Global gridded Chl a concentrations at 9 km spatial resolution, based on the MODIS Aqua sensor, are available from NASA's Ocean Color data portal (http://oceandata.sci.gsfc.nasa.gov). The present work uses monthly, climatological and 8-day composite Chl a data from November to March during the winter seasons from 2002 to 2013. The MODIS Chl a algorithm derives the near-surface Chl a concentration (expressed in milligrams per cubic metre, mg m-3), from remote-sensing reflectance . The climatological dataset is used for the zoning procedure. Monthly data are used in the time series analysis and time-lagged correlations are computed using 8-day composites.

PAR is the quantum energy flux from the sun in the visible spectrum (expressed in einsteins per square metre per day, E m-2 day-1). Under cloud-free conditions PAR is calculated from radiance measurements at the top of the atmosphere, derived from satellite remote sensing data in the visible spectral range and corrected for the effects of clouds . PAR used in this study is also available from the above-mentioned ocean-colour data portal of NASA.

We used MODIS Aqua daytime, 8-day, composite SST at a spatial resolution of 9 km, available from NASA's Ocean Color data portal. The SST is derived from radiance signals in the thermal infrared range at 11 and 12 µm, from the satellite sensor. The brightness temperatures are derived from the observed radiances by inversion (in linear space) of the radiance versus blackbody temperature relationship .

DOT used is calculated utilising the method of and is given as follows: DOT=AOT(fan-f)-AOTma(fan-fma)(fan-fdu), where AOT is the aerosol optical depth, which is obtained from MODIS Aqua (http://oceancolor.gsfc.nasa.gov). AOT represents total aerosol content in the atmospheric column, while DOT indicates just the dust content in the atmospheric column. Here, f is the fraction of AOT contributed by fine particles. reported f to be 0.25 over the northern Indian Ocean. The quantities fan, fma and fdu are respectively the fine-mode fractions of anthropogenic aerosol, maritime aerosols and dust. Following the work of and , fan is taken as 0.90. Similarly, fma is assumed to be 0.47, and fdu is set at 0.25. The fma value is an average value for the period of 2003–2011 over the western part of the equatorial Indian Ocean and fdu is based on satellite values during dust outbreaks in the Middle East. Also, AOTma is the maritime AOT, calculated according to , as follows: AOTma=0.13⋅exp⁡0.09⋅w, where w is the wind speed (m s-1). This study used wind at 10 m obtained from the ERA-Interim reanalysis.

The ERA-Interim reanalysis data of 12-hourly wind components at 10 m simulated by an atmospheric model from the European Centre for Medium-Range Weather Forecasts (ECMWF) at 1.0∘ × 1.0∘ spatial resolution were retrieved . These ERA-Interim wind fields were used to calculate the wind speed.

The MLD for the northern Arabian Sea used in this work is defined as the depth where the temperature is 1 ∘C colder than that at the surface temperature . In this study the vertical temperature-profile data are weekly averages from a Hybrid Coordinate Ocean Model (HYCOM) simulation for the Indian Ocean (; ). HYCOM combines the optimal features of isopycnic-coordinate and fixed vertical grid ocean circulation models in one framework. The adaptive (hybrid) vertical grid conveniently resolves regions of vertical density gradients, such as the thermocline and surface fronts. A detailed analysis and validation of this ocean model for the Indian Ocean can be found in . A comparison of the MLD obtained from the HYCOM modelled data using both temperature and density criteria with the Argo datasets available for winter is carried out. A total of 6256 profiles are collocated for winter for the entire study area. MLD calculated from density criteria have higher RMSD and error percentage (RMSD: 36 m and an error of 68 %) compared with that derived from temperature criteria using 1, 0.5 and 0.2 ∘C (RMSD: 20 m and an error of 28 %). This analysis showed better MLD derivation is with temperature criteria. Hence, a second analysis based on different temperature-based MLD criteria (1, 0.5 and 0.2∘ drop from that at surface) with the Chl a in the six zones were carried out. From this analysis, it was found that MLD calculated using temperature criteria of 1 ∘C explained the Chl a pattern in each of the six selected zones more accurately than those computed using other temperature thresholds. This is the reason for including temperature-based MLD in the present work.

