Variability and trends of ocean acidification in the Southern California Current System: A time series from Santa Monica Bay

Authors

  • A. Leinweber,

    Corresponding author
    1. Institute of Geophysics and Planetary Physics and Department of Atmospheric and Oceanic Sciences, University of California, Los Angeles, California, USA
    • Corresponding author: A. Leinweber, Institute of Geophysics and Planetary Physics, University of California, Los Angeles, CA 90095-1565, USA. (leinweber@igpp.ucla.edu)

    Search for more papers by this author
  • N. Gruber

    1. Environmental Physics, Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland
    Search for more papers by this author

Abstract

[1] We investigate the temporal variability and trends of pH and of the aragonite saturation state, Ωarag, in the southern California Current System on the basis of a 6 year time series from Santa Monica Bay, using biweekly observations of dissolved inorganic carbon and combined calculated and measured alkalinity. Median values of pH and Ωarag in the upper 20 m are comparable to observations from the subtropical gyres, but the temporal variability is at least a factor of 5 larger, primarily driven by short-term upwelling events and mesoscale processes. Ωarag and pH decrease rapidly with depth, such that the saturation horizon is reached already at 130 m, on average, but it occasionally shoals to as low as 30 m. No statistically significant linear trends emerge in the upper 100 m, but Ωarag and pH decrease, on average, at rates of −0.009±0.006 yr−1 and −0.004±0.003 yr−1 in the 100–250 m depth range. These are somewhat larger, but not statistically different from the expected trends based on the recent increase in atmospheric CO2. About half of the variability in the deseasonalized data can be explained by the El Niño Southern Oscillation, with warm phases (El Niño) being associated with above normal pH and Ωarag. The observed variability and trend in Ωarag and pH is well captured by a multiple linear regression model on the basis of a small number of readily observable independent variables. This permits the estimation of these variables for related sites in the region.

1. Introduction

[2] The absorption of anthropogenic CO2 by the oceans leads to a reduction in oceanic pH, an increase in aqueous CO2 [CO2(aq)], and a decrease in the concentration of the carbonate ion math formula [Feely et al., 2008]. The latter also causes a decrease in the saturation state of seawater with regard to mineral phases of CaCO3, commonly computed as the ratio of the in situ carbonate ion concentration over the carbonate ion concentration at saturation, that is, Ω. With atmospheric CO2 bound to continue its increase, these chemical changes will likely continue well into the foreseeable future [Orr et al., 2005; Steinacher et al., 2009; Gruber et al., 2012], with potential major consequences for marine life [Doney et al., 2009] and ocean biogeochemical cycling [Gehlen et al., 2011]. The progression of these chemical changes, further referred to as “ocean acidification,” will occur against a “natural” background variability whose magnitude is not well quantified [Friedrich et al., 2012; Hauri et al., 2013]. This is a critical gap since one can expect that the ability of marine organisms and ecosystems to adapt to progressing ocean acidification will depend on whether they have evolved in a chemically stable or highly variable environment [Hauri et al., 2013].

[3] In the open ocean, surface pH is remarkably uniform in space and time [Feely et al., 2009]. As a result, the ocean acidification-induced reduction in pH is easily detectable at the long-term time series sites in the subtropical gyres with only a few years of observations [Dore et al., 2009; Bates et al., 2012; González-Dávila et al., 2010]. The saturation state with regard to aragonite, Ωarag, varies substantially more in the open ocean. But still, clear long-term secular trends in Ωarag were detected at all subtropical open ocean time series sites [Bates et al., 2012; González-Dávila et al., 2010; Doney et al., 2009]. Given these small “natural” variations and the anticipated progression of ocean acidification, open ocean waters will very soon experience conditions that are far outside this variability range [Friedrich et al., 2012]. This will be much less the case in coastal settings, since the natural background variability there is much higher [Borges et al., 2010]. This is particularly the case for Eastern Boundary Upwelling Systems, such as the California Current System. For example, Wootton et al. [2008] reported seasonal variations in pH of more than one unit from a very near-coastal site off the coast of Washington State in the United States (Tatoosh Island). This seasonal amplitude is an order of magnitude larger than that observed in the subtropical gyres. Frieder et al. [2012] measured diurnal variations in a near-shore kelp forest ecosystem in the southern California Bight off San Diego that are as large as 0.36 units. Also further offshore, the coastal waters are characterized by high variability with overall ranges of more than 0.4 pH units [Alin et al., 2012], that is, a factor of 5 higher than those seen in the open ocean.

