Pan-Arctic analysis of the frequency of under-ice and marginal ice zone phytoplankton blooms, 2003–2021

Historically, phytoplankton in the Arctic Ocean were presumed to contribute to total annual net primary production (NPP) only after sea ice retreat, when shallow mixed layers and the alleviation of light limitation allowed blooms to form (Sakshaug and Skjoldal, 1989; Niebauer, 1991; Perrette et al., 2011). These marginal ice zone (MIZ) blooms, which could rapidly strip surface waters of macronutrients (Niebauer, 1991), accounted for up to 65% of annual NPP in the Barents Sea and on the Bering Shelf (Sakshaug, 2004). However, observations of a massive phytoplankton bloom generated under fully consolidated 1 m thick first-year ice (Arrigo et al., 2012) demonstrated that phytoplankton NPP could be substantial during the sea ice-covered portion of the year. These under-ice (UI) phytoplankton blooms have been observed in most of the Arctic and sub-Arctic seas (Arrigo et al., 2014; Lalande et al., 2014; Mundy et al., 2014; Assmy et al., 2017; Hill et al., 2018; Mayot et al., 2018; Oziel et al., 2019) since at least the late 1950s (Apollonio, 1959; Strass and Nöthig, 1996; Gosselin et al., 1997; Yager et al., 2001; Fortier et al., 2002; Mundy et al., 2009; Apollonio and Matrai, 2011). UI blooms can substantially contribute to total NPP, accounting for approximately 50% of total annual NPP in the Canadian Arctic (Mundy et al., 2009; Oziel et al., 2019) and in the Chukchi (Arrigo et al., 2014) and Greenland (Mayot et al., 2018) seas. Further, these UI blooms likely have large biogeochemical implications, including reducing zooplankton grazing (Payne et al., 2021), which in turn leads to an increase in carbon export flux and nitrogen loss through sedimentary denitrification (Payne and Arrigo, 2022).

Quantifying NPP during the UI bloom period, as well as identifying where in the Arctic Ocean these blooms are generated and how their frequency has changed over time, is critical to understanding how Arctic phytoplankton are responding to the rapid changes to the Arctic environment (Ardyna and Arrigo, 2020). A study modeling light penetration through sea ice (Horvat et al., 2017) suggested that, due to diminishing sea ice thickness across the Arctic (Kwok, 2018), UI light has substantially increased, possibly leading to an increase in UI bloom frequency over time. Modeling work by Clement Kinney et al. (2020) indicated that, in the central Arctic, half of all NPP is generated in partially or fully sea ice-covered waters and that regional NPP has increased by 20% since 1980. However, modeling work in the southern Chukchi Sea (Payne et al., 2022) and across the entirety of the Arctic Ocean (Jin et al., 2016) indicates that increasingly early sea ice loss has led to a reduction in the magnitude of NPP during the UI bloom period.

Because satellites cannot capture NPP generated under sea ice, remote sensing estimates can differ from observational estimates of NPP by an order of magnitude in areas with large UI blooms (Arrigo et al., 2014). Previous work by Lowry et al. (2014), however, used daily chlorophyll a (Chl) concentrations around the time of sea ice retreat to identify parts of the Chukchi Sea that likely hosted UI blooms between 1998 and 2012. Lowry et al. (2014) relied on a conceptual model that assumed that an increase in Chl concentration after sea ice retreat was evidence of the alleviation of light limitation and thus an MIZ bloom (yellow line in Figure 1). Chl concentrations that were initially high but diminished following sea ice retreat (due presumably to nutrient exhaustion during the UI bloom period) were evidence of a previous UI bloom (green line in Figure 1). Here, we present work extending the conceptual model of Lowry et al. (2014) from the Chukchi Sea to the entire seasonally ice-free Arctic Ocean (Figure 2A). We quantify how prevalent UI and MIZ blooms are across the Arctic, identify regions that support the highest proportion of UI blooms, and assess whether there has been a change in the prevalence of UI blooms over time.

