Ocean acidification will progress in an episodic manner, yet this has rarely been considered. Here, we investigate how the intensity, duration, and severity of episodic low aragonite saturation state events along the California continental shelf have changed since ~1750, and how they might change up to 2050 under the A2 scenario from the Special Report on Emissions Scenarios by the Intergovernmental Panel on Climate Change. Our model-based results suggest that between 1750 and 2010, aragonite undersaturation events along the shelf have quadrupled in their number and lengthened in duration, but that even larger changes are bound to occur within the next 20 to 40 years. Undersaturation will become very likely the norm near the seafloor by 2030, and if atmospheric CO2 increases beyond ~500 ppm, this layer will become permanently undersaturated. Combined with a fourfold increase in intensity, the resulting increase in severity of low aragonite saturation state events will substantially affect the viability of calcifying organisms and will alter ecosystem structure.
 As ocean acidification (OA) progresses, the advancement toward low aragonite saturation state (LASS) conditions does not occur gradually in most parts of the ocean, as the long-term trend induced by the uptake of anthropogenic CO2 occurs against a background of naturally occurring variability [e.g., Friedrich et al., 2012; Hauri et al., 2013]. While some calcifying organisms may acclimate to OA [Form and Riebesell, 2012], many are deleteriously affected when the saturation state of aragonite (Ωarag) drops below a species-specific threshold. The episodic nature of LASS events is highly relevant for the potential responses of marine organisms and ecosystems [Waldbusser et al., 2013]. While some organisms, such as corals and brittle stars, may have developed strategies to endure LASS events, allowing them to thrive when the conditions are more favorable [Fine and Tchernov, 2007; Wood et al., 2008] others may be vulnerable to such events. As the mean state of Ωarag decreases, LASS events will become more intense, severe, and longer. Therefore, strategies to endure LASS events may be less effective in the future, as the recovery times become too short or the rate of dissolution becomes too high, for example.
 The episodic nature of LASS events and their evolution under OA has not yet been considered in studies investigating the future progression of OA, as they all relied on monthly or even annual mean data [e.g., Steinacher et al., 2009; Gruber et al., 2012]. Here we take the episodic nature of LASS events explicitly into consideration and investigate the intensity, duration, and severity of LASS events in the waters off central and northern California using high-temporal resolution output from simulations of the U.S. West Coast setup of the Regional Oceanic Modeling System (ROMS) [Gruber et al., 2012; Hauri et al., 2013]. This region is of particular interest because sporadic and short events of aragonite undersaturation (Ωarag < 1) are a natural phenomenon [Hauri et al., 2009; Juranek et al., 2009] and have already increased due to OA [Feely et al., 2008; Juranek et al., 2009].
 Similar metrics to the ones we discuss here have been used to monitor and describe the progression of coral bleaching [Donner et al., 2007] and hypoxic events [Chan et al., 2008], and they are common in climate change risk research [Intergovernmental Panel on Climate Change (IPCC), 2012]. However, due to the lack of high-frequency observations of the inorganic carbon system, our understanding of the intensity, duration, and severity of LASS events and their impact on ecosystem processes is extremely limited [Juranek et al., 2009; Andersson et al., 2011]. Furthermore, it has not yet been studied how these measures of episodic chemical stress will change as OA advances, a gap we address here.
2 Model Setup and Study Area
 The simulations were conducted with a 5 km resolution U.S. West Coast setup of the three-dimensional Regional Oceanic Modeling System (ROMS) coupled to a simple nitrogen-based ecosystem/biogeochemistry model [Gruber et al., 2006]. Ωarag is calculated off-line from the state variables alkalinity, dissolved inorganic carbon (DIC), temperature, salinity, and nutrient concentrations, following the standard carbonate chemistry routines from the Ocean Carbon-Cycle Model Intercomparison Project (http://ocmip5.ipsl.jussieu.fr/OCMIP/). We use 5 years of 2 day averaged model output from a preindustrial run and three periods of 5 year output around 2010, 2030, and 2050 from a transient simulation [Gruber et al., 2012; Hauri et al., 2013]. For the transient simulation, the model was forced with increasing atmospheric CO2 (SRES-A2) [Nakizćenović and Swart, 2000] and altering lateral boundary conditions for DIC extracted from a coarse-resolution Earth System Model (National Center for Atmospheric Research Climate System Model 1.4-carbon model) [Frölicher et al., 2009]. As a result, the ocean's carbonate chemistry changes due to atmospheric CO2 and oceanic DIC increase, while other effects of climate change are not taken into account. Since OA in the upper ocean is mostly determined by the atmospheric CO2 concentration and is not inherent to a specific Special Report on Emissions Scenarios (SRES) scenario [Gruber et al., 2012], we describe our results in the context of the average atmospheric CO2 concentration over each 5 year period and the corresponding year under the SRES A2 scenario.
 We focus our analysis on the nearshore waters off California, from Point Conception (34.5°N, 120.3°W) to Cape Mendocino (40.4°N, 124.4°W). The modeled benthic habitat on the continental shelf is defined by the model layer just above the seafloor (maximum depth: 200 m). The analysis for the waters at 60 m depth and at the surface is constrained to within 10 km offshore. The model results are analyzed separately for each grid point, spanning 5 years in each chosen time period.
