An Assessment of Atmospheric and Meteorological Factors Regulating Red Sea Phytoplankton Growth

: This study considers the various factors that regulate nutrients supply in the Red Sea. Multi-sensor observation and reanalysis datasets are used to examine the relationships among dust deposition, sea surface temperature (SST), and wind speed, as they may contribute to anomalous phytoplankton blooms, through time-series and correlation analyses. A positive correlation was found at 0–3 months lag between chlorophyll- a (Chl- a ) anomalies and dust anomalies over the Red Sea regions. Dust deposition process was further examined with dust aerosols’ vertical distribution using satellite lidar data. Conversely, a negative correlation was found at 0–3 months lag between SST anomalies and Chl- a that was particularly strong in the southern Red Sea during summertime. The negative relationship between SST and phytoplankton is also evident in the continuously low levels of Chl- a during 2015 to 2016, which were the warmest years in the region on record. The overall positive correlation between wind speed and Chl- a relate to the nutritious water supply from the Gulf of Aden to the southern Red Sea and the vertical mixing encountered in the northern part. Ocean Color Climate Change Initiative (OC-CCI) dataset experience some temporal inconsistencies due to the inclusion of different datasets. We addressed those issues in our analysis with a valid interpretation of these complex relationships. ( SV near zero). The peak in the DAOD in the NRS tends to occur in the late boreal spring (April–May), with negative SV in the autumn. These results indicate that different mechanisms are responsible for dust generation in the northern parts of the Red Sea as compared to the southern regions.

The 4 km Advanced Very High Resolution Radiometer (AVHRR) Pathfinder Version 5 SST dataset is a reanalysis of historical AVHRR data that has been improved using extensive calibration, validation, and other information to yield a consistent, research quality time series for global climate studies [38,39]. The collection of National Oceanic and Atmospheric Administration (NOAA) satellite platforms that is used in the AVHRR Pathfinder SST time series includes NOAA-7, NOAA-9, NOAA-11, NOAA-14, NOAA-16, NOAA-17, and NOAA-18. In this study, we used the monthly daytime SST data for the time period from September 1997 to June 2002 to extend our analysis farther back in time than is possible with the MODIS-Aqua SST dataset alone, and to provide better temporal consistency with the OC-CCI dataset. As a check, we compare the AVHRR SST data and the MODIS Aqua SST data from July 2002 to December 2009 in order to characterize the systematic observational differences between the two datasets. We adjust these differences to produce a combined SST dataset for the Red Sea that extends from September 1997 to December 2016.

Modern-Era Retrospective Analysis for Research and Applications Version 2 (MERRA-2) Dust Reanalysis Data
The Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA-2) is an atmospheric reanalysis dataset that is extending from January 1980 to present that is produced by the National Aeronautics and Space Administration (NASA) based on historical analysis using the Goddard Earth Observing System Model, Version 5 (GEOS-5) with its Atmospheric Data Assimilation System (ADAS), version 5.12.4. In order to assess the role of dust-related impact in the Red Sea, we use the NASA Goddard Online Interactive Visualization ANd aNalysis Infrastructure (GIOVANNI) tool [40] in order to obtain and process the monthly mean MERRA-2 data, including the field "Dust Scattering AOT 550 nm", which we designate as dust aerosol optical depth (DAOD) [41], as well as "Dust Dry Deposition" for "Bin-1" through "Bin-5" and "Dust Wet Deposition" "Bin-1" through "Bin-5", and "Surface Wind Speed" [42]. All the MERRA-2 data used is at 0.5 • × 0.625 • resolution covering the period from September 1997 to December 2016.

Multi-Angle Imaging SpectroRadiometer (MISR) Aerosol Optical Depth (AOD) Data
The Multi-angle Imaging SpectroRadiometer (MISR) instrument has been operational on the NASA Terra satellite since early 2000 and it provides observations of the daytime portion of the Earth from nine viewing directions in four spectral bands with an instrument swath of 400 km. In this research, we use the monthly MISR (MIL3MAE.004) AOD product at 558 nm, which is gridded at 0.5 • spatial resolution [43], in order to investigate the relationship between dust aerosols and Chl-a events in the Red Sea for one event, namely June 2010. Studies have shown that MISR AOD retrievals over desert sites tend to have better agreement with Aerosol Robotic Network (AERONET) ground-based measurements than MODIS, especially for low AODs, and MISR provides a useful complement to Remote Sens. 2018, 10, 673 4 of 24 MODIS AODs by capturing individual aerosol events, while still reproducing the overall aerosol climatology [44,45].

Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) AOD Data
The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite [46] has flown in formation with the sun-synchronous, near-polar A-train constellation of Earth observing satellites since 28 April 2006. Developed as a collaboration between the French space agency, Centre National d'Etudes Spatiales (CNES), and NASA, the main scientific objective of the CALIPSO mission is to provide range-resolved information on the vertical distribution of aerosols and clouds. The Cloud Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument uses a two-wavelength elastic backscatter Nd:YAG laser that transmits linear polarized light at 532 nm and 1064 nm, coupled with a receiver telescope of 1 m diameter that measures the perpendicular and parallel components of the attenuated backscatter at 532 nm and the total attenuated backscatter at 1064 nm [47]. The CALIOP Level 2 (L2) data includes information on the aerosol and cloud backscatter coefficient at 532 nm and 1064 nm, and the particle depolarization ratio at 532 nm [48]. Both aerosol and cloud profile datasets are provided at 5 km horizontal and 60 m vertical resolution, for an altitude range of up to 8.3 km above mean sea level (AMSL). The CALIPSO Version 3 (V3) algorithm discriminates aerosols into six subtypes: (i) clean marine; (ii) dust; (iii) polluted continental; (iv) clean continental; (v) polluted dust; and (vi) smoke [49]. The CALIPSO algorithm further assigns a specific Lidar Ratio (LR) to each aerosol subtype, which can used in conjunction with the retrieved backscatter coefficient profiles for the derivation of the extinction coefficient profiles at various levels within the atmosphere [50].
Dust aerosol and the optical properties of dust have been extensively studied using CALIOP data [51][52][53][54][55][56]. In the present study, the CALIOP V3 L2 Aerosol Profile product is implemented to derive pure dust extinction coefficient profiles over the Red Sea for the period of 2007 to 2015. The Climatological Dust Extinction coefficient profiles (Clim-DE) dataset, reported at 1 • × 1 • spatial resolution, are also used [57]. The Clim-DE dataset is based on the CALIPSO pure dust product [54] and the ESA-LIVAS climatology (via http://lidar.space.noa.gr:8080/livas/). The Clim-DE product has been used in previous studies to investigate the vertical structure of Saharan dust transport over Europe [58] and the three-dimensional features of the dust burden over East Asia [59].

