Introduction

Phytoplankton - single-celled drifting primary producers—are the foundation of the marine food web and affect the global carbon (C) cycle by assimilating C in the surface ocean before sinking and sequestering it at depth1,2. In much of the Arctic Ocean, seasonally low light availability due to both high latitude and extensive sea ice cover has historically led to a single short, intense burst of net primary production (NPP; Fig. 1a) by phytoplankton in summer months when sea ice cover retreats3. This phytoplankton bloom supports substantial zooplankton populations4,5 which in turn sustain local and migrating fish, marine mammals, and seabirds6,7,8. Phytoplankton biomass that is not consumed in the pelagic ecosystem is exported to the ocean floor where it supports rich benthic communities on the shallow continental shelves of the Arctic Ocean9,10. Indigenous communities rely on hunted foods for livelihoods, cultural practices, and food sovereignty11 and thus ultimately depend on the summer phytoplankton bloom.

Fig. 1: Idealized annual cycles of changes to Arctic phytoplankton blooms and bathymetric map.
figure 1

Idealized annual cycles of changing Arctic phytoplankton blooms, with original (blue) and changed (red) cycles demonstrating (a) an earlier bloom start and end and a longer bloom, (b) a decrease in the proportion of annual productivity generated during the bloom, and (c) an increase in the daily mean rate of productivity during the bloom, resulting in greater biomass accumulation during the bloom. Bloom start and end are marked with dots. d A bathymetric map of the Arctic Ocean, with the Arctic circle drawn in black, the 1000m isobath (separating the shelf from the deep ocean within the Arctic Ocean, > 66.5∘N, and sub-Arctic, from 50–66.5∘N) marked in maroon, and labels demonstrating the location of the (A) Bering Sea, (B) Chukchi Sea, (C) Beaufort Sea, (D) central Arctic, (E) Hudson Bay, (F) Baffin Bay, (G) Barents Sea, and (H) N. Atlantic.

Full size image

Arctic Ocean phytoplankton populations are experiencing rapid environmental changes due to anthropogenic climate change. As a consequence of polar amplification, temperatures in the Arctic have increased at four times the global lyaveraged rate12. This has coincided with a decline in sea ice extent13,14 and thickness13,15, increasing the light in the surface ocean and allowing the formation of under-ice phytoplankton blooms under consolidated sea ice16 across much of the Arctic Ocean17,18. Sea ice is also both melting out earlier and freezing up later in the year than it did historically13. Between 1998 and 2018, a 25% expansion in open water area in the Arctic Ocean contributed to a 56% increase in phytoplankton NPP19. Sea ice loss may also drive an increase in the wind-driven entrainment of nutrients in the surface ocean in the fall20, allowing for the proliferation of fall blooms3. Changes in the magnitude and timing of phytoplankton NPP are likely to impact food availability for pelagic and benthic upper trophic level organisms as well as the rate of C export to the deep ocean10,21,22.

While previous studies have assessed how the timing of the Arctic phytoplankton bloom has changed over the observational record21,23,24,25,26 or will change by the end of the century27,28, this is the first study that seeks to specifically isolate the impact of anthropogenic climate change on the changes in Arctic Ocean bloom characteristics. Over the relatively short observational record, both anthropogenic climate change and the internal variability of the Earth system influence the physical drivers of phytoplankton productivity. Trends in sea ice extent29 or phytoplankton biomass30 often attributed to climate change can be masked or enhanced by internal climate variability. Isolating the impact of external forcing on changing bloom dynamics is critical to understanding the role of anthropogenic climate change in historical and projected Arctic and sub-Arctic bloom phenology changes.

Here, we separate the externally forced climate change signal from internal climate variability by using the fully coupled Community Earth System Model version 2 Large Ensemble (CESM2-LE31), a suite of 50 simulations with small differences in initial conditions, to assess changes in the mean and variance across many simulations31,32,33. We evaluate how characteristics of the Arctic phytoplankton bloom - start and end timing, duration (see Fig. 1a), relative importance (relative bloom contribution to total NPP; see Fig. 1b, c), species composition, and interannual variability - have changed over the historical record (between 1970 and 2020) and are projected to change by the end of the century. We also use emergence analysis34 to assess whether changes in bloom characteristics are indicative of a new climate state. We find that, by 2100, anthropogenic climate change causes blooms to start earlier in the year and last longer across the Arctic Ocean, and that the importance of these blooms is declining in some regions.

Results

Bloom start, end, and duration

While the mean timing of the start of the phytoplankton bloom in the Arctic Ocean shifts somewhat earlier between 1970 and 2020, climate change drives the bloom to start more than a month earlier by the end of the century, with the greatest changes in bloom start timing observed in the sea ice zone. We define the start of the bloom as the date when water column-integrated C-based biomass first increases above 25% of the maximum biomass for each grid cell and each ensemble member. We find that in 2020, the bloom starts on average 5 days earlier than in 1970 within the Arctic Ocean (> 66.5∘N, Fig. 2a, Table 1). By 2100, the bloom initiates more than a month (34 days) earlier than in 1970 (Fig. 2b, Table 1). Emergence analysis34 reveals that only 1% of the Arctic Ocean has a distinct start timing in 2020 relative to 1970, but that by 2100, 71% of the region has an earlier bloom start date (Table 2; Fig. S1). While the change in bloom start timing between 1970 and 2100 does not differ substantially based on water column depth within the Arctic Ocean, with the bloom shifting 35 days earlier on Arctic shelves (<1000 m) and 32 days earlier in the Arctic deep ocean (> 1000 m), the shelves of the sub-Arctic have a far earlier shift in start time (29 days earlier) than the sub-Arctic deep (9 days earlier). Instead, changes in bloom start timing over both the historical record and by the end of the century are greatest in the sea ice zone. In 2020, areas covered with seasonal sea ice bloom on average 8 days earlier than in 1970 (Fig. 2a). By 2100, this same region experiences a bloom on average 43 days earlier than in 1970. In several areas, most notably Hudson Bay, the northern Chukchi Sea, the Barents Sea, and the region between Greenland and Svalbard, the phytoplankton bloom begins more than 60 days earlier in 2100 than in 1970 (Fig. 2b).