The present study utilises monthly climatological nitrate profiles available from NOAA National Centers for Environmental Information (NCEI)/World Ocean Atlas 2013 (WOA 2013) (http://www.nodc.noaa.gov). WOA 2013 includes global nutrient profiles at 1∘ spatial resolution, which is the average of all unflagged interpolated values from all available in situ observations . Climatological data of nitrate used in this study are objectively analysed values for each winter month, such that nitrate availability in each zone is calculated by averaging nitrate values within the mixed layer determined from HYCOM.

PCA is a statistical method that uses orthogonal transformation to identify the principal components (PCs) contributing to the variance of a signal. This method normalises the dataset and computes covariances, eigenvectors and corresponding eigenvalues for each PC. The eigenvectors are then sorted by decreasing eigenvalues . The first PC is oriented in the direction of the largest variation of the original variables and passes through the centre of the data distribution. The second largest PC lies in the direction of the next largest variation, passes through the centre of the data and is orthogonal to the first PC, and so forth.

The process called k-mean clustering is a signal processing method used to partition a given set of observation vectors into k number of clusters, where k can be any integer greater than 1 . This method generates a set of centroids, one for each of the k clusters. Observation vectors are classified into clusters such that each observation vector is assigned to that cluster for which the total distance from vector to cluster centroid is minimum. For example, if a vector A is closer to centroid i than any other centroids, then A belongs to the cluster i.

Chl a winter variability revealed six distinct ecological zones in the Arabian Sea, which has been compared with the Longhurst biogeographical classification of marine provinces for the study area. The area analysed falls into two Longhurst provinces. These are respectively the north-west Arabian upwelling province (ARAB), covering the west and central part of the study area, and the western India coastal province (INDW) in the eastern part of the study area. The border between these two Longhurst provinces is demarcated with a pink line in Fig. 4a.

Zone 1 is located in the northern part of the study area (Fig. 4a). During winter, moderate (5–10 m s-1) north-easterly winds blow over the area. Intense cooling is reported in this region, which enhances primary production . Similar to zone 1, intense cooling and high production also occur in zone 2 . The first three zones have a similar Chl a pattern, such that peak values occur during the month of either February or March, and for January Chl a concentration is always half of its peak value. These three zones are regions where strong convective mixing occurs during winter, which enhances Chl a concentration in these regions . The cool and dry north-easterly winds blowing onto the region from the adjacent land territories enhance cooling at the ocean surface. Consequently, increased evaporation leads to a decrease in surface temperature and an increase in surface salinity and density, creating convective mixing . All three of these ecological zones are stretched across both the Longhurst provinces, with the majority of the area located in the ARAB province. It should be mentioned here that the ARAB province with upwelling in the Arabian Sea according to Longhurst's classification represents provinces with strong upwelling during summer and with strong convective cooling during winter. The southern part of the study area includes zone 4, where winter cooling is less intense and less marine production occurs compared with zones 1–3 . However, similar to zones 1–3, zone 4 also is split across the two Longhurst provinces, with the western and central parts of zone 4 falling into the north-west Arabian upwelling province and the eastern part of the zone into the western Indian coastal province. Zone 5 includes the coastal area along the Oman coast between 18 and 22.5∘ N, and the coastal region along the west coast of India from 16 to 23∘ N is included under zone 6. These coastal areas are highly productive during winter .

The physical mechanisms in the northern and north-western part of the Arabian Sea are very different from those in the eastern part. Strong convective mixing prevails in the northern and north-western parts of the study area. The strong stratification in the east, due to the presence of low-salinity and high-temperature water limits convective mixing in the eastern part of the study area . The eastern part comprises zone 6, a part of zone 3 and zone 4. Among these three zones, zone 6 is a coastal area and is vulnerable to coastal complex processes. Zone 6 is also an upwelling region .