Daily sea ice concentration was generated from the climate data record of sea ice concentration from passive microwave data (version 4.0; Meier et al., 2021) provided by the National Snow and Ice Data Center in the polar stereographic projection (25 km resolution) for 1998–2021. These data were subsequently regridded to 4 km resolution to allow comparison with Chl data.

Daily Chl concentrations (4 km resolution) were generated from satellite retrievals of remote sensing reflectance (R_RS) using the Arctic Ocean-specific algorithm of Lewis and Arrigo (2020). This algorithm, developed using 501 coincident in situ measurements from 25 cruises, better accounts for the unique bio-optical properties of Arctic waters, such as greater pigment packaging and higher chromophoric dissolved organic matter concentrations, than the standard empirical NASA-Chl algorithm (Matsuoka et al., 2007; Matsuoka et al., 2011; Ben Mustapha et al., 2012; Fichot et al., 2013; Huot et al., 2013; Matsuoka et al., 2017). Daily Level 3 binned R_RS data were acquired from the NASA Ocean Color website (https://oceancolor.gsfc.nasa.gov/) from MODIS-Aqua for the years 2003–2021 (Reprocessing R2018.0). Pixels outside of the Arctic Circle (<66.5°N) were excluded from analysis, as were pixels within 50 km of the coast.

For each year from 2003 to 2021, each pixel in the seasonally ice-free Arctic Ocean (Figure 2A) was classified as being ice-free or as experiencing either a UI or MIZ bloom based on Chl observations near the time of sea ice retreat. If sea ice concentration was below 10% on the date of maximum sea ice extent for the Arctic Ocean for that year, the pixel was considered ice-free. If sea ice concentration was greater than 10%, we calculated the date of sea ice retreat by finding the first date when sea ice concentration diminished below 10% for any given pixel (Figure 2B). We then computed the maximum Chl concentration over two separate periods in the weeks surrounding sea ice retreat. Period one extended from the first Chl observation at that pixel until 2 days after the date of sea ice retreat. Period two extended from 3 days to 21 days after sea ice retreat (Figure 1). Pixels were classified as supporting a UI bloom if the maximum Chl concentration during period one exceeded the maximum Chl concentration observed during period two. The pixel was classified as hosting an MIZ bloom if the maximum Chl concentration was higher during period two than period one (Figure 1). If there was not at least one Chl observation in each of these periods, the pixel was considered to have insufficient observations to classify it as hosting either an MIZ or UI bloom. To estimate the areal coverage of UI and MIZ blooms by region and for the seasonally ice-free Arctic Ocean as a whole, we assumed that the relative frequency of UI and MIZ blooms observed in each region would be maintained across the pixels with insufficient observations to be classified.

This technique of classifying pixels relies on a comparison of Chl concentration from a minimum of two Chl retrievals. When Chl concentrations for those retrievals are similar (<1 mg Chl m⁻³ difference between periods one and two), there is a potential to misclassify blooms. However, we found that the number of pixels where the difference between the Chl retrievals was <1 mg Chl m⁻³ was approximately equal for pixels classified as MIZ and UI blooms, indicating that misclassification should affect both blooms equally. We also tested the sensitivity of our analysis to the difference between Chl retrievals during periods one and two (see Section 2.3).

This analysis does not rely on a Chl threshold to classify Chl observations as UI or MIZ blooms. UI blooms can generate high biomass in surface waters, but due to nutrient depletion, surface Chl can diminish substantially by the time of sea ice retreat (Arrigo et al., 2014; Brown et al., 2015b), meaning that satellite-derived Chl concentrations in the days surrounding sea ice retreat may not be reflective of the original magnitude of the bloom. A previous study by Lowry et al. (2014) that relied upon a Chl threshold of 2.5 mg m⁻³ found that most pixels (59.9%) in the Chukchi Sea never reached the Chl threshold in the weeks surrounding sea ice retreat; consequently, we decided against applying a Chl threshold to identify UI blooms.