Fig. 2: Changes in bloom start, end, and duration.
figure 2

Change in mean phytoplankton (a, b) bloom start date, (c, d) bloom end date, and (e, f) bloom length between 1970 and (a, c, e) 2020 and (b, d, f) 2100. The edge of the seasonal ice zone is indicated in dark green for 1970 and light green for 2020 or 2100, while the edge of the perennial ice zone is colored purple for 1970 and pink for 2020 (there is no perennial ice by 2100). Stippling indicates regions where there are no significant change in bloom timing. The Arctic circle is indicated in black.

Full size image
Table 1 Bloom start (day of year), end (day of year), and length (days) and annual and bloom NPP (Tg C) in 1970, 2020, and 2100, and change in timing (days) or NPP (Tg C) relative to 1970 for the Arctic Ocean (> 66.5°N) or the Arctic and sub-Arctic (> 50°N)
Full size table
Table 2 Proportion of the Arctic Ocean (> 66.5°N) or the Arctic and sub-Arctic (> 50°N) that emerges as having distinct bloom timing or importance metrics in 2020 or 2100 relative to 1970
Full size table

The end of the phytoplankton bloom (the date when integrated biomass first declines below 25% of the maximum) also shifts earlier over both the historical period and by the end of the century, but the change in bloom end timing is relatively uniform across the Arctic and sub-Arctic seas as well as on shelves and in the deep ocean. Between 1970 and 2020, the bloom end shifts earlier by 5 days (Fig. 2c, Table 1), while by 2100, the bloom ends 19 days earlier relative to 1970 (Fig. 2d, Table 1) both within the Arctic Ocean and across the region as a whole. Emergence analysis reveals that, while the bloom ends earlier over less than 1% of the Arctic in 2020, 27% of the Arctic Ocean has a distinct bloom end time in 2100 relative to 1970 (Table 2; Fig. S1). Bloom end timing does not differ substantially between the shelves and the deep ocean, with bloom end timing shifting 18–20 days earlier in both the Arctic and sub-Arctic. Changes in bloom end timing are most stark in the area covered by seasonal sea ice in 1970 but ice-free by 2100, where the bloom ends 31 days earlier by the end of the century (e.g., the Barents Sea and the region between Greenland and Svalbard; Fig. 2d).

While bloom length does not change in 2020 relative to 1970, a far earlier bloom start by 2100 drives a growing season that is on average two weeks longer within the Arctic Circle. Over the historical period, mean bloom length does not change within the Arctic Ocean due to opposing trends within the sea ice zone (Fig. 2e, Table 1). Regions covered in perennial sea ice in the 1970s but seasonal sea ice in the 2020s show an increase in the length of the bloom (4 days), while the growing season shortens (5 days) in areas covered in perennial sea ice in both the 1970s and 2020s (Fig. 2e). By 2100, the bloom is 14 days longer within the Arctic Ocean (Fig. 2f, Table 1). While less than 1% of the Arctic and sub-Arctic emerge as having a distinct bloom length in 2020, bloom length in 2100 is distinct over 9% of the Arctic and the sub-Arctic relative to 1970 (Table 2; Fig. S1). The increase in bloom length is most pronounced within the 2100 sea ice zone, where the bloom is on average 19 days longer than in 1970, and especially in Hudson Bay, the northern Chukchi Sea, and in the coastal Barents Sea (Fig. 2f).

Changes to bloom importance and species composition

While the present-day bloom is considered one of the most productive periods at mid and high latitudes, we find that the importance of the bloom declines by the end of the century driven by a decline in bloom productivity in the ice-free and sub-Arctic parts of the region (Table 1). While the region  >50°N sees a modest increase in annual NPP between 1970 and 2020, annual NPP declines by nearly 300 Tg C yr−1 by the end of the century, driven by a nearly 25% decline in NPP generated during the bloom (Table 1). These large declines are partially offset within the Arctic Circle, where the 15% increase in NPP between 1970 and 2020 is anticipated to persist through the end of the century, with most of that increase driven by a 16% increase in bloom productivity (Table 1). In both the Arctic and sub-Arctic, changes in bloom and total NPP are enhanced in the deep ocean relative to the shelves. Within the Arctic Ocean, total (bloom) NPP is anticipated to increase by 3.2 (2.2) g C m−2 yr−1 on shelves between 1970 and 2100, but total (bloom) NPP will increase by 6.9 (4.6) g C m−2 yr−1 in the deep ocean. In the sub-Arctic, total (bloom) NPP will decrease by 15 (14) g C m−2 yr−1 on the shelves but by 33 (32) g C m−2 yr−1 in the deep ocean.

