Using Satellite-Based Data to Facilitate Consistent Monitoring of the Marine Environment around Ireland

: As an island nation, Ireland needs to ensure effective management measures to protect marine ecosystems and their services, such as the provision of ﬁshery resources. The characterization of marine waters using satellite data can contribute to a better understanding of variations in the upper ocean and, consequently, the effect of their changes on species populations. In this study, nineteen years (1998–2016) of monthly data of essential climate variables (ECVs), chlorophyll (Chl-a), and the diffuse attenuation coefﬁcient (K490) were used, together with previous analyses of sea surface temperature (SST), to investigate the temporal and spatial variability of surface waters around Ireland. The study area was restricted to speciﬁc geographically delineated divisions, as deﬁned by the International Council of the Exploration of the Seas (ICES). The results showed that SST and Chl-a were positively and signiﬁcantly correlated in ICES divisions corresponding to oceanic waters, while in coastal divisions, SST and Chl-a showed a signiﬁcant negative correlation. Chl-a and K490 were positively correlated in all cases, suggesting an important role of phytoplankton in light attenuation. Chl-a and K490 had signiﬁcant trends in most of the divisions, reaching maximum values of 1.45% and 0.08% per year, respectively. The strongest seasonal Chl-a trends were observed in divisions VIId and VIIe (the English Channel), primarily in the summer months, followed by northern divisions VIa (west of Scotland) and VIb (Rockall) in the winter months.


Introduction
The potential of the world's oceans to provide food and nutrition for the growing global population is widely recognised [1], and it is expected to reach 9.7 billion people by 2050 (FAO, 2016). The seas around Ireland are considered one of the most productive and biologically sensitive areas in European Union (EU) waters. As an island nation, fishery resources are highly important to the economy and society. In 2020, Irish vessels landed a total of 188,994 tonnes of fish catch in Irish ports, with an estimated value of EUR 220.5 million [2], and with another EUR 31.5 million landed abroad [3]. According to the Irish Seafood Development Agency, Bord Iascaigh Mhara (BIM), 16,430 people were employed directly and downstream in the Irish seafood sector in 2020.
The value and importance of fishery resources make their responsible management essential. Despite remarkable progress in some areas, the overall downward trend in global marine fish stocks has not improved [1]. Previous attempts to manage either single-or multispecies fish stocks without considering their population dynamics and without accounting for environmental variability have systematically failed, sometimes with catastrophic ecological and economic consequences, such as global declines in fish populations [4] and the collapse of fisheries [5]. Widespread recognition of these shortcomings has led to the NWW waters are very productive, supporting a rich and diverse range of ecosystems, habitats, and species. Some of the most extensive and valuable sea fishery resources in Europe are found off the Irish coast, and the region is considered an ideal location for finfish, shellfish, and seaweed aquaculture [31]. The oceanographic characterisation of these economically important divisions, using satellite data, can contribute to a better understanding of upper ocean environmental changes and, consequently, the effects of these changes on species populations. NWW waters are very productive, supporting a rich and diverse range of ecosystems, habitats, and species. Some of the most extensive and valuable sea fishery resources in Europe are found off the Irish coast, and the region is considered an ideal location for finfish, shellfish, and seaweed aquaculture [31]. The oceanographic characterisation of these economically important divisions, using satellite data, can contribute to a better understanding of upper ocean environmental changes and, consequently, the effects of these changes on species populations.

Satellite-Derived Essential Climate Variables (ECVs)
Nineteen years (January 1998-December 2016) of monthly chlorophyll-a (Chl-a) and diffuse attenuation coefficient for the downwelling spectral irradiance at 490 nm (K490) were used to assess the temporal and spatial variability within the surface waters surrounding Ireland. The Chl-a and K490 products were obtained from the Ocean Colour Climate Change Initiative (OC-CCI) dataset, version 3.1, with a spatial resolution of~4 km/pixel. The OC-CCI dataset is based on ocean colour data, registered by MERIS, MODIS-Aqua and VIIRS, with band shifting, bias correction, and per-pixel uncertainty calculation applied to align the SeaWiFS bands to produce a consistent climate-quality dataset. The biases between sensors from different missions can generate a significant false trend, which could affect findings derived from time-series analysis [32]. Products from OC-CCI have been shown to be more consistent over time, compared to other multi-mission products, than single-mission and other merged ocean colour satellite products [33].
