Coastal processes modify projections of some climate-driven stressors in the California Current System

. Global projections for ocean conditions in 2100 predict that the North Pacific will experience some of the largest changes. Coastal processes that drive variability in the region can alter these projected changes, but are poorly resolved by 15 global coarse resolution models. We quantify the degree to which local processes modify biogeochemical changes in the eastern boundary California Current System (CCS) using multi-model regionally downscaled climate projections of multiple climate-associated stressors (temperature, O 2 , pH, saturation state ( Ω ), and CO 2 ). The downscaled projections predict changes consistent with the directional change from the global projections for the same emissions scenario. However, the magnitude and spatial variability of projected changes are modified in the downscaled projections for carbon variables. Future changes 20 in pCO 2 and surface Ω are amplified while changes in pH and upper 200 meter Ω are dampened relative to the projected change in global models. Surface carbon variable changes are highly correlated to changes in DIC, pCO2 changes over the upper 200 meters are correlated to TA, while changes at the bottom are correlated to DIC and nutrient changes . The correlations in these latter two regions suggest that future changes in carbon variables are influenced by nutrient cycling, changes in benthic/pelagic coupling and TA resolved by the downscaled projections. Within the CCS, differences in global and downscaled climate 25 stressors are spatially variable, and the northern CCS experiences the most intense modification. These projected changes are consistent with the continued reduction in source waters oxygen, increase in source water nutrients, and, combined with solubility-driven changes, altered future upwelled source waters in the CCS . The results presented here suggest projections that resolve coastal processes are necessary for adequate representation of the magnitude of projected change in carbon stressors in the CCS.


Introduction
Greenhouse gas emissions have imparted large physical and biogeochemical modifications on the world's oceans (Friedlingstein et al., 2019;Gattuso et al., 2015;Le Quéré et al., 2018). The oceans have become warmer and stratification average 0.01 units lower than surface open ocean measurements at the Hawaii Ocean Time-series due to the upwelling process, and is also declining at a slightly faster rate (Chavez et al., 2017). The enhanced uptake of CO2 over the CCS shelf amplifies the rate of acidification compared to global rates. 70 Local processes such as upwelling, freshwater delivery, eutrophication, water column metabolism, and sediment interactions drive biogeochemical variability on regional scales (Cai et al., 2011;Feely et al., 2008;Pilcher et al., 2018;Qi et al., 2017;Siedlecki et al., 2017). In the CCS, winds are critical for upwelling variability and are projected to strengthen in response to global warming (Bakun, 1990;Garcia-Reyes et al., 2015;Wang et al., 2015;Rykaczewski et al., 2015;Sydeman 75 et al., 2014). The increased delivery of O2-depleted, carbon-rich waters, with enhanced nutrients and increased productivity has been projected for the CCS with a global simulation (Rykaczewski and Dunne, 2010). High-resolution projections for the CCS reinforced these findings (Dussin et al. 2019;Xiu et al., 2018), but projected that the impact of these altered conditions on productivity varied across the CCS with an increase in the north and a decrease in the south (Xiu et al., 2018), while productivity was identified as driving the biggest change in hypoxia in the region (Dussin et al., 2019). Howard et al. (2020b) 80 found that while alongshore winds intensified in the future, the upwelling response was dampened by increased stratification.
Global projections have coarse spatial resolution, often having only one or two grid cells for the continental shelf, and thus cannot resolve most of the local processes responsible for these observed coastal differences.
In this paper, we focus on the CCS and its known vulnerabilities to climate change by forcing regional models with the Coupled 85 Model Intercomparison Project 5 (CMIP5, Taylor et al., 2012) simulations. We produce multi-model regionally downscaled climate projections of multiple climate-associated stressors (temperature, O2, pH, Ω, and CO2) that resolve coastal processes to create ~100-year projections at resolutions of 12-km and 1.5-km in the northern CCS (N-CCS). First, we quantify the surface to 200-meter depth averaged, sea surface, and bottom condition changes for the climate stressors projected for 2100 at all resolutions (global, 12-km, and 1.5-km). Next, we use the multi-model ensemble to determine the degree to which climate 90 associated stressors are modified relative to global model projections of the CCS considering this signal both spatially, where the models overlap, as well as in different regions of the water column representative of different habitats. Finally, we interpret our results in the context of previous projections for the CCS and suggest drivers of the amplification in the downscaled projections by systematically comparing the projected changes in the winds, source waters, upwelling strengths, and coastal processes in each model system. 95

