Observed and forecasted global warming pressure on coastal hypoxia

Coastal hypoxia is a major environmental problem of increasing severity. A global 40-year observational gridded 5 climate data record and 21 century forecasts from the Community Earth System Model under RCP 8.5 forcing are analyzed for long-term linear trends with a focus on warming-related pressures on coastal oxygen conditions. The forecasted median trends along the global coast are 0.32 C, -1.6 mmol m, and -1.2 mmol m per decade for sea-surface temperature (SST), surface oxygen capacity, and vertical-minimum oxygen concentration, respectively. These trends point to more rapid deterioration in coastal conditions than experienced over the last four decades; the forecasted median coastal trends for SST 10 and oxygen capacity are 48% and 18% faster than the corresponding observed rates. Median rates for the coast and documented hypoxic areas are higher than in the global ocean. Warming and oxygen declines tend to be fastest at high latitudes, one region where new hypoxic areas may emerge as oxygen conditions deteriorate. Over 19% of the coast has extremely rapid forecasted change upwards of 0.60 C per decade warming and -3.0 mmol m per decade oxygen change. There is considerable pressure on current hypoxic areas since future oxygen declines of any magnitude will make hypoxia more severe. The coastal forecasts 15 can inform coastal environmental management strategies to protect future water quality and ecosystem services.


Introduction
Hypoxia in coastal waters is a major environmental problem of increasing severity confronted around the world (Hoegh-global coast to include unknown and potentially emerging hypoxic areas. Results are placed in the context provided by other studies and the implications for future coastal hypoxia are discussed. 65 2 Methods

Observations
The global observational dataset analyzed is the satellite-based SST time-series described in Merchant et al. (2019) and available with updates at the Climate Data Store of the Copernicus Climate Change Service (Embury and Good, 2021). The Level-4 (version 2.0) product combines SST data from several satellite platforms to construct a high-quality climate data 70 record. The dataset is a daily product on a regular grid with 0.05 o (latitude and longitude) resolution. The Level-4 product is gap-filled so that each grid point has an SST value for every day from September 1981 up to within a month of present time.
This study analyzes 40 years of data spanning 1982-2021. Daily data during each August and February are used to represent summer conditions in the northern and southern hemispheres, respectively. The rationale for analyzing these months is they are the summer months in each hemisphere when water temperatures tend to be highest and oxygen levels tend to be lower. 75 Daily data are averaged to create August-averaged and February-averaged SST time series for the northern and southern hemispheres, respectively. Oxygen capacity is calculated with the Garcia and Gordon (1992) oxygen saturation concentration equations using the monthly averaged SST data and a constant 35 salinity. The constant salinity is used because the Merchant et al. (2019) product does not include salinity and because this straightforward approach is sufficient to provide observational context for forecasts. Coastal points are defined as grid points with at least on neighboring land cell (directly to the east, west, 80 north, or south). Techniques for comparing observations to forecasts for overlapping years, for calculating observed linear trends, and extracting information at documented hypoxic areas are described in subsequent sections.

Forecasts
Forecasts of 21 st century water temperatures and oxygen conditions are derived from the CESM Large Ensemble Project that includes ocean biogeochemistry (Kay et al., 2015a), which is a contributor to Climate Model Intercomparison Project phase 5 85 (CMIP5, Taylor et al., 2012). Monthly-averaged results for 2006-2100 are accessed via the Earth System Grid (Kay et al., 2015b). Multiple ensemble members for RCP 8.5 forcing (following CMIP5 protocols) are used: 35 and 27 ensemble members for SST and oxygen capacity, respectively. Runs 1-35 for SST (the "SST" variable) and runs 1, 2, and 9-35 for surface oxygen saturation concentration (the "O2SAT" variable) and vertical-minimum oxygen concentration (the "O2_ZMIN" variable).
Runs 3-8 are omitted for O2SAT and O2_ZMIN because these results are not available on the Earth System Grid. Additional 90 runs (101 and higher) were avoided because of a documented unexplained positive bias in global temperature relative to other ensemble members. All included runs are averaged together to produce ensemble-mean monthly time series of SST, surface oxygen saturation concentration (oxygen capacity), and vertical-minimum oxygen concentrations at each ocean grid cell. https: //doi.org/10.5194/bg-2021-285 Preprint.  For each year at each CESM ocean grid cell (at 1 o latitude nominal resolution), the month with minimum surface oxygen saturation concentration is selected to construct annual time series of minimum oxygen capacity and coincident SST and 95 vertical-minimum oxygen concentrations. August and February are the median months associated with minimum surface oxygen capacity in the northern and southern hemispheres, respectively. The global set of all coastal points includes all CESM ocean grid cells with at least one neighboring land cell (there are 4,899 coastal cells). Coastal points are defined as grid points with at least on neighboring land cell. Globally, there are 4,899 cells coastal cells. Techniques for comparing forecasts and observations for overlapping years, for calculating forecasted linear trends, and extracting information at documented hypoxic 100 areas are described in the following sections.

