Skillful multiyear predictions of ocean acidification in the California Current System

Ocean acidification, caused by the absorption of anthropogenic CO2 by the ocean, is a significant environmental problem impacting marine ecosystems globally. The California Current System (CCS), a highly productive region crucial for US fisheries, is particularly vulnerable due to natural upwelling of corrosive, carbon-rich waters. This upwelling, driven by equatorward winds, brings nutrient-rich waters to the surface, fueling high productivity but also increasing acidity. Anthropogenic CO2 further exacerbates this natural acidity, leading to observed anomalously low surface pH and undersaturation of calcium carbonate minerals. This situation negatively impacts various organisms, including shellfish, which are economically important. The CCS's vulnerability and economic significance make it a priority for studying multiyear biogeochemical predictions. Previous prediction efforts have focused on seasonal forecasts of sea surface temperature and biogeochemical variables as inputs for ecosystem models. While some success has been shown in predicting surface chlorophyll, multiannual to decadal predictions of ocean biogeochemistry in the CCS remain largely unexplored. This timescale is critical for fisheries management, enabling informed decisions on catch limits, fishing closures, and quota adjustments. Persistence forecasts, which extrapolate existing anomalies, provide a baseline. However, initialized predictions, leveraging physically-based models and initial conditions, offer a potentially more powerful forecasting framework if the system is sufficiently deterministic and the model skillful. Ensemble simulations of initialized Earth System Models (ESMs), integrating the physical climate system and biosphere, offer the best current approach for improving on decadal persistence forecasts. This study aims to assess the potential of ESMs for multiyear predictions of ocean acidification in the CCS, using a decadal prediction system to evaluate the skill of its retrospective forecasts.

Numerous studies have documented the effects of ocean acidification on marine ecosystems, particularly in the California Current System (CCS) where the natural upwelling process enhances the impact of anthropogenic CO2. Gruber et al. (2012) highlighted the rapid progression of ocean acidification in the CCS, while Feely et al. (2008) provided evidence for the upwelling of corrosive, acidified water onto the continental shelf. Research by Bednaršek et al. (2014, 2017) investigated the impact on pteropods, showing shell dissolution and vulnerability related to exposure history. The economic significance of these changes, especially for shellfish fisheries, is considerable (National Marine Fisheries Service, 2017). Previous prediction efforts in the CCS have primarily focused on seasonal forecasts using sea surface temperature (SST) (Jacox et al., 2017; Hervieux et al., 2017; Stock et al., 2015) and biogeochemical variables (Siedlecki et al., 2016) as inputs into ecosystem models. Park et al. (2019) demonstrated the potential for skillful initialized predictions of surface chlorophyll, but multiannual to decadal predictions of ocean biogeochemistry in the CCS have been lacking. Studies on decadal predictability of marine productivity and ocean carbon uptake have shown some promise (Séférian et al., 2014, 2018; Li et al., 2016, 2019; Lovenduski et al., 2019), emphasizing the need for sophisticated models like initialized ESMs to account for complex interactions within the climate system. The importance of decadal predictions for fisheries management has also been emphasized (Tommasi et al., 2017), underscoring the need for improved forecasting capabilities at these timescales.

This study employs the Community Earth System Model Decadal Prediction Large Ensemble (CESM-DPLE), an initialized global ESM with embedded ocean biogeochemistry, to assess retrospective forecasts of surface pH anomalies in the CCS from 1955 to 2017. CESM-DPLE uses an ocean model with 1° x 1° horizontal resolution and 60 vertical levels. Forty ensemble members were initialized annually on November 1st from a forced ocean-sea ice reconstruction. These simulations were integrated forward for 10 years. The reconstruction, skillful in representing surface pH variability, serves as a benchmark. The researchers carefully define "potential predictability" (correlations between CESM-DPLE and the reconstruction) and "predictive skill" (comparison of CESM-DPLE to observations). The anomaly correlation coefficient (ACC) quantifies the prediction of anomalies, while the normalized mean absolute error (NMAE) assesses accuracy in predicting anomaly magnitudes. The initialized forecasts are compared to a persistence forecast (extrapolating existing anomalies) and an uninitialized CESM Large Ensemble (CESM-LE) mean. The JMA gridded observational product of surface pH from 1990-2017, based on empirical relationships derived from in situ measurements, is used for skill assessment. The analysis focuses on surface pH anomalies within the California Current Large Marine Ecosystem, removing a second-order polynomial fit to isolate year-to-year variations. The model's representation of physical circulation and carbonate chemistry in the CCS is evaluated, acknowledging that the coarse resolution limits the explicit resolution of fine-scale coastal upwelling processes. Drivers of reconstructed surface pH variability (salinity, alkalinity, SST, DIC) are identified via linear decomposition, determining their contributions to pH variability. Finally, the mechanisms of surface pH predictability are investigated, focusing on the predictability of the variables driving pH changes, including salinity-normalized DIC (sDIC). A budget analysis of DIC in the upper 150 m of the CCS examines the role of advection and mixing in influencing the DIC inventory.