In addition to classifying the seasonally ice-free area that supported UI and MIZ blooms, we quantified the area that supported open water (OW) blooms. We considered a pixel to have generated an OW bloom if Chl exceeded 1 mg m⁻³ in the period from 22 days after ice retreat until the sea ice minimum extent day in any given year. Because UI blooms can substantially deplete the nutrients in surface waters prior to sea ice retreat, Chl concentrations at the time of sea ice retreat should not be compared to the maximum Chl concentration reached over the OW period. Consequently, we evaluated whether any given pixel supported an MIZ or UI bloom around the time of sea ice retreat and separately evaluated whether it supported an OW bloom later in the year (>22 days after retreat).

Following classification (Figure 2C and D), pixels in each category were summed spatially over different areas of interest, including the entire seasonally ice-free Arctic Ocean, 10 geographic regions (Figure 2A), 5° latitudinal bands, and within geographic regions by 5° latitudinal band. We used linear regression to assess trends in the absolute area or proportion of total area classified as one of the three categories between 2003 and 2021. A significance threshold of p < 0.05 was used for the statistical tests presented in this study. All analyses were conducted in Python version 3.8.8.

We evaluated how sensitive our analysis was to several parameters: the sea ice retreat concentration threshold (10% in the primary analysis), the length of time between ice retreat and the end of period one (2 days in the primary analysis), and the difference between the maximum Chl retrievals during periods one and two (0 mg Chl m⁻³ in the primary analysis). We ran our bloom classification analysis using four different sea ice retreat concentration thresholds (50%, 25%, 15%, and 10%) combined with three distinct definitions of the end of period one (ice retreat plus 2 days, 4 days, and 6 days). Additionally, we tested the sensitivity of our analysis to the difference between the maximum Chl retrievals during period one and two. While our primary analysis allowed us to classify UI and MIZ blooms even if the maximum Chl retrievals during periods one and two were virtually identical, this sensitivity analysis required Chl concentrations to differ by at least 0.05, 0.10, or 0.20 mg m⁻³ in order to be classified. For each of these 15 sensitivity runs, we compared the average areal coverage and trends over the time series across the Arctic for the UI and MIZ bloom categories.

Excluding areas within 50 km of the coastline, the Arctic Ocean covers an area of 1.1 × 10⁷ km². Between 2003 and 2021, we analyzed a mean of 7.4 × 10⁶ km² of the Arctic Ocean because the remainder was covered in sea ice year-round. As sea ice retreated over a larger area over time, our area of analysis increased by 1.1 × 10⁵ km² yr⁻¹ (R² = 0.378, p = 0.005). However, the presence of sea ice and cloud cover prevented the analysis of an average of 3.9 × 10⁶ km² (black area in Figure 2C). Consequently, our analysis was based on the 3.5 × 10⁶ km² of the Arctic Ocean that was classifiable as being ice-free or as supporting UI or MIZ blooms, and the size of this classifiable region did not change between 2003 and 2021 across the Arctic Ocean as a whole. Each year, an average of 2.2 × 10⁶ km² of the Arctic Ocean was classified as being ice-free (or mean and standard deviation, SD, of 63.7% ± 3.0%, n = 19, of the classifiable area; Figures 3 and 4C), and this area increased over time by 1.6 × 10⁴ km² yr⁻¹ (R² = 0.212, p = 0.047). A further 8.1 × 10⁵ km² of seasonally ice-free Arctic surface waters experienced MIZ blooms (covering a mean and SD of 23.%1 ± 2.0%, n = 19, of the classifiable area; Figure 4B), and these blooms did not change significantly by proportion or areal coverage between 2003 and 2021 (Figure 3). Finally, 4.6 × 10⁵ km² of the seasonally ice-free Arctic Ocean supported UI blooms (covering a mean and SD of 13.3% ± 1.9%, n = 19, of the classifiable area; Figure 4A). While there was no significant change in the area supporting a UI bloom in any year, there was a 0.2% yr⁻¹ decline in the relative proportion of UI bloom area as a function of total classifiable area between 2003 and 2021 (R² = 0.260, p = 0.026; Figure 3). On average, MIZ and UI blooms covered 63.6% and 36.4%, respectively, of the classifiable seasonally ice-free portion of the Arctic Ocean. UI blooms also declined proportionally within the classifiable seasonally ice-free Arctic, declining from 40.9% coverage in 2003 to 30.9% by 2021 (−0.4% yr⁻¹, R² = 0.284, p = 0.019).

