Dissolved oxygen (DO) is one of the most critical parameters in aquaculture water-quality assessment and monitoring, as its appropriate level is required to support aquatic ecosystems. Fish and other aquatic species are sensitive to reduced DO levels, yet global oceanic DO content has been in steady decline (Schmidtko et al., 2017). The effect of DO on the growth and production of salmonids has been studied for decades, and adequate DO concentrations were found to have a critical role in maintaining their health, productivity and survival (Carter, 2005; Mallya, 2007). DO requirements of aquatic species vary and depend on numerous factors, including fish species, aquaculture system, water temperature, and fish life stages. Direct and indirect effects of reduced DO levels on fish health have been recognized (Mallya, 2007). Direct effects include the changes that significantly affect physiology, growth and developmental stages, feeding activities, reproduction, swimming capability, ability to withstand environmental stressors, behavior, and reproductive abilities of aquatic species (Thomas and Rahman, 2012; Langseth et al., 2014), as well as high-mortality events in severe cases (Shimps et al., 2005). Indirect effects are related to ecological disruption of spatiotemporal interactions between the targeted resources and fish distribution (Stramma et al., 2012). Therefore, minimum acceptable DO concentrations are recommended to prevent these negative impacts of hypoxic episodes in farmed fish.

Monitoring of DO concentrations and other water quality parameters is an important management practice necessary for maintaining the biological health and functioning capacity of aquatic ecosystems (Abdul-Aziz et al., 2007). In particular, DO monitoring serves as an indirect measure of the health of aquaculture systems (Pollock et al., 2007), and the establishment of efficient DO monitoring programs have become an integral part of aquaculture productions (Burke et al., 2021). Real-time continuous monitoring, using DO probes to monitor DO levels 24 h/day and intermittent monitoring with hand-held DO probes at specific times of the day, have become common practices in aquaculture industry (Wei et al., 2019). In both cases, data are repeatedly collected over a period of time and evaluated according to established criteria, some of which include a temporal component when determining thresholds. Our understanding indicates that despite numerous studies investigating the impact of dissolved oxygen (DO) concentration on the health of salmon, only a limited number of these have established a direct link between DO concentration threshold and salmon health. Specifically, it has been generally noted that Atlantic salmon are critically affected when DO concentrations fall below 4 mg/L (Stevens et al., 1998). Furthermore, one set of widely-used and well-established DO criteria in salmon aquaculture, provided by the United States Environmental Protection Agency (US EPA; U.S. Environmental Protection Agency, 1986), describe the impact of several levels of DO concentrations on various salmon life stages, and over various time periods (i.e. 0 [instant measurement], 7 and 30 days).

The availability of DO for aquatic organisms depends on complex interactions between a variety of biotic and abiotic factors. DO levels in an aquatic environment physically depend on the water temperature, with warming sea temperatures causing a significant decline in oxygen solubility and a reduced oxygenation of deeper sea waters (Remen et al., 2016). Atmospheric pressure also influences the exerting capacity of atmospheric wind on the water surface and, ultimately, the solubility of oxygen in the water body (Cox, 2003). The presence of chlorophyll-a pigment in most aquatic plants results in an increased release of oxygen through photosynthesis, which is estimated to have a more significant global impact than that of terrestrial plants (Peña et al., 2010; Correa-González et al., 2014). Concurrently, the coexistence of other living organisms, organic matter, and nitrogenous deposits in aquatic environments creates competition with fish for available DO (Gammons et al., 2011). This intricate interplay between DO levels and various environmental factors highlights the need for a clearer understanding of these local relationships to improve the management of DO levels at salmon aquaculture sites. This improved understanding should be valuable for the development of a mathematical model to predict DO levels in salmon cages (Alver et al., 2022).

A large-scale report of DO levels of salmon aquaculture sites in the Southwestern New Brunswick region of the Bay of Fundy (Page et al., 2005), stressed the importance of coastal water temperatures (with seasonal fluctuations) and movements (tidal), along with wind (air-sea oxygen exchange) as contributing factors. The researchers were interested in building regional predictive models, so also discussed anthropogenic influences from the release of organic waste in certain bays, where local fish processing plants or pulp mill effluent was found to severely reduce DO levels. Their empirical data came from a few single time series taken over the span of a few weeks in surrounding waters (near an aquaculture site) and from one cage within an aquaculture site. Interestingly, they recorded DO concentrations during an algal bloom event that lasted approximately 6 days, demonstrating huge variations (spikes and troughs) in DO levels, presumably from abnormal local levels of algae-induced photosynthesis and respiration.