Model Descriptions
The downscaled regional modeling frameworks both employ the Regional Ocean Modeling System (ROMS, Shchepetkin and McWilliams, 2005). The regional models are forced with realistic atmospheric and ocean boundary conditions to make hindcast 100 simulations as well as future projections. The model domains are shown in Figure 1. The downscaled projections are referred to as "resolving coastal processes" because the historical simulations have been shown to represent observed coastal shelf variability. The model evaluation, provided here and available in other sources (Davis et al. 2014;Deutsch et al. in review;Giddings et al. 2014;Siedlecki et al. 2015), indicate the downscaled models perform better spatially and temporally (seasonal and interannual) than the global models for temperature, oxygen and carbon variables. 105 ~1-degree models: These include the CMIP model fields/global scale model. We focus here on only the representative concentration pathway 8.5 (RCP 8.5) from the Earth system models (ESMs) that make up the CMIP5 modeling framework.
The CMIP5 simulations include biogeochemical components described in Bopp et al. (2013). The CMIP models employed here are further described in Section 2.3. 12-km model: The mid-resolution (12-km) ROMS-based simulation of the CCS is configured for a domain that extends along 110 the North American west coast from 25 deg N to 60 deg N and described in more detail in Howard et al. (2020b). A curvilinear grid is used in the horizontal with close-to-uniform 12-km horizontal resolution and 33 s-coordinate (terrain-following vertical) levels. Atmospheric conditions including air-temperature at the sea surface, precipitation, and downwelling radiation are derived from an uncoupled Weather Research Forecast model output (c3.6.1; Skamarock et al. 2008) as in Renault, Hall, & McWilliams (2016) and Renault et al. (2020) with more information in Howard et al (2020b). To avoid the computational cost 115 of a fully-coupled ocean-atmosphere model, wind and mesoscale current feedbacks are parameterized with a linear function of the surface wind stress as in Renault, Molemaker, et al. (2016). This linear relationship is supported by observations in the CCS (Renault et al. 2017). The biogeochemical model is detailed in Deutsch et al. (2020), and follows Moore et al. (2004).
The model has skillfully simulated the recent interannual to interdecadal biogeochemical variability in the CCS (Howard et al. 2020b), and a similar model setup forced with data-assimilated forcing skillfully simulated O2 variability in the CCS over the 120 last two decades (Durski et al., 2017).
1.5-km model: The highest resolution (1.5-km) simulations of the N-CCS rely on a modeling framework developed by the University of Washington Coastal Modeling Group optimized for the Pacific Northwest "Cascadia" region. The Cascadia model domain encompasses the inland waters of the Salish Sea, coastal waters of the N-CCS (Fig. 1), and includes freshwater and tidal forcing. The grid has a horizontal resolution of 1.5-km on the shelf, 4.5 km far offshore, and 40 s-coordinate (terrain-125 following vertical) levels, with enhanced bottom and surface resolution. The atmospheric conditions from the 12-km model are used to force the model at this resolution. The Cascadia model does not simulate biogeochemistry within the Salish Sea, but yields realistic nitrate outflow from the Strait of Juan de Fuca to the outer coast shelf (Davis et al., 2014). Hindcast experiments from 2004 to 2007 were extensively validated and exhibited skill on all regions of the shelf (Davis et al., 2014;Giddings et al., 2014;Siedlecki et al., 2015). 130

Model Metrics
Carbon variables (e.g. pH, pCO2, and Ω) were computed using model output of dissolved inorganic carbon (DIC), total alkalinity (TA), temperature, and salinity and routines based on the standard OCMIP carbonate chemistry adapted from earlier studies (Orr et al., 2005) using CO2SYS (Lewis and Wallace, 1998). The total pH scale is used for pH throughout.
To compute model means and inter-model comparisons, first a climatological year was generated for each model grid cell, 135 using the 2002-2004 years for the 12km and 1.5km regional models. Then, annual average values for each cell were calculated from the climatology. Finally, spatially-weighted means were calculated from the annual average values. For comparisons between model resolutions, the coarser model was interpolated to the higher resolution model grid. For example, the global values were interpolated onto the 12-km grid prior to averaging the fields within the CCS. Surface conditions were drawn from the surface vertical layer for each simulation. Depth-averaged ocean conditions were calculated over the upper 200 m for 140 all simulations. Where water depth was shallower than 200 m, the entire water column was averaged. For the bottom comparisons, the global simulations have a very different bathymetry than the downscaled simulations because of their coarse resolution, consequently the regional averaging efforts reported here as bottom conditions were isolated to the 0-500 meters depth interval only, to ensure that the global model resolved that depth interval. We also report the downscaled values over the shelf only, limiting the determination of metrics out to the 200 meter isobath. 145 Upwelling intensity -To estimate upwelling, several metrics were employed. The first two rely on the intensity of the winds (e.g. Cumulative Upwelling Index, CUI (Schwing et al., 1996); 8-day wind stress (Austin and Barth, 2002)), the third and fourth rely on measures in the water column itself and are referred to as the Coastal Upwelling Transport Index (CUTI) and Biologically Effective Upwelling Transport Index (BEUTI) (Jacox et al., 2018). The wind-based metrics, CUI and the 8-day, 150 are the same for both downscaled simulations, but CUTI and BEUTI are specific to each ocean model as they are calculated based on ocean measures like vertical transport and nitrate concentrations. CUTI and BEUTI were integrated over bins of 0.5degree latitude spanning 0-50 km offshore.