[4] The saturation state for aragonite, Ωarag, also shows stronger variations in coastal settings compared to the open ocean. In May and June 2007, Feely et al. [2008] observed Ωarag in surface waters along the U.S. West Coast to range between more than 3 and less than 1, that is, they encountered occasional undersaturation in the surface ocean. The depth of the saturation horizon was found to shoal substantially toward the coast, with typical depths at this time of the year ranging between 40 and 120 m, in contrast to the open ocean Pacific, where this horizon is more than 400 m deep [Feely et al., 2004]. Using empirical relationships, Alin et al. [2012] extended upon this work by mapping out the distribution of Ωarag across the entire Southern California Current System. They reported ranges at 20 m between 0.9 and 3.0, and between 0.7 and 2.4 at 125 m depth. With ocean acidification very likely to progress unabatedly, it is critical to develop a better understanding of the distribution and variability of ocean acidification parameters in coastal settings in order to establish a baseline against which future changes can be assessed [Gruber et al., 2012; Hauri et al., 2013]. To this end, we investigate the variations of Ωarag and pH in the upper 300 m of the water column in Santa Monica Bay. Our site is located in the central Southern California Bight, which is part of the California Current System. Although our observations are from a single location, we believe that many of the conclusions apply to related coastal settings as well, especially those located within Eastern Boundary Upwelling regions.

2. Data

[5] The DIC and Alk concentrations were measured at usually 12 m depths in the upper 300 m at the Santa Monica Bay Observatory (SMBO) mooring (33°55.9 N, 118°42.9 W), which is located at a water depth of 450 m. The data reported here were collected on a total of 137 cruises that took place on a biweekly schedule from January 2003 to December 2008. Samples were drawn from Niskin samplers into individually numbered, clean 0.3 dm3 size Pyrex glass reagent bottles, using established gas sampling protocols [Dickson et al., 2007]. DIC followed by Alk samples have been taken first on routine CTD casts, and the time of sampling was usually between 9:30 and 11:00 local time. In general, DIC (1644 samples) was determined immediately after sampling in the laboratory at UCLA using the coulometric SOMMA system described by Johnson et al. [1993]. CO2 gas and DIC reference samples (Certified Reference Material (CRM), produced by A. Dickson, Scripps Institution of Oceanography) were used to calibrate the measurements for each cell individually. The analytical precision of the DIC measurements was determined to be better than ±0.8 μmol kg−1 (based on 48 replicate measurements) and 145 comparisons with the CRMs indicate an accuracy of better than 2.1 μmol kg−1. The analyses of 268 duplicate samples from the time series station show a combined sampling and analysis error of ±1.5 μmol kg−1. Total Alk (696 samples and 58 stations) was determined by open-cell potentiometric titration method [Dickson et al., 2003]. Measurements of 74 CRMs indicate an accuracy of better than 2.5 μmol kg−1. The analyses of 139 duplicate Alk measurements were determined to be better than 2.6 μmol kg−1. Whenever processing could not be completed within 36 h, samples were poisoned using 50 μmol of saturated HgCl2 solution to prevent biological alteration. Nutrient samples were drawn after DIC and Alk, then immediately frozen and analyzed colorimetrically by autoanalysis (Technicon AAII, detection limit 0.1 μmol l−1 [Grasshoff et al., 1999]) at the nutrient laboratory of Scripps Institution of Oceanography.

[6] The parameters pH (on total scale) and Ωarag were computed from the measured DIC, Alk, temperature, salinity, and nutrients employing CO2 system [Lewis and Wallace, 1998] and adopting the CO2 system coefficients of Mehrbach et al. [1973] as refitted by Dickson and Millero [1987]. Since alkalinity measurements started only in December 2005, we extended our alkalinity time series backward in time by using a relationship between Alk and salinity, S, established on the basis of our measurements from December 2005 onward. To find the best fit, we followed Lee et al. [2006] and tested different functions that included salinity, temperature, and a combination of both as input parameters. Our results indicate that a quadratic polynomial fit using solely salinity as input parameter yields the lowest uncertainty. Specifically, we determined Alk for each sample from January 2003 to November 2005 by using:

display math(1)

for the mixed layer (the depth at which the temperature change from the surface is 0.5°C [Levitus, 1982]) (n = 119), with a 1-σ uncertainty of ±6.44 μmol kg−1 (1 σ) and:

display math(2)

for the depth from the bottom of the mixed layer down to about 300 m of the water column (n = 502), with a 1-σ uncertainty of 7.28 μmol kg−1.