To gain an understanding of the changing importance of the bloom, we evaluate the proportion of annual NPP produced during the bloom (see, e.g., Fig. 1b). In 1970, the bloom accounts for less than 30% of total annual NPP off the coast of the western US and Canada and south of Norway, but produces more than 80% of total NPP in Hudson Bay, Baffin Bay, and in the perennially sea ice-covered central Arctic (Fig. 3a). By 2020, the bloom is responsible for a smaller proportion of total annual NPP in the southern Bering Sea, the Chukchi Sea, N. Atlantic, and in parts of the central Arctic, where perennial sea ice cover declines (Fig. 3b). These patterns of declining importance are exacerbated by 2100, when the bloom generates a lower proportion of total annual NPP in the Chukchi, Bering, and Barents Seas and in the N. Atlantic, while much of the central Arctic experiences a resurgence in the importance of the bloom (Fig. 3c). Within the sea ice zone, the bloom produces 72% of total NPP in 1970 and increases to 75% by 2100. However, in ice-free waters, the bloom declines in importance by the end of the century; while 54% of total annual NPP is produced during the bloom in ice-free waters in 1970, only 45% of annual NPP is generated in the bloom by 2100. Changes in bloom importance were greatest in the sub-Arctic and especially in the sub-Arctic deep ocean. Arctic shelves experience a 2% increase while the Arctic deep and sub-Arctic shelves experience a 4% and 6% decline in bloom NPP as a proportion of total NPP, respectively. In the sub-Arctic deep ocean, bloom NPP dropped substantially, from 54% to 40% of total NPP by the end of the century. Relative to 1970, the proportion of NPP produced during the bloom is distinctly different across 6% of the Arctic Ocean in 2020 but 41% of the Arctic Ocean in 2100 (Table 2, Fig. S2).

Fig. 3: Percent annual NPP generated during the bloom and change in average daily NPP during the bloom.
figure 3

The mean proportion of annual net primary production (NPP) generated during the phytoplankton bloom in (a) 1970, (b) 2020, and (c) 2100, and the change in the average rate of bloom NPP between (d) 2020 and (e) 2100 relative to 1970. In a–c, the edge of the seasonal ice zone is highlighted in light green and the edge of the perennial ice zone is marked in purple. Stippling in d and e indicates regions where there is no significant change in the proportion of NPP generated in the bloom.

Full size image

We also evaluate bloom importance by considering daily mean NPP during the bloom. While the bloom is a more concentrated source of productivity in the central Arctic by 2100, mean bloom NPP declines in the N. Atlantic and in the continental shelf regions of the Arctic Ocean relative to 1970 rates. Between 1970 and 2020, daily rates of NPP during the bloom increase by  >100% in much of the central Arctic, with NPP increasing nearly five-fold in the waters north of Greenland (Fig. 3d). By 2100, daily rates of NPP are even greater than in 2020 in the central Arctic, with daily NPP increasing by 65% over the perennial sea ice zone of 1970 (Fig. 3e). However, daily productivity during the bloom declines by  > 25% in the Chukchi Sea, Baffin Bay, the southern Barents Sea, and in the N. Atlantic (Fig. 3e). With the exception of the N. Atlantic, these regions largely show an increase in the length of the bloom, indicating that a lengthening growing season does not always correspond to greater daily NPP.

We further quantify shifts in the species composition of the bloom, finding that Arctic phytoplankton communities are diatom-dominated in both the 1970 and 2020s but shift to mixed diatom and small phytoplankton dominance by the end of the century. While diatoms generate 87.6–90.1% of total NPP in the Arctic and sub-Arctic in 1970 and 2020, diatoms account for only 61.2% and 65.8% of total NPP in 2100 in the Arctic Ocean and Arctic and sub-Arctic, respectively (Fig. 4a, b). Conversely, small phytoplankton, which generate 9.8–12.4% of total NPP in 1970 and 2020, produce more than 34% of total NPP in the Arctic and 38.8% of NPP in the region  > 50°N by the end of the century (Fig. 4c, d). Small phytoplankton generate a far greater proportion of total NPP in Hudson Bay, the Beaufort Sea, southern Greenland, coastal Russia, and the north Atlantic, where they generate more than half of total NPP by the end of the century (Fig. 4d). Diazotrophs account for  <0.2% of total NPP in the Arctic and sub-Arctic in 1970, 2020, and 2100.

Fig. 4: Changes in light and nutrient limitation terms for diatoms and small phytoplankton.
figure 4

Maps of the proportion of total NPP generated by diatoms in (a) 1970 and (b) 2100 and by small phytoplantkon in (c) 1970 and (d) 2100, and of the change in (e–h) light limitation and (i–l) N limitation terms for (e, f, i, j) diatoms and (g, h, k, l) small phytoplankton between (e, g, i, k) 2020 and 1970 and (f, h, j, l) 2100 and 1970.