Methods
To investigate the variation of dust deposition and its impact on oceanic production in the Red Sea region, a multi-sensor approach was used to evaluate Chl-a variability associated with dust deposition using DAOD and other environmental parameters, such as wind speed and SST. Using the MODIS-Aqua Chl-a product, Li et al. [19] revealed a two-month lag correlation positive dust anomalies and high Chl-a anomalies, as well as a two-month lag correlation between negative SST anomalies and high Chl-a anomalies. In this work, we use the OC-CCI dataset to further test and validate the results obtained from Li et al. [19]. Moreover, this research explores the possible factors that are contributing to the influence of aeolian dust on summertime phytoplankton blooms in the Red Sea by investigating other dust-related factors, namely, wind speed, dust sources, and dust deposition type.
A time-averaged map of Chl-a concentration for the Red Sea was generated with OC-CCI data from September 1997 to December 2016 ( Figure 1). The map shows the same meridional pattern of Chl-a concentration, as in previous research [3,19]-high values appear near the entrance of Gulf of Aden in the south and along nearby coastal areas; the northern Red Sea (20 • N to 28 • N) is lower in Chl-a than the southern Red Sea (13 • N to 20 • N), while the region that is south of the Farasan Islands (13 • N to 17 • N) has higher Chl-a amounts than the region just to the north (17 • N to 20 • N). Note, however, that the range of Chl-a values (0.15 to 2.15 mg/m 3 ) from the OC-CCI dataset is smaller than the range that was reported for the MODIS-Aqua dataset (0.1 to 5 mg/m 3 ) [19]. Based on the abundance and spatiotemporal distribution of surface Chl-a, in Figure 1, we partition the Red Sea into four domains at specific, integer latitudinal boundaries for convenience. The four regions, from north to south,  (17°N to 13°N). Note that these designations are slightly different than those that were used in previous studies [3,19]. For each of the four regions, time series analyses of the nearly 20 years (September 1997 to December 2016) monthly data for Chl-a, DAOD, SST, and wind speed were performed using scaled values, anomalies, and lag (cross) correlation. The scaled values, SV, are calculated by finding the mean of the entire time series, xm, subtracting this from each monthly value, xi, and then normalizing the result by the mean of the time series. Formally, this is represented by the equation: The use of the SV allows for time series of different variables to be represented on a similar scale and reveals potential long-term trends in the data, which are represented as deviations from the mean of the time series, xm.
Due to the large seasonality in the Red Sea, it is often helpful to consider the monthly anomalies, as given by: For each of the four regions, time series analyses of the nearly 20 years (September 1997 to December 2016) monthly data for Chl-a, DAOD, SST, and wind speed were performed using scaled values, anomalies, and lag (cross) correlation. The scaled values, SV, are calculated by finding the mean of the entire time series, x m , subtracting this from each monthly value, x i , and then normalizing the result by the mean of the time series. Formally, this is represented by the equation: The use of the SV allows for time series of different variables to be represented on a similar scale and reveals potential long-term trends in the data, which are represented as deviations from the mean of the time series, x m . Due to the large seasonality in the Red Sea, it is often helpful to consider the monthly anomalies, as given by: where x is the mean value that is calculated for that month from the entire dataset. To make the different datasets comparable, the monthly anomalies are scaled by σ, the standard deviation of the values for that month, which we term the Z-score: In order to uncover the relationships between two different parameters, both standard correlation and lag (cross) correlation analyses were done. The standard (Pearson) correlation coefficient, r, between two time series of variables, x and y, with N elements can be expressed as: where, Cov is the covariance function, and σ x and σ y represent the standard deviation of the x and y time series, respectively. The correlation coefficient can assume values between −1 and +1, with −1 representing perfect anti-correlation and +1 representing perfect correlation. Finally, as in Li et al. [19], we consider the lagged (cross) correlation r(k), for lag k, which is given by: where, ccvf is the cross-covariance function. When creating cross-correlations maps, all of the data are re-gridded to the same resolution using a bi-linear re-gridding algorithm. These maps reflect the relationships between phytoplankton-represented by Chl-a-and other atmospheric and oceanic factors in different regions within Red Sea. Lagged correlation maps, in particular, are used to attempt to uncover the mechanisms that are underlying the residence times of dust in the Red Sea and the release of bioavailable Fe into the oceanic mixed layer and its resulting effects on ocean productivity [60,61].

Time Series of Scaled Values for Chl-a, DAOD, and SST
The time series of the SV for three of the datasets, Chl-a from OC-CCI (green), DAOD from MERRA-2 (blue), and SST from blended AVHRR and MODIS-Aqua data (red) are shown in Figure 2 for the different regions of the Red Sea depicted in Figure 1.
A strong seasonal variation in all of the variables is immediately apparent, but is clearest in the SST data. In all four regions, the SST is lowest near the beginning of the calendar year (January-February), and highest in near the end of boreal summer and the beginning of boreal autumn (August-October). Close inspection reveals small, but systematic differences from one region to the next. The SST cools off slightly earlier in the SRS and warms slightly later, especially relative to the NRS. The magnitude of the SST SV is largest in the NRS and smallest SCRS, meaning that larger deviations from the mean are seen in the NRS.
There is also strong seasonality to dust events, particularly in the SRS, as shown in the time series of SV for DAOD. Visually, there is also good correspondence between the SV for DAOD and SST in the SRS, with elevated levels of DAOD corresponding to periods of elevated SST and decreased levels of DAOD corresponding to periods of lower SST. This relationship becomes progressively weaker to the north, with the NRS sometimes showing elevated DAOD events when the SST is near the mean value  Visually, there appears to be little relationship with DAOD in the NRS. In the SCRS, the SV for Chl-a seems to follow a six-month cycle, with a peak in January and another around June. This relationship is also apparent in the SRS. Positive SV values for Chl-a correspond well to positive SV values for DAOD, but with SV values that are near zero for SST. Finally, in the SRS, positive SV values in Chl-a lead the maxima in SV values for SST slightly, and again correspond nearly perfectly with positive SV values for DAOD.
When considering the largest positive excursions in SV for Chl-a, we note that these occur mainly in the southern part of the Red Sea, with the summers of 2011, 2010, 2005, and 2003 having strong events that are larger in magnitude than the typical wintertime Chl-a enhancements. Note that the Chl-a event that was identified in the MODIS-Aqua dataset in summer 2015, and was discussed in detail in Li et al. [19], does not rank as one of the largest Chl-a events in the OC-CCI dataset, but is visible as one of the few outliers in the later part of the time series for the SCRS. The large positive Chl-a events in 2011, 2005, and 2003 are significant in both the SCRS and the SRS, but have small or even negative SV in the NCRS and NRS. The 2010 summertime event in the SCRS is weak in the SRS, but is the largest event in the NCRS. The second largest event in the NCRS is in October 2002, and it appears to be confined primarily to this region.