Future Forcing
To generate future downscaled projections, the global CMIP5 simulations were used to force regional simulations. The 155 methods employed are outlined below.
The 12-km historical simulation forcing is described in Renault et al (in review) and the companion paper, Deutsch et al. (in review). The 12-km projection was forced by adding a monthly climatological difference between CMIP5 RCP 8.5 scenario forcing and the historical run forcing, averaged over 2071-2100and 1971-2000, respectively (Howard et al., 2020b following 160 the delta method commonly applied in dynamical downscaling and described in Alexander et al. (2020). This is done for all variables that influence the surface energy budget including net downward shortwave and longwave radiation, 10-m wind speed (u and v component), air temperature, and specific humidity. CMIP5 models are from GFDL (ESM2M), IPSL (CM5A-LR), Hadley (GEM2-ES), MPI (ESM-LR), NCAR (CESM1(BGC)). A total of six RCP 8.5 scenario runs were conducted: one for each individual CMIP5 model realization, and a final run using the five-member ensemble mean forcing. For this 165 manuscript, we report the output from this final, ensemble mean-forced scenario. However, the output from the five individual CMIP5 model realizations were used to calculate the standard deviation values across the ensemble spread reported in Table   1. Initial and boundary conditions had the same kind of centennial trend addition for temperature, salinity, and all biogeochemical tracers (O2, nitrate, phosphate, silica, iron, dissolved inorganic carbon, alkalinity). More information can be found in Howard et al. (2020b, Table 1). 170 The 1.5-km projection was forced using the open ocean boundary conditions and atmospheric forcing from the 12-km regional simulation described above (Howard et al., 2020b). The boundary conditions included biogeochemical fields from the 12-km model. Because the ecosystem model (BEC) in the 12-km parent grid has more variables than the Cascadia simulation, some of the variables were merged. Specifically, the phytoplankton fields were added together, and the nutrients (ammonia and 175 nitrate) were summed into one nitrogen field. To ensure no biogeochemical model drift between the nested 1.5-km simulation and the 12-km simulation, after one year of spin up, a simulation of 2007 was compared against observations from the region (Fig. 2). The 1.5-km biogeochemical model skill was similar to the original model runs previously published (Davis et al. 2014;Giddings et al. 2014;Siedlecki et al. 2015) for 2007 ( Fig. 2) without any significant drift in time. Temperature and salinity both experienced a significant bias in the upper 200 meters in this configuration, unlike the previously published model 180 runs ( Fig. 2; temperature RMSE 2.43; salinity RMSE 0.627). As we are focused on differences between the modern and the future and we expect the bias to remain the same, we do not bias correct the forcing here.
For the future conditions, atmospheric CO2 concentration (800 ppm), and future atmosphere and ocean forcing from the 12km runs drove the Cascadia simulations. The river forcing was approximated by altering the timing of the freshet of the 2007 185 forcing earlier in the year by two months. This is in line with some historical analyses from the N-CCS region (Riche et al., 2014) as well as some projections of future hydrological conditions for the Fraser River (Morrison et al., 2002). Both of these results suggest that the total precipitation will remain the same, but the increase of rain and decrease in snowpack will shift the freshet earlier ( Figure S2). The river TA was not altered from historical forcing, but the rivers equilibrate with the future atmospheric CO2 concentration (800 ppm). 190

Future Change
For each resolution, simulations were run for a number of years in a base/present state and then compared to a future simulation.
The change is the difference between the future and base/present state, representing a ~100-year anomaly due to climate forcing. The historical/base state for the CMIP5 runs was computed from an ensemble mean spanning 1971-2000. For the 12km simulation, the base/present state spanned 1994-2007. For the 1.5-km simulation, the model was spun up for one year 195 which is part of the appeal of large-scale simulations. The CMIP5 future runs consist of a thirty-year mean spanning 2071-2100. The 12-km model is a late 21st century run spanning 2085-2100. The 1.5-km model is also late 21st century spanning 2094-2096. Results and comparisons for the work presented using both the 12-and 1.5-km-resolution simulations were made using the same year span despite runs existing for a broader range of years for the 12-km simulation. The present state was 200 considered as 2002-2004 for both the 12 and 1.5-km simulations and the future was 2094-2096 (Table 1). The global model ensemble average results represent a 30 year climatology.

Modification
The range of the five ensemble members which forced the 12-km projections is used to bound the potential futures expected.
When the differences provided in Table 1 between the mean downscaled conditions for a region of the CCS (CCS-wide, 205 columns B-C or Cascadia/N-CCS, columns E-G) and the ensemble spread quantified from the 12-km model projections (columns D and H) both exceed the 1 degree model projected change (columns B and E), those regions of the CCS are projected to undergo amplified change. The converse is referred to as dampening. The ensemble spread is provided in Table 1 as the range of the 5-member model spread of the annual average results for the 12-km model projections (columns D and H). The direction with which each variable described above was amplified relative to the global models is highlighted using dark grey 210 (amplified) and light grey (dampened) shading in Table 1.

Results
Model projections of climate-driven stressor variables (temperature, O2, pH, saturation state (Ω), and CO2) in each downscaled projection were compiled for the CCS for three depths (200 m averaged, surface, and bottom < 500 m; Table 1; Fig. 3-5). The global average changes for many of these variables are different from the 1 degree model projected values for the CCS region, 215 but in only a few cases does this difference fall outside of the ensemble spread reported in column D and H of Table 1 (i.e., the signal is amplified or dampened). For each variable and depth, the change between the base state and the projected state is described below and evaluated within the context of the ensemble spread.

Temperature
The surface to 200-meter depth-averaged temperatures at all model resolutions is consistently warmer in the future CCS, both 220 CCS-wide (1.63 and 1.81 degrees C in the 1 degree and 12-km models, respectively, Table 1, columns B and C), and within the N-CCS (1.95 to 2.32 degrees C across the three model resolutions; Figure 3a; Table 1 Columns E-G). The Washington shelf experiences the largest projected differences in the 1.5-km projection (2.32 degrees C, Fig. 3). The 1-degree model projected increase for the CCS (1.63 degrees C) and the N-CCS (2.21 degrees C) falls within the range of warming from the downscaled projections (CCS: 1.38 to 2.24 degrees C; N-CCS: 1.55 to 2.35 degrees C, Columns D and H in Table 1). In both 225 regions of the CCS, the differences between the models are smaller than the range of the 12-km ensemble (Table 1).
At the surface, the SST is warmer in the future in all projections. Spatially, the SST increases most offshore, and increases least near the coast in all simulations of this eastern boundary upwelling system, as a result of upwelling (Fig. 3b). The 1.5 km model projects slightly smaller increases in SST than the 1-degree model or the 12-km model. However, the 1-degree models 230 project SST increases CCS-wide (3.12 degrees C) and in the N-CCS (3.15 degrees C), which fall within the range of SST projections from the 12-km ensemble of downscaled projections (CCS: 2.57 to 4.05 degrees C; N-CCS: 2.42 to 4.18 degrees C, Columns D and H in Table 1).
At the bottom, the temperature increases the most near the coast. The abyssal regions show little to no change in temperature 235 ( Fig. 3c). The shallowest regions of the 1.5-km simulation experience the largest warming -nearly 3 degrees C. In the coastal process-resolving downscaled projections, the projected bottom temperature change is greater (1.75-1.84 degrees C, 2.05 degrees C) than the global projections (1.34-1.65 degrees C). However, the 1-degree model projected increases for the whole CCS (1.65 degrees C) fall within the range of temperature projections from the ensemble of 12-km projections (CCS: 1.47 to 2.21 degrees C, Column D in Table 1), while the N-CCS (1.34 degrees C) is lower than the range of temperature projections 240 from the ensemble of 12-km projections (N-CCS: 1.40 to 2.10 degrees C, Columns H in Table 1).

