; AVHRR) spatial scale resolutions with microwave satellite SST data (AMSR-E, WindSat and AMSR2) in a nominal sampling of 25 km, and with global SST in situ measurements. The MUR product improves the analyzed feature resolution by one order of magnitude over most existing SST analysis products, from~100 km down to~10 km, and is suitable for obtaining accurate SST values displaying smaller biases (<±0.1 • C) and lower root-mean-square errors (<0.6 • C) compared with other satellite and in situ SST datasets [91].

Conclusions
The PB presents high Chl-a values in March, mainly related to the lowest values of the sea surface temperature. This association coincides with a seasonal wind pattern favorable to upwelling and two oceanic cyclonic gyres, both bringing colder waters and nutrients to the sunlit upper layer, favoring phytoplankton growth. Despite this, variations in the water temperature may also have an impact on phytoplankton functioning through changes in the metabolic rates and cell size, i.e., a colder temperature induces a larger cell size and therefore higher Chl-a concentrations. However, we could not test this hypothesis because the temperature was not a factor that was independent of other atmospheric or oceanic processes. By contrast, low Chl-a values in the PB were found in September in association with the seasonal shift of the wind pattern to downwelling-favorable and the low light availability. Besides this, the coastal areas displayed high Chl-a values throughout the year, mainly in the Gulf of Panama and the nearshore region off central-southern Colombia. In March, this high productivity pattern seems to be in phase with the regional patterns favorable for phytoplankton growth, whereas during the rainy season (specifically in September), the input of nutrients by river discharges, together with localized upwelling centers and the geostrophic circulation patterns, is maintaining the high Chl-a values. Although we highlight the physical and chemical variables capable of promoting high and low Chl-a values across the PB, future studies should focus on the phytoplankton community to better assess their changes, year-round, capable of sustaining the high biodiversity of the PB. Additionally, we encourage the combination of different observational (in situ and satellite data) and modeling tools to evaluate these changes, as that will also allow us to assess the phytoplankton variability in association with other spatio-temporal scale processes. We expect that this study will enable a better understanding of the productive dynamics of the PB, leading to the betterment of management decisions regarding fisheries, tourism centers and environmental policies.