Time Series of Z-Scores for Chl-a, DAOD, and SST
The large seasonality evident in all four regions of the Red Sea in SST, DAOD, and Chl-a in Figure 2 tends to obscure significant monthly anomalies in these parameters. For this reason, we show the Z-scores over the different regions of the Red Sea in Figure 3. Recall that the Z-scores represent the monthly values, with the long-term monthly mean subtracted, and divided by the standard deviation of the monthly values (see Equation (3)). A Z-score that is equal to one represents a monthly value that differs from the long-term mean by one standard deviation. If the data are normally distributed, then the expectation is that 95% of the data will have a Z-score ≤2, and only 0.3% of the data will have a Z-score >3. Of the three parameters, the only one that appears to be nearly normally distributed is the SST. Chl-a and DAOD Z-scores are skewed positive in all four regions, meaning that positive anomalies are more likely than negative anomalies.  The DAOD anomalies appear to be primarily negative prior to mid-2000. This happens to be when MODIS-Terra data became available, so the inclusion of this dataset into the MERRA-2 DAOD reanalysis time series likely introduces temporal inhomogeneity. With the exception of a few positive outliers, and the time period between 2010 and 2012, there is no clear pattern to the DAOD anomalies, especially late in the time series. The large positive DAOD anomaly in 2000 occurs in all four regions and it is correlated with a strong negative SST anomaly. However, when comparing across the regions, the largest DAOD anomaly corresponds to the smallest SST anomaly, and vice versa. Other DAOD anomalies appear in late 2010 and 2013 in all four regions, but there does not seem to be any relationship with SST. Finally, the large DAOD anomaly in 2015 only occurs in the SRS and SCRS, where it correlates well with a negative SST anomaly, but there is no DAOD anomaly at all in the NCRS or the NRS.
Even more than the DAOD dataset, the Chl-a Z-score time series shows clear temporal inhomogeneity. Visually, it is possible to separate the Chl-a time series into three distinct periods that are common to all four regions. The first period is prior to mid-2002, where the negative anomalies dominate in all the regions, with the exception of a single peak event in the NCRS and NRS in late and it is correlated with a strong negative SST anomaly. However, when comparing across the regions, the largest DAOD anomaly corresponds to the smallest SST anomaly, and vice versa. Other DAOD anomalies appear in late 2010 and 2013 in all four regions, but there does not seem to be any relationship with SST. Finally, the large DAOD anomaly in 2015 only occurs in the SRS and SCRS, where it correlates well with a negative SST anomaly, but there is no DAOD anomaly at all in the NCRS or the NRS.
Even more than the DAOD dataset, the Chl-a Z-score time series shows clear temporal inhomogeneity. Visually, it is possible to separate the Chl-a time series into three distinct periods that are common to all four regions. The first period is prior to mid-2002, where the negative anomalies dominate in all the regions, with the exception of a single peak event in the NCRS and NRS in late 1998. Also notable in this time period are the gaps in the time series in the SCRS and SRS. Table 1 lists the time period of each of the four sensors used to construct the OC-CCI Chl-a dataset [31]. According to Table 1, prior to mid-2002 only the SeaWiFS sensor was used to generate the OC-CCI Chl-a time series, meaning that persistent clouds or dust could obscure the underlying water surface, thus resulting in a data gap. The introduction of additional sensors helped to mitigate this issue [8,14,30]. The second time period runs from mid-2002 to early 2012 and is characterized by primarily positive Chl-a anomalies. Only the NRS shows any significant negative anomaly during this period, which occurs in early 2010. Table 1 shows that MERIS was no longer included in the OC-CCI Chl-a dataset after April 2012, due to the loss of contact with the Envisat-1 platform. Around the same time, the VIIRS data began to be included. The loss of MERIS and the inclusion of VIIRS corresponds to the third period evident in the time series, which is characterized by primarily negative Chl-a anomalies. Notable exceptions are a strong positive anomaly at the beginning of 2014 that can be seen in all four Red Sea regions, and another large anomaly at the beginning of 2015, which seems to be confined to the SCRS and SRS. The June 2015 event that is discussed in detail in Li et al. [19] is apparent in the Z-score time series, but only in the SCRS.
The time series shown in Figure 3 are suggestive of relationships among these three variables, particularly in the SCRS, and, to a lesser extent, the SRS, as discussed in our earlier work [19]. However, the temporal in homogeneities that are apparent in the both the Chl-a time and DAOD series indicate that caution must be exercised before drawing too strong conclusions from these particular datasets. We explore the representativeness of the datasets for different time periods in our later discussion (Section 3.3).