Oxygen
Annual depth-averaged O2 concentrations at all model resolutions consistently decrease in the future compared with the base state, but the magnitude of the decrease is slightly more severe on average in the downscaled projections (Table 1; Fig. 3). The spatial variability within the CCS region, with more severe declines occurring in the N-CCS, is consistent across models but 245 varies in magnitude. The 1-degree model projected decrease for the CCS (-0.52 ml/l) and the N-CCS (-0.56 ml/l) fall within the range of the ensemble of projections from the 12-km model (Table 1; CCS: -0.52 to -0.72 ml/l; N-CCS: -0.52 to -0.92 ml/l, Table 1).

Columns D and H in
At the surface, O2 declines in all projections, and the degree of change is similar across resolutions (Table 1 The bottom O2 concentration declines in all projections near the coast (Table 1; Fig. 3). The range of change in bottom O2 on the shelf in the 1.5-km projection varies by a factor of two, with the most extreme changes occurring on the outer shelf and in 255 pockets known to experience persistent hypoxia in the present ocean -e.g. near Cape Elizabeth, south of Heceta Bank, and within the region associated with the Juan de Fuca Eddy (Siedlecki et al., 2015). The 1 degree model projected decrease for the CCS (-0.43 ml/l) and the N-CCS (-0.63 ml/l) falls within the ensemble range of the ensemble of bottom O2 concentration projections from the 12-km model (CCS: -0.37 to -0.75 ml/l; N-CCS: -0.40 to -0.92 ml/l, Columns D and H in Table 1). When the bottom is restricted to the shelf in the downscaled simulations (< 200 m isobath; values with asterisk (*) in Table 1), this 260 decrease is more severe but still does not fall outside the range from the comparable depth range of the 1-degree model projection (<500 m).

pCO2
All model projections of pCO2 consistently increase, with larger increases in the downscaled projections than in the global projection (Table 1; Fig. 4). The spatial variability within the CCS region differs across resolutions. All projections show an 265 onshore-offshore gradient in pCO2 with smaller changes closer to the coast and larger changes offshore. In the coastal processresolving downscaled projections, the projected depth-averaged change in pCO2 increases, and the gradient between the nearshore and offshore intensifies. The 1-degree model projected increase for the CCS (492 µatm) and the N-CCS (527 µatm) falls below the ensemble range of downscaled pCO2 projections from the 12-km model (CCS: 682-836 µatm; N-CCS: 780 to 1066 µatm, Columns D and H in Table 1). 270 At the surface, future pCO2 consistently increases in all projections, but varies widely across resolutions (Table 1; Fig. 4). In the downscaled projections, most upwelling areas experience a smaller increase in surface pCO2 than offshore waters. In the 1.5-km projection, regions near the coast of Oregon show the largest surface pCO2 differences between the base and future states, while the region associated with the Columbia River plume shows a much smaller change. Overall, the inclusion of 275 coastal processes contributes to the spatial patterns and magnitudes of projected changes. Consistent with the subsurface signal, At the bottom, pCO2 is consistently higher in all projections, with varying magnitude across the model resolutions (Table 1;

pH 285
The pH averaged over 200-meter depths for all model resolutions consistently decreases, and change is less severe than the global models project for the CCS region (Table 1; Fig. 4). In the 12-km simulation, a slightly smaller pH change (~ -0.26) is observed in the southern CCS than the entire CCS, within the influence of coastal upwelling. In the 1.5-km simulation, the regions of largest decline are on the outer shelf and patches of the Oregon shelf. In the downscaled projections, the depth-averaged 200-meter pH is lower than the 1-degree model projected change in the North and greater than the global change 290 CCS -wide. The 1-degree model projected decrease for the CCS (-0.321) and the N-CCS (-0.332) fall within the ensemble range of projections for the downscaled 12-km model (CCS: -0.310 to -0.357; N-CCS: -0.278 to -0.352).
At the surface, the pH consistently decreases in all projections, and is less severe a decrease in the downscaled projections than for the same region in the global model (Table 1; Fig. 4). The 1-degree model projected decrease for the CCS (-0.319) and 295 the N-CCS (-0.343) is larger than the ensemble range of projections for the downscaled 12-km model (CCS: -0.285 to -0.287; N-CCS: -0.296 to -0.300).
At the bottom, the pH decreases on the shelf in all projections (Table 1; Fig. 4). In the 1.5-km resolution model, the projected conditions show spatial variability on the shelf that is not apparent in the coarser models. The most severe changes in bottom 300 pH correspond with regions that experience the largest changes in bottom O2. The downscaled projections indicate decreases in annual average bottom (< 500 m) pH, but spatial variability exists on the shelves in the coastal process-resolving simulations.
This difference between the 1 degree and downscaled simulations is even greater on the shelves (indicated as the starred values in Table 1 (Table 1).