For comparing Chl a in six zones using Longhurst's provinces, we have classified zone 1, zone 2, zone 3, zone 4 and zone 5 in the Longhurst ARAB province and zone 6 as the INDW province (Fig. 4a). Maximum Chl a observed during February is consistent in both provinces of Longhurst as well as the present six zones. During winter, ARAB (0.5–0.8 mg m-3) and INDW (0.4–0.6 mg m-3) have low values of Chl a with a similar range of variability . However, we have identified a high Chl a concentration (> 0.5 mg m-3) in the entire study area, with significant differences between various parts, particularly higher values to the waters closer to the coast. From our analysis, it is clear that the northern parts have higher concentrations of Chl a, with decreasing concentrations towards the south. Also, the variation between each zone was identified and showed higher concentrations of Chl a in the western zones compared to the more eastern zones. With only two Longhurst provinces, such a spatial difference is not detectable. This spatial difference is due to the difference in physical mechanisms as mentioned in the above paragraph. Longhurst classification is based on 1∘ resolution global Chl a maps in the context of regional meteorological and physical oceanographic variability . It also uses Chl a observations from different time periods. In contrast, the present study utilises primarily Chl a concentration obtained from satellite sensors at about 100 times the resolution used by Longhurst for regional mapping and classification of ecological zones in the northern Arabian Sea. Hence, this regional classification could delineate the spatial Chl a variability better and the obtained zones contain more detailed regional information. This study is restricted to the analysis of data for the winter season. This work intends to characterise a more complete delineation of ecological zones and the mechanisms driving marine production in the study area during winter. Therefore, the influence of other ecological factors such as SST, MLD, PAR, wind and nutrients on Chl a production is included in the interpretations of the Chl a pattern in each of the ecological zones. The Longhurst classification accounts for the differences in physical conditions between the north-western and eastern parts. However, according to , downwelling in the eastern part in winter cannot extend to the northern boundary of West Indian coastal province of Longhurst. Since the present zonal classification limits zone 6 from extending to the entire northern portion of the Longhurst province, our zones seem to be realistic in capturing regional Chl a variability during the winter season.

The annual winter climatology (seasonal average Chl a values over the winter period, November–March, from 2002 to 2013) of Chl a distribution revealed distinct features for each of the identified ecological zones (Fig. 4c). Based on the variability of Chl a concentrations, zone 1 experiences maximum bloom intensity between 1.5 and 9.6 mg m-3 with a mean of ∼ 2.6 mg m-3 and standard deviation of 0.7 mg m-3. Next to zone 1, high Chl a prevails in zone 2, with a range of 1.4 to 7.0 mg m-3 and a mean ∼ 2.8 mg m-3. Standard deviation observed in Chl a is the same for both zones. Moderate values of Chl a (1.3 to 1.9 mg m-3) are observed in zone 3, zone 5 and zone 6. Though similar ranges are observed for these three zones, the temporal evolution is different. In zone 3, Chl a varies between 0.5 and 4.2 mg m-3, with 0.3 mg m-3 standard deviation. Among coastal zones, zone 6 Chl a standard deviation is higher (0.8 mg m-3), with a range of 0.9 to 6.8 mg m-3, than in zone 5 (0.6 mg m-3), with a range between 1.0 and 4.3 mg m-3. Minimum value of Chl a for the winter is observed in zone 4 (0.2 to 1.2 mg m-3); in this zone the lowest mean (0.5 mg m-3) and lowest standard deviation (0.2 mg m-3) are also observed. The Chl a geo-spatial statistical variation in the study area clearly demarcates different ecological zones; however, in situ observations are scarce in the study region, and this limits the possibility to determine the accuracy of the satellite data and thus the accuracy of the delineated zones.

In general, Chl a concentration in zones 1 and 2 follows a typical wintertime cyclic variability with its peak values during the month of February (Fig. 5c). Throughout the study period, the Chl a concentration during February is at least double the concentration during January in these two zones. SST follows an inverse pattern compared with that of Chl a, such that SST minima coincide with Chl a maxima. Surface waters are relatively warm (> 27 ∘C) during November and cool as winter progress, with stronger cooling in zone 1 and 2 (Fig. 5b). By January, SST has reduced by 2.5–3.0 ∘C in both zones with a minimum of ≈ 23–24 ∘C occurring in February. In March, the SST increases to 23–26 ∘C. Although the Chl a range is approximately the same for both zones, comparatively SST is lower in zone 1 than zone 2. The inverse relationship between SST and Chl a have a weak correlation coefficient (correlation coefficients mentioned in this work are statistically significant at 95 % confidence interval) in zone 1 (r=0.39, n=60) and zone 2 (r=0.55, n=60).