Comparision of Sensors Used in OC-CCI Data
When comparing the monthly means of Chl-a between MODIS and OC-CCI, (analysis done but not shown here), a low bias in OC-CCI relative to MODIS-Aqua at large Chl-a values was observed with high outliers in the mean for low Chl-a values. Looking at the monthly median values, the low bias remained, but the high bias is more or less eliminated. We believe that MERIS is the cause of this bias and for that we decided to run a comparative analysis between Chl-a, obtained from the OC-CCI data, against SST, DAOD, and wind observations and anomalies during three different periods amongst the sensors that aee listed in Table 1  As for the DAOD and Chl-a, it is rather complicated because of the lack of observations during major dust events, which make it hard to conclude a definitive correlative behavior. However, we observed an overall negative correlation between DAOD and Chl-a observations (Figure 4(c1-c4)) during all of the time periods, but with rather exceptional positive correlations in the coastal and southern regions of the Red Sea. This positive correlative behavior is not clear during the period 2012-2016 (Figure 4(c4)), mainly because the Chl-a summer observations are insufficient due to major dust events. It is also noteworthy that during 1998-2002, the negative correlation in 23°N-25°N was stronger than that of 2003-2011, something that we need to look into to better understand it. For DAOD and Chl-a anomalies (Figure 4(c5-c8)), a positive correlative behavior is observed over the whole region with a very strong positive correlation in the NRS during 2012-2016 (Figure 4(c8)), which may be attributed to eddy's or wind stirring up influence (Figure 4(b8)) or wet deposition.
In conclusion, we believe that the three different outlined periods more or less behaved similarly when compared to the entire period's behavior, namely during 1998-2016. However, we still acknowledge the presence of few differences that were discussed above. This finding strongly supports the validity of using OC-CCI Chl-a data to study its cross correlative behavior with SST, wind speed, and DAOD. Figure 5 shows the monthly anomaly, ai (see Equation (2)), maps in the Red Sea for June 2010 for Chl-a, SST, and AOD, as well the related meteorological parameters dust dry and wet deposition, and wind speed. The latter three parameters are taken from the MERRA-2 reanalysis. June 2010 was selected because it has a large positive Chl-a anomaly, which is particularly apparent in the NCRS and SCRS time series (see Figure 3). Over the entire time period 1998-2016, SST and Chl-a observations exhibit a consistent negative correlation during the three different periods (Figure 4(a1)), whereas the anomalies (Figure 4(a5)) of both the northern and southern-most regions of the Red Sea stood out. This could be owed to the upwelling of cooler and nutrient-rich seawater [17] and the induction of a cool and nutrient-rich current from Gulf of Aden [3,11,14], respectively. However, during the period 1998-2002, the central Red Sea region (19 • N-23 • N) exhibited a low correlation that is presented by the white colored area (Figure 4(a2)) while the SST and the Chl-a anomalies experienced a positive correlation (Figure 4(a6) due to the influence of eddy activity [2,19], a behavior that is also observed during the period 2012-2016 (Figure 4(a8)).

Anomaly Comparison of Chl-a Concentration and Other Factors for June 2010
A consistent correlative behavior between wind speed and Chl-a observations is clear during all of the time periods, where the northern Red Sea exhibits a negative correlation, while the middle and southern Red Sea exhibits a positive correlation (Figure 4(b1-b4)). This could be attributed to the fact that summer seasons experience the strongest periods of wind speed that can drive the Gulf of Aden currents into the Red Sea, leading to the phytoplankton bloom of the central and southern regions [14]. While the northernmost region shows phytoplankton blooming, mainly during the winter period due to weaker wind. For wind speed and Chl-a anomalies, the entire region exhibits a positive correlation (Figure 4(b5-b8)). An exception is observed during 1998-2002 (Figure 4(b6)), where the northern and central Red Sea (19 • N-23 • N) exhibited a different correlative behavior from the other periods. This can be attributed to the influence of wind stirring up the seawater to enhance nutrient mixing and the eddy activities over the two regions, respectively.
As for the DAOD and Chl-a, it is rather complicated because of the lack of observations during major dust events, which make it hard to conclude a definitive correlative behavior. However, we observed an overall negative correlation between DAOD and Chl-a observations (Figure 4(c1-c4)) during all of the time periods, but with rather exceptional positive correlations in the coastal and southern regions of the Red Sea. This positive correlative behavior is not clear during the period 2012-2016 (Figure 4(c4)), mainly because the Chl-a summer observations are insufficient due to major dust events. It is also noteworthy that during 1998-2002, the negative correlation in 23 • N-25 • N was stronger than that of 2003-2011, something that we need to look into to better understand it. For DAOD and Chl-a anomalies (Figure 4(c5-c8)), a positive correlative behavior is observed over the whole region with a very strong positive correlation in the NRS during 2012-2016 (Figure 4(c8)), which may be attributed to eddy's or wind stirring up influence (Figure 4(b8)) or wet deposition.
In conclusion, we believe that the three different outlined periods more or less behaved similarly when compared to the entire period's behavior, namely during 1998-2016. However, we still acknowledge the presence of few differences that were discussed above. This finding strongly supports the validity of using OC-CCI Chl-a data to study its cross correlative behavior with SST, wind speed, and DAOD. Figure 5 shows the monthly anomaly, a i (see Equation (2)), maps in the Red Sea for June 2010 for Chl-a, SST, and AOD, as well the related meteorological parameters dust dry and wet deposition, and wind speed. The latter three parameters are taken from the MERRA-2 reanalysis. June 2010 was selected because it has a large positive Chl-a anomaly, which is particularly apparent in the NCRS and SCRS time series (see Figure 3). The map of the Chl-a anomaly field in Figure 5a shows a large area of high Chl-a at around 16°N in the SCRS, with a somewhat smaller positive anomaly around 22°N in the NCRS. Note the large region of missing data between about 17° and 18°N, which could be due to persistent clouds or dust. The SST anomalies in Figure 5b show the expected correlation in the SCRS with low SSTs in the area of elevated Chl-a amounts. In the NCRS, however, the SST anomalies are slightly positive. The AOD anomaly map in Figure 5c Figure 5a shows that the positive Chl-a anomalies are not well correlated with the increased wet deposition in these regions, however. Finally, the wind speed anomalies in Figure 5f show the distinct signature of the wind jet that is associated with the Tokar Gap, which is discussed in terms of its relation to the Chl-a anomaly in June 2015 in Li et al. The map of the Chl-a anomaly field in Figure 5a shows a large area of high Chl-a at around 16 • N in the SCRS, with a somewhat smaller positive anomaly around 22 • N in the NCRS. Note the large region of missing data between about 17 • and 18 • N, which could be due to persistent clouds or dust. The SST anomalies in Figure 5b show the expected correlation in the SCRS with low SSTs in the area of elevated Chl-a amounts. In the NCRS, however, the SST anomalies are slightly positive. The AOD anomaly map in Figure 5c shows slightly elevated AODs at around 0.1 over most of the Red Sea during June 2010. Higher AOD anomalies are seen around 19 • N, centered near Port Sudan, Sudan. Interestingly, negative AOD anomalies are more directly associated with the positive Chl-a anomaly in the SCRS. The dust dry deposition map in Figure 5d shows small positive anomalies over most of the region, with the exception of the extreme southern Red Sea. The dust wet deposition in Figure 5e, in contrast, has a large positive anomaly in both the SCRS and SRS, with larger positive anomalies along the eastern coastline. Comparison with Figure 5a shows that the positive Chl-a anomalies are not well correlated with the increased wet deposition in these regions, however. Finally, the wind speed anomalies in Figure 5f show the distinct signature of the wind jet that is associated with the Tokar Gap, which is discussed in terms of its relation to the Chl-a anomaly in June 2015 in Li et al. [19]. In this case, the axis of elevated wind speeds corresponds well with the missing data in the Chl-a map in Figure 5a.