[7] A monthly climatology (at daily interpolated resolution) was produced from the original data by first interpolating them in depth and then binning them in time. To this end, we employed a grid of 12 months by 12 vertical levels (1, 10, 20, 30, 40, 50, 75, 100, 125, 150, 200, and 250 m). We then computed seasonal anomalies by subtracting from the original data the climatological value at this day of the year and the corresponding interpolated depth.

[8] To investigate possible trends at our time series station, we added a straight line fit through the seasonal anomaly data at each standard level using a least-squares regression method. We determined the uncertainty of the trends by bootstrapping [Varian, 2005]. In particular, we created 1000 time series for each depth level by selecting 1000 random subsets of the data and redetermined the trend. The uncertainty of the trend was then computed from the resulting probability density function of the slopes, using the 5th and 95th percentiles. They correspond roughly to the bounds set by the ±2 standard deviations of the trend.

3. Results

[9] Our observations at the SMBO mooring site in the Santa Monica Bay reveal well-established trends with depth for both Ωarag (Figure 1) and pH (Figure 2), masked by a high level of short term to seasonal variability. The variability is maximal in the upper ocean and tends to decrease with depth. Given the non-normal distribution of the data, we analyze them in terms of their median, their interquartile range (IQR), that is, the range between the 25th and 75th percentile, and in terms of their 5th to 95th percentile range.

Figure 1.

Aragonite saturation state, Ωarag, of the upper 250 m of the water column from January 2003 to December 2008 at the Santa Monica Bay Observatory (SMBO) Mooring (33° 56′ N, 118° 43′ W). (a) Time series of average Ωarag over the top 20 m. The red line was calculated using a 45 day running mean. (b) Median profile and quartiles (25–75%, and 5–95%). Red dots are the estimated mean values for the entire Southern California Bight by Alin et al. [2012]. (c) Time-depth plot over the 6 years. The dots indicate the location of the measurements, the white line depicts the depth of the mixed layer, and the black line indicates the saturation horizon for aragonite, that is Ωarag = 1. Two lighter bars in 2008 indicate periods of no data.

Figure 2.

As Figure 1, but for pH. In Figure 2c, the black line indicates a pH of 7.8.

[10] The median Ωarag over the upper 20 m is 2.67, that is, slightly below the mean reported for the global surface ocean of ∼3 (Figures 1a and 1b). The median Ωarag decreases rapidly with depth, reaching the saturation horizon, that is, where Ωarag  = 1, at about 130 m depth (Figure 1b). At all depths, Ωarag varies substantially in time, with an IQR of 0.5 in the upper 20 m, and a 5th to 95th percentile quantile range of 1.2 over the same depth interval. These variability ranges increase slightly until about 30 m depth and then gradually decrease to about 0.1 for the IQR and 0.2 for the 5th to 95th percentile range in the depth range below 150 m. Our median profile is overall similar to the estimated mean profile of Ωarag over the entire southern California Current System (Alin et al., 2012). We disregard here the small difference between median and mean, as the latter data are nearly normally distributed, making mean and median nearly identical. Relative to the data from the entire Southern California Current System, the profile from Santa Monica Bay decreases somewhat faster with depth (Figure 1b), reflecting the overall shoaling of low Ωarag waters toward the coast.

[11] The time series plot (Figure 1a) for the upper 20 m reveals an irregular pattern without a clear seasonal cycle. Periods of relatively constant Ωarag are followed by periods with rapid swings, often initiated by a decrease and followed by a rapid increase of sometimes more than 1 unit. Deeper in the water column, and particularly between 50 and 150 m, a clearer seasonal pattern emerges with lower Ωarag values in spring/early summer (spring: March-May, summer: June-September), and higher values in the fall/winter (fall: October-November, winter: December-February), that is, a seasonal cycle of heaving and lowering of the isolines (Figures 1c and 3c). This is perhaps best visible in the depth of the saturation horizon, which is generally less than 100 m deep in spring/early summer, and deepens to about 200 m in the fall/winter (Figure 1c). Superimposed on this seasonal cycle are periods of rapid shoaling and deepening of the isolines, as illustrated by vacillations of the saturation horizon, which can change by more than 100 m within 2 weeks. In 2003, and in 2005–2007, the saturation horizon shoaled to less than 50 m, reaching 30 m on 31 May 2007, that is, to a depth close to or within the euphotic zone observed in Santa Monica Bay [Leinweber et al., 2009].