Full size image

Relationships between environmental drivers and bloom dynamics changes

To understand what environmental drivers control changes in bloom dynamics, we compare the dimensionless light and nutrient limitation terms (varying from 0-1, with limitation terms close to 0 indicating that growth is highly limited), which scale the rate of growth of phytoplankton within models. While there are only modest changes in phytoplankton light and N limitation terms between 1970 and 2020, large (and opposing) changes by the end of the century appear to allow small phytoplankton to outcompete diatoms in parts of the Arctic Ocean. Over the duration of the bloom in 1970, the mean diatom light limitation term is 0.29 in both the Arctic and sub-Arctic. While light limitation does not change substantially south of the Arctic circle, the mean diatom light limitation term during the bloom increases (indicating more rapid growth is possible) to 0.37 in 2020 and to 0.45 in 2100 within the Arctic Circle (Fig. 4e, f), driven by changes within the 1970 perennial sea ice zone (see Fig. 3a). Similarly, the mean diatom N limitation term during the bloom declines (limiting phytoplankton growth) from 0.79 in 1970 to 0.71 by 2020 and to 0.64 by 2100, with the most pronounced changes in the perennial sea ice zone (Fig. 4i, j). Small phytoplankton, which have a mean light limitation term of 0.16 during the bloom in the Arctic and sub-Arctic in 1970, see a modest increase in the light limitation term in 2020 (0.20), but by 2100, the light limitation term has more than doubled (0.43) within the Arctic Circle (Fig. 4g and h). N limitation for small phytoplankton follows a similar but opposing pattern, with the mean N limitation term dropping slightly from 0.94 in 1970 to 0.92 in 2020 and falling substantially, to 0.75, by the end of the century. Changes in both light and N limitation terms appear to be driven by changes across the seasonal sea ice zone in 1970 (see Fig. 3a).

Between 1970 and 2100, as sea ice cover diminishes in the Arctic Ocean, increases in surface ocean light lead to an increase in the average bloom light limitation term and appear to allow an earlier bloom start. In 1970, the Arctic Ocean experiences an average of 171 days with light above the compensation irradiance35, the light intensity at which phytoplankton photosynthesis equals respiration. While this period with light above the compensation irradiance threshold increases by 10 days by 2020, it increases by a total of 31 days (to 202 days per year) by 2100, indicating that phytoplankton growth can occur over a longer period. Although the change in bloom start timing is correlated with the change in the number of days of light above the compensation irradiance, bloom start timing is more closely correlated with the change in annually averaged sea ice cover between 1970 and 2100 (Table 3). Both of these light-related variables are far more weakly correlated to changes in bloom end timing (Table 3). Additionally, the proportion of the year each grid cell is sea ice covered and the number of days with light above the compensation irradiance are both closely correlated with the ensemble variability in both bloom start and end timing in 1970 and 2100 (Table 3).

Table 3 R2 values for the linear regressions between the change in (2100-1970) or variability of (1970 or 2100) environmental variables and bloom start and end timing
Full size table

During this same period, N inventories in the Arctic and sub-Arctic experience large changes. March surface ocean nitrate (NO3-) concentrations decline from 7.8 mmol m−3 in 1970 to 7.3 mmol m−3 in 2020 and drop to 4.3 mmol m−3 by 2100 across the region as a whole. While these changes are starkest in the sub-Arctic, with the sub-Arctic deep ocean seeing a 4.6 mmol m−3 change in NO3- by 2100, they result in very low end-of-century surface NO3- concentrations in the Arctic, where the shelves and deep ocean have mean March concentrations of 2.1 and 2.6 mmol m−3, respectively, compared to 4.6 and 7.1 mmol m−3 in the sub-Arctic shelves and deep ocean in 1970, respectively. This decline in surface N inventory could be driven by changes in the mixed layer depth (MLD) within the Arctic, with a shoaling of the mixed layer resulting in a reduction in nutrient entrainment into the surface ocean, or by changes in the advection of N into the Arctic from the sub-Arctic. Between 1970 and 2100, there are minimal changes to the Arctic MLD, with the mean shelf MLD deepening from 54 to 58 m on Arctic shelves and no significant change in the MLD in the Arctic deep ocean (100 m). However, in the sub-Arctic, the MLD shoals substantially, from 73 to 62 m on the shelves and from 360 to 120 m in the sub-Arctic deep ocean, implying that the decline in surface NO3- concentrations in the Arctic is driven by a decline in N transport into the region from the sub-Arctic, rather than a reduction in local N entrainment. While changes in surface NO3- concentrations are correlated with changes in bloom start timing and inter-ensemble variability in NO3- concentrations are an important control on bloom start timing variability in 2100 (Table 3), changes in N inventories appear to be not as important a driver of changes to bloom timing as changes in surface light conditions.

Case studies of shifts in bloom dynamics

Large spatial variability in the timing and importance of the phytoplankton bloom, dictated by both latitude and changes in sea ice conditions, motivates our investigation of changes at individual locations in the Arctic and sub-Arctic (Fig. 5a). We contrast the responses at four single grid cell locations, each of which are not representative of the entirety of the region they are located within but rather are used to illustrate the different possible responses of bloom dynamics to climate change across locations at different latitudes and with distinct historical sea ice conditions.