The generation of the OC-CCI v3.1 Chl-a algorithm involved the blending of the Ocean Colour Index (OCI) [34], the OC5 algorithm [35], adopted for coastal regions, and the OC3 [36] and OC4 [37] algorithms. The blended Chl-a algorithm, used in versions v3.0 and v3.1, attempts to weight the outputs of the best-performing algorithms (OC3, OCI (OC4 + CI), and OC5), which are based on the water types present, further improving performance in Case-2 waters compared to earlier versions, which were mainly open-ocean focused [38]. Some filters were applied to the products, considering the realistic values of each parameter. In the case of Chl-a, all the values less than 0.001 were set to 0.001 mg/m 3 , while the values greater than 100 mg/m 3 were set to 100 mg/m 3 [34]. The diffusive attenuation coefficient at 490 nm for downwelling irradiance (K490), which is an apparent optical property (AOP), was computed from the inherent optical properties (IOPs) at 490 nm using Lee et al.'s (2005) algorithm. K490 can be considered an indicator of turbidity, directly related to the presence of scattering particles in the water column. It indicates how strongly light is attenuated within the water column at a specific wavelength-in this case, 490 nm. Considering IOPs, all values greater than 10 (1/m) were discarded [34]. More detailed information about the OC-CCI v3.1 dataset can be found in the product user guide [38]. In November 2020, ESA Ocean Colour project released the v5.0 dataset, which incorporates Sentinel-3 A-OLCI data. The new v5.0 OC-CCI product extends the data availability until the end of June 2020, including associated per-pixel uncertainty information. The dataset used in this study does not correspond to the latest v5.0 version. However, this issue does not affect the application of our methodology and the results derived from it, as the main difference between both datasets corresponds to the time spans.
In addition to the Chl-a and K490 OC-CCI products (Figure 2), daily sea surface temperature (SST) data, covering the same nineteen years (January 1998-December 2016), were obtained from the NOAA Reynolds Optimal Interpolation SST (OISST) dataset, version v2.0, with a spatial resolution of 25 km. Selecting this dataset instead of the one produced by OC-CCI corresponds to a practical reason, since SST values had already been extracted and analysed for the same study area [29]. The OISST version v2.0 is a global dataset created using the International Comprehensive Ocean-Atmosphere Data Set (ICOADS) and AVHRR Pathfinder data. In this dataset, the operational AVHRR products were used along with in situ data from ships and buoys to allow for the largescale adjustment of satellite biases [39]. More detailed information on this dataset can be found in previous studies [39][40][41][42].

Statistical Analyses
Firstly, an exploratory analysis was performed to summarise the main characteristics of the Chl-a, K490, and SST data in the study area. Subsequently, spatial and temporal variability analyses were performed with respect to the 19 years of monthly Chl-a and K490 data. Trend analyses were also performed for both parameters, complementing previous temporal and spatial SST analyses [29].

Statistical Analyses
Firstly, an exploratory analysis was performed to summarise the main characteristics of the Chl-a, K490, and SST data in the study area. Subsequently, spatial and temporal variability analyses were performed with respect to the 19 years of monthly Chl-a and K490 data. Trend analyses were also performed for both parameters, complementing previous temporal and spatial SST analyses [29].

Time Series Decomposition
A time series can be defined as a collection of observations on a variable of interest. Most time-series analyses, based on remote sensing data, use the classical decomposition X = S + T + I, where S, T, and I represent the seasonal, trend, and irregular (or residual) components, respectively [43]. However, this approach does not allow for a flexible specification of the seasonal component, while the trend component is represented by a deterministic function of time that is easily affected by outlying observations. In this study, the time series were decomposed using the stl.fit () procedure [44,45], built on the Seasonal-Trend decomposition based on Loess (STL, local polynomial regression fitting) [46], where Loess is a nonlinear estimating method [46]. One of the advantages of the "stl.fit ()" function is its capability to identify a seasonal component that changes over time, responds to nonlinear trends, and is robust to outliers. This is achieved by smoothing the time series, using Loess, in two loops: an inner loop, which iterates between the seasonal and trend smoothing; and the outer loop, which minimises the effect of outliers, locating them in the irregular, or residual component [46]. The irregular/residual component is derived by subtracting the seasonal and trend time series components [46]. STL, as other non-parametric regression methods, requires the subjective selection of seasonal and trend smoothing parameters [46]. The Loess smoothing parameters, applicable to the STL analysis, must be defined in advance and set according to the user's knowledge of the data [47,48].

Time Series Decomposition
A time series can be defined as a collection of observations on a variable of interest. Most time-series analyses, based on remote sensing data, use the classical decomposition X = S + T + I, where S, T, and I represent the seasonal, trend, and irregular (or residual) components, respectively [43]. However, this approach does not allow for a flexible specification of the seasonal component, while the trend component is represented by a deterministic function of time that is easily affected by outlying observations. In this study, the time series were decomposed using the stl.fit () procedure [44,45], built on the Seasonal-Trend decomposition based on Loess (STL, local polynomial regression fitting) [46], where Loess is a nonlinear estimating method [46]. One of the advantages of the "stl.fit ()" function is its capability to identify a seasonal component that changes over time, responds to nonlinear trends, and is robust to outliers. This is achieved by smoothing the time series, using Loess, in two loops: an inner loop, which iterates between the seasonal and trend smoothing; and the outer loop, which minimises the effect of outliers, locating them in the irregular, or residual component [46]. The irregular/residual component is derived by subtracting the seasonal and trend time series components [46]. STL, as other non-parametric regression methods, requires the subjective selection of seasonal and trend smoothing parameters [46]. The Loess smoothing parameters, applicable to the STL analysis, must be defined in advance and set according to the user's knowledge of the data [47,48]. The STL algorithm [44], applied in this study, allows a selective choice of the STL smoothing parameters and selects the best STL model based on the lowest error (in our case, the root mean squared error). This fact is beneficial for studies of time series that need to evaluate interannual variability. We direct the reader to previous studies for more information on the STL approach [44][45][46].

