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

The Barents Sea supports one of the world’s largest cod stocks and is of high socioeconomic importance. Its ecosystem is strongly influenced by Atlantic Water inflow from the Subpolar North Atlantic (SPNA) via the North and Norwegian Atlantic currents with a 2–10 year advective lag. Heat carried by this inflow controls regional temperature and sea-ice extent, shaping environmental conditions for marine organisms. Observed co-variability links low sea-ice extent to higher phytoplankton productivity, and warmer ocean temperatures to larger cod stocks. Prior studies demonstrated multi-year predictability of Barents Sea physical conditions: winter sea-ice cover (up to ~2 years) and temperatures (several years to a decade), with rapid Arctic winter sea-ice decline in 1997–2007 predicted 5–7 years in advance, consistent with SPNA-Barents advective timescales. These couplings suggest potential for ecosystem predictions. While cod stock has been predicted statistically up to a decade using upstream hydrography, no dynamic coupled physical-ecosystem predictions for the Barents Sea have been attempted. This study provides a first step by assessing whether phytoplankton abundance (primary production proxy) is predictable using decadal hindcasts from the Norwegian Climate Prediction Model (NorCPM1) with biogeochemistry (HAMOCC), initialized from a temperature-salinity reanalysis. By comparison with satellite chlorophyll and in situ hydrography/nutrients, the study evaluates predictability and mechanisms in two Barents Sea domains.

Background literature establishes: (i) strong influence of Atlantic Water on Barents Sea climate and ecosystems; (ii) robust links between sea-ice extent and phytoplankton productivity, and between temperature and cod stock size; (iii) demonstrated predictive skill for Barents Sea winter sea-ice and temperatures on interannual-to-decadal scales using statistical and dynamical models; (iv) skillful statistical forecasts of cod using upstream anomalies; (v) limited prior work on dynamic prediction of biogeochemistry and phytoplankton, with challenges in predicting summer hydrography and sea ice. Additional studies relate SPNA variability and Subpolar Gyre (SPG) dynamics to heat and nutrient anomalies advected into the Nordic/Barents Seas, affecting ecosystem drivers. These works motivate testing whether initialized coupled physical-biogeochemical models can predict multi-year phytoplankton variability and identify the underlying advective mechanisms.

Modeling framework: NorCPM1 (NorESM1-based) includes CAM4-OSLO atmosphere (~2°), MICOM ocean (~1°), CICE sea ice, CLM4 land, and HAMOCC ocean biogeochemistry (C, N, P, Si, Fe; one generic phytoplankton and zooplankton). Phytoplankton growth is limited by light, nitrate, phosphate, dissolved iron, and temperature. Simulations: (1) 30-member historical run (CMIP6 historical forcing 1850–2014, extended to 2029 with SSP2-4.5); (2) 30-member reanalysis (1950–2018) with monthly assimilation of temperature and salinity anomalies via Ensemble Kalman Filter, updating ocean and sea ice states; no biogeochemical assimilation; (3) annual decadal hindcasts (10 members) initialized Nov 1 each year from the reanalysis. Observations: Satellite chlorophyll (OC-CCI, 1998–present), HadISST sea-ice concentration and SST, in situ macro-nutrients (nitrate, phosphate; IMR 1980–2017), CTD temperature/salinity (ICES), and preprocessed Atlantic Water (50–200 m) temperature/salinity at Svinøy, Gimsøy, and Fugløya–Bjørnøya sections (ICES). Domains and seasons: Two Barents Sea regions analyzed: Polar Domain (seasonally ice-covered; 74.5–77.5°N, 27.5–52.5°E) and Atlantic Domain (ice-free; 70.5–72.5°N, 19.5–32.5°E). Seasonal means: winter (Jan–Mar), summer (May–Jul). Surface properties from model upper 0–5 m and in situ 0–30 m; deep properties from 50–200 m. Upstream sections (Svinøy 63°N,3°E; Gimsøy 69°N,12°E; BSO 73°N,20°E) used for pathway analysis. Data processing and skill assessment: Model and satellite fields regridded to 1×1°. Satellite chlorophyll converted to phytoplankton carbon using C:Chl ratio of 120. Time series were linearly detrended to remove long-term trends beyond the record’s length and smoothed with a 4-year running mean to emphasize multi-year variability. Predictive skill evaluated via anomaly correlation between detrended observations and hindcasts and by variance comparison; 95% confidence via 1000-iteration bootstrap. To attribute skill to internal variability rather than external forcing, hindcasts were compared against the historical ensemble mean: correlations significantly different and/or variance contrast used to exclude externally forced signals. Residual maps (hindcast minus historical) of 50–200 m temperature and nitrate were used to diagnose advective pathways and lead–lag propagation.

