Phytoplankton abundance in the Barents Sea is predictable up to five years in advance

The Barents Sea, home to one of the world's largest cod stocks, is a region of significant economic importance. Accurate interannual-to-decadal predictions of its ecosystem are crucial for effective marine resource management. The ecosystem's structure and dynamics are strongly influenced by the inflow of Atlantic Water through the Barents Sea Opening. This warm, saline water, originating from the Subpolar North Atlantic (SPNA), is transported by the North and Norwegian Atlantic currents, with an advective lag of 2–10 years. The advected heat significantly affects the Barents Sea's temperature and sea ice extent, shaping the environmental conditions for its inhabitants. Strong relationships have been observed between phytoplankton activity and sea ice extent, with reduced ice cover often associated with higher productivity. Similarly, ocean temperature is strongly linked to cod stock size, with warmer waters generally correlating with larger stocks. The thermal and dynamic memory inherent in the advected Atlantic Water creates potential for predicting Barents Sea physical conditions years in advance. Hindcasts have shown sea ice cover predictability up to 2 years, and dynamical climate prediction models have successfully predicted Barents Sea winter temperatures for several years to a decade. The long-term predictability of the Barents Sea's physical state, coupled with known physical-biological interactions, suggests the possibility of predicting ecosystem dynamics. While statistical models have shown decadal predictability of the Barents Sea cod stock using upstream hydrographic anomalies, dynamic predictions using coupled physical-ecosystem models remain largely unexplored. A critical first step is achieving skillful predictions of phytoplankton primary production, requiring accurate prediction of summer hydrography and sea ice. This study utilizes retrospective decadal hindcasts from the Norwegian Climate Prediction Model (NorCPM1), incorporating the biogeochemical model HAMOCC, to address this challenge. The hindcasts were initialized from a reanalysis assimilating observed temperature and salinity data to closely represent the real climate system. By comparing hindcasts with satellite-derived chlorophyll and hydrographic measurements, the study aims to demonstrate the predictability of phytoplankton abundance and identify the underlying mechanisms.

Previous research has established strong links between the physical and biological aspects of the Barents Sea ecosystem. Studies have shown the predictability of winter sea ice cover (up to 2 years) using statistical models based on observed heat transport. Dynamical climate models have also demonstrated skillful predictions of Barents Sea winter temperatures (from a few years to a decade). The rapid decline in winter Arctic sea ice extent (1997–2007) was also shown to be predictable 5–7 years in advance, consistent with the advective timescale between the SPNA and the Barents Sea. The Barents Sea cod stock has been shown to be predictable up to a decade in advance using statistical models and upstream hydrographic data. However, this study is novel in its use of coupled physical-ecosystem numerical models to dynamically predict ecosystem components, focusing specifically on phytoplankton primary production as the base of the food web. Existing literature highlights the challenges in predicting summer hydrography and sea ice in this region.

This study employed simulations from the Norwegian Climate Prediction Model (NorCPM1), a coupled climate model that includes the atmospheric model CAM4-OSLO, the ocean circulation model MICOM, the biogeochemical model HAMOCC, the land model CLM4, and the sea ice model CICE. HAMOCC simulates the dynamics of several biogeochemical elements, including carbon, nitrogen, phosphorus, silicate, and iron, with a simplified representation of phytoplankton and zooplankton. The simulations included an extended historical simulation (30 realizations), a reanalysis (30 members) with monthly assimilation of temperature and salinity anomalies using an Ensemble Kalman Filter, and decadal retrospective predictions (hindcasts, 10 members each) initialized from the reanalysis. The historical simulation used CMIP6 historical forcing from 1850 to 2014, extended to 2029 using SSP2-4.5 scenario forcing. The reanalysis assimilated temperature and salinity data from 1950 to 2018. Satellite-derived chlorophyll concentration from the OC-CCI dataset (1998-present), satellite-derived sea ice concentration and SST from the HadISST dataset, and in situ measurements of macronutrients (nitrate and phosphate) from the Norwegian Institute of Marine Research were used for comparison. Predictability was assessed by evaluating the correlation between observed and simulated phytoplankton concentrations (after detrending to remove long-term trends). The significance of correlations was determined using a bootstrap approach. To identify drivers of predictability, time series of key variables (temperature, sea ice, stratification, nutrients) were analyzed for two regions: a Polar Domain (seasonally ice-covered) and an Atlantic Domain (ice-free). Analysis included comparing the model results with the historical simulation to isolate variability related to internal climate dynamics. A C:Chl ratio of 120 was used to convert chlorophyll concentration to carbon concentration. Annual mean properties at upstream hydrographic sections in the Nordic Seas were also analyzed to investigate advective pathways.