A deepening of the MLD during winter is seen in both zones (Fig. 5b). During November, the MLD is shallow (≈ 35 m), and as winter progresses, MLD deepens to ≈ 80 m in January and in zone 1 to 90–110 m during February. In general, the MLD in zone 2 is about 10 m shallower than in zone 1 during January and February. The MLD starts to become shallow again in March. The peak concentrations of Chl a coincide with the deepest MLD. However, MLD and Chl a in zone 1 and 2 are moderately correlated (correlation coefficient, r=0.28).

During winter in the study area, SST cooling initiates MLD deepening. The decrease in SST is mainly due to evaporation, which has dual effects, i.e. increase in salinity and reduction in temperature, causing increased density of surface water (Naqvi et al., 2006), as a consequence of which convective overturning takes place. As winter progresses, SST drops and convective overturning occurs (salinity and temperature effect), increasing the MLD . MLD also influences SST variability. For example, when the MLD deepens, SST will decrease as cool water is mixed toward the surface. On the contrary, during a shallow MLD, SST is generally higher . Hence, both of these environmental parameters are dependent on each other and both influence marine primary production in the study area. Wind speed fluctuates strongly for zones 1 and 2. In zone 1, the maximum variability (0.5–3.0 m s-1) is seen during November and December and for zone 2, the wind varies strongly throughout winter, with maximum wind speed (0.5–3.0 m s-1) for December and January (Fig. 5a). High inter-annual variability is seen in wind speeds along the two zones, with peak wind speed in any one of the months between November and February. Only in certain years did moderate wind (< 3 m s-1) coincide with high Chl a, and the correlation coefficient confirms that Chl a and wind speed are not correlated in zone 1 and 2 (r=0.09, n=60).

PAR follows the seasonal cycle of incoming solar radiation . PAR is the waveband of light that is used in photosynthesis, and it is closely correlated with total incoming solar radiation heating the water column. Hence, an increase in PAR is accompanied by higher surface temperatures and associated with enhanced stratification, which results in reduced mixing and vice versa . Decreasing PAR (33–36 E m-2 day-1) prevailed in the study area during November to December for both zones, which corresponds to a decreasing trend in temperature and a deepening MLD cycle. Contrarily, when PAR increased after December, surface temperature started increasing and mixing was reduced. However, there is a 1-month time lag between the onset of increasing PAR and the onset of increasing SST. The increase also coincides with a reduction of the MLD. This is due to the high heat capacity of water and the large amount of energy required to heat the water column when the MLD is deep. Low SST and deep MLD favours increased nutrient supply to the euphotic zone and hence production increases in these zones by February. These transitions, in terms of a reduction in SST and peak MLD, initialise algal blooms. Hence, PAR also influences production indirectly, by affecting the stratification that controls nutrient availability. There is stronger correlation between Chl a concentration and PAR in zone 1 (r=0.69, n=60) compared to zone 2 (r=0.49, n=60).

Peak Chl a concentrations occurring during February coincided with the lowest SST (< 25 ∘C) and deepest MLD (90–110 m). Thus, the increased amount of Chl a is found to be directly related to sea-surface temperature variability (i.e. cooling) and the deepening of the mixed layer. A similar inverse relation between productivity and SST is observed in the Indian Ocean by Singh and Ramesh (2015). PAR was found to have an indirect influence on primary production in these zones. PAR increases surface temperature and as a result mixing gets reduced and vice versa. During the month of January, PAR increase coincides with SST reduction and MLD deepening, with the increase in PAR starting about 1 month before the reduction in SST and deepening of MLD. SST reduction and MLD deepening increases nutrient supply to the mixed layer, thus enhancing production. Nitrate is high at 100 m depth (> 12 µmol L-1), and thus a deepening of the mixed layer beyond 100 m will mix up nutrients towards the surface . This suggests that the highest Chl a concentrations are due to the increase in nutrients by a deepening of the MLD triggered by cooling (Figs. 5, 6, 7, 9) in December and January. Wind influence is not strong in these zones. Though the wind speed is relatively low (< 2 m s-1) during most years, certain cases with moderate wind speed (≈ 3 m s-1) are observed in these zones. Moderate wind (5 m s-1) occurring during January to February could have enhanced mixing and thus production in zone 2.