315
At the surface, Ω consistently decreases in all projections (Table 1; Fig. 5). Spatially, the 12-km projection shows the largest declines in surface Ω in the southern domain with little gradient between the shelf and offshore. At the 1.5-km resolution, the N-CCS projected declines are lowest offshore, and the declines are even larger than the 12-km projected changes even when considering the ensemble spread. The 1-degree projected decrease for the CCS (-0.96) is larger than the range of the ensemble of projections from the downscaled 12-km model (CCS: -0.86 to -0.94). The 1 degree projected decrease for the N-CCS (-320 0.76) is smaller or less severe than the range of the ensemble of projections from the downscaled 12-km model (N-CCS: -0.82 to -0.88). The same is true for the surface decline Ωcalcite (Table 1).
At the bottom, Ω decreases in all projections near the coast on the shelves (Table 1; Fig. 5). At the 1.5-km resolution, spatial variability in the magnitude of the projected conditions exists on the shelf. The most severe changes in bottom Ω correspond 325 with regions that experience the largest changes in bottom O2. In the N-CCS, the 1.5-km projected declines are even larger than the 12-km projected declines even when considering the ensemble range. The 1-degree model projected decline in Ωarag for the CCS (-0.47) is more severe than the downscaled ensemble range (CCS: -0.38 to -0.46) for the downscaled 12-km model.
In the N-CCS, the 1-degree model projected decline in Ωarag for the N-CCS (-0.32) falls within the range of the ensemble of projections from the downscaled 12-km model (-0.30 to -0.40). In the 1.5-km model projections, the decline is greater than the 330 1-degree model projects and falls well outside the range of the 12-km projections. This difference between the global and downscaled simulations is even greater on the shelves (indicated as the starred values in Table 1). The same is true for the bottom decline in Ωcalcite (Table 1).

Themes across projected changes for the CCS 335
All climate-associated stressor variables agree with the 1-degree projections in terms of the direction of the trend, but not the magnitude of the change. The 1-degree model projections for the CCS are largely consistent with the 1 degree model projected global trends with some differences in the nearshore upwelling areas (Fig. S1). All carbon variables are sensitive to the inclusion of coastal processes which both downscaled projections provide. In addition, all of the projections suggest greater change in most variables in northern regions of the CCS, and in the upwelling regions. 340 Nitrate increases over much of the domain in the upper 200 m (Fig. 6, Table S1) in both the high-and medium-resolution downscaled simulations. Nitrate on average increases in the global simulations, but the magnitude and direction varies widely across the ensemble members, a result consistent with Howard et al. (2020b). In addition to the projected small increase in nutrient concentrations in the upwelling system of the CCS, the winds are slightly more intense (2 % increase in the magnitude 345 of the wind stress) during the upwelling season (April -September) in the future years. The timing of the onset and duration of the upwelling season in the northern CCS remains the same in the future in these projections. Despite these differences from wind-based upwelling metrics, the in-water upwelling metric CUTI indicates no net change in the upwelling intensity of the future N-CCS in the simulations evaluated here. When nitrate is included in the upwelling measure, as in BEUTI, there is a slight decline in the upwelling of nitrate (1-2%), consistent with a decrease in nitrate at the surface in the N-CCS (Fig. 6). 350 This result is sensitive to the distance offshore (20-75km) over which the index is calculated. The direction of the trend does not change, but BEUTI for example, further declines (4%) as bin boundaries move closer to shore. The further offshore the bin extends, the weaker the signal becomes. Both of these measures suggest that the upwelling is not intensified in our projected future, despite the slight increase in winds. This result is consistent with the results of Howard et al. (2020b), where increased stratification in the future simulations impeded increases in upwelling intensity. 355 The projected temperature change affects the solubility of the gases, generating a solubility-driven decline, and the increased nutrient content would correspond to a stoichiometric oxygen loss as well. Over the entire CCS, the decrease in oxygen over the upper 200 meters was 0.45 ml/l or 20.10 mmol/m 3 (Table 1). The solubility-driven change accounts for most (~67%) of this change (13.42 mmol/m 3 using 1.92 change in temperature from Table 1). The additional nitrate brought into the region 360 from the large-scale models (0.79 mmol/m 3 ) corresponds to an additional drawdown of 6.81 mmol/m 3 of oxygen. In the N-CCS, the change in oxygen is a bit larger than across the entire CCS -0.69 ml/l (30.82 mmol/m 3 , Table 1). The solubility driven changes contribute a bit less (~44%) in the N-CCS than the entire CCS, but the nitrate signal is larger in the N-CCS corresponding to 11.21 mmol/m 3 change in oxygen from an increase of 1.3 mmol/m 3 of nitrate in the upper 200 meters of the water column. The result is that the solubility driven changes combined with the increased supply of nutrients to the upper 200 365 meters of the N-CCS account for 80% of the projected oxygen change in the N-CCS.
Similarly, we would expect carbon content to increase in source waters commensurate with an increase in nutrients and lower oxygen concentrations. The corresponding stoichiometric increase in DIC to the increase in nutrients (5 mmol/m 3 ) would only account for a small decline in pH (0.02) or Ω (0.03). The majority of the pH and Ω decline is due to anthropogenic carbon 370 content increase in DIC associated with the RCP 8.5 scenario forcing (~95 mmol/m 3 ). Spatial variability in ∆DIC corresponds with variability in the water buffer capacity and Revelle factor, with southern CCS regions of relatively high buffer capacity having the greatest rates of DIC uptake (Fig. 7, Table S1).
TA increases in the future in the subsurface on the shelves of the CCS and even more so on the upper slope (Fig. 7, Table S1), 375 but it mostly decreases across domains. At the surface, it declines, and these two changes offset each other in the depthaveraged 200 meter change. Overall, the decrease in TA in the projections from the downscaled simulations is substantially smaller than the decrease from the 1 degree models (Table S1). These changes in their corresponding depth ranges contribute to the results for the carbon variables in Table 1, impacting different carbon variables differently, for example, pCO2 is amplified while pH is not modified over much of the CCS, and Ω is amplified at the bottom and the surface but dampened 380 over the upper water column. These patterns will be examined more closely in Section 3.7 that focuses on modification and which follows.
On the shelves of the downscaled simulations, the source waters are further modified by coastal processes including increased benthic-pelagic coupling, freshwater delivery and denitrification. The inclusion of these processes causes the bottom waters 385 on the shelf to experience a more severe increase in pCO2 and declines in oxygen, pH, and Ω than observed in the shallowest regions of the global models (starred values in Table 1). The difference between the bottom estimated changes in the CCS in the depth ranges resolved by the global models and on the downscaled shelves is greatest for the carbon variables.