Each seasonally ice-free pixel was evaluated separately to determine if it supported an OW bloom. Between 2003 and 2021, OW blooms covered an average of 2.5 × 10⁶ km² (Figure 4D) and their areal coverage increased by 5.2 × 10⁴ km² yr⁻¹ (R² = 0.752, p < 0.001). OW blooms were rarely generated in areas that had insufficient observations to be classified as supporting UI or MIZ blooms during the sea ice retreat period, covering only 12.0% of the insufficiently observed area. Regions that supported UI and MIZ blooms infrequently went on to generate OW blooms, which covered a mere 20.4% of the area that was classified as supporting a UI bloom and only 25.9% of the area classified as supporting an MIZ bloom. However, OW blooms were commonly generated in regions that were always ice-free and covered 77.7% of the ice-free region, on average. OW blooms were also frequently observed along the shelf-break, indicating the possible role of nutrient upwelling at the shelf break in providing nutrients for these blooms (Figure 4D).

At the most southerly latitudinal band of 66.5°N–70°N (9.0 × 10⁵ km² classifiable area), ice-free waters were most common, covering 75.1% of the total classifiable area (Figure 5A), and this ice-free area expanded by 5.0 × 10⁴ km² between 2003 and 2021 (Table 1). For the seasonally ice-covered part of this latitudinal band, sea ice retreated on average (± SD) on day of year (DOY) 163 ± 6 days (n = 19, mid-June). Between 2003 and 2021, the average date of sea ice retreat shifted earlier by 0.56 d yr⁻¹ (R² = 238, p = 0.034), while the area that experienced seasonal sea ice retreat diminished by 4.3 × 10³ km² yr⁻¹ (R² = 0.472, p = 0.001) as the ice-free area expanded. UI blooms accounted for an average of 10.9% of the classifiable area in this latitudinal band, but these blooms declined by both proportion and area (Tables 1 and 2). MIZ blooms, which covered 14.0% of the classifiable area between 66.5°N and 70.0°N, did not change significantly between 2003 and 2021.

For the 70°N–75°N latitudinal band (1.8 × 10⁶ km² classifiable area, or twice the size of the next largest latitudinal band), most of the classifiable area (66.0%) was ice-free (Figure 5B). Between 2003 and 2021, ice-free area expanded by 1.4 × 10⁵ km² (Table 1). For the seasonally ice-free pixels in this latitudinal band, sea ice retreated on average on DOY 194 ± 5 days (n = 19, mid-July) but neither the timing of sea ice retreat nor the area over which sea ice retreated changed significantly. MIZ and UI blooms were proportionally higher than to the south, covering 21.2% and 12.8% of the classifiable area, respectively. Over this latitudinal band, ice-free area increased its areal coverage between 2003 and 2021 (Table 1).

Between 75°N and 80°N (7.5 × 10⁵ km² classifiable area), ice-free area still dominated, covering 47.5% of the classifiable area (Figure 5C). Sea ice retreated on average on DOY 190 ± 9 days (n = 19, mid-July) for the seasonally ice-free area in this latitude band, and the seasonally ice-free area expanded by 4.8 × 10⁴ km² yr⁻¹ (R² = 0.648, p < 0.001). MIZ and UI blooms covered a larger proportion of the classifiable area than at lower latitudes, accounting for 36.5% and 16.0% of classifiable area, respectively. An increase in classifiable area in this latitudinal band was driven by an increase in MIZ blooms (Table 1).

