Sea-ice algal phenology in a warmer Arctic
The Arctic sea-ice decline is among the most emblematic manifestations of climate change and is occurring before we understand its ecological consequences. We investigated future changes in algal productivity combining a biogeochemical model for sympagic algae with sea-ice drivers from an ensemble of 18 CMIP5 climate models.
Sea-ice algae form a large fraction of sea-ice (sympagic) biomass (1). Before the onset of the seasonal phytoplankton bloom, sea ice provides critical habitat for marine algae and the upper trophic levels for whom sea-ice algae are their sole food-source during this period. Together with sub-ice phytoplankton, sea-ice algae represent the foundation of ecological interactions in the sea-ice biome (2). Algal growth in sea ice is limited by the strong seasonality of light, nutrients, and brine connectivity and, in first-year ice, by the relatively narrow time window during which sea ice exists. At high latitudes, light conditions in the sea-ice habitat are subject to extreme seasonal changes. Light availability is largely regulated by the photoperiod, which depends on latitude and time of the year, and by the albedo and light attenuation of different sea-ice surfaces (e.g., snow-covered ice, bare ice, and ponded ice). Nutrients in sea ice are replenished through brine connectivity with the underlying water column and through local regeneration processes. Brine connectivity depends on brine permeability, a function of both sea-ice temperature and salinity. The warmer and less salty the ice, the greater its permeability and thus its habitable space.
We expect future higher sea-ice surface temperature (3) and thus warmer and more permeable sea ice. However, steeper temperature gradients through thinner ice/snow, together with the expected general freshening of the Arctic Ocean (4, 5), might have a counterbalancing effect on sea-ice permeability. Large uncertainties in future nutrient dynamics (i.e., how mixed layer depth might change and be sustained) limit our understanding of the future Arctic Ocean ecosystem (6, 7).
In sea ice, the phenology of the narrow time window for biological growth is constrained by the incidence of suitable light conditions and by the length of the ice season (1, 8–10). While sea-ice freeze-up (breakup) is overall expected to occur later (earlier) in the season at all latitudes (3), there are more uncertainties in the future timing of sympagic algal blooms (11–13), with snow changes considered to be the single most important predictor of future sea-ice algal phenology (14). A few conceptual and qualitative models have been proposed to describe the anticipated response of the seasonal ice zone biota to a warmer climate (2, 15–19), and some more quantitative works have used climate model projections and earth system models to analyze the effects of a fresher, ice-free Arctic on phytoplankton primary production (6, 7). Melt onset trends have shown considerable advances on pan-Arctic scale with marked regional differences and the southerly regions melting earlier (20). The strongest trend since 1979 has occurred in the Barents Sea and East Greenland Sea (25 and 30 days, respectively), while less rapid trend is observed in the Baffin Bay, Kara Sea, and Hudson Bay (22, 15, and 10 days, respectively), and the weakest trend is found in the Central Arctic Sea (5 days). There is increasing consensus that these advances in melt onset, together with thinner sea ice and increased stratification, have led to conditions generally more favorable for sub-ice pelagic blooms in the last 30 years (21), suggesting that the advancement of the phytoplankton bloom peak has made the annual season of primary productivity overall longer as well (22).
However, less is known of the conditions that lead to sea-ice algal blooms, which are also dependent on other factors such as brine connectivity. Changes in sea-ice nutrient concentrations are expected to affect the overall amplitude (8, 9, 23) but are less likely to influence the timing of the growth of sea-ice biota (14, 24). Thus, sympagic and pelagic dynamics might not be directly related (25), and the onset of seasonal primary production in the Arctic marine food web may occur in subsequent waves made of sub-ice and open-water phytoplankton blooms (22) on top of sea-ice algal blooms. These pulses of primary production trigger higher trophic levels, from zooplankton to fish to top predators such as whales and polar bears. These phenological cascades are expected to be stronger in the Arctic Ocean where there is tight coupling between abiotic conditions and the timing of primary production and where the food chains are short (22).