 A LASS event is defined as a consecutive sequence of days of duration (D, days) with Ωarag less than a specific threshold (T). Adopting the approach described in Sheffield and Wood , we define the intensity (I, unitless) of such an event as
where Ωmean is the mean Ωarag during the event. The intensity describes the mean magnitude of below-threshold conditions over the duration of the event. Because both the intensity and duration of LASS events have an integrated impact on organismal health [Beesley et al., 2008; Wood et al., 2008], we combine these two parameters into an index of severity (S):
with units of days.
 Here we analyze the intensity, duration, and severity of LASS events at the surface, 60 m depth, and bottom waters, with a particular focus on aragonite undersaturation events near the seafloor using the well-established thermodynamic threshold of T = 1. To account for the fact that many organisms show negative responses to LASS at Ωarag values other than the thermodynamic threshold, we also vary T between 0.6 and 1.4 to explore its effect on the intensity, duration, and severity of LASS events.
4 Intensity, Duration, and Severity of Low Aragonite Saturation State Events
 Our results suggest that aragonite undersaturation events (T = 1) in nearshore bottom waters off California have quadrupled in number since the preindustrial era. In 1750, 16% of the benthic area was not exposed to aragonite undersaturation at all, while elsewhere, aragonite undersaturation events occurred on average once per year (Table 1). Under present-day conditions, the entire benthos experiences aragonite undersaturation at some point during the year.
Table 1. Number of Low Aragonite Saturation State Eventsa
T = 0.6
T = 0.8
T = 1
T = 1.2
T = 1.4
aTotal number of low aragonite saturation state events per year and cell that go below a certain threshold T in nearshore bottom waters during the 5 year periods around 1750, 2010, 2030, and 2050.
 While their numbers increased, these aragonite undersaturation events have barely intensified between 1750 and now (median intensity of 0.02 in 1750 and 0.03 in 2010). This changes drastically in the near future, as their median intensities are projected to increase to 0.16 in 2030 = 452 ppm) and to more than 0.30 in 2050 = 533 ppm, Table 1 and Figure 1a). This rapid increase of the intensity of LASS events is not a consequence of an acceleration in the rate of change of the saturation state. It is a result of the fact that as the median saturation state approaches the threshold, the intensity of the LASS events increases nonlinearly, particularly once the lower end of the variability envelope, defined as the range given by ±1 temporal standard deviation around the median, crosses the threshold.
 The durations of the aragonite undersaturation events at the seafloor lengthen considerably between 1750 and present-day and are projected to increase in length to near permanency between 2030 and 2050 : 452–533 ppm, Figure 1b). In the preindustrial simulation, aragonite undersaturation events are short and sporadic. Eighty percent of all events are shorter than 14 days (Figure 1b). The median duration shifts from 6 days in 1750 to 16 days in 2010. The duration of aragonite undersaturation events in 2010 lasts from a few days (47%) to the entire year (3%) (Figure 1b, orange). Twenty-five percent of the events are longer than 6 weeks. By 2030 = 452 ppm), the majority (53%) of the aragonite undersaturation events last a year or longer (Figure 1b, blue). Nevertheless, about 30% of the events in 2030 are still shorter than 2 months and typically occur during the transition between periods of weak oversaturation and strong undersaturation. By 2050 = 533 ppm), the model projects year-round aragonite undersaturation for the benthic ecosystems off central and northern California (Figure 1b, red).
 The mean severity (the product of length and intensity) is projected to increase by 3 orders of magnitude until 2050 = 533 ppm, Figure 1c). This metric increases nearly fourfold from 0.14 days in 1750 to 0.51 days in 2010. In 1750, only 25% of the aragonite undersaturation events have a mean severity higher than 0.5 days, while in 2010, more than 50% are higher than 0.5 days. The relative frequency distribution of the severity modeled for 2030 shows one peak at around 50 days and a second peak at around 350 days (Figure 1c, blue). This illustrates that already by 2030 = 452 ppm), the severities of the majority of aragonite undersaturation events (60%) are higher than in 2010. The intensification and simultaneous lengthening of aragonite undersaturation events between 2030 and 2050 : 452 = 533 ppm) further amplify the severity of aragonite undersaturation events to about 550 days (Figure 1c, red). These temporal changes in severity are so substantial that as atmospheric CO2 rises from 390 ppm to 500 ppm and beyond, the distribution of severities are non-overlapping.
 The analysis using a wider range of thresholds suggests that the relative direction of the changes between 1750, the present, and the future remain the same as for the T = 1 case but that the actual numbers are very sensitive to the chosen threshold (Figures 1d–1f). For thresholds T > 1, the intensity, duration, and severity of events strongly increased between 1750 and 2010 (Figures 1d–1f and Table 1). The intensity of these events has more than doubled since 1750 (Figure 1d, black and blue). While the majority of events with T = 1.2 lasted less than 2 weeks in the preindustrial simulation, most of these events lasted 1 year or longer in 2010. By 2030, more than 25% of the events with a threshold as low as T = 0.8 are projected to last a third of the year or longer = 452 ppm, Figure 1e, red). However, due to the low intensities of these events (< 0.1), their severities are low (< 10 days) compared to the severities of events with a higher threshold (Figure 1f, black, blue, and orange).