The northernmost latitudinal band, 80°N–85°N, had a far smaller classifiable area (only 5.2 × 10⁴ km² on average) than more southerly latitudes due to greater sea ice coverage. The classifiable area nearly doubled over time at this latitudinal band (Table 1) as sea ice retreated. For the seasonally ice-covered portion of this latitudinal band, sea ice retreated on average on DOY 213 ± 19 days (n = 19, early August), and this area expanded by 2.7 × 10⁴ km² yr⁻¹ (R² = 0.490, p < 0.001). MIZ blooms, which on average covered 49.6% of classifiable pixels across this latitudinal band, were generated across the largest area (Figure 5D). UI blooms covered an average of 36.5% of the classifiable area, while 13.9% of the classifiable area was ice-free.

Most of the central Arctic (89.8%) had insufficient Chl observations to be classified as supporting either MIZ or UI blooms between 2003 and 2021. However, UI blooms covered 34.0% of the area classifiable as supporting MIZ or UI blooms (Figure 5). The central Arctic had a moderate increase in classifiable area in the 75°N–80°N latitudinal band, which was driven largely by an increase in MIZ bloom area, although UI bloom area also increased (Table 3).

Baffin Bay had the second highest mean areal coverage of both MIZ blooms and UI blooms (Figure 5F, Table 3). However, while UI blooms covered nearly 40% of the classifiable area of Baffin Bay (Figure 5F), UI blooms declined by 0.5% yr⁻¹ between 2003 and 2021 as a proportion of total classifiable area and were replaced by MIZ blooms (Table 2). These changes were driven largely by a decline in UI bloom area in the 75°N–80°N latitudinal band (Table 1).

While the Canadian Arctic Archipelago had the smallest classifiable area, the region had the second highest proportion of UI bloom area (45.3% of classifiable area; Figure 5G). There were no significant trends in UI or MIZ bloom area or proportion over time.

UI blooms were most prevalent in the Chukchi Sea, accounting for the highest proportion of classifiable area (55.5%; Figure 5I) and covering both the largest area and second largest estimated area of all the regions of the Arctic Ocean (Figure 4A, Table 3). However, the region also saw large-scale declines in UI bloom areal extent, which fell by nearly 50% from 1.3 × 10⁵ km² in 2003 to 6.6 × 10⁴ km² by 2021 (Table 1). UI blooms declined and MIZ blooms expanded as a proportion of classifiable area at a rate of 1.1% yr⁻¹ (Table 2). This change in UI bloom proportion was driven by a reduction in UI area in the 66.5°N–70.0°N latitudinal band (Table 2).

In the Laptev Sea (Figure 5K), UI blooms covered less of the classifiable seasonally ice-free area than any other regions (25.0%; Table 3). Further, in the southernmost latitudinal band (70°N–75°N), UI blooms declined and were replaced by MIZ blooms (0.7% yr⁻¹; Table 2). An increase in the classifiable area from 75°N–80°N was driven by an increase in MIZ bloom area (Table 1).

MIZ blooms were prevalent in the Kara Sea, covering 71.8% of the classifiable part of the region (Figures 4B and 5L, Table 3). In the 80°N–85°N latitudinal band, a small expansion of the classifiable area led to an expansion of both MIZ and UI bloom area (Table 1).

The Barents and Nordic seas were dominated by ice-free area (89.0% and 92.8% of classifiable area, respectively; Figures 4C, 5M and N, Table 3). While only observed over a small percentage of classifiable area, UI and MIZ blooms still had sizeable areal coverage in the Barents Sea (Table 3). In the Nordic Sea, ice-free area expanded substantially during the period 2003–2021, driven relatively equally by changes from 66.5°N–70°N and 70°N–75°N (Table 1). This expansion of ice-free area drove declines in both UI and MIZ blooms in the southern latitudinal band (66.5°N–70°N) and in MIZ blooms in the northern latitudinal band (70°N–75°N) over this time series (Tables 1 and 2).