Correlation Analysis of Chl-a Anomalies with SST and DAOD Anomalies
To further investigate the relationships between SST and the MERRA-2 DAOD with Chl-a, the correlation coefficients, r (see Equation (4)), for the anomalies in these parameters for each month and each region within the Red Sea are given in Table 2. Table 2. SST/Chl-a and DAOD/Chl-a monthly anomaly correlation coefficients for the four Regions of Red Sea. Light blue indicates negative correlation between −0.5 and −0.7, dark blue indicates negative correlation <−0.7. Light red indicates positive correlation between +0.5 and +0.7. Dark red indicates positive correlation >+0.7. Significant negative correlations (<−0.5) are colored blue and significant positive correlations (>+0.5) are shaded red. Absolute values that are greater than or equal to 0.5 are shaded in light colors, and absolute values greater than 0.7 are shaded in darker colors. These limits are chosen because the coefficient of determination, r 2 , is 0.25 for r = 0.5 and approximately 0.5 for r = 0.7. This means that 25% and 50% of the association is explained by a linear relationship between the terms for these two thresholds, respectively. This same information is plotted in Figure 6 as a monthly time series for the SST/Chl-a anomaly relationship (Figure 6a) and the DAOD/Chl-a anomaly relationship (Figure 6b).

NRS NCRS SCRS SRS
As seen in Table 2, and Figure 6, SST anomalies are typically negatively correlated with Chl-a anomalies, while DAOD anomalies are positively correlated with Chl-a anomalies. These relationships are not unexpected, as phytoplankton tends to thrive in relatively cooler environments, while dust deposition can provide necessary nutrients, including Fe, which helps to foster phytoplankton growth, particularly in oligotrophic waters, like the Red Sea. The correlation with the largest magnitude is −0.785, found between SST anomalies and Chl-a anomalies in February in the SCRS. This is followed closely by the large positive correlation of +0.736 between DAOD anomalies and Chl-a anomalies in March in the NCRS. Other significant negative correlations are found in the SRS between SST anomalies and Chl-a anomalies in January and November.

December
SST/Chl-a −0.319 −0.151 −0.542 −0.569 DAOD/Chl-a +0.095 +0.083 +0.269 +0.212 Significant negative correlations (<−0.5) are colored blue and significant positive correlations (>+0.5) are shaded red. Absolute values that are greater than or equal to 0.5 are shaded in light colors, and absolute values greater than 0.7 are shaded in darker colors. These limits are chosen because the coefficient of determination, r 2 , is 0.25 for r = 0.5 and approximately 0.5 for r = 0.7. This means that 25% and 50% of the association is explained by a linear relationship between the terms for these two thresholds, respectively. This same information is plotted in Figure 6 as a monthly time series for the SST/Chl-a anomaly relationship (Figure 6a) and the DAOD/Chl-a anomaly relationship (Figure 6b). As seen in Table 2, and Figure 6, SST anomalies are typically negatively correlated with Chl-a anomalies, while DAOD anomalies are positively correlated with Chl-a anomalies. These relationships are not unexpected, as phytoplankton tends to thrive in relatively cooler environments, When considering the overall temporal structure of the correlations between SST and Chl-a anomalies that are shown in Figure 6a The correlation between DAOD anomalies and Chl-a anomalies in Figure 6b shows a different pattern than Figure 6a. First, the NRS has large positive correlations in February, May, and October. The SCRS and SRS are nearly identical, with the exception of June, which may be due to the residual issues in the SRS due to clouds or persistent dust in this month. The correlation in these two regions peaks in February, falls rapidly in April, peaks again in July, then falls in October, but peaks again in November. The NCRS shows characteristics of both the NRS and the two southern regions with a broad peak in correlation centered on March, declining until July-August, with a second peak in the correlation in September.
February and March are the months that are showing the largest number of significant correlations in all regions, while August is the only month that shows no significant correlations. Significant positive anomaly correlations between DAOD and Chl-a outnumbered the significant negative anomaly correlations between SST and Chl-a. With the exception of February and March, the former tend to be concentrated in the boreal summer months, while the latter are more frequent in the boreal winter. Finally, the number of months with significant anomalies of either sign is largest is the SRS (9), followed by the SCRS (8), the NRS (7), and the NCRS (5). Note, however, that these results are at least partly due to the selection of the threshold for significance.