- Phytoplankton abundance in the Barents Sea is predictable up to 5 years in advance using NorCPM1 hindcasts, with domain-dependent mechanisms. - Polar Domain (seasonally ice-covered): Significant positive correlation between observed chlorophyll and predicted phytoplankton for lead years 2–9 arises largely from a multi-year positive trend (2005–2014). After detrending, hindcasts still skillfully predict summer phytoplankton anomalies up to lead year 5. The 2005–2014 high phytoplankton period coincides with anomalously low summer sea-ice concentration and higher SST, enhancing light availability; this event is absent in the historical simulation, indicating internal variability and successful initialization. - Atlantic Domain (ice-free): A pronounced early-2000s chlorophyll peak is robust in satellite, in situ section data at the Barents Sea Opening, and the reanalysis, and is skillfully predicted at 5-year lead; it is absent in the historical ensemble, again indicating internal variability and initialization-driven skill. A later (~early 2010s) satellite chlorophyll peak is not captured by the hindcasts, reducing overall skill for 1998–2018. - Mechanisms: • Polar Domain: Predictability is linked to advection of heat from the SPNA, which reduces winter sea-ice growth and leads to anomalously low winter sea-ice extent that persists into summer, increasing light for phytoplankton. NorCPM1 predicts anomalously high oceanic heat loss and reduced winter sea-ice during 2005–2014; residual temperature maps show a warm anomaly propagating from SPNA to the Nordic Seas and into the Barents Sea over 3–5 years. • Atlantic Domain: Predictability is driven by advection of positive nitrate anomalies from the SPNA, increasing winter nutrient inventories and fueling higher summer phytoplankton. Residual nitrate maps show a positive anomaly forming in eastern SPNA (lead years 1–2), entering the eastern Nordic Seas and Norwegian coast (lead years 3–4), and reaching the Barents Sea (lead year 5). In situ nitrate peaks at Svinøy (2000) and Gimsøy (2001) corroborate the advective signal and are predicted at appropriate leads. - Biomass–nutrient consistency: Observed nitrate drawdown of ~0.7 µmol N l⁻¹ (modeled ~0.4 µmol N l⁻¹) implies organic carbon of ~56 mg C m⁻³ (modeled ~32 mg C m⁻³) using Redfield 106:16, consistent with observed (modeled) phytoplankton anomalies of ~40 (20) mg C m⁻³ in the Atlantic Domain early 2000s. - Temperature vs nitrate timing: Heat and nitrate anomalies show different timing along the pathway, likely reflecting different source waters and/or modification by air–sea fluxes. The sequence is consistent with SPG contraction/expansion modulating both heat and nutrient properties of inflowing waters. - Some key local drivers (shortwave radiation, mixed layer depth, stratification) show little predictability relative to advection-driven signals. - Spatial skill pattern: Stronger synchronization and predictability upstream (SPNA) than in the Nordic Seas may explain higher skill at longer leads in the Barents Sea.

The study demonstrates that multi-year predictability of Barents Sea phytoplankton arises from dynamically predictable upstream ocean processes transmitted via advection. In the Polar Domain, heat anomalies from the SPNA reduce winter sea-ice growth, and the resultant winter sea-ice anomalies persist into summer, enhancing light and phytoplankton abundance—yielding up to 5-year predictability. In the Atlantic Domain, nitrate anomalies advected from the SPNA precondition winter nutrient inventories, enabling predictable spring–summer blooms at similar lead times. The contrast in mechanisms underscores the importance of correctly initializing and simulating large-scale circulation pathways and their biogeochemical properties. The lack of the early-2010s Atlantic Domain chlorophyll peak in hindcasts and absence of skill in some intermediate lead years indicate sensitivity to timing errors, local processes, and limitations in assimilation (no biogeochemical assimilation) and model physics. Differences in the timing of temperature and nitrate anomalies along Svinøy and Gimsøy sections suggest contributions from distinct source waters and SPG state, consistent with prior findings that SPG contraction brings warmer, nutrient-poor waters and expansion brings cooler, nutrient-richer waters. The demonstrated initialization-driven skill, contrasted against the historical ensemble, confirms that these predictions reflect internal variability rather than external forcing. Given ongoing sea-ice decline, the relative importance of heat-related predictability may shift northward into the Arctic interior, while nutrient-advection control may remain central for the Barents Sea. These results support the feasibility of near-term ecosystem predictions and motivate exploring predictability propagation to higher trophic levels and fisheries applications.

Two events of anomalously high phytoplankton abundance in the Barents Sea are predictable up to five years in advance using an initialized coupled physical–biogeochemical model. Predictability arises via two robust mechanisms: (i) advection of nutrient anomalies from the SPNA in the ice-free Atlantic Domain, and (ii) advection of heat from the SPNA leading to reduced sea ice and enhanced light in the Polar Domain, with winter anomalies persisting into summer. These mechanisms are consistent with known influences of North Atlantic Subpolar Gyre variability on the properties of Atlantic inflow and are likely to persist in future climates, although diminishing sea ice may shift the locus of heat-related predictability northward. This work provides a foundation for decadal ecosystem prediction in the Barents Sea and supports broader efforts toward operational marine ecosystem forecasts. Future research should investigate translating lower-trophic predictability to higher trophic levels (e.g., zooplankton, fish), assess benefits of assimilating biogeochemical variables, improve representation of regional processes, and evaluate robustness under different external forcing scenarios.

- Biogeochemical variables were not assimilated; predictability relies on physical initialization and model biogeochemistry, potentially limiting accuracy of nutrient and biomass states. - In situ observations in the Polar Domain are sparse (mostly August–September), limiting direct validation of early-season nutrient and phytoplankton states; deep nutrient observations were used as proxies for winter surface conditions. - Satellite chlorophyll to carbon conversion uses a fixed C:Chl ratio (120), whereas this ratio is variable; this introduces uncertainty in comparing chlorophyll-based observations to model phytoplankton carbon. - Some drivers (shortwave radiation, mixed layer depth, stratification) showed limited predictability, and skill was absent in some lead years, indicating sensitivity to local processes and timing errors. - Early-1990s nutrient measurements may contain biases, reducing confidence in older peaks; overall measurement uncertainties (e.g., nitrate ~2%) constrain detection of small anomalies. - The historical ensemble represents external forcing only; separation of internal vs external variability assumes negligible forced variability at the event timescales and regions considered. - Spatial differences in data assimilation performance (better synchronization upstream in SPNA than in Nordic Seas) may affect the lead-time structure of skill in the Barents Sea.