Comparison of observations and forecasts
To characterize the reliability of CESM in coastal areas, observations and forecasts are compared for global coastal points.
There are 16 forecast years spanning 2006-2021 that now overlap with observed conditions. Every CESM coastal point is matched with the closest coastal point on the observational grid, which has higher spatial resolution than the forecast grid. The 105 summer month values (calculated as described in previous section) for the overlapping period are averaged together to determine mean observed and forecasted values at each coastal point. The resulting SST and oxygen capacity values are quantitatively compared to each other. Linear regression is completed to indicate the global relationship between observed and forecasted SST and oxygen capacity for coastal waters. Regression results are included in the supporting dataset (Whitney, 2021) and are reported with the slope, offset, p-value (p) of the F-statistic, correlation coefficient squared (r 2 ), and root mean 110 square error (RMSE). The level of agreement between observed and forecasted mean values for the overlapping period assesses the quality of CESM performance in coastal waters. Observed and forecasted temporal trends are not compared for the overlapping period because observations have high interannual variability that can obscure longer-term trends over shorter periods. Individual ensemble members have comparably large interannual variability, but the ensemble-mean forecast results have much less interannual variability. 115

Trend analysis
Linear regression analysis is applied to characterize long-term temporal trends. For observations, the entire 40-year observational period is regressed with time for SST and surface oxygen capacity. For forecasts, the entire 94-year forecast period is regressed for SST, surface oxygen capacity, and oxygen concentrations. The regression slopes are reported as rates of change associated with each variable and are included in the supporting dataset (Whitney, 2021). Regression statistics are 120 calculated. The p-values of the F-statistic are used to categorize results depending on whether or not they are statistically significant at the 90% confidence level (p=0.10). Correlation coefficients are calculated but not emphasized because ensemblemean forecasts average out interannual variability present in ensemble members and therefore tend to have high r 2 values; while the observational record has higher interannual variability and correspondingly lower r 2 values. Rates of change are https: //doi.org/10.5194/bg-2021-285 Preprint.  shown for the entire global ocean and emphasis is placed on change in coastal waters. The documented coastal hypoxic areas 125 are further highlighted and are identified as described in the next section.

Documented coastal hypoxic areas
The 532

Observed trends
The 40-year observational SST record (updated from Merchant et al., 2019) indicates warming has occurred throughout the world ocean (Fig. 1a). SST rates are stronger in the Northern Hemisphere, with the most rapid warming occurring in Arctic 140 areas near coasts. Just over half (55%) of the ocean grid points have linear trends with p≤0.10 (Table 1). P-values are lower where calculated rates are lower in parts of the Atlantic, much of the South Pacific, and most of the Southern Ocean. Where p≤0.10, the mean r 2 =0.20; indicating that linear trends describe an appreciable part of the observed variance, but interannual variability is larger. The observed global median SST trend (including only points with p≤0.10) is 0.22 o C per decade ( Table   1). The median trend with all points included (regardless of p-value) is 0.13 o C per decade. This rate is consistent with the 145 global mean SST rate of 0.1 o C per decade from 1982-2013 (Pershing et al., 2015).
Observed rates along the global coast indicate conditions relevant to coastal hypoxia. The median SST trend for global coastal points (0.27 o C per decade, for points with p≤0.10) is 22% faster than the observed ocean median rate (Table 1). SST rates tend to be near the median coastal value from 60 o S to 30 o N (Fig. 1b) and increase towards higher latitudes. The observed median trend for documented hypoxic areas (0.24 o C per decade) is 6% faster than the ocean median rate (Table 1), but is 150 lower than the median global coastal rate because there are few documented hypoxic areas at high latitudes where warming is fastest. The histograms of SST rates (Fig. 1c) indicate the documented hypoxic areas have a narrower spectrum than the global coastal points.
The global distribution of oxygen capacity (saturation concentration) trends (Fig. 2a) at the surface is tied to observed SST rates (Fig. 1a). Since oxygen saturation concentrations decrease nonlinearly with temperature with a greater response at lower temperatures, oxygen capacity decreases are amplified at high latitudes where waters are colder and forecasted warming rates are faster (Weiss, 1970;Garcia and Gordon, 1992;Altieri and Gedan, 2015). Oxygen capacity linear trends with p≤0.10 occur in the same locations as for SST trends and r 2 values are similar to those for SST. The observed global median oxygen capacity trend at the surface (including only points with p≤0.10) is -0.9 mmol m -3 per decade (Table 1). This rate is several times faster 165 https: //doi.org/10.5194/bg-2021-285 Preprint. Discussion started: 3 November 2021 c Author(s) 2021. CC BY 4.0 License. than the median rate of -0.2 mmol m -3 per decade observed in offshore (farther than 100 km from the coast) upper-ocean (0-170 300 m depth) waters for 1976-2000 .
The observed median oxygen capacity trend for global coastal points (-1.4 mmol m -3 per decade, for points with p≤0.10) is 62% faster than the surface ocean median rate (Table 1). The observed median coastal rate is half of the rate calculated for a global coastal band (within 30 km of the coast) for 1976-2000 . Rates tend to be near the median coastal value from 60 o S to 30 o N (Fig. 2b). At higher latitudes, rates tend to increase with latitude. The median oxygen capacity trend 175 for documented hypoxic areas is -0.8 mmol m -3 per decade (Table 1). This rate is lower magnitude than for all global coastal points because of little high-latitude coverage, where oxygen capacity rates are largest. Similar to SST rates, the histograms for oxygen capacity (Fig. 2c) indicate the documented hypoxic areas have a narrower spectrum than the global coastal points.
Overall, the observational SST and oxygen capacity analysis provides context for the forecasts and new information about global coastal conditions influencing coastal hypoxia. 180