Lag (Cross) Correlation Maps of SST, Wind Speed, and DAOD with Chl-a
In Li et al. [19], we discovered a clear two-month lag correlation between Chl-a and DAOD anomalies using a MODIS data. To expand this major finding through the longer time and to gain further understanding of how the factors discussed here influence Chl-a, the MERRA-2 data of DAOD and Surface Wind Speed as well as MODIS Aqua SST data are re-gridded to 4 km spatial resolution with linear re-gridding algorithm. This re-gridding algorithm is already applied and presented in Figure 4a-c. Through the current lag analysis, SST & Chl-a observations, and anomalies exhibits a gradually changing behavior (Figure 7(a1-a8)). For instance, we found that the negative correlation between Chl-a and SST of the northern region gradually flips to a positive correlation at lag 3 due to a seasonal change (Figure 7(a1-a4)). Moreover, the lag 3 map (Figure 7(a4)) shows a south-north positive trend implying on the inflow of the nutritious waters from the Gulf of Aden to the Red Sea environment. Slowly dissipating/increasing behavior in the anomalies of Chl-a and SST is observed as we move from 0 to 3 months lag (Figure 7(a5-a8)). It is worth noting that the SST/Chl-a anomalies (Figure 7(a5-a8)) use MODIS Aqua SST data during the period of 2003 to 2016, whereas the SST/Chl-a anomalies' plot during the period of 1998 to 2002 uses the AVHRR SST data (Figure 4(a6)). The northern region of the Red Sea shows an increasing negative behavior from lag 0 to 3 months lag between the Chl-a and wind speed observations (Figure 7(b1-b4)). This finding shed the light of the decline of the wind driven vertical mixing with seasonal change. On the other hand, a declining positive correlative behavior from 0 to 3 months lag is clearly observed over the northern and southern Red Sea regions between the wind speed & Chl-a anomalies (Figure 7(b5-b8)).
It is quite evident that the Red Sea environment exhibits a consistent negative correlative behavior between the Chl-a and the DAOD observations (Figure 7(c1-c4)) with clear positive correlative exceptions dominating the southern region and the coastal areas. This can be cautiously attributed to possible deposition scenarios. However, it is noteworthy that the Chl-a and DAOD anomalies shows a positive correlation at zero lag (Figure 7(c5)) and moving north-south trend from lag 1 to lag 3 ( Figure 7(c6-c8)). The exceptional positively correlated round-shaped area at SCRS at lag-2 ( Figure  7(c7)) agrees with the findings of Li et al. [19] that showed the bloom at the same region during June 2015 because of the eddy's contribution as well as the Tokar gap dust event. This indicates that the bloom of June 2015 was so profound to dominate the correlation plot, showing that the two-month lag high correlation may have resulted from this event. However, all of the plots are all about the statistical relationship, not necessarily proving the causation between the observations.

Calipso-Based 3D Climatology of Desert Dust Aerosol over the Red Sea
In addition to the horizontal variability of AVHRR and MODIS Aqua SST, MERRA-2 Dust Reanalysis Data, MISR AOD, and OC-CCI Chl-a product, the CALIPSO based ESA-EARLINET LIVAS Pure-Dust and Clim-DE products are used to understand the structure of dust aerosol layers over the study domain. The combined use of Pure Dust AOD and Clim-DE products provides the full three-dimensional desert dust aerosols over the Red Sea. The northern region of the Red Sea shows an increasing negative behavior from lag 0 to 3 months lag between the Chl-a and wind speed observations (Figure 7(b1-b4)). This finding shed the light of the decline of the wind driven vertical mixing with seasonal change. On the other hand, a declining positive correlative behavior from 0 to 3 months lag is clearly observed over the northern and southern Red Sea regions between the wind speed & Chl-a anomalies (Figure 7(b5-b8)).
It is quite evident that the Red Sea environment exhibits a consistent negative correlative behavior between the Chl-a and the DAOD observations (Figure 7(c1-c4)) with clear positive correlative exceptions dominating the southern region and the coastal areas. This can be cautiously attributed to possible deposition scenarios. However, it is noteworthy that the Chl-a and DAOD anomalies shows a positive correlation at zero lag (Figure 7(c5)) and moving north-south trend from lag 1 to lag 3 (Figure 7(c6-c8)). The exceptional positively correlated round-shaped area at SCRS at lag-2 (Figure 7(c7)) agrees with the findings of Li et al. [19] that showed the bloom at the same region during June 2015 because of the eddy's contribution as well as the Tokar gap dust event. This indicates that the bloom of June 2015 was so profound to dominate the correlation plot, showing that the two-month lag high correlation may have resulted from this event. However, all of the plots are all about the statistical relationship, not necessarily proving the causation between the observations.