We first tested the sensitivity of our analysis to two parameters in combination: the threshold for sea ice concentration (%) where we considered sea ice to have retreated, and the number of days between sea ice retreat and the end of period one. Ice retreat thresholds of 10%–15% resulted in almost no change in the areal coverage of UI and MIZ blooms, nor in the trends in UI area from 2003 to 2021 (Table 4). However, ice retreat thresholds of 25% and 50% resulted in insufficient observations for most pixels to be classified, reducing the classifiable area in the seasonally sea ice-covered Arctic Ocean. This approach also drove a 6%–50% reduction in UI and MIZ area. Due to the reduced proportion of analyzable area classified as supporting UI or MIZ blooms, there were no trends in the proportion of UI bloom area at the 25% or 50% ice retreat threshold, but there was an increasing trend in MIZ areal coverage over time (Table 4). Our analysis was also relatively insensitive to changes in the length of time between ice retreat and the end of period one. Added days disproportionately increased UI bloom area but had little effect on trends in UI or MIZ blooms (Table 4).

We also evaluated the sensitivity of our analysis to the difference between the maximum Chl retrievals during periods one and two. As the difference between Chl retrievals increased, there was a decline in the analyzable area, reducing the areal coverage of UI and MIZ blooms by 20%–60% (Table 4). Across all the Chl difference thresholds that we evaluated, there was a declining trend in the proportion of the classifiable area that supported UI blooms (Table 4). When the difference between the Chl retreivals increased above 0.05, there was also a statistically significant increase in MIZ bloom areal coverage across the Arctic (Table 4).

In the Arctic Ocean, water column NPP is partitioned between three distinct periods defined by sea ice cover—the UI, MIZ, and OW periods. While NPP during the UI period was considered negligible historically (Hameedi, 1978; Perrette et al., 2011), observations of massive UI phytoplankton blooms in recent years (Mundy et al., 2009; Arrigo et al., 2014; Mayot et al., 2018; Oziel et al., 2019) demonstrate that this period can be highly productive. Changes in sea ice conditions, particularly the thinning of sea ice and the proliferation of melt ponds on the surface of the ice (Webster et al., 2015; Kwok, 2018) have changed UI light availability substantially (Light et al., 2015), possibly increasing areal coverage (Horvat et al., 2017) or productivity (Zhang et al., 2015) of UI blooms. The MIZ period, on the other hand, has long been considered responsible for the largest pulse of NPP in the Arctic Ocean (Sakshaug and Skjoldal, 1989; Niebauer, 1991; Sakshaug, 2004; Perrette et al., 2011). Following the MIZ and UI periods, nutrients are stripped from the surface ocean and NPP during the OW period typically is concentrated in a subsurface Chl maximum (Hill and Cota, 2005; Martin et al., 2010; Arrigo et al., 2014; Brown et al., 2015a). The OW period may be gaining in importance; increases in NPP across the Arctic Ocean appear to be driven in part by increases in the length of the OW period (Lewis et al., 2020). Our analysis indicated that the area of the seasonally ice-free Arctic Ocean supporting an OW bloom of at least 1 mg Chl m⁻³ increased substantially between 2003 and 2021. However, our analysis, which seeks primarily to understand the relative frequency of UI and MIZ blooms, uses Chl observations from around the time of sea ice retreat that may not represent the maximum Chl concentration achieved during the UI period. Consequently, we could not directly compare the Chl concentrations of UI, MIZ, and OW blooms. Our analysis was further limited by sea ice and cloud cover, which prevented the classification of nearly three quarters of the seasonally ice-free Arctic Ocean (3.9 × 10⁶ km²) in any given year.