Retrospective forecasts reveal a potential for predicting surface pH up to 5 years in advance, significantly exceeding the skill of a simple persistence forecast. While persistence is useful at lead year 1 in some areas, the initialized forecast shows statistically significant improvement nearly everywhere. At lead year 2, persistence becomes less skillful in the southern CCS, while the initialized forecast maintains predictability. The initialized forecast retains significant predictive skill in the central and southern CCS through lead year 5. Area-weighted analysis confirms the superiority of the initialized forecast over persistence and the uninitialized forecast across all 5 lead years. The lead year 1 ACC of 0.72 is comparable to or better than skill achieved by seasonal forecasts of SST in the CCS. The NMAE is lower than both persistence and the uninitialized forecast. Predictability in salinity-normalized DIC (sDIC) is the major contributor to surface pH predictability across all lead years, surpassing the combined predictability of SST and salinity-normalized alkalinity. The predictability in sDIC is primarily driven by anomaly persistence but is further enhanced by initializations. A DIC budget analysis indicates that variability in vertical and lateral DIC advection is crucial in determining the DIC inventory. Subsurface and basin-wide initializations of DIC are key to making skillful predictions. Skill assessment using observational data confirms positive ACCs throughout much of the CCS at lead year 1, with skill maintained through lead year 4 south of Cape Mendocino. Persistence in observed surface pH is less effective than the initialized predictions.

This study demonstrates, for the first time, the potential of an initialized ESM to predict surface pH multiple years in advance in a complex and economically important region. Although the spatial resolution doesn't allow direct aid in coastal fisheries management, the results confirm the feasibility of multiannual to decadal surface pH predictions. The ESM forecasts can serve as boundary conditions to improve regional biogeochemical forecasting and extend lead times. High-resolution downscaled forecasts could significantly enhance fisheries management. While the focus is on surface pH, other parameters like calcium carbonate saturation states are likely to be predictable due to their dependence on dissolved CO2. The analysis of interannual variations in surface pH, after detrending to remove the dominant long-term ocean acidification signal, demonstrates the value of initialization. The study acknowledges limitations such as the coarse resolution, preventing explicit resolution of fine-scale coastal upwelling, and the limited timeframe for skill assessment due to data constraints.

This research demonstrates the potential for skillful multiyear predictions of ocean acidification in the California Current System using an initialized Earth System Model. The findings highlight the importance of initializing dissolved inorganic carbon anomalies and the role of advection in driving predictability. While limitations exist regarding spatial resolution and data availability, this study represents a significant advancement in forecasting ocean acidification and its impact on economically vital marine ecosystems. Future research could focus on improving model resolution and incorporating more comprehensive observational data to further refine the predictive capabilities and extend the forecasting horizon. Dynamically downscaled decadal forecasts using high-resolution regional models could offer more precise insights and contribute directly to improved fisheries management strategies.

The study's limitations include the relatively coarse spatial resolution (~100 km x 100 km) of the ESM, which does not explicitly resolve fine-scale coastal upwelling processes. The use of subgrid parameterizations to represent eddy-induced offshore flux might also impact the accuracy of predictions. Although the model reasonably captures key large-scale features like alongshore winds, air-sea CO2 fluxes, and surface pH, finer-scale variability might be underrepresented. Furthermore, the analysis of predictive skill is restricted by the limited temporal coverage of gridded surface pH observations (1990–2005), affecting statistical significance assessments. The use of a smooth, scenario-based projection of atmospheric CO2 after 2005, as opposed to observed data, might have also influenced the results. Finally, the focus on annual average surface pH anomalies, while showing higher predictability than monthly resolution, limits the assessment of intra-annual variability.