Calipso-Based 3D Climatology of Desert Dust Aerosol over the Red Sea
In addition to the horizontal variability of AVHRR and MODIS Aqua SST, MERRA-2 Dust Reanalysis Data, MISR AOD, and OC-CCI Chl-a product, the CALIPSO based ESA-EARLINET LIVAS Pure-Dust and Clim-DE products are used to understand the structure of dust aerosol layers over the study domain. The combined use of Pure Dust AOD and Clim-DE products provides the full three-dimensional desert dust aerosols over the Red Sea. Figure 8 shows the seasonal spatial patterns of CALIPSO columnar Dust AOD (DAOD) at 532 nm, over the Red Sea and the surrounding areas averaged for December-February (DJF), March-May (MAM), June-August (JJA), and September-November (SON) between 01/2007 and 12/2015. It must be clarified that the seasons have been defined based on the temporal and spatial characteristics of Aeolian dust activity over the Red Sea [10,63]. From the geographical distributions of DAOD, it is revealed that the dust activity over the Red Sea peaks in summer, while it diminishes during winter. Moreover, a strong north-to-south and west-to-east DAOD increasing gradient is apparent over the study domain, throughout the year. The maximum DAODs (up to 0.5) are recorded over the southern parts of the Red Sea and Saudi Arabia in summer and spring, respectively. In autumn (SON), DAODs do not exceed 0.3, while in winter (DJF), the minimum DAODs (less than 0.1) throughout the year are observed over the eastern parts of the Sahara Desert.
Remote Sens. 2018, 10, x FOR PEER REVIEW 17 of 24 Figure 8 shows the seasonal spatial patterns of CALIPSO columnar Dust AOD (DAOD) at 532 nm, over the Red Sea and the surrounding areas averaged for December-February (DJF), March-May (MAM), June-August (JJA), and September-November (SON) between 01/2007 and 12/2015. It must be clarified that the seasons have been defined based on the temporal and spatial characteristics of Aeolian dust activity over the Red Sea [10,63]. From the geographical distributions of DAOD, it is revealed that the dust activity over the Red Sea peaks in summer, while it diminishes during winter. Moreover, a strong north-to-south and west-to-east DAOD increasing gradient is apparent over the study domain, throughout the year. The maximum DAODs (up to 0.5) are recorded over the southern parts of the Red Sea and Saudi Arabia in summer and spring, respectively. In autumn (SON), DAODs do not exceed 0.3, while in winter (DJF), the minimum DAODs (less than 0.1) throughout the year are observed over the eastern parts of the Sahara Desert. In order to illustrate the vertical distribution of dust aerosols over the Red Sea, the seasonal extinction coefficient profiles (Clim-DE), which were averaged over the period 2007-2015, have been produced for each sub region (NRS, NCRS, SCRS, SRS). As indicated in Figure 9, throughout the year, the main portion of dust aerosols is confined in the lowest troposphere (lower than 1 km), with maximum Clim-DE values ranging from 0.08 (Figure 9d) to 0.24 (Figure 9o) km −1 , depending on the sub region. Mean dust extinction coefficient gradually decreases with height, reaching values that are less than 0.01 km −1 up to 6 km −1 . Furthermore, elevated dust aerosol layers (2-5 km) are evident mainly between March and August and for latitudes southern to 20°N, encompassing the SCRS and SRS regions, while during autumn and winter, dust aerosol layers are in general suppressed below 5 km height. Among the sub regions, the highest extinction coefficients are found during MAM in the northern domains (NRS and NCRS), while the corresponding values in SCRS and SRS are observed during JJA. The synergistic implementation of the mean DAOD product (Figure 7) and the climatological extinction coefficient profiles (Figure 9) provides an insight regarding the seasonal variation of dust aerosols three-dimensional distribution above the Red Sea. In order to illustrate the vertical distribution of dust aerosols over the Red Sea, the seasonal extinction coefficient profiles (Clim-DE), which were averaged over the period 2007-2015, have been produced for each sub region (NRS, NCRS, SCRS, SRS). As indicated in Figure 9, throughout the year, the main portion of dust aerosols is confined in the lowest troposphere (lower than 1 km), with maximum Clim-DE values ranging from 0.08 (Figure 9d) to 0.24 (Figure 9o) km −1 , depending on the sub region. Mean dust extinction coefficient gradually decreases with height, reaching values that are less than 0.01 km −1 up to 6 km −1 . Furthermore, elevated dust aerosol layers (2-5 km) are evident mainly between March and August and for latitudes southern to 20 • N, encompassing the SCRS and SRS regions, while during autumn and winter, dust aerosol layers are in general suppressed below 5 km height. Among the sub regions, the highest extinction coefficients are found during MAM in the northern domains (NRS and NCRS), while the corresponding values in SCRS and SRS are observed during JJA. The synergistic implementation of the mean DAOD product (Figure 7) and the climatological extinction coefficient profiles (Figure 9) provides an insight regarding the seasonal variation of dust aerosols three-dimensional distribution above the Red Sea. The strong spatial and temporal variability of DAOD across the study domain is driven by the geographical features as well as on the different meteorological mechanisms favoring dust mobilization and uplifting over arid areas of the eastern Saharan Desert and the Arabian Peninsula [63][64][65]. Dust generating dynamical processes include meteorological mechanisms of different scales, such as cyclones and anticyclones [66,67], Nocturnal Low-Level Jets [68,69], and Haboobs [70,71]. Between February and April, mid-latitude Mediterranean cyclones associated with cold fronts passages result in a long-range transport of Saharan dust aerosol towards the northern Red Sea [63,72]. By contrast, during summer the major sources of dust aerosols are the Arabian Peninsula deserts (An-Nafud, Ad-Dahna, Rub-Al-Khali) and the local sources of Eritrea and the Republic of Sudan. Under favorable synoptic conditions, dust events take place over the southern Red Sea [63,73]. In addition, the dust aerosol transport and deposition processes are highly dependent on the local topography in the vicinity of the Red Sea. At the northern parts (NRS, NCRS), the less complex surface elevation and the flatter topography allow for the free transport of dust aerosol plumes. On the contrary, in the southern parts of the Red Sea (SCRS, SRS) mountain ridges along both the African and Arabian Peninsula coastlines confine dust aerosol flows creating persistent dust layers for extended periods [16]. The removal processes of dust aerosol particles include both dry and wet deposition [74,75]. Dry deposition (or gravitational settling) is the predominant removal mechanism of mineral particles from the atmosphere in the north Red Sea (NRS, NCRS), while over its southern parts, wet deposition plays a key role [72]. The higher dust aerosol load over the southern part of Red Sea [10] in combination with the predominance of wet removal processes results in an increased dust aerosol deposition and input to the southern parts of the Red Sea with respect to its northern parts.

Discussion
In this work, we studied a combination of atmospheric and meteorological factors, including SST, dust deposition (DAOD), and wind activities regulating Red Sea phytoplankton growth using The strong spatial and temporal variability of DAOD across the study domain is driven by the geographical features as well as on the different meteorological mechanisms favoring dust mobilization and uplifting over arid areas of the eastern Saharan Desert and the Arabian Peninsula [63][64][65]. Dust generating dynamical processes include meteorological mechanisms of different scales, such as cyclones and anticyclones [66,67], Nocturnal Low-Level Jets [68,69], and Haboobs [70,71]. Between February and April, mid-latitude Mediterranean cyclones associated with cold fronts passages result in a long-range transport of Saharan dust aerosol towards the northern Red Sea [63,72]. By contrast, during summer the major sources of dust aerosols are the Arabian Peninsula deserts (An-Nafud, Ad-Dahna, Rub-Al-Khali) and the local sources of Eritrea and the Republic of Sudan. Under favorable synoptic conditions, dust events take place over the southern Red Sea [63,73]. In addition, the dust aerosol transport and deposition processes are highly dependent on the local topography in the vicinity of the Red Sea. At the northern parts (NRS, NCRS), the less complex surface elevation and the flatter topography allow for the free transport of dust aerosol plumes. On the contrary, in the southern parts of the Red Sea (SCRS, SRS) mountain ridges along both the African and Arabian Peninsula coastlines confine dust aerosol flows creating persistent dust layers for extended periods [16]. The removal processes of dust aerosol particles include both dry and wet deposition [74,75]. Dry deposition (or gravitational settling) is the predominant removal mechanism of mineral particles from the atmosphere in the north Red Sea (NRS, NCRS), while over its southern parts, wet deposition plays a key role [72]. The higher dust aerosol load over the southern part of Red Sea [10] in combination with the predominance of wet removal processes results in an increased dust aerosol deposition and input to the southern parts of the Red Sea with respect to its northern parts.