UI phytoplankton blooms have been observed throughout the Arctic Ocean (Figure 4A; Ardyna et al., 2020). Our analysis found evidence of UI blooms over an area of 0.46 million km²; by using the frequency of UI blooms for each region, we estimate that a total of 1.9 million km², or nearly 40% of the seasonally ice-free Arctic Ocean, may have supported UI blooms between 2003 and 2021. We found that the Chukchi Sea had the largest area and highest proportion of UI blooms of any region of the Arctic Ocean. This finding is perhaps not surprising given that the UI blooms observed in the Chukchi Sea have been among the most productive blooms ever observed (Arrigo et al., 2012), as well as the relatively large number of years (1998, 2011, 2013, 2014, 2018) where in situ studies found evidence of UI blooms (Yager et al., 2001; Arrigo et al., 2012; Hill et al., 2018; Lowry et al., 2018; Lalande et al., 2020). Two previous studies assessed UI bloom prevalence across the Chukchi Sea. Using remote sensing and a minimum Chl threshold of 2.5 mg Chl m⁻³, Lowry et al. (2014) found that 72% of the area of the Chukchi Sea showed evidence of producing UI blooms in any given year between 1998 and 2012. A modeling study by Zhang et al. (2015) found that UI blooms were generated across 46% of the Chukchi Sea from 1988 to 2013. Our analysis indicated that an average of 56% of the classifiable area in the Chukchi Sea hosted UI blooms between 2003 and 2021, in line with previous estimates.

The relative ubiquity of UI blooms across the Arctic Ocean in our analysis stands in contrast to the findings of Perrette et al. (2011), who used similar satellite-derived estimates of Chl to assess the importance of MIZ blooms. However, the analysis of Perrette et al. (2011) was published in the same year that massive UI blooms were first observed in the Chukchi Sea (Arrigo et al., 2012) and as a result did not differentiate between UI and MIZ blooms. Similar to Perrette et al. (2011), our analysis found that MIZ blooms are widespread, with evidence that they covered an area of 0.80 million km² and a total estimated extent of 3.2 million km² of the seasonally sea-ice free Arctic Ocean each year from 2003 to 2021. We found that MIZ blooms are likely generated across 63% of the seasonally ice-free Arctic Ocean. MIZ blooms were revealed to be particularly widespread in Baffin Bay and in the Kara and Laptev seas, where they accounted for >60% of the classifiable, seasonally ice-free area.

We also evaluated how UI and MIZ blooms in the seasonally ice-free Arctic Ocean changed between 2003 and 2021. On a pan-Arctic scale, we found that the areal coverage of UI blooms did not change substantially, but UI blooms diminished in their proportional coverage of the seasonally ice-free Arctic Ocean by 7% between 2003 and 2021 as the ice-free area of the Arctic Ocean increased. While our analysis cannot be directly related to studies that quantify NPP during the UI period, these findings are qualitatively similar to a previous model intercomparison study by Jin et al. (2016), who found that UI blooms declined in their magnitude across the Arctic from 1980 to 2009. Jin et al. (2016) also found that declines in NPP during the UI period coincided with a substantial 3.2 to 8.0 Tg C yr⁻¹ increase in annual NPP across the Arctic Ocean between 1980 and 2009. A similar rate of increase in NPP across the Arctic was observed in a remote sensing study by Lewis et al. (2020), who found that annual NPP increased by 6.8 Tg C yr⁻¹ between 1998 to 2018. While changes in NPP during the UI and MIZ periods were not evaluated by Lewis et al. (2020), both of these studies found that increases in total NPP were attributable to an increase in the NPP generated during the OW period.