 We also analyzed the intensity, duration, and severity of undersaturation events in nearshore waters at the surface and at 60 m depth (Table S1). The model indicates that these metrics are significantly less acute at these depths than they are in bottom waters (Table 1). This general pattern is a consequence of upwelling of deep CO2-rich waters onto the inner shelf, the remineralization of organic matter at the seafloor, as well as the biological carbon pump, which removes inorganic carbon from the surface and exports it to depth.
 The fast and large changes in aragonite undersaturation events predicted by the model are likely to have substantial implications for the benthic community of aragonite calcifiers sensitive to these conditions. Sensitive organisms will be forced to either develop tolerance to extreme conditions [Tunnicliffe et al., 2009] or migrate to regions with lower stress.
 The progression of these OA metrics is likely to have differential effects on various organisms depending on their life stages, energy reserves, and acclimation strategies. The intensity of aragonite undersaturation events influences the rate of dissolution and thus directly the viability of some organisms. Aragonite undersaturation intensities as projected by our model for 2050 may cause death to juvenile clams due to dissolution [Green et al., 2009]. Based on the projected change in undersaturation intensity from 0.02 to 0.3 between 2010 and 2050, and using the formula and constants from Acker et al. , we estimated that dissolution rates of pteropod shells would increase by a factor of more than 150. While some organisms are able to counteract dissolution in undersaturated conditions by up-regulating their calcification rates, this defensive response could impair other essential processes such as growth, reproduction, and protein synthesis [Wood et al., 2008; Andersson et al., 2011]. A longer exposure to or a shorter recovery time between aragonite undersaturation events may decrease the time available to recover from dissolution [Fine and Tchernov, 2007] or to restore their energy resources for growth and reproduction, making organisms more vulnerable to undersaturated waters during the subsequent event. Of particular concern are the developmental and reproductive life stages of some benthic species, as they tend to be more vulnerable to undersaturated conditions than their adult forms [Lee et al., 2006; Waldbusser et al., 2013].
 The main shortcomings of our model simulations are (i) that it overestimates the observed Ωarag [Hauri et al., 2013], particularly in the nearshore surface waters, and (ii) that it is forced with climatological atmospheric boundary conditions. The limited observations indicate a potential overestimation of the modeled Ωarag of 0.2 [Hauri et al., 2013], although this may serve as an upper boundary, as the observations stem from a single survey with very intense upwelling. Accounting for this overestimation of Ωarag, the mean intensity of aragonite undersaturation events would be higher than presented here. In contrast, the climatological forcing may not affect the mean Ωarag that much, but its variance, leading to little change in the intensity of the LASS, but a shortening of their duration, and hence a reduction in their severity. We cannot assess how the low and high biases will affect the final result in detail, but we expect that in reality, the intensity, duration, and severity may be changing even more than we have modeled.
 Because our simulation only accounts for rising CO2 and DIC concentrations, the effects of other global change-induced processes such as increased temperature, stronger winds, or ecosystem shifts are not accounted for. Initial results reveal a complex response, owing to potential compensatory effects between changes in wind, temperature, and ocean productivity [Lachkar and Gruber, 2012].
 The intensity, duration, and severity of LASS events are important new biogeochemical metrics that can be used as a quantitative tool to track the episodic exposure of organisms to OA. These metrics can be employed in the analysis of newly available high-frequency carbon chemistry observations [e.g., Hofmann et al., 2011]. While these metrics have not yet been used to describe how organisms and ecosystems will respond to OA, similar metrics have proven powerful tools to quantify thermal stress in corals [Donner et al., 2007] or risks associated with extreme events in climate change research [IPCC, 2012]. Hence, it is imperative that new experimental perturbation studies be designed within the context of the intensity, duration, and severity of LASS events in order to quantify the effect of these changes on biology.
 The results presented here demonstrate that although the organisms along the U.S. West Coast were already exposed to periods of undersaturation during the preindustrial era, the nature of LASS events has already changed and is projected to undergo significant further changes over the next 40 years. This process could force large shifts in the physiology of individual organisms and the structure of ecosystems, thereby driving feedbacks into fisheries, global biogeochemical cycles, and climate.
 We are grateful to Mark Payne, Damian Loher, Zouhair Lachkar, Thomas Frölicher, Fortunat Joos, and Marco Steinacher for their support. C.H. was supported by the European Project of Ocean Acidification, which received funding from the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreement 211384. EPOCA is endorsed by the international programs Integrated Marine Biogeochemistry and Ecosystem Research, Land-Ocean Interactions in the Coastal Zone, and Surface Ocean Lower Atmosphere Study. N.G, A.McD., and M.V. acknowledge funding from ETH Zurich.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.