Figure 3.

Climatological mean seasonal cycle of (a, c) Ωarag, (b, d) pH, (e) temperature, and (f) salinity at the SMBO site. Figures 3a and 3b depict the mean seasonal cycle for the upper 20 m of the water column, while Figure 3c–3f depict the climatology of the interior ocean.

[12] The median pH over the top 20 m is 8.08 with an IQR of 0.09 and a 5th to 95th percentile range of 0.22 (Figures 2a and 2b). As is the case for Ωarag, this median value is slightly below the mean pH reported for the global surface ocean. The median pH decreases rapidly with depth, plunging below 7.8 at around 75 m and below 7.7 at about 180 m (Figure 2b). This median profile is again comparable to the mean estimated profile for the entire southern California Current System [Alin et al., 2012] and also with the tendency of a steeper gradient with depth as a result of the onshore shoaling of low pH waters. The pH in the upper 20 m of the water column fluctuates strongly without a clear seasonal signal, but with many intermittent short-term events (Figures 2a and 3b). As was the case for Ωarag, surface pH often show swings that begin with a drop of about 0.1 units or more, followed by a rapid increase of up to 0.4 units. In fact, surface pH and Ωarag are quite strongly positively correlated in the upper 20 m (r = 0.92) (Figure 4a). Deeper in the water column, a more pronounced seasonal cycle appears with spring/early summer periods characterized by lower pH and the fall/winter period characterized by higher pH (Figure 3d). Here pH and Ωarag are even more strongly correlated (r = 0.98) (Figure 4a).

Figure 4.

Relationship between pH, Ωarag, DIC, and Alk in the upper 250 m at the SMBO site for the data interpolated to standard depths (1, 10, 20, 30, 40, 50, 75, 100, 125, 150, 200, and 250m). (a) pH versus Ωarag, (b) Ωarag versus DIC, (c) Ωarag versus Alk, and (d) pH versus DIC. The colors indicate the depth range.

4. Discussion

[13] A number of questions emerge from the data: What are the causes for the observed variations? What is the contribution of the seasonal versus the nonseasonal variability to the total variability? Do the data show any trends? We attempt to provide answers to these questions next.

[14] The key driver of the variability in pH and Ωarag is the variability of DIC, while the contribution of temperature, Alk, and salinity are small to negligible. This conclusion is based on the tight correlation of DIC with pH and Ωarag, with DIC explaining more than 95% of the total variance in pH and Ωarag (Figures 4b and 4d). Thus, to first order, the question of what drives the variations in Ωarag and pH is answered by explaining the drivers for DIC. We do this separately for the seasonal and the nonseasonal components of the variations.

4.1. Seasonal Variability

[15] The climatological seasonal cycle for Ωarag and pH emerge clearly from the data, despite the fact that the seasonal cycle is responsible for only about 20% of the total temporal variance in the upper ocean and for about 40% over all depths (Figure 3).

[16] In the upper 50 m (with the exception of the surface data), the seasonal minimum is reached in April/May, that is, during the peak of the upwelling season in Santa Monica Bay. This upwelling shoals the isolines of all quantities, thereby bringing cold, DIC-rich, and hence low pH and low Ωarag waters closer to the surface (Figure 3). As the season progresses, the waters in Santa Monica Bay stratify, and the excess of photosynthesis over respiration, that is, net community production, causes a net drawdown of CO2, which causes DIC to decrease and hence pH and Ωarag to increase. Biology is likely also responsible for the absence of the seasonal upwelling induced minimum of pH/Ωarag in the top few meters of the water column. With upwelling occurring episodically in Santa Monica Bay, and with the resulting phytoplankton bloom causing rapid increases in pH and Ωarag after the minima, the resulting zigzag (see also Figures 1a and 2a and discussion below) is averaged out when computing the long-term seasonal average in the upper ocean. This does not occur further down in the water column, as these blooms are strongly skewed toward the surface, explaining the dominance of the upwelling signal in the upper thermocline.