Climate model projections are the only available tools that can be used to quantitatively assess potential long-term changes in sea-ice biogeochemical dynamics in a warmer Arctic. Here, we use a near-mechanistic modeling approach that combines sea-ice drivers obtained from an ensemble of 18 CMIP5 [Climate Model Intercomparison Project Phase 5 (26)] climate models with a state-of-the-art sea-ice biogeochemical model (24) to provide a quantitative overview of the effect of different physical drivers on the potential phenological changes in pan-Arctic first-year ice primary production. We focus our study on first-year ice, which is becoming the dominant type of sea ice in the Arctic Ocean (3, 20). In this work, we consider the Arctic first-year ice as a single pan-Arctic physico-biogeochemical unit. This is motivated by the inherent constraints of climate projections and by the sparse availability of recurring studies on seasonal sea-ice biogeochemistry. We seek to elucidate the first-order impacts of large-scale, climate-driven ice changes on sea-ice algal production rather than considering the more local and subregional components, e.g., as done in (27) to explain the coupling between sea ice and tundra in the Svalbard Archipelago since it would require observational efforts that are not yet available on a pan-Arctic scale.
Changes in sea-ice physical drivers of biological responses
Daily and monthly simulations of physical properties of sea ice were used to derive empirical probability density functions of sea-ice drivers in latitudinal bands of 2°, based on 18 CMIP5 climate models (table S1) for recent historical (1961–2005) and business-as-usual representative concentration pathways [RCP8.5, 2061–2100 (28)] scenarios (see Materials and Methods for details). This method is superior to the use of zonal averages when dealing with variable events such as the day of freeze-up and breakup of sea ice, and it has been successfully used in climate model analyses (29, 30). It also allows us to derive statistical descriptors of the extremes of the distribution, without any assumption on the shape of the distribution itself. Models were selected according to the availability of high-frequency daily data at the time of analysis. Sea-ice drivers considered were date of freeze-up, date of breakup, maximum sea-ice thickness, first-year ice extent, first month of snow cover, last month of snow cover, maximum snow depth, and minimum near-surface atmospheric temperature, as well as the dates at which the maxima and minima occurred. The climate data were obtained from the same CMIP5 dataset used in earlier works on future Arctic changes under the RCP8.5 scenario [e.g., (31)]. However, here, we considered minimum air temperature rather than mean air temperature, focused on first-year ice only rather than on all Arctic sea ice, and explored daily data of sea-ice drivers rather than monthly averages, providing an original contribution to the analysis of CMIP5 model outputs.
We initially verified the ensemble means of first-year ice freeze-up and breakup days and the area extent for the period 1979–2005 from the 18 single distributions against remote sensing data (Fig. 1). The observations of sea-ice concentration for current climate conditions were obtained from passive microwave data (32). The sea-ice drivers of phenology (freeze-ups and breakups) from the climate model ensemble distributions compare well with the distributions computed from the satellite data (Fig. 1, A and B). In particular, the simulated medians of the days of freeze-up and breakup match the observations at all latitudes [root mean square differences (RMSDs) are 4 and 11 days, respectively]. The spread of the modeled distribution with respect to observations is similar for the freeze-up (Fig. 1A) but narrower for the modeled breakup, indicating less variability in CMIP5 models (Fig. 1B). Less agreement is found when comparing the medians of observed and modeled sea-ice extent (RMSD = 0.2 × 106 km2; Fig. 1C), likely because of the known biases of CMIP5 models to simulate this sea-ice driver (33, 34). The difference is particularly large in the latitudinal band between 70°N and 78°N. Our statistical approach demonstrates acceptable performances of the climate models, especially in terms of making the sea-ice drivers of phenology usable for the analysis of changes in seasonal sea-ice and their impact on ice algal phenology. The larger spread in sea-ice extent will instead be taken into consideration and discussed when integrating our latitudinal-based results on a pan-Arctic scale.