Modification
In general, the CCS experiences a greater change for most variables at all resolutions than the global ocean. However, only 390 the carbon variables emerge as amplified or dampened by the downscaled simulations. Across the spread of ensemble members for the entire CCS in the 12km simulation, the downscaled projected increase for pCO2 (columns C, F, G) is amplified relative to the 1-degree models (columns B, E) in all three depth ranges (Table 1). The N-CCS pCO2 increase is amplified in both downscaled simulations (12-km, 1.5km), at all depth ranges (Table 1, columns F and G). In the 200 m depth averaged changes, the pCO2 change is correlated to the TA changes (Table S2). At the surface the changes in pCO2 are correlated to the 395 temperature, and at the bottom, DIC, TA, and nutrient changes (Table S2).
The downscaled decrease in pH at the surface and the bottom is modified relative to what the 1-degree models project for the CCS and N-CCS in the downscaled projections. The surface change is less severe than in the 1-degree model, and so is considered dampened relative to global change. At the bottom, pH in the N-CCS is dampened relative to the 1-degree model. 400 In both downscaled projections (12-km, 1.5km), the surface and bottom pH decrease is dampened in the N-CCS relative to the surface pH decline in the 1 degree model for that region (column E). While pH is not modified in the 200 m depth averaged changes, at the surface, the pH changes are correlated to the DIC and TA (Table S2).
The decrease in Ω over the upper 200 meters is less severe than the 1-degree model projection for the N-CCS, and falls outside 405 of the ensemble range, so is considered dampened relative to global change (Table 1). The N-CCS 200 m depth averaged Ω decrease is dampened in both downscaled simulations (12-km, 1.5km), and amplified at the surface (Table 1). At the surface, the changes in saturation state are highly correlated to DIC changes. At the bottom, the changes are also correlated to the changes in nutrients (Table S2). In the N-CCS region, more of the water column is modified for the carbon variables.

Discussion 410
Globally, under RCP 8.5, the future oceans simulated using CMIP5 are projected on average to be warmer (SST, mean ± 1 SD: 2.73 ± 0.72°C), higher in pCO2, lower in O2 content (RCP8.5: -3.45 ± 0.44%), and more acidified (surface pH -0.33 ± 0.003 units) (Gattuso et al., 2015). Regionally, the CMIP5 models project the North Pacific to be one of the regions to experience the most warming, most severe O2 declines, and largest extents of corrosive conditions (Bopp et al., 2013;Feely et al., 2009;Gattuso et al., 2015;Gruber et al., 2012;Hauri et al., 2013;Long et al., 2016). Here, we present one of the first 415 downscaled multivariable projections of environmental change out to the end of the century in the CCS driven by a suite of CMIP5 forcings instead of a single global model. While the downscaled projections show changes that are similar in direction to those of the global simulation, the magnitude and spatial variability of the change differs in the coastal process-resolving downscaled projections and to a varying degree depending on the variable, depth range, and subregion of interest.

