Low-pass filtering was performed with a Lanczos cosine filter (Emery and Thomson, 2001) with quarter-power (-6dB) cutoff at five days (120 hours) to remove tidal and inertial period variability (DiMarco and Reid, 1998). Unfiltered (i.e., raw) and low-pass filtered versions of the current velocity data were analyzed. Velocity data were initially oriented to geographic coordinates, i.e., north-south and east-west. For analyses presented in Section 3.1, current velocity data were transformed to an across-shelf and along-shelf coordinate system by rotating them 45 degree clockwise to be consistent with shelf and monsoon winds direction. Although vertical velocities were measured by RDCPs, no reliable upwelling/downwelling signals could be derived from the current measurements due to the large variability in the vertical velocity data and contamination by vertical migration of organisms (Wang et al., 2014).

The RDCPs were configured with additional sensors for temperature, conductivity (salinity), pressure, dissolved oxygen, and/or turbidity. Unlike the current sensor, all additional sensors provided single-point measurements at the depths of the instruments as shown by numbers in Figure 1B . Table 1 gives specifications for the additional sensors deployed on RDCP600s, including model number, sensor type, measurement range, sensor resolution, response time and accuracy. The instruments were calibrated by the manufacturer before deployment; calibration coefficients were used to convert sensor voltages and counts to engineering units prior to analysis. The temperature sensors were configured for the expected temperature range for the instrument deployment depth based on historical data (Wyrtki, 1971). Dissolved oxygen concentration was measured with Aanderaa Optode optical sensors. The Optodes were set for an expected range of 0-500 μM (micromolar), with a resolution of < 1 µM and an accuracy of better than 8 μM or 5%, whichever is greater. All moored data went through extensive quality control protocols (DiMarco et al., 2001; Bender and DiMarco, 2010) to identify outliers and other quality related issues prior to analysis. The percentages of “good” data passed all the quality control steps were > 89.7% for all the parameters except for the salinity data, which was > 84.1%.

Several maintenance services were performed on the array following its initial installation in August 2005. The LORI system was serviced by remotely operated underwater vehicles (ROV) in June 2006 and January 2009. The M2 array node was replaced in June 2006 and the entire mooring system was refurbished and replaced with new instrumentation on 27 January 2007 and 22 September 2010. During the first two years of operations, moored data were compared against CTD casts during the mooring maintenance and servicing cruises and were found to be within acceptable tolerances for both temperature and salinity. The temperature differences were found to be always less than 0.5°C for all moored instruments and salinity differences were all less than 0.3.

No significant sensor drift was noticed in the dissolved oxygen data during the five years deployment (linear drift of less than 3 μM/year and not statistically significant). Part of the reason is due to the prominent seasonal cycle in dissolved oxygen, which makes the relatively small sensor shift negligible in our analysis.

A salinity (conductivity) drift (< 0.51/year) was seen by comparing the data before and after replacing the conductivity sensor. A linear fit was applied to remove the small salinity trend that was present. Salinity estimates were derived from conductivity observations using the Practical Salinity Scale (Millero et al., 2008); following the convention of McDougall and Barker (2011). Salinity estimates were represented as unitless in figures.

Hourly wind speed and direction observations were collected at the Port of Sohar wind station (24.4°N, 56.63°E), at Sohar Majis, Oman ( Figure 1A ), for the time period 1989 – 2010. Raw wind data were quality controlled prior to analysis. Wind data were range checked against historical and climatological limits; statistical outliers were removed from the database prior to analysis. When appropriate, the winds were rotated 45° clockwise to form the cross-shelf and along-shelf velocity components. Low-pass filtering of wind data was accomplished using cosine-Lanczos filter weights with cut-off at five days (Emery and Thomson, 2001).

The 5-year time series (July 2005-July 2010) of low-pass filtered along-shelf winds at Sohar Majis, Oman, show a pronounced annual signal, with winds from the northwest during the NE monsoon and from the southeast during the SW monsoon ( Figure 2 , blue line). Inter-monsoonal winds at Sohar have decreased speed, consistent with known seasonal monsoonal variability in the region (Reynolds, 1993; Morrison et al., 1998). Along the northern coast of Oman between 22.5°N and 25.5°N, northwest winds (i.e., winds from the northwest) are downwelling favorable as the integrated surface Ekman transport is directed shoreward; southeast winds are upwelling favorable. Note the wind directions are different from the traditional monsoon wind directions (SW/NE), since in the Sea of Oman they are almost parallel to the Omani coastline.

The seasonal pattern of the northern Oman coastal currents and monsoonal wind variability correlate well with a lag of about one to two months ( Figure 2 ). In late boreal winter and spring, i.e., the transition to the NE monsoon and downwelling favorable winds, the along-shelf currents at the bottom instrument (207-m depth) at the M2 location approach 60 cm s^-1 and are directed to the southeast ( Figure 2 , top panel). During upwelling favorable conditions of fall and early winter, current velocities are smallest (<10 cm s^-1). Superimposed on the general annual pattern of currents are variations with periods of 5-10 days. The cross-shelf currents were small throughout the year, typically less than 10 cm s^-1 (not shown).

The strongest seasonal variability in any record at the LORI cabled observatory is that seen in the dissolved oxygen data. The seasonal pattern is characterized by an abrupt transition from hypoxic (concentrations less than 60 μM; here generally <10 μM) to more saturated values, i.e., greater than 100 μM ( Figure 2 , red line, 120h low-pass filtered). The transition occurs in early to mid-winter ( Figure 3 ) and coincides with the transition to downwelling favorable winds and the southeast currents ( Figure 2 ). Following the steep increase in oxygen, there is a gradual and steady decreasing trend until the fall ( Figure 3 ). At the M4 location, the dissolved oxygen concentrations at the two deepest instruments (541- and 1046-m below surface) were consistently anoxic. At the shallowest M4 instrument (235 m below surface), the temporal pattern of dissolved oxygen is similar to that found at the shallow inshore locations, but the maximum concentration was considerably less than that found at M1 and M2 ( Figure 3 ). The instruments at M43 are positioned close to the OMZ boundary and the low oxygen extends thru December and into January as a result.

There is short-term variability (days to weeks) superimposed over this seasonal cycle at the deepwater (>400m) locations (M31, M32 and M43) that indicates that hypoxic and anoxic conditions can occur near the shelf edge at any time of the year ( Figure 3 ). However, the persistence, that is the prolonged duration of the low-oxygen conditions, is longer (order of 2-4 weeks) in late summer and fall, while non-fall hypoxic events tend to be shorter (order of 1-8 days).

The rate of oxygen decline for the period defined by the abrupt winter transition (peak oxygen values) and the fall hypoxic period was estimated for the years 2005-2009 using a least-squares linear regression of the raw (unfiltered) oxygen time-series for Stations M11 and M21 ( Figure 4 ). Generally, the oxygen decline rate is nearly constant for each location and water depth between February and September (e.g. M11 and M21, Figure 4 ) and the magnitude of the decline rate increases between September to mid-November. At M11, the average oxygen decline rate of years 2006-2009 is approximately -11 μM/month between February and September; the rate becomes approximately -28 μM/month between September and November in most years 2005-2009 except for 2008 ( Figure 4A ). At M21, the average rate of years 2006-2009 is approximately -13 μM/month between February and September and increases to about -25 μM/month between September and mid-November ( Figure 4B ). The 2005 records only cover the September-November time period, when the rate appears to be somewhat higher (~25 µM/month).

The other locations do not show a consistent gradual oxygen decline during the year. We note that M11 and M21 (see Figure 1B for locations) are the bottom instruments and are positioned above the permanent oxycline that marks the oxygen minimum zone in the Indian Ocean. The bottom instrument at M31 is near the top of the OMZ and M41 is well within the OMZ. The instruments at M22, M32, and M43 are in the interior of the water-column ( Figure 1B ). The reason why the oxygen decline rates speed up between September and November is unclear and needs more investigation. It might be associated with the decomposition of algal blooms that occurred in the summer (Harrison et al., 2017). It is tempting to conclude that the relatively constant rate of oxygen decline during the year ( Figure 4 ) suggests steady rates of Ekman pumping that continuously bring low oxygen water towards the surface. However, other processes, such as local respiration and episodic wind events (i.e., weather) are also most certainly involved. The observed variability around the annual rate of decline in all years, and short-term increases in oxygen concentration are evidence of additional unresolved processes in play. Further work is needed to quantify the relative importance of physical and biological processes and their effects.

The five-year long temperature records show prominent temperature increases starting in November until February at M1, M2 and the top sensor of M3 stations ( Figure 5 , vertical gray bars indicate the time period of 15 October - 15 December), which is the same period of the winter monsoon. The temperature increase is likely caused by the downwelling generated by NE monsoon winds. The downwelling drives the upper warm water deeper and causes the temperature increase at the depth of the mooring sensors in the winter. At the same time, atmospheric-derived dissolved oxygen from the surface is mixed into the deep water and elevates the oxygen content from hypoxic levels to more saturated values. The highest value of dissolved oxygen occurs about a month after the temperature increase suggesting that it takes about a month for the high oxygen water near the sea surface to reach the sensor depth by downwelling. The calculated dissolved oxygen saturations (not shown) suggest the water is under saturated throughout the year at all depths and locations, except in February at M11 when the oxygen concentration is the highest (~200 µM). M11 is the shallowest station and the oxygen-saturated water measured at 66 m in February is likely brought down from the surface by the NE monsoon driven downwelling. Periodic temperature decreases are found in the 5-year temperature data during the summer monsoon at the three onshore moorings. For example, in July 2007, there are persistent temperature decreases at all top sensors ( Figure 5 ). During the same period, low oxygen water is seen at all top sensors ( Figure 3 ). This is a good indicator of the upwelling caused by the persistent summer monsoon.

A scatterplot of current velocity ( Figure 6A ) shows that the strong southeast current flow is associated with high oxygen concentrations at M21. Currents are largely quiescent (<10 cm/s) the rest of the year, with the low current velocities generally being associated with low dissolved oxygen concentrations (highlighted in red in Figure 6A ). The correlation coefficient between the along-shelf current velocity and oxygen records at M21 (shown in Figure 2 ) is r = 0.71 (p-value < 0.01).

Examination of the temperature records at M21 indicates that the low-oxygen water is associated with a drop in ocean temperature ( Figure 6B ). As previous work has shown (Olson et al., 1993), this indicates that upwelling processes that uplift waters already low in dissolved oxygen content are more likely responsible for the variability than in situ respiration (water column or benthic). The occurrence of relatively warm temperatures is coincident with waters relatively high in dissolved oxygen content ( Figure 6B ). This occurs during the NE monsoon, when the southeast currents associated with the downwelling are generated.

The Persian Gulf outflow is also associated with high dissolved oxygen (e.g., Wang et al., 2012; Wang et al., 2013a), which might contribute to the short-period (2-10 days) variations in the dissolved oxygen time series in the study region. Given the high salinity and density of the Persian Gulf outflow, and the fact that it sinks to a depth of 150-350 m as it enters the Sea of Oman and flows south along the shelf edge, ventilating the layer between σ₀ 26.0-26.8 kg m^-3 (Olson et al., 1993; Wang et al., 2013a; Al-Yamani and Naqvi, 2019; Lachkar et al., 2019), it might be thought that this could adversely affect the oxygen content. However, this water is essentially saturated with oxygen as the Persian Gulf is very shallow, and will contain more dissolved oxygen during winter, when the outflow is stronger and denser than in summer (Bower et al., 2000; Swift and Bower, 2003). There is also considerable oxygenation of the dense outflow through tidal mixing within the Strait of Hormuz (Bower et al., 2000). According to Banse (1997), the incoming Persian Gulf water can add about 5 µM of oxygen to the water in the Sea of Oman, and also reduces denitrification. The incoming water stays close to the shelf edge in winter, but spreads across the Sea of Oman in summer, which can result in changes in the depth of the oxycline and affect fisheries (Queste et al., 2018). A good example of this was seen in the WOCE line (I-07) in July-August 2005 between 23° and 24°N, which showed an intrusion of relatively high oxygen water (20-50 µM) within the OMZ at depths between 80 and 300 m associated with a bolus of high salinity water (36.6-37.3) from the Persian Gulf at stations 843-848 (Talley, 2014).

Our interpretation of the moored data is that the downwelling conditions prevalent during winter abruptly move the OMZ front seaward while ventilating the inshore and shallow waters of the shelf. In the late summer and fall, upwelling conditions persist and allow the OMZ to gradually push shoreward. Thus, the seasonal wind pattern sets up an oscillation of the OMZ front along the north Omani coast, which regularly ventilates the shelf. Peak ventilation occurs in February/March and peak hypoxic conditions occur in November and December ( Figures 3 , 4 ). This is borne out also by the climatology of the offshore region in the Sea of Oman (see Figure 7 ), which shows oxygen data, taken from the World Ocean Atlas (Garcia et al., 2019), from degree squares centered at 24.5°N, 57.5°E and 25.5°N, 57.5°E. At both latitudes, maximum ventilation to below 300m depth is apparent in the March-May period, while hypoxic conditions are at their most severe and reach closest to the surface (~50 m) during October-December.

Claereboudt et al. (2001) and Thangaraja et al. (2007) have shown the presence of fish mortality along the northern Oman coast. There is suspicion and some circumstantial evidence that low dissolved oxygen concentration is a possible cause of at least some of the fish mortality in the northern Arabian Sea (Claereboudt et al., 2001; Piontkovski et al., 2012; Harrison et al., 2017). A bimodal pattern of fish kill events is seen; with fish kills being prevalent during winter/early spring (February-April) and late summer/fall (August-November) and with no fish kill events observed in May through July ( Table 2 ). Unfortunately, there are few direct observations of oxygen concentration that demonstrate conclusively that low oxygen levels are responsible for the fish kill events, and there is also no obvious relationship between the available fish kill survey data and the moored observations ( Table 2 ). However, based on the limited data, a larger numbers of fish kills occur in late summer and fall when the dissolved oxygen concentrations are lowest and dissolved oxygen less than 60 μM frequently reaches depths as shallow as 66 m ( Table 2 ; see also Figure 7 ). The likelihood of a fish kill event occurring in any month is greatest (46–75%) in this period, correlates with the period during which the OMZ impinges on the shelf and slope during the SW Monsoon ( Figures 2 - 4 ). Although the summer/fall fish kill and low oxygen are coincident in time, this does not constitute a direct causal relationship. However, this can direct future investigation and we recommend a more targeted observational study to more fully determine whether a direct mechanism exists.

In spring, however, fish kills are only moderately likely (11–40%) and appear unrelated to changes in the depth of the <60 μM oxygen surface. At this time, fish kills occur when oxygen values are high due to the SE currents that ventilate the lower waters and are thus not attributable to low oxygen. We therefore speculate that they may be associated with Harmful Algae Bloom (HAB, Harrison et al., 2017) or other phytoplankton bloom events (Claereboudt et al., 2001; Al-Gheilani et al., 2011; and Gomes et al., 2014).

Increases in chlorophyll a concentrations in the Sea of Oman are found in August-September and January-February at the end of the monsoon periods (Al-Azri et al., 2010). While these authors state, based on fortnightly sampling from two coastal locations, that concentrations at the end of the SW monsoon are generally higher, peak concentrations can occur at almost any time. More widespread studies of Moderate Resolution Imaging Spectroradiometer (MODIS) data by Abuelgasim and Alhosani (2014); Bakhtiar et al. (2020), and Piontkovski and Al-Oufi (2014) that cover the whole Sea of Oman showed highest concentrations in February-March, which Abuelgasim and Alhosani (2014) attributed to the outflow of nutrient-rich water from the Persian Gulf. Such increased nutrient inputs could support more frequent HAB outbreaks. Again, more targeted studies are required to confirm or disprove this speculation.

Recent fish assemblage surveys using video techniques to characterize the habitat and fish abundance and diversity along the continental shelf margin (77~164 m) near Muscat, Oman (within 180 km from the cabled array) found species diversity increased by 96% in March 2006 compared to November 2005, a pattern consistent at all three types of habitats: sand, reef and megabenthos (McIlwain et al., 2011). McIlwain et al. (2011) also found all observed fish prefer demersal habitats along the Omani continental shelf. They suspect that the seasonal changes in fish assemblage are likely caused by the monsoon-generated upwelling events (Smith and Bottero, 1977; Lee et al., 2000; Shi et al., 2000), yet no hydrographic data were measured during their surveys to examine this. The seasonal hypoxia shown in this study provides strong evidence to support their hypothesis. As shown in Figures 3 , 4 , the lowest oxygen level is measured in November, while oxygen is typically not hypoxic in March on the Oman continental shelf. Because most fish prefer demersal habitats, the deoxygenation of the near-bottom water will displace the fish community along the coast when hypoxia is formed and cause the reduction of both abundance and diversity of coastal fish as shown in McIlwain et al. (2011).

Although the interannual variability of the date of onset of monsoons, ventilation of the deep shelf, or the rate of decline of oxygen during 2005-2010 are limited, ongoing and future climate change could lead to changes in the timing and rates that are observed. For example, the relationship between dissolved oxygen concentration and fisheries underscores the impact that large-scale atmospheric patterns can have on coastal ecosystems (IPCC, 2021). Changes to atmospheric forcing of the northern Indian Ocean may alter monsoonal wind patterns and could lead to changes in duration of the upwelling/downwelling seasons of the Sea of Oman (Goes et al., 2005). The systematic decrease in spring snow and its persistence over large parts of southwest Asia and the Himalayan-Tibetan Plateau region in recent years has enhanced the land-sea pressure gradient, which in turn has strengthened southwest monsoonal winds resulting in intensified wind-driven coastal upwelling in the Sea of Oman during summer and fall (Goes et al., 2005). This could potentially lead to a longer hypoxic season, which may further threaten fisheries of the region.

Increased global warming is likely to increase the buoyancy of the incoming Persian Gulf water, thus reducing the ventilation and oxygen concentration within the northern Sea of Oman (Queste et al., 2018; Lachkar et al., 2019; Naqvi, 2021), although the former is counteracted to some extent by increasing salinity in the outflow as freshwater flow into the Persian Gulf is reduced (Al-Yamani and Naqvi, 2019). Decreasing ventilation will lead to increased denitrification in the Sea of Oman, thus reducing its oxygen concentration. Thus, the Omani coast may be doubly susceptible to large-scale atmospheric and oceanographic change. Indeed, the development of hypoxia along the Oregon coast (Grantham et al., 2004; Adams et al., 2016) may already be the evidence of similar change on the US west coast. Sustained ocean-monitoring systems such as the one described here are necessary to fully characterize the long-term response of oceanographic systems to climate variability (IOC/UNESCO, 2009; Falkowski et al., 2011).