[17] From 50 m downward, the seasonal cycle is largely determined by changes in the water masses present at this site. The seasonal minima tend to be shifted by a month later relative to the upper 50 m, that is, to June, associated with a minimum in temperature and maximum in salinity (Figures 3e and 3f) and the associated minima and maxima in DIC and alkalinity (data not shown). The dominant contribution of the water mass changes to the seasonal cycle in the layers below 50 m also likely explains the lack of a remineralization signal from the spring to the fall. From the export of the organic matter produced in the upper 0 to ∼20 m and its subsequent remineralization, one would expect a spring to fall increase in DIC and hence a decrease in Ωarag and pH in the 50 to 200 m depth range. Instead, Ωarag and pH mostly increase from spring through the fall in this depth range, consistent with the warming and freshening of the waters (Figures 3e and 3f) and the associated changes in DIC and Alk. Thus, at these depths, the seasonal changes in the water masses appear to dwarf the seasonality of the locally generated biological signal.

4.2. Nonseasonal Variability

[18] The nonseasonal variability of Ωarag and pH is substantial and tends to be highly correlated with the corresponding T and S anomalies (Table 1). The largest contribution to the nonseasonal variability stems from short-term intermittent events that are driven by a combination of upwelling and relaxation events and the passage of eddies and other mesoscale structures (Figures 5b and 5c). The latter tend to produce vertically more coherent anomalies compared to upwelling events, which produce more variability at the surface. This is because the new injection of nutrients will cause a phytoplankton bloom in the near surface waters that will lead to a zigzag pattern of initially low values reflecting the upwelling, followed by a steep increase in response to the strong net CO2 uptake by phytoplankton. The clearest such signal can be seen in May 2007, but can be found in most other years as well during the spring season, when intermittent upwelling occurs. These upwelling/bloom-induced zigzags in pH and Ωarag tend to decouple the seasonal anomalies of these two parameters from those of T and S, such that the correlations are smallest in the top 20 m. In contrast, the vertically more coherent short-term anomalies associated with eddies tend to be highly correlated with the T and S anomalies.

Table 1. Correlation Coefficients of Deseasonalized pH, Ωarag, Temperature (T), and Salinity (S) Anomalies at the SMBO Site and the Oceanic Niño Index (ONI) Over the Period From 2003 to 2008a
 pH AnomalyΩarag AnomalyT AnomalyS AnomalyONI
  1. a

    Correlation coefficients not statistically significant are marked in italic.

Averaged Over the Upper 20 m of the Water Column
pH Anomaly10.950.44−0.260.13
Ωarag Anomaly0.9510.68−0.290.18
T Anomaly0.440.681−0.260.24
S Anomaly−0.26−0.29−0.261−0.35
ONI0.130.180.24−0.351
Averaged Over 30–40 m of the Water Column
pH Anomaly10.960.69−0.530.35
Ωarag Anomaly0.9610.83−0.540.37
T Anomaly0.690.831−0.460.30
S Anomaly−0.53−0.54−0.461−0.48
ONI0.350.370.30−0.481
Averaged Over 50–125 m of the Water Column
pH Anomaly10.980.56−0.740.48
Ωarag Anomaly0.9810.68−0.760.49
T Anomaly0.560.681−0.540.33
S Anomaly−0.74−0.76−0.541−0.53
ONI0.480.490.33−0.531
Figure 5.

Time series of (a) Ocean Niño Index (ONI), and (b) deseasonalized Ωarag, (c) pH, (d) temperature, and (e) salinity. The data were deseasonalized by removing from the original measurements from the corresponding seasonal climatological value (shown in Figure 3).