420
The CCS experiences a greater change for many variables at all resolutions than the global ocean, however this change is modified in the CCS in both downscaled simulations (12-km, 1.5km), for pCO2, Ωarag and Ωcalcite, and pH. Amplification of global trends within the CCS upwelling systems in the future has been shown before for oxygen specifically (Dussin et al., 2019) and was identified through the response of the downscaled model to a series of idealized experiments with perturbations in the source water oxygen and nutrient concentrations. Source water changes in oxygen drove a two-fold larger change in 425 oxygen than nutrient supply alone, and both of these drivers were determined to be more important than intensifying winds.
In our projections, the more realistic winds were different than in Dussin et al. (2019), with a small intensification of about 2%. The source waters were lower, but the oxygen decrease in the N-CCS fell within the range of the ensemble members explored here. Dussin et al. (2019) only explored one global model (GFDL) as a driver, and as such the definition of amplification differs from the one we use here. Much like experiments conducted by Liu et al. (2012Liu et al. ( , 2015 for ocean 430 temperatures and described in Alexander et al. (2020), using a multi-model mean to drive a downscaled ocean model retains the linear component of the climate change forcing only and is not able to assess the range of the response. Fundamentally, our definition of amplification relies on the range of the ocean condition responses. The mechanism of remote biogeochemical redistribution and influence via boundary conditions in the CCS remains influential for the carbon variables that were identified as amplified here. While climate stressor variables have been identified as amplified historically using records spanning 435 several decades or more (Chavez et al., 2017;Osbourne et al., 2020), regional future projections have focused on multi-model mean conditions projected over 100 years into the future. Although the downscaled model projected a small intensification in the projected upwelling-favorable wind stress of 2% which is consistent with prior work on this topic (Bakun, 1990;Garcia-Reyes et al., 2015;Rykaczewski and Dunne, 2010;440 Rykaczewski et al., 2015), no change was quantified in the upwelling fluxes in the water column using the CUTI, and a slight decline was observed in BEUTI (nutrient flux) measures of upwelling. This is likely due to the compensation from increasing stratification in the future simulations as noted in Evans et al. (2020). We observe lower oxygen, higher nutrient content in source waters, but this change does not make it to the surface. Within the CCS, solubility-driven oxygen changes are important, which is consistent with oxygen escape from the ocean being increasingly important in future projections (Li et al. 2020). 445 Moving south within the CCS, solubility increasingly outcompetes nutrient-driven changes in oxygen drawdown. However, the projections here indicate that the magnitude of oxygen decrease was within the bounds of the ensemble range, and so was not amplified relative to the global models, unlike the carbon variables.
The projected change in pH is consistent with prior pH projections for the CCS downscaled with the same RCP 8.5 scenario 450 using a ROMS model (Gruber et al., 2012;Hauri et al., 2013;Marshall et al., 2017;Turi et al., 2016). These projections were performed with different biogeochemical models described in Gruber et al. (2012), and Fennel et al. (2006 and relied on multi-model means or individual ensemble members for the global models. The pH values we obtained are lower than the projections of Rykaczewski and Dunne (2010) for the CCS. They used GFDL Earth System Model 2.1, a different forcing downscaled projections provided here of pCO2, Ω, and pH, the biogeochemical model formulations differ across resolutions and may be contributing to the amplified signals. Differences include parameterizations of gas exchange, detritus classes, sinking velocities, as well as benthic boundary conditions; the latter has been identified in prior work to be important for O2 models in coastal regions (Moriarty et al., 2018;Siedlecki et al., 2015). Any of these could contribute to the differences observed across model resolutions. CMIP5 results are based on an ensemble average of many models, which all utilize different 460 formulations and complexities for their ecosystem models, further contributing to the uncertainty provided from the biogeochemical boundary conditions. Overall, the projected annual, depth-averaged change in pH appears to depend mostly on the anthropogenic forcing scenario, and all the models agree on the direction and relative magnitude despite these differences.

465
Projected changes in surface and bottom pH, Ω, and pCO2 are modified by the inclusion of coastal processes when downscaling is employed. Coastal processes that influence the variability of carbon variables occur in these regions of the water column include freshwater delivery and sediment water interactions. At the surface, pH, Ω and pCO2 change differently. TA declines at the surface, while DIC increases. DIC increases due to increased storage of carbon from the increase in carbon in the future atmosphere. The TA changes are driven in part by the altered timing of the freshet in the N-CCS as well as the presence of a 470 river plume in an upwelling regime. Freshwater in the region is known to be corrosive due to naturally low TA, which impacts the buffering capacity of the surface waters. This result can be seen in the surface difference plots for pCO2 near the Columbia River plume (Fig. 4) and the surface TA change (Fig. 7). The 12-km projections include climatological freshwater fluxes as precipitation along the coastline instead of resolving river plumes like the 1.5-km projections, but despite these different freshwater parameterizations, both models indicate modification of carbon variables in the N-CCS Cascadia domain with 475 different directions for different variables. While the freshwater amplifies the global rate of change for the surface pCO2 and Ω, over the entire water column (200 m average), the pH change is dampened as the DIC and TA changes are offset by the temperature changes for that variable. In our regional downscaled simulations, the change in temperature and TA act together to offset the increase in the DIC signal 480 in the coastal upwelling regime (Fig. 7), and for pH these changes offset each other in the upper 200 m of the water column.
The global models show very little change in TA in the region. While the downscaled bottom TA change is small (20-50 mmol/m 3 , Fig. 7, Table S1), this amounts to an increase in pH of 0.07-0.18 and an increase in Ω of 0.15-0.21 -enough to offset 40-60% of the reduction in Ω due to increased atmospheric CO2 concentrations. At the bottom, the increase in TA modifies the projected pH change in the N-CCS by reducing it. 485 Both biogeochemical models include denitrification at the sediment water interface which impacts both the nitrogen and TA cycling in the model. Denitrification is a source of TA (Chen et al., 2002) and has been shown to impact shelf wide TA in other regions (Fennel et al., 2008). In the CCS, denitrification has been observed on the shelf and slope, peaks on the slope, and is greater off the Washington coast than off Mexico (Hartnett and Devol, 2003). As the source waters become lower in 490 oxygen content, denitrification should increase, providing an additional feedback on the nutrients and carbon variables, and dampening the global ocean acidification signal. Bottom Ω is not amplified in Table 1, but if calculated solely over the shelf region (asterisks in Table 1), then bottom Ω changes are amplified in both domains. While this source of alkalinity has not been observed directly in the modern ocean, a source of TA was identified and interpreted as calcium carbonate dissolution in the CCS (Fassbender et al., 2011). These results suggest future projections should consider salinity and TA forcing feedbacks 495 when performing regional projections as these can alter regional impacts of carbon variables.
Different carbon variables are sensitive to different physical climate forcings, a result that is consistent with recent work suggesting that competing sensitivities may dampen the variability of pH in the future (Jiang et al., 2019;Kwiatkowski and Orr, 2018;Salisbury and Jonsson, 2018;Takahashi et al., 2014). The sensitivities of the various carbon variables to thermal 500 and geochemical (carbon dioxide) forcing was explored in an idealized simulation of future conditions from CMIP5 models in Kwiatkowski and Orr (2018), in the modern ocean in Takahashi et al. (2014) and Jiang et al. (2019), and in the Gulf of Maine by Salisbury and Jonsson (2018). Kwiatkowski and Orr (2018) determined that the balance between the change in DIC and TA drives the variability of pH, in combination with temperature in mid-to-high latitudes --effects that largely cancel each other out. While the processes that control TA and DIC are similar, the addition of atmospheric CO2 changes DIC alone, 505 altering the balance between these two reservoirs. In the Gulf of Maine, temperature and salinity anomalies in combination with TA variability offset the long-term OA trend (Salisbury and Jonsson, 2018). Their analysis indicated that Ω is more sensitive to temperature and salinity variations than pH, and this result was confirmed globally in Jiang et al. (2019). Kwiatkowski and Orr (2018) also found the seasonal amplitude of Ωarag is expected to strengthen in some regions and attenuate in others due to the high sensitivity of this variable to temperature. 510 By examining the relationships between the carbon variables and other regional variables, the relative importance of different processes responsible for the modification of carbon variables by coastal systems can be inferred. Themes emerge in the correlations with carbon variables across different regions of the water column: at the surface, DIC changes are important, and at the bottom, changes in DIC, TA, and nutrient content changes have the highest correlations with more than one carbon 515 variable. Over the 200 m water column, TA is important. The TA decreases in all the projections but by an order of magnitude less than the 1 degree models in the downscaled simulations (Table S1). A dampened reduction in TA is consistent with changes in the biological metabolism or benthic pelagic coupling alongside the sedimentary processes in the region -all of which can act as a source of TA on shelves.