Non-Parametric Trend Analysis
Once the seasonal component was extracted, Sen's slope was calculated using the Mann-Kendall test [49,50]. The Mann-Kendall test is used to statistically evaluate a monotonic upward or downward trend of the variable of interest over time. A monotonic upward (downward) trend means that the variable is consistently increasing (decreasing) over time; however, the trend may or may not be linear [49,50]. The Mann-Kendall test does not require a normal distribution of the residuals resulting from the fitted regression line [49,50]. Sen's slope is a nonparametric estimator that uses a median value to estimate the slope and is quite robust in the presence of outliers.
Additionally, to identify cases where structural changes (i.e., points where the changes are more pronounced) might occur, we used the statistical R package "strucchange" [51] and its "breakpoints ()" function. The "breakpoints ()" function allows to identify breakpoints that minimise the residual sum of squares of a linear model with m + 1 segments. The optimal number of structural breaks is determined using the Bayesian information criterion (BIC) [51]. This analysis allowed the identification of significant changes (e.g., trend shifts) in the time series of each ICES division.
Monthly trends and significance levels were also calculated, using the "wql" R package [52] and the "seasonTrend ()" function, which calculates the signs and magnitudes of trends for individual months [53]. "SeasonTrend ()" is based on the Mann-Kendall test and calculates the actual and relative Sen s slope, the p-values for the trend, and the fraction of missing slopes encompassing the first and last fifths of the data [53]. This analysis can be informative because some time series may not show a significant trend overall but may show a significant trend for individual months, such that the direction of the trend varies from month to month.

Similarities between ICES Divisions
To examine similarities among divisions with respect to Chl-a, K490, and SST, an agglomerative hierarchical cluster analysis using the climatology values for each ICES division was performed. In hierarchical cluster analysis, data are grouped into homogeneous clusters by merging data points, or classes, one at a time in a series of sequential steps [54,55]. Here, we use agglomerative hierarchical cluster analysis, also known as bottom-up clustering. In the initial stage of the algorithm, each n data point is considered as a single cluster [56]. Then, in each step, the two most similar data points or clusters are combined into a larger cluster, resulting in n − 1 clusters [56]. Similarities or proximities are determined by a distance function, in our case the Euclidean distance, which corresponds to the squared distance between data points or vectors. For details on agglomerative hierarchical algorithms, we refer the reader to previous overviews [57].

Exploratory Statistical Analyses
Exploratory analysis, including all ICES divisions (Figure 3), showed a median SST of 12.04 • C, with the most common values between 9 • C and 12 • C ( Figure 3). The data distribution considerably changed in the case of Chl-a and K490. Both variables showed a highly skewed distribution to the right with a median value of 0.74 mg/m 3 for Chl-a and 0.12 1/m for K490. The most common values for Chl-a in the time series were lower than 1 mg/m 3 , while for K490, the most common values were lower than 0.1 1/m, indicating a general first optical depth (Z 90 = 1/K490) of 10 m. This value (Z 90 ) represents the layer depth from which 90% of the water leaves radiance that the satellite register comes from.
The three ECVs-Chl-a, K490, and SST-showed a significant correlation in almost all divisions (Table 1). SST and Chl-a showed significant correlation in all divisions except for division VIIj (southwest of Ireland-east). In most cases, there is an inverse relationship between SST and Chl-a. Significant and positive SST-Chl-a correlations were found within oceanic divisions, such as divisions VIa (west of Scotland) and VIb (Rockall), as well as in ICES divisions located on the west coast of Ireland such as divisions VIIc (Porcupine Bank) and VIIk (south of Ireland-west) ( Table 1). The highest SST-Chl-a correlations were negative and found in divisions VIIa (Irish Sea), VIIg (north Celtic Sea), and VIIf (Bristol Channel), all of them highly influenced by coastal processes and dynamics (e.g., tidal mixing, stratification intensity, freshwater input/river runoff). Overall, SST and K490 showed a significant positive correlation in most of the divisions, except for divisions located in more oceanic waters, where the relationship was primarily negative. Chl-a and K490 showed significant and positive relations in all ICES divisions, suggesting that the attenuation of light in the first meters of the water column could mainly depend on phytoplankton biomass.