The pan-Arctic decline in the proportion of UI bloom area in our analysis appeared to be driven by both early sea ice loss at lower latitudes and local processes. Jin et al. (2016) hypothesized that, although thinner sea ice cover and an increase in first year ice has increased the light penetration and thus increased UI NPP, reductions in sea ice cover and earlier sea ice retreat have driven an overall reduction in the areal coverage of large UI blooms. One might expect that a reduction in the relative importance of UI blooms might be driven by changes in UI bloom frequency at lower latitudes, where sea ice loss happens far earlier in the year. We found, on a pan-Arctic scale, that only the southernmost latitudinal band (66.5°N–70.0°N) experienced a decline in UI bloom areal extent and proportional coverage. On average, sea ice retreated 10 days earlier in 2021 than 2003 across this latitudinal band, and sea ice retreated nearly a month earlier than for more northerly latitudinal bands. Further, our analysis also found that changes in UI bloom coverage were driven by changes in two regions—the Chukchi Sea (–1.1% yr⁻¹) and Baffin Bay (–0.5% yr⁻¹)—that had the highest areal coverage of UI blooms. This declining trend for UI blooms in the Chukchi Sea contrasts with a modeling study by Zhang et al. (2015) who found that UI blooms increased in areal coverage by 2% yr⁻¹ between 1988 and 2013. However, two other Chukchi Sea-based studies found evidence of the waning importance of UI blooms in the Chukchi Sea. Lowry et al. (2014) found an increase in MIZ blooms (which increased from 13.8% during 1998–2000 to 36.0% during 2010–2012) and a decline in “probable” UI blooms in the Chukchi Sea from 1998–2012. Similarly, a one-dimensional model implemented in the southern Chukchi Sea by Payne et al. (2022) found a large reduction in the magnitude of UI NPP between 1988 and 2018 (with UI blooms generating less than 34% of the NPP between 2014 and 2018 that they did between 1988 and 1992). They attributed this reduction in the importance of UI blooms to earlier sea ice retreat and a shorter UI period at their model location (Payne et al., 2022). Further, they found that a reduction in NPP during the UI period did not affect total NPP in the southern Chukchi Sea, which was 22% greater than in the northern Chukchi Sea (Payne et al., 2022). This higher rate of annual NPP was driven most substantially by higher NPP during the OW period, reminiscent of the findings of Jin et al. (2016).

Although UI blooms may be diminishing in their importance as a contributor to total NPP, our analysis indicates that presently nearly 40% of the seasonally ice-free Arctic may support these blooms. Because UI blooms can contribute substantially to local annual NPP (Mundy et al., 2009; Arrigo et al., 2014; Mayot et al., 2018; Oziel et al., 2019), analyses such as ours can be used to identify regions where satellite-based estimates of NPP may represent a substantial underestimate of annual NPP because of the unquantified contributions of UI blooms. Further, UI blooms have been found to have important ecosystem consequences, including reducing zooplankton grazing (Payne et al., 2021), altering the partitioning of organic matter between benthic and pelagic ecosystems (Arrigo et al., 2014), and affecting nitrogen cycling, particularly by changing the rate of sedimentary denitrification (Payne and Arrigo, 2022). Critical next steps include leveraging new technologies, such as moorings, floats, and autonomous underwater vehicles, to better quantify the importance of UI productivity in relation to total annual NPP, and increasing in situ studies of the impacts of UI productivity on food availability to upper trophic organisms, as well as changes in biogeochemical cycling that may result in areas with UI blooms.

The scripts and output files used to generate this work can be found at https://doi.org/10.25740/xw652dc8052.

The authors would like to thank Matt Mills, Stephanie Lim, Claudette Proctor, and James Lauer for their feedback on earlier drafts of this work.

Courtney Payne was funded by a fellowship from the ARCS (Achieving Rewards for College Scientists) Foundation.

The authors have no competing interests, as defined by Elementa, that might be perceived to influence the research presented in this manuscript. Kevin R. Arrigo is an associate editor at Elementa and was not involved in the review process of this article.

Contributed to conception and design: CMP, KRA.

Contributed to acquisition of data: CMP, GLvD.

Contributed to analysis and interpretation of data: CMP, GLvD, KRA.

Drafted and/or revised the article: CMP, GLvD, KRA.

Approved the submission: CMP, GLvD, KRA.

How to cite this article: Payne, CM, van Dijken, GL, Arrigo, KR. 2024. Pan-Arctic analysis of the frequency of under-ice and marginal ice zone phytoplankton blooms, 2003–2021. Elementa: Science of the Anthropocene 12(1). DOI: https://doi.org/10.1525/elementa.2023.00076

Domain Editor-in-Chief: Jody W. Deming, University of Washington, Seattle, WA, USA

Associate Editor: Jean-Éric Tremblay, Department of Biology, Université Laval, Québec, Canada

Knowledge Domain: Ocean Science