Persistent spatial structuring of coastal ocean acidification in the California Current System

The near-term progression of ocean acidification (OA) is projected to bring about sharp changes in the chemistry of coastal upwelling ecosystems. The distribution of OA exposure across these early-impact systems, however, is highly uncertain and limits our understanding of whether and how spatial management actions can be deployed to ameliorate future impacts. Through a novel coastal OA observing network, we have uncovered a remarkably persistent spatial mosaic in the penetration of acidified waters into ecologically-important nearshore habitats across 1,000 km of the California Current Large Marine Ecosystem. In the most severe exposure hotspots, suboptimal conditions for calcifying organisms encompassed up to 56% of the summer season, and were accompanied by some of the lowest and most variable pH environments known for the surface ocean. Persistent refuge areas were also found, highlighting new opportunities for local adaptation to address the global challenge of OA in productive coastal systems.


Additional details on calculation of Ω arag-pH
For a given pH value, possible solutions to Ω arag can vary widely with S, T, and A T , and we explored the effects of uncertainties in these parameters. Salinity can influence calculated values of Ω arag (equation 1) through its effects on the solubility product constant (K sp ) (equation 2), [Ca ++ ] (equation 3) and carbon speciation (Dickson 2010). k 1 and k 2 , respectively, under those same T and S conditions. At pH = 8, DIC = 2200, this translates into a change in [CO 3 2-] of 2.3%. Collectively, the uncertainty associated with the use of a mean S is on the order of +/-0.04 units of Ω arag for every 1 unit deviation in S. Temperature influences Ω arag through its effects on K sp and carbon speciation. For every 3°C change (approximately 25% of the observed dynamic range), K sp shifts by 0.6%. Temperature has a much larger impact on Ω arag via its influence on k 1 and k 2 , which change by 7% and 12%, respectively, with a 3°C shift. This results in a change in Ω arag by as much as 15% depending on initial T. The importance of T in constraining Ω arag-pH can be seen in Fig S2 where solutions for Ω arag-pH narrow considerably if T is known.
Error in measurement of in-situ T is an additional source of uncertainty. We quantified drift in sensor T measurement by cross-calibrating a subset of sensors that were previously calibrated against factory-calibrated Seabird SBE-37 temperature and conductivity sensors. Over a 6 month period, cross-calibration at monthly intervals identified no directional drift between sensors with net between sensor fluctuations that ranged from 0.001 to 0.059°C. Error in T can affect our estimates of Ω arag-pH through effects of T on calculation of pH from sensor voltage and through the effects of T on pH and Ω arag . Observed range in sensor T error translates into an error in pH calculation from sensor voltage by a maximum of 0.00069 units, well outside the precision of CRM-based calibrations employed and sensor performance. We then applied the effects of sensor T error to our calculation of Ω arag . The net effects of the observed range in T error translate into a maximum of 0.5% error in Ω arag . Error in measurement of S in-situ represents another source of uncertainty. Salinity measurements varied in precision among the research groups, depending on whether samples are analyzed as bottle samples on high precision laboratory salinometers (Autosal Guideline Instruments) or by lower precision field sensors (Yellow Springs Instrument -YSI conductivity meters). In repeated cross-calibrations against factory-calibrated Seabird SBE-37 temperature and conductivity sensors and YSI sensors, a maximum error in measurement of S of 2.7% (as residuals from calibration line) was encountered. At S of 33, this translates into a maximum error of 0.9, a value that translates into a maximum error of +/-0.04 units of Ω arag . We note that this is a maximum error and because of non-linearities in the carbonate system, this maximal error is suppressed under more acidified conditions, declining to +/-0.009 units of Ω arag when Ω arag approaches minimum values for our system. Error associated with pH measurement in-situ represents another source of uncertainty. Because sensors were calibrated against CRM or CRM-based spectrophotometric measurements, we consider sensor drift between intervals of calibration to the key source of uncertainty. We used pre-and post-calibration values as a measure of the uncertainty associated with sensor drift. Across deployments, mean drift (! = -0.002, 95% C.I. = 0.017 pH units, ) did not differ significantly from zero. Mean absolute drift was 0.032, +/-0.037 s.d.. This translates into an error of up to +/-0.34 in estimate of Ω arag at the upper range limits of pH observed in our study. However, because of the nonlinearities in the carbonate system, the effects of instrument drift translates into error of +/-0.08 when Ω arag becomes corrosive, declining further to +/-0.04 at the lower range limits of exposure.
Uncertainty in A T is a remaining unknown in estimating Ω arag from pH. Our analysis suggests that accurate Ω arag-pH values require the collection of discrete bottle samples so that a mean A T can be determined. With an estimated mean A T , differences between Ω arag-pH and Ω arag will reflect deviations in the sample A T from mean A T . This error varies as a function of Ω arag with increasing variance at high Ω arag (Fig S5). Because the buffer factor for Ω arag with respect to changes in A T reaches a minimum as DIC approaches A T (Egleston et al. 2010), one expectation would be for differences between Ω arag-pH and Ω arag to increase at low values of Ω arag where DIC ≈ A T . At this point, [CO 3 2-] is very small and small changes in A T can result in large relative changes in [CO 3 2-]. Such changes are, however, small in the absolute sense, as a 20% change in [CO 3 2-] when DIC and AT =2200 is +/-9 µmol kg -1 . We estimate that at Ω arag between 2 and 4, error from uncertainty in A T is approximately 0.2 for every 100 µmol kg -1 of A T (Fig S5).
Between Ω arag values of 2 and 1, this error diminishes to 0.1 for every 100 µmol kg -1 of A T . Below Ω arag values of 1, this error is reduced even further to 0.05 for every 100 µmol kg -1 of A T . The realized deviations from Ω arag will increase considerably in systems such as estuaries where A T can vary widely. Because A T has a relatively narrow window of variability (mean = 2203, s.d. = ±69) in our study, we estimate maximum deviations in Ω arag-pH from Ω arag to be 0.15 in our system. While full determination of carbon system parameters provide data of highest resolution and are often critical for studies of ocean carbon inventories and fluxes, our analyses suggest that where full determination is not possible, Ω arag-pH can serve as a robust proxy for Ω arag .
Density-based estimates of anthropogenic DIC reflect the fundamental first order relationship between density, the time of last ventilation, and the corresponding levels of DIC ant at equilibration with the atmosphere. A ventilation age older than the mean yields an overestimate of the current effects of anthropogenic CO 2 , but translates into greater future declines. If ventilation age is underestimated, future declines in Ω arag will be less than projected but the effects of DIC ant would contribute to a larger portion of undersaturation events than the system faces currently. We note that because the increase in DIC ant is set by differences in equilibration at 280 and 400 ppm, the expected amplitude of change from pre-industrial conditions is unaffected by uncertainties in ventilation age. We further explored the effects of uncertainties in our estimate of DIC ant by examining the effects of subtracting DIC ant across a window of ± 3 standard deviations. For each site (Fig 5A-C, Table S1), the observed cumulative distributions of Ω arag sit outside the 3 S.D. window for our estimates of pre-industrial Ω arag .