Remote Sens. 2022, 14, x FOR PEER REVIEW 7 general first optical depth (Z90 = 1/K490) of 10 m. This value (Z90) represents the layer d from which 90% of the water leaves radiance that the satellite register comes from. The three ECVs-Chl-a, K490, and SST-showed a significant correlation in alm all divisions (Table 1). SST and Chl-a showed significant correlation in all divisions ex for division VIIj (southwest of Ireland-east). In most cases, there is an inverse relat ship between SST and Chl-a. Significant and positive SST-Chl-a correlations were fo within oceanic divisions, such as divisions VIa (west of Scotland) and VIb (Rockall well as in ICES divisions located on the west coast of Ireland such as divisions VIIc ( cupine Bank) and VIIk (south of Ireland-west) ( Table 1). The highest SST-Chl-a cor tions were negative and found in divisions VIIa (Irish Sea), VIIg (north Celtic Sea), VIIf (Bristol Channel), all of them highly influenced by coastal processes and dynam (e.g., tidal mixing, stratification intensity, freshwater input/river runoff). Overall, SST K490 showed a significant positive correlation in most of the divisions, except for d sions located in more oceanic waters, where the relationship was primarily negative. a and K490 showed significant and positive relations in all ICES divisions, suggesting the attenuation of light in the first meters of the water column could mainly depend phytoplankton biomass.

Chl-a Temporal Trends
Nineteen years of monthly data (January 1998-December 2016) were used to determine temporal trends of Chl-a in each of the ICES divisions. Most of the ICES divisions showed a significant (p-value ≤ 0.05) and positive trend, indicating an increase in Chl-a concentration ( Table 2). The strongest trends were observed in ICES divisions located in the southeastern part of the study area, VIIa (Irish Sea) and VIId (eastern English Channel), with a Sen's Slope (SS) of 0.014 mg/m 3 /year and 0.010 mg/m 3 /year, respectively. The lowest values were found in ICES divisions located in the most western areas such as VIIc (Porcupine Bank) with a trend of 0.004 mg/m 3 /year followed by the ICES divisions VIIk (southwest of Ireland-west) and VIIh (south Celtic Sea) with a trend of 0.005 mg/m 3 /year in both cases. ICES divisions located in the northern areas, VIa (west of Scotland) and VIb (Rockhall), did not show any significant Chl-a trend. Most of the ICES divisions showed at least one structural change in Chl-a concentration (  The signs and magnitudes of trends for individual months were also computed because they may behave differently from the general computed trend ( Figure 6). The monthly analysis can provide a more comprehensive idea of how Chl-a values behave between seasons. The ICES divisions VIIa (Irish Sea), VIIf (Bristol Channel), and VIIg (north Celtic Sea), all of them located close to each other, did not show any significant trend between seasons. The ICES divisions VIIb (west of Ireland) and VIIh (south Celtic Sea) showed the same pattern with positive trends in the winter months. Similar behaviour was found in VIIj (southwest of Ireland-east) and VIIk (southwest of Ireland-west), but these ICES divisions also showed a significant and positive trend in late summer (September). The ICES divisions located in northern areas, VIa (west of Scotland) and VIb (Rockall), showed increasing Chl-a values in winter months. However, VIa (west of Scotland) showed a significant negative trend in late spring-early summer. The ICES divisions corresponding to the English Channel showed different behaviour between them. VIId (western English Channel), most influenced by surrounding coastal areas, showed a significant and positive trend in summer months, whereas VIIe (eastern English Channel) showed positive trends in spring and summer months but negative trends in autumn (September). The signs and magnitudes of trends for individual months were also computed because they may behave differently from the general computed trend ( Figure 6). The monthly analysis can provide a more comprehensive idea of how Chl-a values behave between seasons. The ICES divisions VIIa (Irish Sea), VIIf (Bristol Channel), and VIIg (north Celtic Sea), all of them located close to each other, did not show any significant trend between seasons. The ICES divisions VIIb (west of Ireland) and VIIh (south Celtic Sea) showed the same pattern with positive trends in the winter months. Similar behaviour was found in VIIj (southwest of Ireland-east) and VIIk (southwest of Ireland-west), but these ICES divisions also showed a significant and positive trend in late summer (September). The ICES divisions located in northern areas, VIa (west of Scotland) and VIb (Rockall), showed increasing Chl-a values in winter months. However, VIa (west of Scotland) showed a significant negative trend in late spring-early summer. The ICES divisions corresponding to the English Channel showed different behaviour between them. VIId (western English Channel), most influenced by surrounding coastal areas, showed a significant and positive trend in summer months, whereas VIIe (eastern English Channel) showed positive trends in spring and summer months but negative trends in autumn (September). The magnitude of Chl-a trends was different between ICES divisions. The strongest trends were found in ICES divisions VIId and VIIe (the English Channel), mainly in summer months, followed by the northern ICES divisions VIa (west of Scotland) and VIb The magnitude of Chl-a trends was different between ICES divisions. The strongest trends were found in ICES divisions VIId and VIIe (the English Channel), mainly in summer months, followed by the northern ICES divisions VIa (west of Scotland) and VIb (Rockhall) in winter months. High increments in Chl-a trend values were also found in the west of Ireland (VIIb) in the summer months.