Fig. 5: Annual cycles in daily NPP for 1970, 2020, and 2100 at four locations.
figure 5

Maps of the (a) change in the length of the phytoplankton bloom from 1970 to 2100 and (b) change in the standard deviation (S.D.) in the length of the phytoplankton bloom between 1970 and 2100, with circles demonstrating the location of the featured grid cells. Mean (line) and S.D. (shaded region) daily NPP (g C m−2 d−1) in 1970 (blue), 2020 (purple) and 2100 (red) for individual grid cells in (c) the Bering Sea (61.0∘N, 178.2∘W), (d) the Chukchi Sea (72.5∘N, 171.0∘W), (e) Hudson Bay (58.9∘N, 84.4∘W), and (f) the central Arctic (85.9∘N, 70.6∘W), with mean bloom start and end (dots) and S.D. in bloom start and end (lines) for 1970 (blue), 2020 (purple), and 2100 (red) above.

Full size image

Because of differences in patterns of sea ice loss, locations in the Chukchi and eastern Bering Seas have very different changes in bloom dynamics. In our Bering Sea location (830 m deep), which is only occasionally covered by seasonal sea ice in 1970 and is entirely sea ice-free by 2020, there is an 8% increase in the mean light in the surface ocean by 2100 (Fig. S3a), yielding 16 additional days (336 in 1970, 352 in 2100) when light is greater than the phytoplankton compensation irradiance. Consequently, the phytoplankton bloom starts in late April, with no significant change in the mean start date between 1970, 2020 and 2100 (Fig. 5c) or in the proportion of total NPP generated by diatoms (>97%). However, the bloom ends on average in late July in 1970, mid-July in 2020, and in mid-June by 2100, as March NO3- concentrations drop from 13.6 to 11.9 mmol m−3 even as the March mixed layer deepens, from 117 to 133 m. Thus, by 2100, the eastern Bering Sea bloom is 38 days shorter and less variable in length than in 1970 (Fig. 5b, Fig. S4b). In contrast, our Chukchi Sea location (60 m deep) transitions from intermittent perennial sea ice coverage in 1970 (20% of ensemble members) to ice-free conditions in 28% of ensemble members in 2100. Consequently, there is a 160% increase in the mean light in the surface ocean (Fig. S3b) and a 61 day increase in the number of days with light above the compensation irradiance (169 d in 1970, 230 d in 2100). There, the phytoplankton bloom, which starts on average on June 22nd in 1970, shifts earlier by nearly two months to April 23rd by 2100 (Fig. 5d). However, there is substantial interannual variability in this start date in 2100, with a standard deviation (S.D.) of 23 days rather than 9 days in 1970 (Fig. 5d). While the end of the bloom shifts earlier by 6 days over the historical period and by 9 days by the end of the century, this much earlier bloom start yields a bloom that is 51 days longer, on average, in 2100 than in 1970 (Fig. 5a), although the length of the bloom is also more variable (Fig. 5b). While total NPP in the Chukchi Sea location is diatom-dominated in 1970 and 2020 (96–97%), diatoms only accounted for 77% of total NPP by 2100 as March surface NO3- concentrations decline from 4.1 to 3.4 to 1.1 mmol m-3 over this period. As a consequence of the shortened length of the bloom in the Bering Sea, there is a 33% decline in NPP generated during the bloom and a 8% reduction in annual NPP relative to 1970, resulting in a 19% decline in the proportion of NPP generated during the bloom. In the Chukchi Sea, however, there are increases in both bloom NPP (39%) and total NPP (25%) by the end of the century, leading to a bloom that is a slightly greater contributor to total NPP than in 1970. However, mean daily NPP during the bloom declines by 24% by 2100 as the bloom length increases.

In our Hudson Bay (130 m deep) and the central Arctic (1500 m deep) locations, both latitude and reduced sea ice cover influence changes in bloom dynamics. While our Hudson Bay location (59∘N) is always covered in seasonal sea ice in 1970 and 2020, by 2100 it is sea ice-free in almost all ensemble members (94%), yielding a 155% increase in light in the surface ocean (Figure S3c) and a three month increase in the number of days with light above the compensation irradiance (225 d in 1970, 349 d in 2100). Between 1970 and 2020, the bloom start shifts 12 days earlier and ends a week later, resulting in nearly three week longer bloom period (Fig. 5e). By the end of the century, the bloom starts 100 days earlier and ends in mid-September, more than three weeks earlier, leading to a bloom period 80 days longer than in 1970 (Fig. 5a). Despite a deepening mixed layer (which increases from 54 m to 83 m between 1970 and 2100), February and March surface NO3- concentrations are 50% lower in 2100 than in 1970, driving a shift in phytoplankton community composition from diatom-dominated (accounting for 76% of total NPP in 1970) to small phytoplankton-dominated (accounting for 86% of total NPP in 2100). In contrast, our central Arctic location (84∘N) transitions from always to usually (90% of ensemble members) experiencing perennial sea ice cover between 1970 and 2020, but is only covered in seasonal sea ice by 2100. While these changes increase mean light in the surface ocean by nearly 300% (Figure S3d), the high latitude of this location still limits annually integrated light to 24% of that of the Hudson Bay location, and only four additional weeks with light in the surface ocean above the compensation irradiance (122 d in 1970, 150 d in 2100). Consequently, the bloom start date, which occurs in early July in 1970 but is highly variable (S.D. of 14 d), shifts earlier by nearly one month and becomes far less variable by 2100 (S.D. of 4 d; Fig. 5f, Figure S4b). The bloom also ends eight days earlier, resulting in a bloom that is three weeks longer and less variable in length in 2100 than in 1970 (Fig. 5b). These changes in bloom timing are largely unrealized over the historical period, with no significant change in bloom start or length but a six day earlier bloom end between 1970 and 2020. Further, at this central Arctic location, community composition becomes slightly less diatom-dominated (with diatom NPP shifting from 82% to 75% of total NPP) as March surface NO3- concentrations decline (from 5.3 to 1.8 mmol m-3) between 1970 and 2100, despite a slight deepening of the mixed layer (76 to 83 m). Changes in the timing and length of the bloom have vastly different consequences on the bloom importance at these two locations. In Hudson Bay, annual and bloom NPP increase moderately (15–19%) over the historical period, but by 2100 both increase by  ~ 55%, yielding no change in the relative importance of the bloom but a slightly lower daily rate of NPP during the bloom (Fig. 3e). In contrast, in the central Arctic, both bloom and total annual NPP more than double by 2020 (Fig. 3d) despite minimal changes in the timing of the bloom. By the end of the century, these changes are even more exaggerated, as annual and bloom NPP tripled compared to 1970 (Fig. 3e). While the bloom did not substantially change in importance between 1970 and 2100, daily mean rates of NPP in 2020 and 2100 were more than double the rates of NPP in 1970.