[19] On interannual time scales, the periods from fall 2004 until spring 2005, from fall 2005 until spring 2006, and from fall 2007 until spring 2008 stand out due to their persistent anomalies, that is, positive anomalies in Ωarag, and pH in 2004/2005 and negative ones in 2005/2006 and 2007/2008 (Figure 5). These anomalies are well correlated with anomalies in temperature and salinity (Figures 5d and 5e), suggesting that they are the result of larger-scale changes in the circulation of the Southern California Bight, possibly in association with El Niño and La Niña events (Figure 5a). In fact, with the exception of the upper 20 m, where we find no significant correlation with the Oceanic Niño Index (ONI) (http://www.cpc.ncep.noaa.gov/), pH and Ωarag in the water column at the SMBO site reveal significant positive relationships with ONI. The strongest positive correlation between ONI, and the anomalies in pH and Ωarag are found between 50 and 125 m, where ONI explains nearly 50% of the variance (R2 = 0.23 for pH and R2 = 0.24 for Ωarag). Most of this correlation comes from the 2005/2006 and the 2007/2008 low pH and Ωarag events that are clearly related to the La Niña conditions prevailing during these winters, which bring colder and saltier waters to the Southern California Bight. In contrast, the warm/high pH/Ωarag anomalies over the winter 2004/2005 are not associated with a high ONI, requiring another explanation. Alin et al. (2012) reported a similar positive correlations among ONI, pH, and Ωarag at a site further south in the Southern California Bight, although their time series was not based on direct observations, but actually derived by extrapolation. They thereby actually assumed that one can use relationships derived from the spatial co-occurrence of parameters to estimate the temporal co-occurrence. Although this is a very far-reaching assumption, our results would actually lend support to this approach.

4.3. Linear Trends

[20] Over the span of the 6 years reported here, no significant trend can be detected in pH or Ωarag over the top 100 m (Figure 6). This is largely due to the high nonseasonal variability that masks any potential underlying trend. In contrast, below 100 m, statistically significant trends emerge for both quantities. Median trends for Ωarag range between −0.005 and −0.012 yr−1, with an 100 to 250 m average of −0.009 ± 0.006 yr−1, and median trends for pH range between −0.003 and −0.005 yr−1, with an average of −0.004 ± 0.003 yr−1. In order to determine the sensitivity of the trends to the deseasonalization, we computed the trends also for the original, but vertically interpolated data. This gave the expected larger uncertainties, but the same pattern of trends with depth (Figure 6). These trends in Santa Monica Bay are larger in magnitude than most of those reported so far (e.g., ∼ −0.003 pH units yr−1 in Monterey Bay [Borges et al., 2010]). They are also larger than those expected on the basis of the recent trend in atmospheric CO2, but the differences are not statistically significant.

Figure 6.

Linear trends in (a) Ωarag and (b) pH at the SMBO time series site. The trends were determined from either the deseasonalized data (darker color) or the original data (lighter color). The median trends and the 5th to 95th percentiles of the probability density function of the trends computed by a Monte Carlo method with bootstrapping are shown.

5. Empirical Relationships for Estimating pH and Ωarag

[21] Recent studies have presented robust empirical relationships for estimating the carbonate system based on proxy data such as temperature, salinity, and oxygen along the U.S. West Coast [Juranek et al., 2009; Alin et al., 2012]. Although there are several shortcomings as discussed by Alin et al. [2012], these empirical equations can be used to estimate the carbon system over larger areas and can reproduce seasonal to interannual cycles, particularly for conditions for which the relevant carbonate system parameters are missing. Here we employ a similar multiple linear regression approach to estimate pH and Ωarag at our time series site, based on a variety of input parameters and first-, second-, and third-order combinations to find the best models to fit the data. As possible input parameters, we chose temperature (T), salinity (S), nitrate (NO3), and phosphate (PO4). We then use the root mean squared error (RMSE), the R2, as well as the Akaike information criterion [Burnham and Anderson, 2002] in order to assess the goodness of the fit and to reject or select a particular model.

[22] Our results indicate that the addition of the third-order polynomial term did result in only a marginally better goodness of the fit, while increasing the degrees of freedom of the model substantially. We therefore rejected all third-order models and selected only second-order polynomial models. The overall “best” models, that is, those with the highest explained variance while having the lowest degrees of freedom, were second-order polynomial models with T and PO4 as independent parameters (Figures 7 and 8). These models achieve excellent fits to the data with an R2 = 0.95 and RMSE of 0.17 for Ωarag and an R2 = 0.92 and RMSE of 0.049 for pH (Table 2). Inclusion of more terms, such as S, would have resulted in slightly better fits, but at the cost of increasing the likelihood of overfitting. The residual structure of these models is random over nearly the entire ranges, but show substantial biases at their upper ends (Figures 7 and 8). These high residuals occur near the surface and are associated with low DIC conditions occurring after upwelling events, that is, the result of a strong biological drawdown of DIC. This appears to occur in the absence of a concomitant drawdown of phosphate, so that our largely thermocline-based relationship with phosphate is unable to capture such events. Given these large biases at the upper end, we restrict the valid range of our relationship for Ωarag to 0.6–3.0, and for pH to 7.55–8.10. We also note that the RMSE is substantially higher at the upper end of the range, while the RMSE at the lower end is actually very good. These strengths and weaknesses should be kept in mind when applying the proposed models.