Discussion
In this work, we studied a combination of atmospheric and meteorological factors, including SST, dust deposition (DAOD), and wind activities regulating Red Sea phytoplankton growth using 20 years (1997-2016) observation and reanalysis data. We discovered a bias in the OC-CCI data owed to the data merge from different sensors leading to some inconsistency over different time periods. The systematic high bias values occurred mainly during the service period of MERIS sensor. We presented this issue of inconsistency in the time series variation of Chl-a in Figure 3 and discussed it in Section 3.3. To overcome this problem, we divided the data into three periods: before MERIS (1998MERIS ( -2002, during MERIS (2003-2011), and after MERIS (2012-2016), and compared each time frame with the entire time period (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016). Our results confirmed that except for the central Red Sea during (1998)(1999)(2000)(2001)(2002) and for NRS during (2012-2016), the correlation maps of most Red Sea regions match those of the entire time period. This is an important conclusion that supports the validity of using OC-CCI data for the long-term correlative analysis that is the focus of this work. An evident negative relationship between Chl-a and SST anomalies is observed and a major decline in Chl-a is correlated with the warmer waters of 2015 and thereafter. Hence, we argue against previous research [7] proposing that the warmer climate conditions could make the Red Sea ecosystem more productive, since phytoplankton could not get underlying nutrition [76]. The vertical analysis of dust aerosol optical depth proved that DAOD is a reliable indicator of dust deposition over the Red Sea ecosystem. We found that dust deposition contributed at different extents over different regions of the Red Sea to the Chl-a anomalies. This is clear from the positive correlation between DAOD and Chl-a at different lags. It is noteworthy that the SCRS exhibits a two-month lag, thus confirming the impact of the anomalous event during June 2015 [19]. However, not all of the dust events could induce phytoplankton blooms, due to the varying nature of the dust sources. For instance, SRS received more wet deposition as compared to the NRS (see Figure 5), resulting in more bioavailable nutrient (e.g., Fe, Si) for ocean ecosystems. Meanwhile, the positive correlation between wind speed and Chl-a verified that preferred wind patterns could enhance the horizontal intrusion of nutrient-rich water to the southern Red Sea [3]. We found that stronger winds brought more nutritious water from Gulf of Aden into the Red Sea (see SRS in Figure 4(b5-b8)). Furthermore, strong wind also enhanced vertical mixing, thus bringing deeper water that is nutrient rich to be brought into the euphotic zone [3]. This is clear from the positive correlation between Chl-a and wind speed anomalies over the northernmost Red Sea as shown in Figure 4(b8). Ocean circulation is a quite important factor influencing the phytoplankton blooms distribution [3,19]. Hence, it is expected that the inflow of nutrient-rich seawater from the Gulf of Aden can affect most regions of the Red Sea, including NCRS, SCRS, and SRS. Our 19 years analysis showed a similar monthly correlation variability between Chl-a and SST anomalies for the central and southern regions, but not for the NRS, suggesting the role that the inflow played here (See Figure 6a). Moreover, since mesoscale anti-cyclonic eddies affect phytoplankton blooms through transferring nutrients and/or Chl-a to the open waters in the central Red Sea, our cross correlation maps showed the eddies' role, resulting in the October 2002 bloom that wsa reported in Sofianos and Johns [2] (see Figure 4(a2,a6,b2,b6)) and the June 2015 bloom reported in Li et al. [19] (see Figure 7, Chl-a and DAOD anomalies at lag 2). The CALIPSO AOD climatology supported the dust contribution to the Chl-a, as discussed above and showed by the positive correlation coefficients (Table 2), since the dust activity peaked during summer and diminished during winter (see Figure 8). The extinction coefficient vertical profiles validated the dust aerosols contribution since the elevated dust layers (2-5 km) were evident during summer season mainly over the SCRS and SRS regions (see Figure 9).

Conclusions
This study considers a combination of different factors that are regulating nutrient supply in the Red Sea environment that may contribute to the observed anomalous phytoplankton outbreaks. We found that the Red Sea environment experienced, as expected, a negative correlation between SST and Chl-a observations. This negative correlation is plausible since phytoplankton normally blooms Remote Sens. 2018, 10, 673 20 of 24 during winter. However, we also found that this negative relationship still holds between SST and Chl-a anomalies, with few exceptions that can be attributed to some eddy activities or possible wet dust deposition. It is clear that anomalously cooler water, either from upwelling or intruding current from Gulf of Aden, could bring nutrients leading to enhanced phytoplankton blooming. On the other hand anomalously warmer water experiences reduced phytoplankton blooming because of nutrients shortage that is induced by stronger stratification and reduced mixing layer depth. This is evident from the extremely high SST anomalies paired with extremely low Chl-a anomalies during 2015-2016. Hence, we believe that warmer climate conditions could make the Red Sea ecosystem less productive. The extent of the June, 2010 anomalous event over the entire Red Sea environment made it quite interesting and it was deeply investigated. A combination of factors, ranging from SST anomalies, wet and dry dust deposition, and elevated wind speeds all contributed to the extent of that event. The strong north-to-south, west-to-east increasing gradient of the DAOD explains the higher dust activity on the southern regions of the Red Sea environment during summer and spring seasons that agrees with the higher productivity. This is evident from the correlation maps between Chl-a and DAOD anomalies at specifically two and three months lags over the southern Red Sea yet still with some contributions to the other regions as well at 0 lag. Our multi sensor approach together with the correlative analyses enhanced our understating of varying contributing atmospheric and meteorological factors into the phytoplankton blooms. Further research will be conducted for specific cases of anomalous blooms over the Arabian Gulf to investigate the regional extent of contributing factors to other neighboring water bodies.
Author Contributions: Wenzhao Li implemented the methods, analyzed the data, and provided written sections to the corresponding author for revision and inclusion. Hesham El-Askary led the entire research effort and came up with the idea and plan of work. He was involved in all steps of data rendering, analysis, manuscript writing and revising. Mohamed Qurban provided suggestions and insights about eddies and productivity as well as participating actively in the discussions and manuscript preparation. Emmanouil Proestakis, Vassilis Amiridis, Antonis Gkikas and Eleni Marinou from National Observatory of Athens and University of Patras contributed CALIPSO Dust AOD analysis and pure dust climatological extinction coefficient profiles. Michael Garay performed the time series analysis helped with the lag correlation, OC-CCI data correction and assisted with the writing, particularly the revisions. Olga Kalashnikova provided technical opinions assisted with the Giovanni data collection and time series analysis. Thomas Piechota provided technical opinions to the overall idea and helped in the language editing and writing process. K. P. Manikandan contributed to the field investigation for future cross comparison with satellite observations as well as in the data processing and in the group discussions. All authors read and approved the final manuscript.