Fig S1
. Ω arag for discrete surf-zone samples calculated from 2 measured carbon parameters vs. Ω arag-pH modeled using pH and an assumed A T of 2200 mmol kg -1 .    Error in performance of Ω arag-pH (as absolute difference from Ω arag ) in relation to uncertainty in A T (as the absolute difference in assigned mean A T and measured A T ) for the NACP (A) and GLODAP (B) datasets used in Fig S5 Color scale denotes Ω arag and illustrates the reduced error in Ω arag-pH when Ω arag is low.

Stability of results with respect to choice of summary statistics on pH and Ω arag exposure serverity.
In Fig 1, we used the lower 5 th percentile of pH values as a metric of exposure severity. To evaluate the stability of the reported spatial pattern to percentile choice, we've plotted the latitudinal pattern of exposure as indexed by other percentiles at 5% increments for the lower quartile values (FigS6). The latitude vs. percentile curves are all highly congruent with each other, suggesting that our finding of spatial patterning in exposure is a robust feature of the system. The stability of our results also holds for Ω arag (Fig S7). In Fig 4, we've plotted shipbased offshore near-bottom pH vs. lower 5 th percentile pH from intertidal stations. To evaluate the effects of using a lower 5 th percentile over other percentile thresholds, we performed the same regression for other percentiles at 5% increment up to the 95 th percentile. The R 2 of regressions reach a maximum (0.87) at the 35 th percentile (Fig S8a) and suggests that Fig 4 represents a conservative test of the strength of the relationship between offshore and intertidal pH exposure. R 2 values decline after the 35 th percentile and regression slopes are not significant after the 55 th percentile (i.e. for higher pH encountered in the intertidal). This is expected as shipbased near-bottom measures represent the local minima in source pH and not a mean value, and because low intertidal pH values arise from upwelling events that bring cold, low pH water from depth over the shelf to the shore (Fig 3). The connection between bottom shelf waters and the shore is further evident in a cross-shelf section (Fig S9) from the NOAA West Coast Ocean Acidification WCOA2011 34 from 44.20°N that is combined with intertidal measurement from the SH site (44.25°N). The cross-shelf section took place during a period (Aug 18 th 2011) of upwelling favorable winds that resulted in the shoreward bottom flows and shoaling of density, pH, and Ω arag layers (Fig S9a). We do note that while the relationship is quite strong (max R 2 of 0.87, p<0.005), additional observations will be important for fully evaluating the persistence of the offshore shelf patterns in bottom pH, and to further improve the predictive precision and geographic generality of cross-shelf coupling in low pH exposure.
Whereas upwelling circulation connects offshore waters from depth to the shore to influence the severity of coastal low pH exposure (Fig S8a), we also explored the strength of shelf to shore coupling in describing exposure to high pH values. During wind relaxation or downwellingfavorable event, offshore surface waters flow shoreward bringing warm and relatively high pH values to the coast. Surface pH values from the NOAA shelf stations were significantly predictive of how the upper range of pH values (80 th to 95 th percentile) varied across the 7 stations of the 2011 intertidal observing network (Fig S8b). We note however that the strength of the relationship between offshore bottom measurements and low pH exposure, and that for offshore surface measurement and high pH exposure do not appear to be asymmetrical. Offshore surface values significantly account for a narrower range (80 th to 95 th percentile vs 5 th to 55 th percentile), lower portion of the variation (max R 2 of 0.68 vs. 0.87), and lower statistical significance (min p-value of 0.02 vs. 0.002 for test of slope) in upper pH exposure relative to the performance of deep values in accounting for lower pH exposure (Fig 8a,b). The lower strength of surface to shore coupling point to additional factors such as nearshore productivity, and heterogeneity in offshore source water properties that can modulate the cross-shelf pH connectivity. Fig S6. Consistency in the spatial pattern of intertidal pH across different metrics of low pH exposure from lower 5 th to 25 th percentile. Fig S7. Consistency in the spatial pattern of intertidal Ω arag-pH across different metrics of low Ω arag-pH exposure from lower 5 th to 25 th percentile. Fig S8. (a) Effects of percentile choice on the coefficient of determination (R 2 ) value for regressions between intertidal and offshore (ship-based) pH measurements from depth. (b) same as (a) but with offshore surface value as independent variable in regressions. Color-filled symbols reflect significance test of slopes for p-values of <.05 (yellow), <.01 (orange), <.005 (red). Open symbols are not significant.