K490 Temporal Trends
As in the case of Chl-a, 19 years of monthly data (January 1998-December 2016) were used to extract the temporal trends of K490 in each of the ICES divisions. A significant K490 trend was detected in almost all divisions, except for VIIa (Irish Sea), VIId (eastern English Channel), and VIIf (Bristol Channel). Due to the nature of the data (very low values), the percentage per year was used for comparison ( Table 3). The highest trends were found in division VIIe (eastern English Channel) with 0.125% per year, followed by VIIg (north Celtic Sea) and VIIj (south of Ireland-east) with 0.078% per year, respectively ( Table 3). The lowest significant trends in K490 values were found in divisions VIIh (south Celtic Sea) with a trend of 0.040% per year and VIIc (Porcupine Bank) and VIIb (Rockall) with a value of 0.042% per year. All significant trends were positive, indicating an increment of K490 values and, consequently, a decrease in the attenuation depth and water quality. The monthly K490 time series were also analysed to detect unusual shifts in K490 values (Figure 7). Only in division VIIg (north Celtic Sea) was no shift in K490 values detected, as was the case for Chl-a, indicating temporal stability in this area. The behaviour of K490 in the remaining ICES divisions was quite variable, but one or two breakpoints were observed in most of them. Breakpoints in K490 trends were found in coastal divisions, such as VIIa (Irish Sea), VIIf (Bristol Channel), and VIId (eastern English Channel). The breakpoints occurred at different times, suggesting that they could be caused by local conditions. Most of the K490 breakpoints were coincident with the ones obtained for Chl-a values. However, new breakpoints were found for in VIIf (Bristol Channel) and VIId (eastern English Channel). These results suggest that factors other than phytoplankton concentration influence the water attenuation and turbidity. Light penetration is strongly attenuated by particulate and dissolved matter originating from river runoff, or which could be re-suspended from the seabed due to strong vertical tidal mixing [58]. The Bristol Channel and its adjacent coastal seas, for example, are considered to be one of the most energetic European continental shelf areas, exhibiting some of the largest spring tides and tide ranges, which contribute to sediment resuspension [59].  Analysis of monthly trends for K490 showed variability among ICES divisions ( Figure 8). As with the Chl-a time series analysis, there were no significant trends for K490 in some ICES divisions. These divisions were VIIa (Irish Sea), VIIf (Bristol Channel), and VIIg (north Celtic Sea). ICES divisions located in the southwest of Ireland such as VIIb (west of Ireland), VIIc (Porcupine), VIIh (south Celtic Sea), VIIj (southwest of Ireland-east), and VIIk (southwest of Ireland-west) showed positive and significant trends (p-value ≤ 0.05) in winter months, with some slight differences between them. Moreover, ICES divisions located in oceanic waters such as VIa (west of Scotland) and VIb (Rockall) showed negative and significant trends in summer months. In the case of VIa (west of Scotland), it showed negative trends in summer and positive trends in spring and autumn. This could indicate a possible reduction in the summer phytoplankton bloom in this area and a shift to the spring and autumn months, as Chl-a and K490 parameters are strongly correlated in this ICES division (r = 0.86, p-value ≤ 0.001). A significant and positive trend was observed in spring in the case of VIId and VIIe, both corresponding to the English Channel, while a negative and significant trend was observed in autumn. In the case of VIIe (eastern English Channel), the results were coincident with the trends of Chl-a. The strongest trends were found in spring in the English Channel and west of Ireland. Analysis of monthly trends for K490 showed variability among ICES divisions (Figure 8). As with the Chl-a time series analysis, there were no significant trends for K490 in some ICES divisions. These divisions were VIIa (Irish Sea), VIIf (Bristol Channel), and VIIg (north Celtic Sea). ICES divisions located in the southwest of Ireland such as VIIb (west of Ireland), VIIc (Porcupine), VIIh (south Celtic Sea), VIIj (southwest of Irelandeast), and VIIk (southwest of Ireland-west) showed positive and significant trends (pvalue ≤ 0.05) in winter months, with some slight differences between them. Moreover, ICES divisions located in oceanic waters such as VIa (west of Scotland) and VIb (Rockall) showed negative and significant trends in summer months. In the case of VIa (west of Scotland), it showed negative trends in summer and positive trends in spring and autumn. This could indicate a possible reduction in the summer phytoplankton bloom in this area and a shift to the spring and autumn months, as Chl-a and K490 parameters are strongly correlated in this ICES division (r = 0.86, p-value ≤ 0.001). A significant and positive trend was observed in spring in the case of VIId and VIIe, both corresponding to the English Channel, while a negative and significant trend was observed in autumn. In the case of VIIe (eastern English Channel), the results were coincident with the trends of Chl-a. The strongest trends were found in spring in the English Channel and west of Ireland.