Figure 7.

Analysis of multiple linear regression fit of Ωarag. (a) Predicted versus measured Ωarag. (b) Residual (predicted minus measured) Ωarag versus measured Ωarag. (c) Residual Ωarag versus depth. (d) Residual Ωarag versus time. The results from the second-degree polynomial model with temperature and phosphate as independent parameters are shown.

Figure 8.

As Figure 7, but for pH.

Table 2. Regression Models in the Santa Monica Bay Using Temperature (T), salinity (S), NO3, and PO4 as Input Parametera
Regression ModelR2RMSE
  1. a

    The models with T and PO4 are considered as the “best” models. Note also the validity range of the regression models.

For 0.6 < Ωarag< 3.0, All Depths
Ωarag (T, PO4) = 0.06 + 0.29T – 0.0067T2 – 1.22PO4 + 0.29PO420.950.17
Ωarag (T) = −3.21 + 0.54T – 0.012T20.910.23
Ωarag (T, S) = 760.70 + 0.54T − 0.011T2 − 45.26S + 0.67S20.920.21
Ωarag (T, S, NO3) = 63.76 + 0.18T − 0.0023T2 − 3.52S + 0.049S2 – 0.068NO3 + 0.0015NO320.940.18
For 7.55 < pH < 8.10, All Depths
pH(T, PO4) = 7.14 + 0.13T – 0.0041T2 − 0.24PO4 + 0.044PO420.920.049
pH(T) = 6.18 + 0.22T – 0.0062T20.890.058
pH(T, S) = 97.28 + 0.16T – 0.0043T2 – 5.27S + 0.077S20.900.055
pH(T, S, NO3) = −62.64 + 0.064T – 0.0018T2 + 4.30S − 0.066S2 − 0.017NO3 + 0.00033NO320.930.047

[23] In addition to the T- and PO4-based regression models, we also provide equations for several other models with other combinations of the independent variables, permitting researchers to estimate pH and Ωarag in Santa Monica Bay and possibly beyond for situations when phosphate measurements are not available (Table 2). Most of the alternative models are less accurate and precise than our “best” models, but still reasonably good for many practical applications.

6. Summary and Conclusions

[24] The temporal variations of pH and aragonite saturation in the upper 300 m of the water column at SMBO are at least a factor of 5 more pronounced than those seen at the subtropical time series sites [Dore et al., 2009; Bates et al., 2012; González-Dávila et al., 2010]. This high variability is largely driven by nonseasonal variations, which are caused by intermittent upwelling and relaxation events and mesoscale processes, but also includes interannual variability associated in part with El Niño Southern Oscillation. It is reasonable to assume that the marine ecosystems in Santa Monica Bay and the California Current System in general have adapted to such a high degree of variability in the relevant ocean acidification parameters. Whether this will make them more resilient to the already reported changes in the carbonate chemistry and those anticipated in the coming decades is not at all clear. Furthermore, despite this high level of variability, the California Current System is projected to enter territory outside the present variability envelope within the next few decades [Hauri et al., 2013], with potentially large stretches of the nearshore upper ocean becoming undersaturated over the entire summer [Gruber et al., 2012]. The continuation and expansion of coastal ocean acidification monitoring activities therefore seems prudent [Borges et al., 2010].

Acknowledgments

[25] This research was supported by grants from the University of California Marine Council, U.S.NSF, and ETH Zurich. We are deeply indebted to all of the people who have helped through the years with sampling and the analyses: Heather Coleman, Andrew McDonnell, Rebecca Rooke, Carmen Hill-Lindsay, Jeffrey Mendez, Jaynel Santos, Jannelle Doi, Richard Carlos, Sarah Diringer, and Tabitha Esther. Special thanks to Andrew Dickson, who supported us in the setup of the alkalinity system, and to the captains and crews of the more than 130 cruises we undertook to create this time series. We thank Wiley Evans and a second reviewer for their helpful and supportive comments.

Ancillary