The study found that NorCPM1 skillfully predicted past interannual variability in phytoplankton concentration in the Barents Sea, with spatial variations in predictability. In the Polar Domain, significant positive correlations were found 2–9 years after hindcast initialization. In the Atlantic Domain, the model showed no predictive skill using the initial skill score. Analysis of time series revealed that in the Polar Domain, a positive phytoplankton anomaly during the 2000s and 2010s was successfully predicted up to lead year 5. This predictability was linked to the skillful prediction of anomalously low sea ice concentration, which increased light availability. The historical simulation did not show a similar anomaly, indicating that the predictability arose from skillful initialization rather than external forcing like climate change. In the Atlantic Domain, two peaks in chlorophyll concentration (early 2000s and early 2010s) were observed. The early 2000s peak was skillfully predicted in the hindcasts at lead year 5, again linked to internal climate variability and not external forcing. Analysis of drivers indicated that in the Polar Domain, the positive phytoplankton anomaly was associated with decreased sea ice and a slight weakening of vertical stratification. In the Atlantic Domain, the positive phytoplankton anomaly correlated with positive anomalies in winter nutrient concentrations (nitrate and phosphate), suggesting that the prediction of these nutrient levels drove the phytoplankton prediction. The advection of these anomalies from the Subpolar North Atlantic is the likely mechanism. Maps of temperature and nitrate anomalies from hindcasts showed a progression from the SPNA northward into the Barents Sea, supporting the advection hypothesis. In situ measurements from hydrographic sections corroborated the advective nature of these anomalies. The study suggests that the predictability of the temperature and nitrate anomalies are linked to separate water masses, possibly related to shifts in the North Atlantic Subpolar Gyre.

The findings demonstrate multi-year predictability of phytoplankton abundance in the Barents Sea, a significant result given the challenges in predicting marine ecosystems. The identification of two distinct prediction mechanisms – advection of nutrients from the SPNA and the prediction of sea ice concentration – highlights the complex interplay between physical and biological processes. The results are robust, as they are not driven by external forcing, confirming the role of internal climate variability. The links to the North Atlantic Subpolar Gyre provide broader context for understanding these variations. While the study focuses on phytoplankton, the findings suggest potential for extending this predictability to higher trophic levels, contributing significantly to improved ecosystem forecasting in the Barents Sea. The decreasing sea ice cover may alter the relative importance of heat and nutrient advection in driving future phytoplankton dynamics.

This study demonstrates, for the first time, the multi-year predictability of phytoplankton abundance in the Barents Sea, using a coupled physical-biogeochemical model. Two distinct mechanisms, advection of nutrients and sea ice prediction, are identified. These findings contribute significantly to our understanding of Barents Sea ecosystem dynamics and offer a promising foundation for future ecosystem prediction efforts. Future research should investigate how this predictability translates to higher trophic levels and consider how changing climate conditions might alter these relationships.

The study uses a simplified representation of the marine ecosystem (one phytoplankton and one zooplankton group) within the biogeochemical model, which might limit the accuracy of the predictions. The availability and quality of observational data, particularly in situ measurements, could influence the assessment of model skill. The analysis focuses on two specific events of high phytoplankton abundance, limiting the generalizability of the findings to all phytoplankton variability.