Discussion

This analysis demonstrates that, while bloom timing and magnitude has changed over the historical period due to anthropogenic climate change, phytoplankton bloom dynamics over much of the Arctic and sub-Arctic will shift to a distinctly new state by the end of the century. In 2020 only 1% of the Arctic Ocean has emerged and has a distinct bloom start in 2020 relative to 1970, but by 2100 the bloom start has emerged and starts earlier over 71% of the Arctic. Consequently, the bloom starts more than a month earlier and lasts two weeks longer than in 1970, driven largely by changes in sea ice coverage. While our work is the first to separate internal climate variability from the externally forced climate change signal, our results reinforce previous studies, which find that sea ice loss in the 2000s and 2010s yielded changes in bloom timing. Satellite analysis24 found evidence that, between 1997 and 2009, phytoplankton blooms shifted earlier across 11% of the Arctic Ocean, and that regions with earlier blooms were located in the areas with the greatest sea ice loss. A modeling hindcast study (2006-2013 simulation)25 found that early sea ice melt in 2007 and 2012 triggered a spring phytoplankton bloom that started one month earlier and terminated earlier than in years with more extensive sea ice. Similarly, we find that changes in bloom start and end timing between 1970 and 2100 are greatest in areas where sea ice cover is lost, and changes in bloom start timing are more closely correlated to annual mean sea ice concentration and to changes in surface ocean light than to environmental factors related to nutrient inventories, indicating that sea ice loss continues to drive these changes at the end of the century.

We assess how the proportion of NPP produced during the bloom shifts between 1970 and 2100, finding that the bloom declines in importance in the sub-Arctic deep ocean, and is consequently declining in its relative importance in the Bering Sea and N. Atlantic and in the southern Chukchi and the Barents seas, two major inflow shelves into the Arctic Ocean. Between 1970 and 2020, we find that the bloom retains its importance on Arctic shelves but declines in importance in the central Arctic, a finding consistent with a modeling and remote sensing study investigating bloom timing and importance between 1998 and 201836. However, we find that these initial changes in bloom importance are reversed by the end of the century, with most annual NPP generated during the bloom in the central Arctic but a weakened bloom on Arctic inflow shelves. Previous modeling and remote sensing studies3,25,37 have found rising evidence of fall blooms in these same areas, indicating that annual NPP becomes less concentrated in the spring bloom in the early 21st century. Similarly, a remote sensing study19 found that inflow shelf regions showed the largest changes in both biomass and NPP between 1998 and 2018. There has been an increase in waters flowing into the Arctic through these gateways38,39 which can alter nutrient inventories directly or indirectly (by altering stratification and making nutrients more or less accessible to Arctic phytoplankton)36,40,41. Our results indicate that inflow shelves may continue to see NPP distributed across more of the year by the end of the century as nutrient inflows into the Arctic decline. However, there is substantial disagreement between Earth system models in future total annual NPP in the Arctic, driven largely by disagreement in future N inventory in the region42, indicating that this finding may not be consistent across other models. We also find that phytoplankton NPP by the end of the century is more dominated by small, flagellate phytoplankton than in 1970 or 2020, a finding similar to previous modeling and observational studies and often linked to reduced N inventories in the Arctic and associated with changes in food webs and carbon export43,44,45,46.

Overall, changes in the timing and magnitude of the phytoplankton bloom, while distinct spatially depending on latitude and sea ice loss, are likely to significantly impact Arctic ecosystems. Our work indicates that, by the end of the century, the bloom may represent a less concentrated source of productivity and thus continuous but lower food concentrations22 available to the species reliant on the bloom6,7,8,10. Shifts from diatoms to smaller flagellates are likely to also have substantial impacts on ecosystems and C export, as flagellates such as Phaeocystis spp. can form large, gelatinous matrices and can often remain ungrazed and disproportionately sink to the sediments44,47. The mismatch in timing between phytoplankton blooms and zooplankton production48 may be enhanced, which has been previously found to lead to suboptimal feeding conditions for zooplankton21. Changes in bloom timing and duration have been linked to changes in the partitioning of food between the pelagic and benthic ecosystems49 and even to an increase in fish recruitment failures by the end of the century27, indicating that these changes to the bloom will have consequences that ripple through Arctic ecosystems and could be particularly detrimental to the Indigenous communities that rely on hunted marine foods11.