Discussion
Chl-a, K490, and SST parameters can provide a comprehensive picture of the biophysical properties of both oceanic and coastal surface waters, e.g., [11,60,61]. Chl-a and SST are considered ECVs, which, together with turbidity, are used by the Marine Strategy
As previously mentioned, pelagic fisheries such as mackerel, herring, and blue whiting are among Ireland s most important seafood resources. Ireland's 2020 environmental assessment, conducted by the Environmental Protection Agency (EPA), reported that the status of commercial fish and shellfish stocks was not fully compatible with Good Environmental Status [69]. The Irish Fisheries Ecosystems Advisory Services (FEAS) found that the number of sustainably fished stocks has slightly decreased, and 13 stocks remained overfished in 2020 [70]. On the other hand, fisheries are one of the economic sectors that will be most affected by Brexit. More than 70% of the total Irish fishing fleet over 12 m in length, which operate in both whitefish and pelagic fisheries, would be affected by the loss of access to UK waters [71]. Therefore, understanding species distribution is key to optimizing fishing efforts and the sustainability of these resources in the future.
Satellite imagery has been used as a tool to support fish population studies and their changes, e.g., [7,11,72]. This study showed that SST was significantly and negatively related to Chl-a in most of the ICES divisions, indicating a decrease in Chl-a concentration with an increase in SST. This relationship was more pronounced in divisions located close to the coast, such as divisions VIIa (Irish Sea), VIIf (Bristol Channel), and VIId (eastern English Channel). The most likely cause of this relationship could be an increase in the stratification and a deepening of the thermocline leading to a reduction in nutrients in the euphotic zone [73]. Significant and positive correlations between SST and Chl-a were detected in oceanic divisions, such as VIa (west of Scotland) and VIb (Rockall), and in divisions located off the west coast of Ireland, such as VIIc (Porcupine Bank) and VIIk (southwest of Ireland-west). The significant positive correlations between SST and Chl-a could be explained by the temporal cycles of phytoplankton, which strongly depend on seasonal changes in the vertical water column structure, e.g., [74,75]. Seasonal solar stratification plays a crucial role in determining the light and nutrient requirements for phytoplankton growth and the onset of blooms throughout the year, e.g., [74]. As the spring season progresses, solar heating generates a buoyance flux that stabilizes the water column and, together with increased solar radiation, triggers the spring phytoplankton bloom in waters around Ireland, e.g., [72,74]. However, prolonged stratification would impair nutrient regeneration and thus negatively affect primary production. Significant increases in SST have been observed in some areas around Ireland [29], with trends ranging from 0.28 • C to 0.20 • C per decade [29]. If this warming trend continues in the future, it could lead to persistent stratification of the water column, resulting in a decrease in Chl-a concentration and, consequently, affecting local phytoplankton abundance and the food web that depends on it.
The results presented in this study also showed that surface waters around Ireland exhibit high spatial variability in Chl-a and K490 values. The highest Chl-a and K490 values were found in coastal divisions, such as divisions VIIa (Irish Sea), VIIf (Bristol Channel), and VIId (eastern English Channel). These results were consistent with in situ oceanographic data obtained through scientific surveys, e.g., the PELTIC12 survey [76]. Coastal areas are highly influenced by terrestrial runoff and associated inputs of nutrients, such as nitrogen and phosphorus, which can lead to increases in phytoplankton biomass. High Chl-a levels often indicate poor water quality, and the long-term persistence of these high values is considered an indicator of an environmental problem such as eutrophication [77]. Other studies, e.g., [78], have also found an increase in Chl-a levels in the northeast Atlantic between 52 • N and 58 • N since the mid-1980s. Our study suggests that this trend is continuing, and that Chl-a values are increasing in these areas.
Seasonal variability of Chl-a in Irish waters shows a high peak in spring and a low peak in autumn [75]. In winter, low light conditions inhibit phytoplankton growth, but in spring (March to April), phytoplankton biomass increases in response to increasing solar irradiance levels [75]. After the spring bloom, Chl-a levels decrease due to a combination of factors, including zooplankton grazing, nutrient depletion, and sedimentation [75]. An autumn bloom, which is shorter and less intense than a spring bloom, usually marks the end of the autumn seasonal growing period [75]. In our analysis, this pattern was evident in divisions VIIg and VIIh (the Celtic Sea), VIIb, (west of Ireland), and VIIj (southwest of Ireland-east). However, other patterns were also detected. Several ICES divisions were characterised by a single and prolonged peak in Chl-a values with its maximum detected in spring and summer months. The areas that showed this behaviour mainly corresponded to ICES divisions located in the west of Ireland, such as the divisions VIb (Rockall), VIIc (Porcupine Bank), and VIIk (southwest of Ireland-west). This single and long-lasting peak could be explained by a succession of different phytoplankton groups, such as the abundance of diatoms in spring and autumn and the dominance of dinoflagellates in summer [75]. Other authors have described the same behaviour in Chl-a values in the eastern Atlantic Ocean (a single peak in summer months), caused by above-average concentrations of particulate inorganic carbon (PIC) [79]. In coastal ICES divisions, such as divisions VIIa (the Irish Sea) and VIId (Bristol Channel), Chl-a did not show any particular patterns, except for minimum Chl-a values detected during the summer months. These coastal divisions often receive