Comparison of observations and forecasts
The observational SST record is compared to CESM forecasted coastal conditions for the overlapping 16 years spanning 2006-2021 (Fig. 3a). There is an essentially one-to-one relationship between observed and forecasted coastal SST with a low p-value (p<0.001) and high correlation (r 2 =0.97). There is approximate 5 o C scatter around the linear relationship, but the RMSE is small (0.03 o C). This level of agreement indicates that CESM results are broadly representative of global coastal SST 185 conditions. The comparison of observed and forecasted coastal oxygen capacities averaged over the overlapping period (Fig.   3b) show a near one-to-one relationship with a low p-value (p<0.001) and high correlation (r 2 =0.94). The scatter away from the regression line is larger at higher oxygen capacities (in colder waters). The nonlinear relationship between temperature and oxygen capacity (noted above) means that temperature scatter in colder waters translates into more oxygen capacity scatter.

Forecasted trends
The CESM Large Ensemble Project ensemble-mean forecast for the RCP 8.5 scenario indicates SST will appreciably increase 205 throughout the world ocean over the 21 st century (Fig. 4a). All of the ocean cells have linear trends with p≤0.10 (Table 1). SST rates account for more than 90% of the variance in the ensemble-mean forecast (r 2 >0.90) for most of the ocean; the only exceptions are some areas to the north and south of Greenland and near parts of Antarctica. The forecasted global median SST trend is 0.35 o C per decade (Table 1). Global distributions of SST warming have been studied in detail for multiple models and RCP scenarios (e.g. Bopp et al., 2013). Bopp et al. (2013) includes CESM simulations in an analysis of ten models running 210 the RCP 8.5 scenario and finds the global average SST increase is 0.27 o C per decade (from the 1990s to 2090s) when averaged across all included models. The CESM forecasted global median SST trend is 61% higher than the observed global warming rate (0.22 o C per decade) calculated in the previous section (Table 1). Under the RCP 8.5 scenario, the ocean SST will increase considerably faster than the observed linear trend over the last four decades.
The global distribution of SST warming indicates variations among oceans, with the Arctic Ocean forecasted to experience at 215 least twice the median warming rates (Fig. 4a). The influence of ocean circulation patterns on warming rates also is evident.
Observations (Fig. 1a) also indicate rapid SST increases near Arctic coasts and influences of ocean circulation are evident.
There are, however, clear differences in the spatial structure of forecasted and observed rates. Differences away from the coast https: //doi.org/10.5194/bg-2021-285 Preprint. Discussion started: 3 November 2021 c Author(s) 2021. CC BY 4.0 License.