This study was initiated as part of active water quality monitoring of all MOWI aquaculture sites in British Columbia (BC), Canada. Previous observations of intermittently low DO measurements in some aquaculture sites in BC brought up the need for a more detailed description and detection of any significant deviation from acceptable standards and recommended DO concentrations for the protection of various salmonid life-stages (Fisheries and Oceans Canada, 2015). Also, our study was motivated by the limited knowledge of DO patterns in aquaculture sites in our study area, and the unique large-scale availability of DO data in the region during the span of an entire farmed salmonid production cycle. The objectives of this study were two folds: first we wanted to describe the measurement of DO, based on the US EPA recommended DO concentrations (U.S. Environmental Protection Agency, 1986), evaluating low DO (hypoxic) episodes from aquaculture sites, and better understand the within- and among-site patterns of DO. Then, we investigated the potential effects of environmental factors such as temperature, wind speed, wind direction, precipitation, and remotely-sensed (sea surface) estimated absorption from phytoplankton, on DO levels on aquaculture sites.

By using in-cage DO logger data concurrently with historical remotely-sensed and regional weather data, we were able to analyze spatiotemporal patterns of DO concentrations in relation to environmental variables. The frequency of hypoxic episodes was substantially different among aquaculture sites and among seasons. The time-series regression modelling results emphasized the significance of seasonal effect on DO concentrations, and how much the effect of environmental variables on DO concentrations varied among the aquaculture sites.

The results from the two-number criteria, demonstrating the frequency of hypoxic episodes from 7-day minimum, 7-day mean and 30-day mean thresholds, suggests that both acute and chronic exposures to low DO were occurring simultaneously in aquaculture production systems. This is an important finding, especially when considering that the 30-day mean criterion is deemed to be a period sufficient to produce negative effects on the growth and reproduction of adult salmonids (U.S. Environmental Protection Agency, 1986). In our study, the chronic exposures (30-day mean) to low DO occurred at any moment in all of three cages in 14 of the 20 sites with data for all three cages (Sites 2, 4, 7, 8, 9, 11, 13, 14, 15, 17, 18, 19, 20, and 21; excluded Site 6 with only two cages) during the study period. The low DO concentration resulted from either a series of frequent short-term exposures or prolonged exposures to low DO concentration over a long period of time. Continuous exposure of fish to DO concentrations less than 6.5 mg/L in benthic waters or in vertical stratification have been reported to alter fish physiological processes and spatial distribution (Zhang et al., 2015). To comprehend the impact of low dissolved oxygen (DO) concentrations on overall aquaculture performance, it is crucial to assess fish health, growth and mortality. Unfortunately, we did not have access to health events or mortality data, fish weight information, or feed conversion rate data for the aquaculture sites in our study. Consequently, the effect of chronic exposure to low DO concentrations on the growth and reproductive capacity of salmonids in the study sites remains purely speculative and should be approached with caution. Producers, however, have observed increased mortalities associated with lower DO in fish with pre-existing conditions (e.g. gill damage, deformities, etc.), and during fish-handling events (e.g. sea lice treatments and harvesting). Nevertheless, the observation of short- and long-term hypoxic episodes in these 14 sites are worth noting and likely have an impact on the suitability of these sites for growing salmon, either because of a direct influence on mortality and productivity, or simply by having populations of fish more susceptible to sudden fluctuations in oxygen levels in coastal waters.

Spatial variation in DO concentrations across aquaculture areas and sites was observed in this study, and the variation was more obvious for low DO episodes, as defined by the production impairment classification and the two-number criteria. Previous publications also found substantial spatial variation in DO levels of salmon sea cages (Johansson et al., 2006; Johansson et al., 2007; Oppedal et al., 2011; Solstorm et al., 2018). This variation might be related to aquaculture management practices, geological or geographical locations of the aquaculture sites (Page et al., 2005), environmental and climatic factors, and the oxygen fluctuation and consumption conditions within sites. Geographically, we found that the proportion of hypoxic episodes, based on the production impairment classification, varied among areas ( Figure 3A ), suggesting that some of the causes of low DO may be occurring at a larger regional scale.