a considerable amount of terrestrial and riverine runoff, most notably in winter months, resulting in nutrient enrichment. Other studies reported different behaviour of Chl-a values in these areas. For example, some authors [80] reported an increase in Chl-a concentrations in the spring in the Irish Sea, which corresponds to division VIIa in this study. These authors reported that Chl-a levels in the Irish Sea reach a peak between March and May, or June, with concentrations higher than 40 mg/m 3 [80]. In this study, we did not find such high concentrations, probably because we used monthly averages rather than isolated in situ measurements, or due to recent and continuous changes in environmental conditions. Both Chl-a and K490 showed significant increasing trends in most ICES divisions. This result differs from our preliminary analyses [81], which found no trends in Chl-a values, using the same time series of data. This fact highlights the importance of considering the presence of outliers when calculating temporal trends of oceanographic variables such as Chl-a. To decompose the Chl-a and K490 time series and extract the temporal trends, the Seasonal and Trend decomposition using Loess (STL), incorporated in the R software function "stl.fit ()" [44], was used. The STL method has been shown to be robust to extreme values and unusual observations, i.e., outliers, and avoids effects on seasonal and trend components without reducing the size of the dataset [44]. The significant trends reported in this study were based on a nineteen-year, monthly time series. However, some studies have reported that a minimum length of the Chl-a time series is required to obtain a reliable trend value. For example, Beaulieu et al. [33] considered that continuous Chl-a time series of at least 27 years are required, while Henson et al. [82] estimated the minimum period to be 39 years. Beaulieu et al. [33] also stated that a minimum period of 43 years is required if the time series is interrupted. In any case, satellite data time series are constantly growing, and new missions will continue to maintain the temporal continuity of these data in the future, offering an unprecedented source of information for trend analysis.
One of the most important ecological insights from multidecadal observations is that changes can occur abruptly as discrete shifts from one state to another [53]. These changes can reveal important information about trends that cannot be detected with simple trend analysis [83]. In our case, most of the ICES divisions exhibited some structural change during these 19 years, with the exception of division VIIg (north Celtic Sea). This fact could be explained by the oceanographic nature of the Celtic Sea, which is surrounded by several fronts, such as the Irish Sea Front to the north, and the Ushant Front to the east. These fronts, along with the continental shelf in the southern part and some eddies produced by tidal currents, may provide some stability to this area. Even these assumptions should be studied in more detail, it is known that oceanographic fronts can separate water masses that may change little over hundreds of kilometres.
The highest number of structural changes was detected in division VIIb (west of Ireland). This high variability could be explained by the influence of freshwater input and nutrients from the Shannon, Clare (Corrib), and Laune rivers. In some ICES divisions, which include oceanic or intermediate depth waters, Chl-a exhibited multiple structural changes that coincided in time. In general, most Chl-a breakpoints within oceanic divisions were found in wintertime, between October and December. However, it is quite difficult to determine the causes of these shifts due to the lack of comprehensive monitoring programmes that would allow in-depth investigation of specific processes. For example, the observed structural Chl-a change between October and December 2002 spans over divisions VIa (west of Scotland), VIb (Rockall), VIIa (Irish Sea), VIIb (west of Ireland), and VIIf (Bristol Channel), indicating a relatively large-scale process. The mixing of upper ocean waters is strongly stimulated by the passage of weather systems during winter and early spring, disrupting the development of water-column stratification, bringing up nutrients, and thus influencing the timing of the seasonal increase in phytoplankton [84]. The variability and intensity of such mixing events are associated with atmospheric synoptic events, or storms [84], which would directly affect the variability and intensity of phytoplankton blooms. A chain of low-pressure systems (a persistent large low-pressure system) have been recorded in October 2002 [85]. The passage of these low-pressure systems could have initiated and promoted phytoplankton blooms over an extensive area, resulting in the observed October-December 2002 Chl-a breakpoints. The breakpoints in both the K490 and Chl-a time series were coincident in time, suggesting, as previously discussed, a direct relationship between both parameters and dependence between water attenuation and Chl-a values.
As expected, the analyses presented here have shown that Chl-a and K490 exhibit similar cycles and patterns in different ICES divisions, generally within divisions that are spatially close to each other. Marine waters are a fluid and ever-changing environment that do not fit strictly within the boundaries of the ICES-defined divisions and sub-areas, primarily outlined as geo-political delaminations. Considering Chl-a, K490, and SST parameters, a hierarchical cluster analysis revealed that the divisions considered could be categorised into three main groups (Figure 9). Two of the main groups could be classified as oceanic (VIa and VIb) and coastal (VIIa, VIIf, and VIId) waters, while the largest number of divisions could be designated as intermediate waters (VIIb, VIIc, VIIe, VIIg, VIIh, VIIj, and VIIk). This fact suggests that considering each division as an isolated and discrete management area could lead to a mismatch between environmental processes and associated fish population behaviour.