Figure 4: Forecasted SST trends: (a) spatial distribution for all CESM ocean grid cells, (b) latitudinal dependence for global coastal points and documented coastal hypoxic areas, and (c) histograms of global coastal points and documented coastal hypoxic areas.
The format and color coding follow Fig. 1. in the Arctic Ocean are immediately apparent. The low p-values of observed SST trends in much of the Southern Ocean (p>0.10) preclude comparisons of forecasted and observed spatial structure in this region. 225 Focusing on warming in global coastal areas reveals new information directly relevant to coastal hypoxia. The forecasted median SST trend for global coastal points (0.39 o C per decade) is 11% faster than the forecasted ocean median rate and 48% higher than the observed median coastal rate (Table 1). SST rates tend to be near the median coastal value from 60 o S to 30 o N https: //doi.org/10.5194/bg-2021-285 Preprint. Discussion started: 3 November 2021 c Author(s) 2021. CC BY 4.0 License. (Fig. 4b). Above 30 o N, warming rates tend to increase with latitude and variability among coasts increases. The latitudinal patterns are broadly similar for forecasted and observed coastal SST rates (Fig. 4b and Fig. 1b). The forecasted median SST 230 trend for documented hypoxic areas (0.40 o C per decade) is 14% faster than the forecasted ocean median rate (Table 1). For documented hypoxic areas, the forecasted median trend is 70% higher than the median observed rate (Table 1) and 75% faster than the median trend under moderate A1B emissions scenario (Altieri and Gedan, 2015). The documented hypoxic areas sample much of the variability in coastal SST rates from 45 o S to 60 o N. The hypoxic area database, however, has little coverage at higher latitudes, where the most rapid warming is forecasted. Histograms (Fig. 4c) indicate most coastal locations (92%) 235 and documented hypoxic areas (96%) are forecasted to warm more than 0.3 o C per decade; 0.1 o C per decade faster than the A1B results in Altieri and Gedan (2015). A significant portion of the global coast (18%) are forecasted to warm faster than 0.6 o C per decade, but few of the documented hypoxic sites (2%) are in this range because most are not at high latitudes.
The global distribution of forecasted oxygen capacity (saturation concentration) trends (Fig. 5a) at the surface is tightly linked to SST rates (Fig. 4a). All of the ocean cells have linear trends with p≤0.10 ( Table 1). The forecasted oxygen capacity trends 240 have r 2 >0.90 over most of the ocean with exceptions in the areas where SST rates have r 2 ≤0.90. The forecasted global median oxygen capacity trend at the surface is -1.2 mmol m -3 per decade (Table 1). This rate is 50% higher than the observed global median trend (Table 1). The forecasted median rate for ocean waters above 60 o N (-5.3 mmol m -3 per decade) is several times higher than the total ocean median rate.
The forecasted median oxygen capacity trend for global coastal points (-1.6 mmol m -3 per decade) is 28% faster than the 245 surface ocean median rate (Table 1). The forecasted median coastal rate is 18% faster than observed. Rates tend to be near the median value from 30 o S to 30 o N (Fig. 5b). Outside that latitude range, rates tend to increase with latitude and variability among coasts increases, particularly in the northern hemisphere where rates can exceed -10.0 mmol m -3 per decade. The latitudinal pattern in coastal oxygen capacity trends (Fig. 5b) is similar to the observed coastal pattern (Fig. 2b). The median oxygen capacity trend for documented hypoxic areas is -1.4 mmol m -3 per decade (Table 1). This rate is lower magnitude than 250 for all global coastal points because of little high-latitude coverage, where oxygen capacity rates are largest. The forecasted median trend for documented hypoxic areas is 78% higher than observed (Table 1). Histograms (Fig. 5c) indicate for most coastal locations (95%) and all documented hypoxic areas the oxygen capacity trend is forecasted to be faster than -0.9 mmol m -3 per decade. A significant portion of the global coast (28%) has forecasted rates faster than -3.0 mmol m -3 per decade, but few of the documented hypoxic sites (4%) are in this range because most are not at high latitudes. 255 The forecasted global distribution of vertical-minimum oxygen concentration changes (Fig. 6a) is influenced by warming, oxygen capacity declines, ocean circulation changes (including vertical exchange), and ecosystem changes. In CESM and in nature, the depth of the vertical-minimum oxygen concentration tends to be near-bottom in coastal areas and is bathymetrically constrained to be relatively close to the surface, whereas the minimum oxygen concentration can occur much deeper in the open ocean and consequently have a more remote connection to surface oxygen capacity. The forecasts for most ocean cells 260 have linear trends with p≤0.10 ( Table 1). The global median trend in oxygen is -0.7 mmol m -3 per decade (Table 1). The global distribution of vertical-minimum oxygen concentration rates (Fig. 6a) is broadly consistent with the global distribution of the https: //doi.org/10.5194/bg-2021-285 Preprint. Discussion started: 3 November 2021 c Author(s) 2021. CC BY 4.0 License. RCP 8.5 model-average forecasted changes in oxygen concentrations (at 200-600 m depth) shown in Bopp et al. (2013). The total ocean oxygen content decrease (from the 1990s to 2090s) calculated in Bopp et al. (2013) translates to a -0.6 mmol m -3 270 per decade trend, which is close to the current results.
The median trend in vertical-minimum oxygen concentrations for global coastal points (-1.2 mmol m -3 per decade) is 77% faster than the ocean median rate (Table 1). Similar to oxygen capacity, oxygen concentrations rates tend to be near the median value from 30 o S to 30 o N and increase at higher latitudes (exceeding exceed -10.0 mmol m -3 per decade), but there is a lot of scatter (Fig. 6b). The median trend for documented hypoxic areas (-1.4 mmol m -3 per decade) is lower than for all global points 275 https: //doi.org/10.5194/bg-2021-285 Preprint. Discussion started: 3 November 2021 c Author(s) 2021. CC BY 4.0 License.
because of little high-latitude coverage and approximately the same as the corresponding oxygen capacity decrease (Table 1). Histograms (Fig. 6c) indicate most coastal locations (72%) and documented hypoxic areas (89%) have forecasted oxygen concentration declines. Some coastal (19%) and documented hypoxic areas (8%) have forecasted trends faster than -3.0 mmol m -3 per decade. 285