Methods

Model simulations

We use fully coupled Community Earth System Model version 2 (CESM2) simulations for this analysis. CESM2 includes explicit atmosphere (Community Atmosphere Model version 6; CAM6), terrestrial (Community Land Model version 5; CLM5), ocean (Parallel Ocean Program version 2; POP2), sea ice (CICE version 5.1.2;50), and ocean biogeochemistry (Marine Biogeochemistry Library, or MARBL) model components run at 1∘ horizontal resolution51. We assess changes in biomass and NPP among the three phytoplankton functional types represented in MARBL (diatoms, diazotrophs, and picoplankton;52). MARBL also includes a generic zooplankton functional type and the global cycles of oxygen, C, and nutrients (including N, P, Si, and Fe). To represent the heterogeneity of sea ice, thickness is calculated on a sub-grid scale50, and light availability for photosynthesis by different phytoplankton functional types is adjusted based on the coverage of sea ice in each grid cell53.

Here, we use daily output from an ensemble of 50 simulations from the CESM2 large ensemble (CESM2-LE;31). These simulations are initialized in 1850 from a pre-industrial control and are subsequently run for historical (1850-2014) and future (2015-2100) scenario periods using historical and SSP3-7.0 forcing protocols from the 6th Coupled Model Intercomparison Project (CMIP;54). While the ensemble members have identical prescribed forcing, four initial ocean states are used to approximate the impact of Atlantic meridional overturning circulation phase on climate and air temperatures vary due to a roundoff level perturbation (10-14 K). Consequently, each member of the CESM2-LE is an equally likely realization of a future climate projection. Thus, the ensemble mean captures the response of the system to external forcing alone, while differences between ensemble members demonstrate the importance of internal climate variability32. We evaluate how the timing of the Arctic phytoplankton bloom has changed by the end of the century and over the observational period, from 1970-2020.

Analysis

To describe changes in bloom timing, we calculate phytoplankton bloom start and end dates for each modeled grid cell and for each ensemble member for 1970, 2020, and 2100. Several Arctic (> 66.5∘N) and sub-Arctic (50–66.5∘N) remote sensing studies define the start of the phytoplankton bloom as the date when a proportion (15–25%) of the maximum amplitude of a Gaussian curve fitted to Chlorophyll-a (Chl) observations is first reached55,56,57. Here, we define the start of the phytoplankton bloom as the date when water column-integrated C-based biomass first increases above 25% of the maximum biomass. We consider the bloom to end on the date when integrated C-based biomass first diminishes below 25% of the maximum following the peak in biomass, and consider the bloom length to stretch from the bloom start to the bloom end. While we rely on the 25% threshold in the main text of this paper, we also report bloom start, end, and length changes between 1970 and 2100 using a 15% threshold in Table S1 (comparable to Table 1). A 15% threshold leads to an earlier start and later end to the bloom in each year, but the changes over time are similar to our findings with a 25% threshold, with modest changes between 1970 and 2020 and a far earlier bloom start, slightly earlier bloom end, and longer bloom by the end of the century (Table S1). We also choose to report changes in bloom dynamics using standalone years in this paper, but in Figure S5 we show that the 1970-2100 linear trend in changes in bloom start, end, and length is remarkably similar to the difference between 2100 and 1970 in these bloom timing metrics (Fig. 2). Finally, we quantify changes in bloom timing metrics using water column-integrated C-based biomass, rather than modeled surface Chl. Because C:Chl ratios vary based on the light history of phytoplankton and are highly variable in the Arctic Ocean58, we believe depth-integrated C-based biomass to be the more relevant metric of bloom timing to regional biogeochemical cycling and to the ecosystem. However, we rely on surface Chl concentrations when evaluating modeled bloom dynamics timing against satellite Chl-derived bloom timing metrics (see section “Bloom timing evaluation”).

We use two metrics - the proportion of total NPP produced during the bloom and the average daily rate of productivity during the bloom - to evaluate changes in the importance and productivity of the bloom. To do this, we quantify annual NPP and ‘bloom NPP,’ whereby we integrate NPP generated between the bloom start and end date. We also evaluate changes in species composition by quantifying how the NPP produced by diatoms and small phytoplankton shifts over time.

We present changes in sea ice conditions, light in the surface ocean, and light and nutrient limitation terms to provide context about the environmental changes driving changes in bloom dynamics. Sea ice cover is classified as either seasonal or perennial based on monthly sea ice concentration in March (the month of the sea ice maximum) and September (the month of the sea ice minimum). A grid cell is considered to have perennial sea ice cover if the mean sea ice concentration is ≥15% in more than half of the 50 CESM2-LE ensemble members in both March and September. The grid cell instead has seasonal sea ice cover if the mean sea ice concentration is ≥15% in March but is  < 15% in September in more than half of the 50 CESM2-LE ensemble members. We also describe changes in light in the surface ocean and calculate the number of days in which shortwave radiation in the surface layer of the ocean exceeds a compensation irradiance for phytoplankton growth of 4.5 W m-235. Finally, we present changes in the light and nitrogen (N, the nutrient that most limits phytoplankton growth in the Arctic Ocean59) limitation terms for diatoms (the most abundant phytoplankton functional type in the Arctic in this model60) and small phytoplankton between 1970 and 2100. These limitation terms, which are available on a monthly basis from the CESM2-LE, vary from 0−1 and scale the rate of phytoplankton growth, such that limitation terms close to 0 result in very low growth. We take monthly mean light and N limitation terms (calculated for each functional type over the top 100 m and weighted by C-based biomass distribution) and compute the mean light and N limitation terms over the duration of the bloom for each grid cell and each ensemble member.