The transition from single-species fishery management to an ecosystem approach continues to progress, as environmental factors are incorporated into fish stock assessments, resulting in more appropriate and more effective management tools and systems, e.g., [7]. EAFM implementation requires a high temporal frequency of sampling, often over large regions. This study demonstrates that satellite data can provide a cost-effective and timeefficient approach to qualitatively and quantitatively assessing biophysical conditions and changes in the marine surface waters. SST trends are already influencing the onset of spawning, migration, and distribution of blue whiting, northeast Atlantic mackerel, and western horse mackerel as well as the recruitment of some gadoids in the Irish Sea, Celtic Sea, and west of Scotland [86]. The recently reported SST increase in surface waters around Ireland over the period 1982-2015 [29] and the changes in Chl-a trends reported in this study may already have implications for the life cycles and dynamics of fish populations. Pelagic fish species, such as herring, blue whiting, northeast Atlantic mackerel, and horse mackerel, are keystones of the food web, and changes in their abundance can have significant impacts throughout the marine food chain.  The transition from single-species fishery management to an ecosystem approach continues to progress, as environmental factors are incorporated into fish stock assessments, resulting in more appropriate and more effective management tools and systems, e.g., [7]. EAFM implementation requires a high temporal frequency of sampling, often over large regions. This study demonstrates that satellite data can provide a cost-effective and time-efficient approach to qualitatively and quantitatively assessing biophysical conditions and changes in the marine surface waters. SST trends are already influencing the onset of spawning, migration, and distribution of blue whiting, northeast Atlantic mackerel, and western horse mackerel as well as the recruitment of some gadoids in the Irish Sea, Celtic Sea, and west of Scotland [86]. The recently reported SST increase in surface waters around Ireland over the period 1982-2015 [29] and the changes in Chl-a trends reported in this study may already have implications for the life cycles and dynamics of fish populations. Pelagic fish species, such as herring, blue whiting, northeast Atlantic mackerel, and horse mackerel, are keystones of the food web, and changes in their abundance can have significant impacts throughout the marine food chain.
A systematic monitoring of biophysical changes and trends in the marine environments will be key to understanding the complex dynamics of marine ecosystems, especially bottom-up (primary production) forcing. Satellite-based data, in conjunction with conventional ship-based data, will enable the development and use of informative and reliable marine ecosystem indicators and variables (e.g., SST, Chl-a, turbidity) and support continuous assessment of the biotic and abiotic components of the marine ecosystem. Despite the recognised interactions between marine species and their environment, spatial management approaches remain largely limited to static boundaries and isolated sampling stations [87]. Consequently, any introduced fixed-time area closures would not always encompass the core habitat of species of concern and could unjustifiably restrict (or allow) fishing activity [87]. Static or seasonal area closures are usually based on historical data on species distribution, thus putting these areas at risk of losing ecological relevance as species' distributions shift with changing climate and human activity [87]. Satellite data could add value to species distribution models, where the major focus is to identify spatial patterns in datasets [88]. In addition, satellite-based data could be useful in developing A systematic monitoring of biophysical changes and trends in the marine environments will be key to understanding the complex dynamics of marine ecosystems, especially bottom-up (primary production) forcing. Satellite-based data, in conjunction with conventional ship-based data, will enable the development and use of informative and reliable marine ecosystem indicators and variables (e.g., SST, Chl-a, turbidity) and support continuous assessment of the biotic and abiotic components of the marine ecosystem. Despite the recognised interactions between marine species and their environment, spatial management approaches remain largely limited to static boundaries and isolated sampling stations [87]. Consequently, any introduced fixed-time area closures would not always encompass the core habitat of species of concern and could unjustifiably restrict (or allow) fishing activity [87]. Static or seasonal area closures are usually based on historical data on species distribution, thus putting these areas at risk of losing ecological relevance as species' distributions shift with changing climate and human activity [87]. Satellite data could add value to species distribution models, where the major focus is to identify spatial patterns in datasets [88]. In addition, satellite-based data could be useful in developing ecological response models to better understand the responses of species to changes in environmental variables [89,90].

Conclusions
In recent years, legislation and policy sectors have emphasised the need for a holistic approach to decision making for the protection and management of marine areas and living resources. The primary goal of EAFM is to maintain the health of marine ecosystems while ensuring appropriate use of marine resources for the benefit of present and future generations. A major limitation in the practical implementation of EAFM for fishery management is the sparse availability of information on the status of the overall ecosystem structure and processes. To overcome this limitation, some countries have made significant investments to support and maintain routine monitoring of the marine environment and fish populations.
In this context, satellite-based information can help fill existing data gaps and facilitate surface water assessment and monitoring. No other observational strategy can provide synoptic views of upper ocean optical properties with such a high spatial and temporal resolution and extent. This study demonstrates the advantages and benefits of satellite-based data in providing up-to-date knowledge of the ecosystem status of Irish waters, confided within ICES-defined divisions. The study also contribute to promote the incorporation of satellite-based data into living marine resource management, such as fishery management.