Chl a, SST, MLD and PAR values in zones 3 and 4 followed similar seasonal patterns of variability to those in zones 1 and 2 (Fig. 6). However, the range of values are different in these zones compared with the ecological zones further north. The magnitude of Chl a concentration in zone 3 is 2 to 3 times less than in zones 1 and 2 and Chl a in zone 4 is about half of the Chl a concentration in zone 3. Thus, the Chl a concentrations occurring in zones 3 and 4 differ from the first two zones. The inter-annual variability of SST, MLD and PAR is higher in the third and fourth zones compared to the two zones further north. In zone 3, as it is closer to the Equator, PAR is 3–4 E m-2 day-1 higher, SST is 1.5 ∘C warmer and MLD ≈ 10–15 m shallower than in zones 1 and 2. Furthermore, PAR in zone 4 is 2–3 E m-2 day-1 higher, SST is 1.0 ∘C higher and MLD is 10 m shallower than in zone 3. The fact that Chl a concentration in zone 3 is less than in zones 1 and 2, and Chl a in zone 4 is still lower than that in zone 3, affirms that variation in Chl a production in the northern Arabian Sea is strongly related to physical parameters viz. SST, MLD and PAR. SST is an indirect indicator of favourable conditions for algal blooms. Low SSTs can be the result of intensified convection, which will also be manifested by increased mixed-layer depth and entrainment of waters rich in nutrients . The warmer SSTs in zone 3 and 4 indicate that the rate of convection is most likely weak in zone 3 and even weaker in zone 4 compared to zones 1 and 2. This indicates that phytoplankton production in these zones could be linked to the convectional strength and increase in nutrients and that the production is limited by nutrients in the period before and after the bloom. In zone 3 and zone 4 the inverse correlation between SST and Chl a is stronger compared to zone 1 and 2, with r=0.62 and 0.70, respectively.

The availability of nutrients is the prime factor influencing production. A deepening of the mixed layer will increase nutrient availability, but the magnitude depends both on the depth of the MLD and on concentration of nutrients below the mixed layer. The relatively shallow MLD and high SST in zones 3 and 4 compared with zones 1 and 2, suggests low transport of nutrients into the mixed layer in zones 3 and 4 compared to the other two zones. MLD and Chl a productivity in zones 3 and 4 are correlated (r=0.50, n=60 and r=0.56, n=60, respectively). The indirect influence of solar radiation in maintaining SST and MLD and thus nutrient availability is evident from higher PAR, higher SST and more shallow MLD values in zones 3 and 4 compared with zones 1 and 2. Hence, a direct dependence of SST cooling and deepening of the MLD and indirect dependence of PAR on the primary production is evident in the first four zones. In zone 3, a weak correlation exists between PAR and Chl a (r=0.41, n=60), and in zone 4 these two parameters are not correlated at all. The increasing wind speed pattern prevalent during winter indicates that wind mixing could be the prime factor governing the ecological dynamics in this zone during winter. Chl a has a weak positive correlation with the wind speed in zone 3 (r=0.30, n=60) and moderately correlated in zone 4 (r=0.47, n=60). A relatively warm surface and shallow mixed layer in zone 4 indicate weak convective overturning . Hence, wind-induced mixing in this zone can influence production.

Elevated values of Chl a (> 2.5 mg m-3) persist in zones 5 and 6 throughout the winter season, with high levels of variability. This suggests that the dynamics in the coastal and continental shelf zones 5 and 6 are more complex than open-ocean waters (zones 1, 2, 3 and 4). Chl a in zone 5 shows significant inter-annual variability for the winter period with its peak value during February. In zone 6 there is low variability of Chl a for the winter period in January, while the range of variability is high for both December and February. In zones 5 and 6, MLD maxima and SST minima occurred either during January or February (Fig. 7b). During January and February, in zone 5 MLD varied between 70 and 80 m, and in zone 6, MLD varied between 80 and 100 m. A comparison between MLD values in zones 5 and 6 with those in zones 1 and 2 shows that MLD is shallower in zone 5, whereas the variability of MLD in zone 6 is comparable to that in the first two zones. MLD variability for the winter is consistent with Longhurst's observations; however, the range of MLD in Longhurst is smaller (range from 40 to 70 m) compared to present study (40–110 m) for the winter period. Additionally, SST is higher during January–February in zones 5 and 6 (> 24.5 ∘C) compared to SST values in zones 1 and 2 (23.5–24.5 ∘C). The correlation coefficient between MLD and Chl a is lower in zone 5 (r=0.53, n=60) than in zone 6 (r=0.69, n=60). However, the inverse relation between SST and Chl a concentration is higher in zone 5 (r=-0.64, n=60) compared to zone 6 (r=-0.54, n=60). PAR ranges for December and January in zones 5 and 6 are almost equal to PAR ranges in zones 3 and 4 (36–38 E m-2 day-1) and are higher compared to zones 1 and 2 (30–34 E m-2 day-1) for the same period. A weak inverse correlation (r=-0.27, n=60) exists between PAR and Chl a in zone 6, while in zone 5 these parameters are not correlated (r=0.12, n=60). For zone 5, wind and Chl a production are weakly correlated (r=0.30, n=60), while in zone 6, these parameters are not correlated (r=-0.09, n=60).

In zone 5, low SST prevails, which is an indicator of strong convective activity. Again, in zone 6, high SST coincides with deep MLD and strong wind. Wind is reported as the one of the main forcing factors in the INWM by , which is consistent with our present study. Comparatively warm surface water indicates convective overturning is weak, and in addition the presence of strong wind suggests production in this zone could be controlled by upwelling induced by wind. Production in zone 6 was found to be more complex with the influence of wind. and indicate that nitrogen fixation may provide a significant contribution to production in zone 6. However, the results are based on experiments with limited temporal coverage.

Multiple linear regression analysis is carried out to understand the combined effect of all chosen environmental parameters on Chl a production. Multiple linear regression (MLR) is performed here on normalised values of selected parameters, such that individual values are subtracted from the minimum value of observation and then divided by the range of observation. MLR equations with r values are tabulated in Table 1. The MLR equations for all six zones are found to be statistically significant and are carried out using 60 data points. WND represents wind speed and WNDa the wind speed alongshore component in the coastal zones (zones 5 and 6).

In general, MLR analysis confirms that production is controlled by surface cooling, enhancement of PAR and deepening of MLD. However, the dependence of each of these variables differs within each zone. A negative impact of wind is observed in the first three zones and a positive influence of wind in the last three zones. As for zone 5 and 6, alongshore wind components are considered, and the southward component enhances production in these cases. However, the negative wind speed coefficient observed for first three cases may be due to the fact that the time span considered in this work is monthly and the effect of wind occurs on shorter timescales. In the first zone, for a 1 % (0.01) increase in normalised light availability as well as a 1 % normalised cooling increased the Chl a concentration by 0.0047 and 0.0023 mg m-3 respectively. In the second zone PAR had the most influence on production (for each 1 % increase in PAR chlorophyll increased by 0.0041 mg m-3), followed by MLD and SST. For the third zone, MLD and PAR have the highest influence on Chl a. Cooling has a major influence in the fourth zone. In the fifth zone, MLD, PAR and SST have major control of production, similar to zone 2; however, the rate of enhancement differs between these zones. In zone 6, cooling also has a major influence, which is followed by a PAR decrease and MLD deepening. An inverse relation of PAR to Chl a is observed in this zone. It should be mentioned here that, unlike in the other five zones, two Chl a maxima, SST minima and a deepening of MLD with an initial peak are observed during December. PAR has its minima during this initial bloom month and an increasing trend corresponding to the second peak of Chl a, i.e. the bloom occurs during its minimum and maximum values of PAR winter cycles. This implies PAR is not a limiting factor for production in this zone.