520
If we take a specific example of Ω, the Ω changes are correlated with nutrient content changes, in addition to the DIC changes that result from the emissions scenario forcing. As a result, bottom Ω on the shelf is modified by the TA generation organic matter remineralization and via sedimentary processes. Consistent with this result, the bottom change in DIC is greater in the downscaled simulations than the 1 degree models (Table S1). The 1 degree models do not resolve the shelf bathymetry, and thus the sedimentary processes that seem likely to play an important role in this coastal setting. 525 These results are consistent with the idea that the changes in different regions of the water column are driven by different mechanisms -surface changes in the carbon variables are driven by the RCP and modified by coastal biological cycles, bottom conditions are influenced more by benthic pelagic coupling and sedimentary processing as suggested by the TA and nutrients correlations, and the water column modification is a combination of all of these. 530 Spatially, within the CCS, the projections suggest greater change in most variables in northern regions of the CCS, and on the shelves in the annual averages. Currently, the N-CCS is the most productive region of the CCS (Davis et al., 2014;Hickey and Banas, 2008) and experiences the most prevalent hypoxia (Connolly et al., 2010;Peterson et al., 2013) and corrosive events (Feely et al., 2018;. One aspect that differentiates the region from the rest of the CCS is that the N-CCS experiences 535 both seasonal upwelling and downwelling currently (Hickey and Banas, 2008), which is not projected to change in 2100 under RCP8.5. During the summer upwelling season in the N-CCS, oxygen declines on the shelf over the season as organic material and respiration increases on the shelf. Coincident with the oxygen decline, carbon parameters are impacted in kind (e.g. pH and Ω decreases, pCO2 increases; Siedlecki et al. 2016;Feely et al. 2018). The dramatic fall transition on the Washington and Oregon shelves is observed in the modern ocean (Adams et al. 2013;Connolly et al. 2011;Hales et al. 2006;Siedlecki et al. 540 2015), fully quantified by an oxygen budget in Siedlecki et al. (2015), and is projected to continue in the future projections examined here. The continued existence of a downwelling season in the future impacts some variables more than others. The projected change for oxygen and carbon variables is different seasonally. The fall transition and winter mixing re-oxygenates shelf waters (Siedlecki et al., 2015), and this pattern continues into the future. Hypoxia will continue to exist in the summer months, but for a longer period of the summer. Corrosive, low pH conditions, however, will occur year-round. The fraction of 545 the year during which bottom water on the shelf is corrosive (Ω <1) or low pH (pH <7.65) nearly doubles. This asymmetry has been identified previously in the Salish Sea (Ianson et al., 2016), and exists because of the difference in equilibration timescales at the surface for oxygen and carbon dioxide.

Conclusions
We present one of the first multi-model, downscaled multivariable projections of changing temperature, pH, pCO2, Ω, and O2 550 in the CCS. The downscaled projections are driven by the same delta forcing as the global simulation, and projected changes for the CCS are consistent with the directional trends indicated by the global model for scenario RCP 8.5 -warmer, more acidified, higher carbon content, and lower oxygen concentration. However, the magnitude and spatial variability of the change differs in the coastal process-resolving downscaled projections and to a varying degree depending on the variable of interest.
Changes in pCO2 concentrations, Ω, and pH are modified in the downscaled projections relative to the projected global 555 simulation, suggesting downscaled projections are necessary to more accurately project future conditions of these variables.
Carbon variables are likely modified by the inclusion of shelf processes including benthic pelagic coupling and sedimentary processing as suggested by the TA and nutrients correlations. The diversity of these projected changes of future ocean conditions emphasizes the need to improve our understanding of mechanisms by which coastal processes interact with these large-scale drivers of change or properly simulate and capture these feedbacks in projections. 560

Code and Data Availability
Archived model fields will be available from the Zenodo library at upon acceptance of this publication. Codes used for 12 km downscaled model simulations are available online (https://github.com/UCLA-ROMS/Code