Changes in bloom timing and magnitude metrics and in environmental variables are described across several regions: within the Arctic Ocean (> 66.5∘N), in the Arctic and sub-Arctic (> 50∘N), and on shelves (< 1000 m deep) and in the deep (> 1000 m deep) Arctic (> 66.5∘N) and sub-Arctic (50-66.5∘N). We use paired t tests (using α=0.05) to determine whether the ensemble mean change between 1970, 2020, and 2100 is significant relative to the internal variability. To assess whether the climate states of 2020 and 2100 are significantly different from that of the 1970 reference year, we rely on time of emergence analysis34, whereby a new climate state is considered to have emerged if the ensemble mean exceeds that of the reference year by more than 2 S.D. for any given variable. All statistical analyses are conducted in Python version 3.10.12.

Model evaluation

Observational and reanalysis products have been used to validate CESM2 globally51,52 and in the Arctic specifically. CESM2-modeled sea surface temperatures51 as well as changes in sea ice extent and cloud cover from 1979-201461,62 compare well with observations. However, CESM2 is known to produce thinner, less extensive sea ice61 and less snow on ice63 than is typically observed, possibly resulting in an inaccurate representation of phytoplankton productivity in ice-covered waters. N concentrations compare favorably to observations in the Arctic Ocean and in most sub-Arctic regions, with the exception of the sub-Arctic Pacific, where modeled N concentrations are lower than observations52. While different satellite algorithms produce very different estimates of NPP in the Arctic Ocean and in sub-Arctic seas, modeled and observed seasonal cycles of NPP in the Arctic as well as regional patterns of productivity are in reasonable agreement. Although near-coastal NPP appears to be underestimated by the model60, satellites often overestimate NPP in coastal regions due to the high concentrations of colored dissolved organic matter in these waters64.

Bloom timing evaluation

In an effort to evaluate the ability of the CESM2-LE to capture the start and end of the phytoplankton bloom, we compute the start and end of the phytoplankton bloom in the CESM2-LE using the dates that surface Chl surpasses or diminishes below 25% of the maximum Chl concentration for the years 2005-2014. We compare this to the start and end of the phytoplankton bloom using daily Chl concentrations produced by an Arctic-specific ocean color algorithm65 for the years 2005-2014. For the year 2010, we evaluate the difference between the Chl-based and C-based bloom start and end timing across the large ensemble within the Arctic Ocean (> 66.5∘N). We find that the bloom starts at a similar time between the two methods (DOY 158 ± 2 for the C-based method and DOY 163 ± 2 for the Chl-based method). However, the bloom end date is earlier and more variable using the Chl-based method (DOY 203 ± 5) than the C-based method (DOY 225 ± 2), likely because phytoplankton biomass is more concentrated at depth, where nutrients are higher, by the end of the growing season.

A comparison of satellite-derived and modeled bloom start timing reveals that, while CESM2 broadly produces similar spatial patterns of bloom timing, modeled blooms typically start earlier in the year than satellite-observed ones, likely partially due to missing observational data. Between 2005 and 2014, a single representative ensemble member from the CESM2-LE has an average chlorophyll a (Chl)-derived bloom start date of June 10 (Figure S6a) across the Arctic Ocean (> 66.5∘N). In contrast, the mean satellite Chl-derived start date of the phytoplankton bloom is July 5 (Figure S6c). This almost month earlier modeled start date is particularly notable because a bloom start date could not be determined from satellite observations for the central Arctic, where sea ice cover prevents ocean color observations but where the bloom likely starts later in the year. The modeled bloom start date may be earlier in the model than in observations for a number of reasons. CESM2 is known to produce sea ice thinner than observations in the Arctic61, allowing for earlier phytoplankton growth as light availability increases earlier in the year. However, low solar zenith angle as well as cloud and sea ice cover result in a paucity of daily Chl satellite observations in the Arctic Ocean18. As a consequence, satellite-derived bloom start dates are likely to be an overestimate of the true bloom start date. The spatial variation in bloom timing is similar for both modeled and satellite-observed start dates, with earlier bloom starts in the Barents Sea, the southern Chukchi Sea, and coastal waters, and later bloom starts at higher latitude in the central Arctic.

While the modeled and satellite-derived end of the phytoplankton bloom are similar, a lack of satellite data again complicates our evaluation of model skill. The model produces a phytoplankton bloom that ends on average on July 18th (Figure S6b) between 2005 and 2014. Similarly, satellite observations indicate that the phytoplankton bloom ends on average on July 21st (Figure S6d). However, a lack of data means that for many pixels, especially those at higher latitudes, there is no observation of the end of the phytoplankton bloom. Despite the patchiness of satellite estimates, CESM2-derived bloom end date is in good spatial agreement with the observations, with an earlier bloom end date in the Barents Sea and central Baffin Bay and a later bloom end at higher latitudes.