Discussion
The forecasted warming and declining oxygen conditions will exert considerable pressure on current hypoxic areas. The documented hypoxic areas already experience oxygen concentrations at or below the 63 mmol m -3 threshold (applied in Diaz and Rosenberg, 2008). If concentrations were at this threshold in 2000, the forecasted median trend in oxygen capacity and concentration represents a 20% reduction by 2100. Furthermore, future oxygen declines of any magnitude will have deleterious 290 ecosystem effects as hypoxia worsens (in intensity and duration) in existing hypoxic areas. The forecasted oxygen declines can erode oxygen gains achieved in systems improved by wastewater treatment and nutrient management. For example, oxygen conditions in Long Island Sound have improved with reduced nutrient loading, but current and forecasted future warming favor deteriorating oxygen conditions (Whitney and Vlahos, 2021).
Declines in oxygen conditions along the global coast will create emerging hypoxic areas. Forecasted warming and oxygen 295 declines are particularly severe at high latitudes in the northern hemisphere. Two adjacent fjords in Norway (Trysfjord and Ofotfjord) are the only sites above the Arctic circle within the documented hypoxic area database (Dommasnes et al., 1994;Diaz et al., 2011), but it is likely new Arctic coastal hypoxic areas will emerge. For instance, recent observations suggest Jago Lagoon, Alaska may be on the brink of hypoxia (Smith, 2012;Beaufort Lagoon Ecosystems LTER, 2020). High-latitude waters are colder and therefore tend to have higher saturation concentrations farther from the hypoxic threshold, but oxygen 300 conditions are forecasted to decline most rapidly in these coastal waters. Emphasis also should be placed on rapidly growing coastal megacities at low and mid-latitudes, which tend to struggle with wastewater treatment infrastructure as populations increase and are likely to experience emerging or worsening hypoxia (von Glasow et al., 2013;Varis et al., 2006).
It is important to note that ecosystem problems can arise before conditions deteriorate down to the canonical hypoxic threshold (63 mmol m -3 ), as many organisms experience physiological stresses above this threshold and have differing tolerances for 305 low oxygen conditions (Vaquer-Sunyer and Duarte, 2008). In addition to directly reducing oxygen capacities, warming also increases metabolic rates, related biological oxygen demand, and thermal stratification (Brown et al., 2004;Cloern, 2001;Breitburg et al., 2018). Thus, attention should be paid to all coastal areas with lowering oxygen conditions, not just the areas already experiencing seasonal hypoxia.
The results of this study provide a global perspective on some of the climate pressures confronting existing and emerging 310 coastal hypoxic areas. Observations indicate the warming and reduced oxygen capacities that coastal waters have been experiencing and the CESM forecast for the RCP 8.5 scenario points to even more rapid warming and oxygen declines throughout the 21 st Century. It is encouraging that CESM coastal performance is broadly consistent with SST and oxygen https: //doi.org/10.5194/bg-2021-285 Preprint. Discussion started: 3 November 2021 c Author(s) 2021. CC BY 4.0 License.