Although we did not have information on the stocking density of each cage in our study, presence of fish within cages and their stocking density have been found to relate to variations in DO levels and cage-level oxygen consumption dynamics (Johansson et al., 2007). Larger cages with increased total biomass is linked to a decrease in DO levels and an increased risk of hypoxia due to greater oxygen consumption rate (Alver et al., 2023). Also, other factors such as the horizontal variability in water current speed and direction, sea cage physical structure (including netting), net fouling, and even the position of the cages within aquaculture sites may be responsible for some of the spatial and cage-level variations in hypoxic patterns (Johansson et al., 2006; Johansson et al., 2007; Solstorm et al., 2018). When producers observe hypoxic events on sites and/or in specific cages, they likely will adapt their feeding strategies, run air diffusers (passing air bubbles through deeper waters to oxygenate and create water movement) and increase net cleaning frequencies to improve water circulation.

The application of the time-series regression modeling approach to water quality and other environmental data allows the assessment of potential relationships between water quality and environmental parameters (Huang and Schmitt, 2014). The variables that we examined in the time-series regression analyses have been previously recognized as potential influential factors for DO concentrations within sea cages (Li et al., 2018). Importantly, the chosen time scale for a study can also have a potentially large impact on the observed associations between the environmental variables and DO concentrations (Rajwa-Kuligiewicz et al., 2015). Thus, for our study to focus on the short-term association between DO concentration and the environmental predictors, we controlled for long-term trends (i.e. seasonality) by a piecewise polynomial function in our time-series regression analyses, as demonstrated in Figure 4 with smoothed curves of DO concentration over time (Bhaskaran et al., 2013).

Previously published studies reported that low levels of DO concentration were commonly observed during the warmer months (Page et al., 2005; Mansour et al., 2008), and it is a well-known fact that the amount of oxygen that can dissolve in water is inversely proportional to the temperature of water. However, our findings show a pattern for the higher temperature ranges, both in the summer months ( Table 2 ) and with a linear spline (knot at 10°C; Figure 7 ), that there is more dissolved oxygen (measured as daily average minimum DO) as the water gets warmer ( Figure 4 ). Similarly, the seasonal trend of DO concentration was not necessarily the lowest during summer (Huang and Schmitt, 2014). Interestingly, Rajwa-Kuligiewicz et al. (2015) have also reported that higher concentration of DO was observed with higher water temperature over both short and long-time scales, in which the relation of DO concentration and water temperature appeared to make a hysteretic loop, rather than an inversely proportional linear relation. This loop may be caused by the cyclical patterns of other climatic conditions, such as a light-driven photosynthesis of algae, than from strictly water temperatures. In our study, looking at seasonal patterns for DO and environmental predictors ( Figure 5 ), we can see relatively consistent temperature profiles in spring and autumn, while we observed high DO concentration during spring and low DO concentration during autumn, in conjunction with relatively high a _ph(443) in spring and low a _ph(443) in autumn, and completely opposite wind directions between the two seasons – all of which may have contributed to these results. Therefore, while oxygen solubility and DO concentration in aquatic ecosystems are primarily temperature-dependent (Remen et al., 2014), the effect of temperature on DO concentration in salmon aquaculture sites in BC, Canada may be reversed by other driving factors, in what are most likely multifactorial hypoxic episodes.

In our study, no significant association was observed between daily average minimum DO concentration and total precipitation, which is consistent with previous studies (Tyler et al., 2009; Rajwa-Kuligiewicz et al., 2015). From our time-series regression analysis, however, maximum wind gust speed >55 km/h tended to increase daily average minimum DO concentration across the sites ( Figure 7 ). This finding further supports the evidence that increased wind-speed, which improves the air-water oxygen transfer process, does lead to a better DO supply in the water column. Effects of wind mixing include the diffusion of atmospheric oxygen through air-water interface, vertical mixing, and breaking up of any vertical stratification of the water column, all of which may lead to relieving hypoxia (Scully, 2010; Jonasson et al., 2012; Ni et al., 2016).

Lastly, the association between DO concentration and a _ph(443) was found to be significant. The ability to successfully capture remotely sensed data, such as sea surface a _ph(443) coefficients, depends heavily on having clear (cloud-free) skies. During the study period, we had a _ph(443) data for only about 28% of the days, and the missing data were replaced with a constant minimum value of 0.003 (representing halfway between the minimum value and zero). Considering that phytoplankton activity is related to sunlight, this was deemed biologically appropriate for these analyses, however, knowledge of the effects of clouds on a _ph(443) coefficients would better inform these assumptions. For example, future adjustments could be based not only on the use of missing absorption data, but also on using cumulative or historical sunlight data to better capture both temporal exposure and intensity to solar radiation. Ideally, researchers would be able to estimate regional amounts of phytoplankton biomass and know if this biomass is alive and undergoing photosynthesis or respiration, or if it is dead/dying and consuming oxygen as it naturally decays in the coastal waters. Part of the issue also relates to defining an appropriate geographical region, and currently, the resolution is set at 4 km, which is problematic for coastal waters surrounded by land. One solution, as was done for this study, is to average all a _ph(443) measurements taken in the larger study region, but ideally, future resolution of remotely-sensed data would allow for site-level information to be captured.

One limitation of this study is the somewhat irregular interval in the measurement of DO concentration across the sites. There were many missing data points and temporal gaps over days and weeks, and at some sites and cages there were gaps of missing data for months. These temporal gaps may have introduced some biases into our analysis, especially in the analysis of seasonal periodicity in the DO time-series (Burke et al., 2021). Consequently, due to limited availability of data, we investigated the effects of the environmental factors only on the average daily average minimum DO concentrations, not on the hourly DO concentration, in the time-series regression modelling. Other explanatory variables such as water flow, salinity, microalgal biomass, salmon stocking density, feeding patterns, disease incidence, amount of available sunlight (solar radiation), pH, inorganic nutrients, and other aquaculture management practices (including cage-level aeration with diffusers and increased net cleaning frequencies) may affect DO dynamics and thus ultimately salmon health and productivity. This study also highlighted the importance of measuring DO over a long period, across multiple seasons, to capture and better understand long-term seasonal trends in DO time-series. Moreover, the availability of ambient dissolved oxygen (DO) concentration data would enable a valuable comparison with the DO observations within fish cages, potentially enhancing the analytical depth of our study. This is due to the fact that the distribution of DO levels within a fish cage is contingent on ambient DO levels, which can exhibit seasonal variations (Alver et al., 2022).

In summary, our study confirms that certain aquaculture areas, in British Columbia, do experience repeated short- and long-term hypoxic episodes, below the recommended optimal concentrations (>7 mg/L), with some sites reaching ‘stressed’ levels of DO concentration (<4 mg/L). The DO dynamics among aquaculture sites showed substantial variations across areas and across cages within sites. Our study provided some evidence that DO concentration is not necessarily inversely proportional to temperature in aquaculture sites, but rather that it depends on multiple other environmental factors. Using time-series regression analysis, temperature, wind (gust) speed, and absorption due to phytoplankton [a _ph(443)] were found to be important variables for predicting DO levels. Exploring DO dynamics is not only necessary for understanding the current biological state and detecting deviation from recommended concentrations to safeguard salmonids but also useful to plan for active monitoring of water quality parameters. Furthermore, the information from this study can help inform baseline parameters for predictive and simulation-type models for regional DO concentrations (Eze and Ajmal, 2020).

The data analyzed in this study is subject to the following licenses/restrictions: The dataset used in the study is proprietary data generated by salmon farms. Requests to access these datasets should be directed to Krishna Thakur kthakur@upei.ca.

JJ: Data curation, Formal Analysis, Investigation, Visualization, Writing – original draft. BA: Data curation, Writing – original draft. KT: Conceptualization, Formal Analysis, Project administration, Resources, Writing – original draft, Writing – review & editing. HS: Methodology, Validation, Writing – review & editing. BB: Data curation, Writing – review & editing. RV: Conceptualization, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing.