We selected four study regions distributed along a longitudinal gradient of the subtropical and tropical South Pacific Ocean. These regions were sampled as part of the basin-scale survey conducted during the Tara Pacific expedition 2016–2018;. The selected regions span a gradient of oceanographic conditions, ranging from the ultra-oligotrophic eastern basin (near Rapa Nui) to the moderately oligotrophic Western Pacific Basin (near Fiji and Tonga). Island densities, size, and geomorphic types vary between and within these regions (Fig. ). For instance, the region studied around Rapa Nui in the Eastern Pacific encompasses 2 islands and 3 submerged reefs, while the region studied around Fiji and Tonga in the Western Pacific encompasses 158 islands and 306 submerged reefs.

The spatial and temporal dynamics of IMEs and their associated chlorophyll a enhancement have been characterized for each of these case study regions . Here, we extend this analysis to include physiological indicators derived from bio-optical and remote sensing measurements to gain new insights about the underlying mechanisms that contribute to the observed chlorophyll a enhancement.

We measured inherent optical properties (IOPs), including absorption (a), attenuation (c), and backscattering (bb), along the Tara Pacific transect, using Seabird's ACs spectrophotometer and ECO-BB3 integrated into a continuous flow-through system. We calculated particulate absorption (ap), particulate attenuation (cp), and particulate backscattering (bbp) by correcting a, c, and bb for instrument drift, bio-fouling, and the influence of dissolved matter, based on hourly measurements of 0.2 μm filtered seawater . We derived bio-optical proxies for phytoplankton biomass and community composition from ap and cp, benefiting from the high sampling resolution of the continuous sampling system to detect gradients in optical properties along the sampling transect. We calculated the line-height of the ap peak of [Chl a] at 676 nm (ap676LH) and decomposed all ap spectra into Gaussian functions aligned with the light absorption peaks of specific accessory pigments . Each day, around 10:30 a.m. local time, we collected surface water samples for pigment analysis via high-pressure liquid chromatography (HPLC). We estimated [Chl a] from ap by correlating total [Chl a] measured via HPLC with the amplitude of ap676LH (Fig. ). Similarly, we estimated accessory pigment concentrations from continuous underway measurements by relating concurrent pigment concentrations derived from HPLC to the amplitudes of ap Gaussian functions (Fig. ). Photo-protective carotenoid concentrations ([PPC]) were estimated from the Gaussian function centered on 492 nm, photosynthetic carotenoid concentrations ([PSC]) were estimated from the Gaussian function centered on 523 nm, chlorophyll c concentrations ([Chl c]) were estimated from the Gaussian function centered on 638 nm, and chlorophyll b concentrations ([Chl b]) were estimated from the Gaussian function centered on 660 nm (Table ). We investigated changes in pigment composition along the Tara Pacific transect sampled using the ratio of these accessory pigments to [Chl a] (e.g. [PSC]/[Chla]; Table ). Phycoerythrin concentrations were not measured as part of the HPLC analysis, thus we estimated the relative concentrations of phycoerythrin using the ratio of the Gaussian absorption function centered on the phycoerythrin absorption peak i.e. 550 nm; to the Gaussian absorption function centered on 676 nm, the [Chl a] absorption peaks (i.e. agauss550/agauss676). Although not a central objective of this study, we tracked variations in ratios of these accessory pigments to [Chl a] (e.g. [PSC]/[Chla]; Table 2) as a crude indicator of changes in phytoplankton community composition. To quantify changes in the mean size of suspended particles, we computed the slope exponent of cp i.e. γcp which is inversely proportional to mean particle size;. Additionally, we measured sea surface temperature and salinity (SSS) with a Seabird SBE45 thermo-salinograph integrated into the flow-through system. We also continuously recorded above-water instantaneous photosynthetically active radiation (iPAR) from the aft deck of Tara, using a cosine PAR sensor (QCP2150; Biospherical Instruments) positioned 5 m above sea level. All variables derived from in situ continuous underway data are described in Tables  and and are publicly available . Finally, we collected daily discrete samples of surface seawater to estimate the concentrations of dissolved inorganic nitrogen (i.e., nitrates and nitrites), phosphate, silicate, and total iron see sampling protocols in.

We computed level-2 (L2) satellite products from the Moderate Resolution Imaging Spectroradiometer (MODIS), the Visible Infrared Imaging Radiometer Suite (VIIRS), and the Ocean and Land Colour Imager (OLCI) level-1 data following method; we built two sets of satellite data, a “calibration dataset” and a “study dataset”. We downloaded the “calibration dataset” for the entire area covered by the Tara Pacific transect May 2016 to October 2018, see. The “study dataset” consisted of four six-month-long sequences of satellite images in the vicinity of the islands of interest (see Sect.  for additional information on temporal coverage and resolution). We applied a polynomial-based atmospheric correction POLYMER version v4.17beta2; on both datasets to compute 1 km spatial resolution L2 remote sensing reflectance data (Rrs) using ancillary data from the European Centre for Medium-Range Weather Forecasts reanalysis model version 5 (i.e. ERA5). We removed poor-quality data pixels by applying the flag recommendations of POLYMER see reference and projected each satellite scene onto an equally spaced 1 km spatial resolution plate-carré reference grid using a nearest-neighbor interpolation from Python's SciPy library. We corrected Rrs for Raman scattering and derived particulate back-scattering coefficient bbp at 430 and 550 nm using the quasi-analytical algorithm V6 . We estimated [Chl a] using the blended CI-OCx algorithm applying the coefficients derived for each sensor . We estimated phytoplankton carbon (Cphyto) using the empirical relationship between bbp(470 nm) and Cphyto . Following , we computed surface-area integrated [Chl a] as a proxy for surface phytoplankton biomass integrated over entire IME and background ocean (BO) zones (see Sect.  for IME and BO zones definition), in two-dimensional metric tons of chlorophyll a (t m⁻¹), by summing the [Chl a] of each pixel within IME and BO zones multiplied by the area of that pixel: 1∑[Chla]IME=∑n=1NpixelIME[Chla]n×areapixeln All variables derived from satellite data are described in Table .

We computed two biomass-independent phytoplankton nutrient-stress indices. The fluorescence quantum yield (ΦSat) is indicative of the physiological stress of phytoplankton cells due to iron limitation (higher ΦSat = more iron stress) and was computed using intermediate products of [Chl a], normalized fluorescence line height (nFLH), and iPAR . iPAR was obtained from the output of the function l2gen of SeaDAS , and the computation of nFLH is detailed in Sect.  in the Appendix.

raised concerns about the limitations of using MODIS-Aqua data to compute ΦSat and infer phytoplankton iron stress. MODIS-Aqua overpass time (∼ 13:30 local at the equator) coincides with the time of the day when non-photochemical quenching (NPQ) is maximal, significantly reducing the signal-to-noise ratio of chlorophyll fluorescence after NPQ normalization (that is, FChl a_NPQ) and, consequently, ΦSat. In the present study, we also computed ΦSat from MODIS-Terra and OLCI Sentinel-3a, which sample the equator around 10 and 10:30 local time when the NPQ is likely different from the NPQ at the time of MODIS-Aqua overpass. Because the final product ΦSat is merged from MODIS-Aqua, MODIS-Terra, and OLCI Sentinel-3a, it benefits from an improved signal-to-noise ratio due to the combination of measurements taken at different times of the day. The quality of ΦSat computed here also benefited from the use of the POLYMER atmospheric correction scheme, which improves data retrieval around clouds and in sun-glint conditions. This is particularly important because the intermediate product of nFLH often shows artificially high values near clouds, likely due to the adjacency effect.

We derived a second physiological proxy that is indicative of phytoplankton stress due to macronutrient limitation. When incident light decreases, phytoplankton cells maintain their growth rate by upregulating chlorophyll pigment synthesis to compensate for the decreased solar energy available . In contrast, phytoplankton cells under macronutrient stress downregulate chlorophyll pigments synthesis . The ratio of [Chl a] to Cphyto for a given growth irradiance represents the total physiological stress (ηobs=[Chla][Cphyto]), and encompasses the physiological stress due to SST, light, and macronutrients availability (; ; ). Although small changes in [Chla]/[Cphyto] have been attributed to temperature variations for a single species in ex situ experiments , it is likely that the species present in natural assemblages are adapted to their ambient temperature conditions. Given that SST is relatively homogeneous in this region on the timescale of physiological adaptation (i.e. a few days), we assume that all species are adapted to their ambient temperature conditions, making the impact of SST on [Chla]/[Cphyto] negligible at the community level. Consequently, we can isolate the physiological stress due to macronutrient limitation (η′) by normalizing ηobs by the photoadaptation effect. As part of this process, we computed the median light in the mixed layer (IML) as follows: 2IML=PAR×e-0.5×MLD×KdPAR[Em-2d-1] where MLD is the mixed layer depth extracted from the Copernicus Marine Service Global Ocean Physics Reanalysis products , PAR is the standard daily PAR output of the function l2gen of SeaDAS , and KdPAR is the diffuse attenuation coefficient of photosynthetically active radiance. KdPAR was approximated from the 1 % light horizon (Zeu): 3KdPAR=-ln⁡(0.01)Zeu[m-1] where Zeu is estimated from [Chl a] following Eq. (10): 4log⁡10Zeu=1.524-0.436X-0.0145X2+0.0186X3[m] with X=log⁡10[Chla]. We partitioned all IML and ηobs of the four case study regions into two-dimensional bins spanning the entire dynamic range of IML and ηobs of the entire dataset. Because the quantity of level-2 data is too large for our computing capacity, we used 8 d medians merged products of all variables used in this computation (see merging method Sect. ). We identified the 1st and 95th percentiles of ηobs for each IML bin to which we fitted a simple exponential model (Fig. ): 5η^obs=A+B×exp⁡-C×IML The model equations fitted on the 1st and 95th percentiles of ηobs represent the range of acclimation for the entire dataset.

For each pixel, we computed the difference between the macronutrient replete state (ηobsmax^) and the observed state ηobs relative to the full span of η for the pixel's given IML to provide a quantitative metric of the degree of macronutrient stress: 6η′=ηobsmax^-ηobsηobsmax^-ηobsmin^[unitless] Thus, η′ is scaled to the parameter space of the specific remote sensing dataset used for this study, and is representative of the relative nutrient stress in the studied regions over the 6-month time series. We arbitrarily defined three macronutrient stress categories, classifying phytoplankton populations as macronutrient-replete when η′≤0.2 (including negative values), moderately stressed when 0.2<η′≤0.6, and highly stressed when η′>0.6.

Level-2 satellite estimates of SST, [Chl a], and iPAR (different from the daily PAR used in η′ computation) were individually calibrated against in situ measurements obtained from the underway system of the ship to minimize inter-sensor variability and biases. We performed match-ups and robust linear regressions between these variables measured in situ and their satellite counterparts derived from the calibration dataset (Appendix ). We used the parameters from their respective robust linear regressions (see Table ) to produce “calibrated” products before merging them. For parameters derived from remote sensing data with no comparable in situ measurements (e.g. ΦSat), we adjusted the values from all sensors to align with those of MODIS-Aqua, minimizing inter-sensor discrepancies. This alternative method improved the smoothness of the final merged product and the delineation of spatial patterns. However, potential biases associated with the retrieval of products not calibrated against in situ data remained unconstrained. Although in situ bbp data were available, the relation between in situ bbp and satellite bbp was very sensitive to match-up criteria, and the linear regressions were not well constrained (see Appendix ). The coefficients used to align the data from different satellite sensors with in situ bbp were inaccurate, leading to noisy merged satellite bbp products. Moreover, an artifact was discovered in MODISA and MODIST bbp maps in ultra-oligotrophic regions ([Chla]<0.04 mg m⁻³; see Appendices  and ). Therefore, only OLCI and VIIRS-SNPP bbp were binned into the 8 d median products, and OLCI was nudged to best match VIIRS-SNPP's values. We performed this cross-satellite sensor nudging only when: (1) at least 10 % of the total pixels of the adjusted sensor and the reference sensor were valid (i.e. unflagged), (2) the slope of the fit was positive, (3) R2≥0.9, and (4) nRMSE ≤10 %. Before computing the merged products of a given 8 d period and a given region, we grouped all re-projected level-2 images and removed outliers based on the distribution of all individual measurements of the grouped scenes following the same outlier removal method as inAppendix C. For each case study presented here, we produced a 6-month-long time series of 8 d medians of each of the variables presented in Table  (i.e. Animations S1, S2, S3, and S4 in the Supplement), following the approach of . Each case study region was centered geographically on an island sampled during the Tara Pacific expedition (Fig. ), and each 6-month time series was centered temporally on the day of in situ sampling by Tara . We propagated the error associated with the satellite product retrieval, nudging, and merging throughout each step to represent the final binned product uncertainty denoted as the standard error of mean (i.e. SEM) of the merged product (e.g. SEM[ΦSat]IMEf; see Appendix ). We used this final uncertainty to determine if changes associated with an IME are significant.

We detected IMEs on each 8 d median map using an iterative method to define a [Chl a] contour around each individual IME see method in. We used the islands and submerged reefs database from , which was consolidated from the General Bathymetric Chart of the Oceans (GEBCO) database , the high-resolution global island database , and the submerged reef database from . Following the methodology in and , submerged topographic features shallower than 30 m depth are treated as islands in the IME detection algorithm; therefore, the term IME can also be used in this study to qualify enhanced [Chl a] zones associated with seamounts. We used the modeled daily surface currents data for the detection algorithm i.e. global ocean ensemble physics reanalysis products distributed by Copernicus Marine Services;. As in previous studies , we define the background ocean (BO) reference zone associated with each IME zone as an area equal in size to the corresponding IME zone but located outside of it, closest to the island or submerged reef mask. We focused our analysis on archipelagos located near the center of each study region. We excluded IMEs detected near regional domain boundaries from the analysis because one of the stopping criteria of the detection algorithm could be triggered prematurely near the domain boundaries, thus underestimating the area of that IME . Large IME patches – such as those associated with the Society Islands, Samoa, and Fiji – frequently stem from the combined effects of multiple islands, each of which may generate its own local IME superimposed on the broader signal of the main island . To account for this, we identified all islands that were included at least once within the IME associated with each main archipelago during the analyzed 6-month period and selected all IME events linked to this set of islands for each region. Using this approach, we detected a total of 2025 individual IME realizations associated with the four studied archipelagos over the six-month study periods. We then extracted bio-optical properties derived from satellite data for all these 2025 IMEs and their associated BO areas. To identify consistent patterns across the South Pacific Subtropical Basin and all seasons sampled, we performed a principal component analysis (PCA) using the average bio-optical properties of all the 2025 individual IME realizations detected and their associated BO areas. Only variables derived from satellite measurements and assimilative models were used in this analysis to ensure sufficient data coverage (i.e. [Chl a], bbp, [Chla]/[Cphyto], SST, MLD, ΦSat, and η′). Additionally, we identified in situ measurements located within IMEs contours to validate patterns we observed in satellite data and complement the interpretation with information on pigment ratios as proxies for phytoplankton community composition, as well as macronutrient and iron concentrations. To examine the differences between coastal IMEs and advected IMEs, we categorized all in situ observations (i.e. continuous underway measurements, macronutrient concentrations, and iron concentrations) into three groups: background ocean (BO), coastal IME, and advected IME. In situ data located within an IME contour detected from satellite observations and over bathymetry shallower than 100 m were classified as coastal IME, whereas all remaining in situ data within IME contours were classified as advected IME (see Appendix ). We limited this categorization to in situ data because satellite data in shallow areas are, in general, not retrieved in clear water conditions due to the impact of bottom reflectance on radiometric estimates.

The IME detected around Fiji was characterized by a surface integrated [Chl a] enhancement of about 55 t m⁻¹ and covered a surface of about 500 000 km² at the time of sampling . The increase of cp660 (i.e. proxy for particulate organic carbon concentration; see Table ) by a factor of ∼15 observed over a ∼10 km transect approaching shore, west of the archipelago, suggests the [Chl a] enhancement in the IME was associated with a significant increase in phytoplankton biomass Fig. 6 in. A distinct and synchronized decrease in SST and increase in SSS on 30 May ∼10:00 UTC indicated a change in seawater physical properties measured along the inbound transect to the Fiji archipelago (Fig. B and C). This change in water mass physical properties was associated with a gradual decrease in the proportion of photoprotective carotenoids ([PPC]/[Chla]) and a gradual increase in [Chla]/[Cphyto] (Fig. D and E). These two independent metrics of light acclimation had opposite trends, suggesting that the same forcing was impacting both of them. The proportion of [PPC] can be modulated by phytoplankton in response to changes in ambient light to prevent intracellular photo-oxidative stress . Likewise, the proportion [Chl a] to [Cphyto] is modulated in response to light conditions, resulting in phytoplankton cells at the surface of the ocean characterized by lower [Chla]/[Cphyto] and higher [PPC]/[Chla] than cells residing at depth, where low levels of ambient light require more [Chl a] for photosynthesis but also cause less photo-oxidative stress, therefore requiring lower concentrations of photoprotective pigments. The lowest SST and [PPC]/[Chla] and the highest [Chla]/[Cphyto] of the inbound transect were measured just adjacent to the island's shelf break. These parameters changed suddenly when Tara sailed across the shelf through the Navula Passage west of the Fiji archipelago (i.e. coastal zone Fig. D and E). Together, these synchronized trends observed with four independent measurements suggest an upward entrainment of water parcels and low-light adapted phytoplankton cells, an observation that is consistent with the occurrence of an upwelling event. Based on the spatial variability of these parameters, we estimate that this upwelling occurred in a ∼170 km wide band adjacent to the island shelf west of Viti Levu at the time of sampling. [PPC]/[Chla] was marginally lower, and [Chla]/[Cphyto] was marginally higher, but overall heterogeneous in this IME detected in the outbound transect, suggesting a weaker upwelling of cells that are acclimated to low light to the surface than in the inbound transect. The surface currents were flowing southward at the time of sampling and the week before, suggesting that the IME detected south of Fiji was located downstream of Fiji. The inbound transect to Fiji crossed an IME episode associated with a submerged seamount located north of Fiji and not connected to the island shelf (15°39^′38.2^′′ S, 175°51^′59.8^′′ E; the IME is visible in the inbound panel of Fig.  but is outside the domain of the zoomed maps in Fig. ). This IME episode was also characterized by a synchronized decrease in [PPC]/[Chla] and an increase in [Chla]/[Cphyto], consistent with the entrainment of low-light-adapted phytoplankton into the surface layer. This pattern suggests enhanced vertical mixing around the seamount, likely driven by the interaction between ambient currents and the seamount topography . Similar signals were detected southwest of Rapa Nui and Tahiti-Mo'orea, along the outbound transects, indicating that sub-surface phytoplankton populations were recently upwelled to the surface (see Appendix ). In contrast, there is no clear evidence that upwelling occurred around Samoa at the time of sampling (see Appendix ). Indeed, in situ data show that the IME advected offshore had the same characteristics as the water mass closest to shore (that is, higher SST, lower SSS, and no significant differences in [PPC]/[Chla] and [Chla]/[Cphyto]; see Appendix ), therefore the main source of nutrient in Samoa's IME were likely associated with terrigenous processes.

We extracted satellite estimates of [Chl a], [Chla]/[Cphyto], and SST along the ship track (i.e. dashed line Fig. ) to assess if satellite-based estimates can also inform us about processes involved in IME. Although satellite-derived [Chl a] and SST closely matched the magnitude of in situ underway measurements, the satellite [Chla]/[Cphyto] values along the ship track were consistently lower than those estimated from in situ underway data. Satellite [Chl a] and SST were nudged to agree with in situ underway estimates across the Pacific Ocean see Sect.  and, however, bbp was not adjusted because of the low correlation between in situ and satellite data (see Sect. ). This suggests that the Quasi-Analytical Algorithm QAA; we used to invert bbp from Rrs may not perform well in these ultra-oligotrophic regions since all bbp measurements used to develop this inversion algorithm are higher than the in situ bbp of the BO zones we measured in these regions (bbp443<0.001 m⁻¹). Moreover, bbp inverted from satellite data using QAA was shown to be overestimated in these regions when compared to bbp estimated from profiling floats when bbp700<0.001 m⁻¹;. In this case, an overestimation of the satellite bbp would lead to an underestimation of [Chla]/[Cphyto] and would explain the observed discrepancies between in situ and satellite estimates. Despite this difference in magnitude, the spatial variability of [Chla]/[Cphyto] estimated from satellite data captured the same spatial variability as the situ estimates, although slightly smoother due to the 8 d time binning applied (i.e. dashed line Fig. ). These results show the potential of satellite data to discriminate between biomass enhancement and signatures of coastal upwellings around islands and improve the mechanistic understanding of IME. However, it requires minimizing the satellite temporal binning to capture short-lived events that would otherwise be undetectable.

The satellite map around Fiji shows that the IME patch is characterized by higher [Chla]/[Cphyto], as well as lower ΦSat and η′ indices, suggesting enrichment in iron and macronutrients, and enhanced upwelling of subsurface phytoplankton population within the IME relative to BO (Figs.  and ). Total iron concentrations measured at the sampling stations were the highest in Fiji's IME (Fig. ). The iron stress index (ΦSat) decreased along the inbound transect, with the steeper decrease detected over the island shelf, past the detected upwelling zone. This suggests that iron enrichment was driven by processes associated with the archipelago itself (e.g. leaching from sediments, runoff, or shallow hydrothermal activity) rather than the upwelling event (Fig. ). The macronutrient stress index (η′) also decreased along the inbound transect with a minimum coinciding with the upwelling signal and higher values over the shelf area, suggesting that the upwelling west of Fiji was an important source of macronutrients to the euphotic zone further away from shore. η′ was also consistently lower in the entire IME zone relative to the BO, with lower values detected as far as ∼150 km offshore on the outbound transect, suggesting a macronutrient enrichment across Fiji-Tonga's IME. Differences in spatial variability between macronutrient or iron concentrations and their respective stress indices can result from two (or more) factors. (1) Macronutrient and iron concentrations reflect standing stocks, whereas the stress indices are more indicative of the macronutrient and iron fluxes experienced by cells. In other words, although high macronutrient and trace metal concentrations likely imply increased supply, low concentrations do not necessarily imply limiting flux. (2) The spatial resolution of sampling differs between measurement types: the macronutrient and iron stress indices (respectively η′ and Φsat) are retrieved at 1 km² resolution, while discrete sampling of macronutrients and iron were collected at a much coarser spatial resolution (on average every 290 km). Despite these differences, the combined use of both concentrations and stress indices is informative because a disparity between them is suggestive of rapid uptake of macronutrients or iron by local phytoplankton populations.

The concentrations of nitrogen and phosphate were, in general, higher at coastal stations compared to offshore stations (advected IME and BO; Fig. ), indicating that terrigenous inputs, through runoff and organic macronutrient production in coral reefs, were a significant source of macronutrients close to shore. These concentrations were rarely higher in the advected IME compared to the BO. For example, despite a strong enrichment in nitrogen and phosphate in the coastal IME of Fiji (Fig. ), concentrations are already low at the closest station to the island shore downstream of Fiji (Fig. ), suggesting that a large proportion of these macronutrients were rapidly depleted relatively close to shore. The upstream water masses, which were impacted by local coastal upwelling and terrigenous processes, were probably diluted south of Fiji through mixing with the oligotrophic BO and modified by local upwelling and downwelling associated with positive and negative vorticity . Interactions of currents with islands and seamount topography can also result in doming isopycnals and increased vertical mixing downstream of islands . This enhanced vertical mixing could have sustained a small influx of macronutrients to the euphotic zone in the IME, as shown by the consistently lower macronutrient stress (i.e. η′≤0.6) in the IME zone of the outbound transect. Although phosphate concentrations measured at stations located in the IME of Fiji-Tonga were consistently higher than in the BO, nitrogen concentrations were lower at the offshore station closest to Fiji on the outbound transect (Fig. ). Phytoplankton biomass was higher at this station, particles were larger on average (lower γcp), and pigment ratios were characteristic of a higher proportion of diatoms (higher [PSC]/[Chla]), suggesting that nitrogen (nitrate + nitrites) could have been quickly consumed by the phytoplankton community (Figs.  and ).

Relatively high total iron concentrations were detected downstream of Fiji (south), further away from the coast, yet ΦSat increased rapidly along the outbound transect. Automated microscopy analyses show a high prevalence of Trichodesmium spp. in the IME zone along the outbound transect , and this bloom was visible to the naked eye during sampling (Bourdin, personal observation). Iron requirements vary between phytoplankton taxa; for example, Trichodesmium spp. is known to have higher iron requirements compared to other taxa . Furthermore, there is a high inter-species variability in the ratio of chlorophyll fluorescence to chlorophyll concentration (i.e. FChl / [Chl a]) , which could impact the estimation of ΦSat and explain the difference in the relationship between the total iron concentration and ΦSat at these stations. Trichodesmium spp. colonies are composed of “bright cells” that have twice the basal fluorescence of other cells in a non-diazotrophy state and three times as many when in a diazotrophy state . The bloom of Trichodesmium spp. observed on the outbound transect at sampling stations where the total iron concentration was an order of magnitude higher than in BO indicates that diazotrophy was likely occurring in this bloom. Trichodesmium spp. diazotrophy can be a significant nitrogen source in the western subtropical Pacific Ocean, where the influx of nitrogen into the euphotic zone from depth is limited due to low mesoscale mixing . We did observe an increase in the concentration of dissolved inorganic nitrogen (nitrate + nitrite) coinciding with the highest Trichodesmium spp. biomass in the advected IME zone see microscopy count in that could indicate a biogenic input. Together, these observations suggest high diazotrophic activity in the IME zone along the outbound that could explain the observed high ΦSat values in an iron-enriched zone. In addition, not all iron forms in the measured total iron are bioavailable and, therefore, must be interpreted with caution in the context of the physiological response of phytoplankton to iron availability.

For the three other case studies, total iron concentrations generally increased while ΦSat decreased. The magnitude of the increase in total iron concentration shows a longitudinal gradient with the lowest increase around Rapa Nui and the highest increase around Fiji (Fig. ). This observation is consistent with the gradient in iron enrichment associated with IMEs detected across the SPSG. Fiji and Tonga's IMEs were, on average, characterized by the largest difference in ΦSat between the IME and the BO zone, and the smallest difference was observed for Rapa Nui's IME (Fig. ). The decrease in η′ was more marginal in the other three case studies, particularly around Rapa Nui and the Society Islands, where it only decreased significantly at the border of the coastal zone. At this distance from shore, the 8 km spatial resolution of the MLD product used for the calculation of η′ is likely too coarse to capture the complexities of coastal submesoscale dynamics. Similarly, macronutrient concentrations around Fiji were generally higher in coastal zones, but declined sharply with increasing distance from the island (Fig. ). This pattern suggests that macronutrients supplied by terrigenous and reef-associated processes are rapidly consumed by phytoplankton near the shore. This observation supports the hypothesis of , who proposed that nitrogen is quickly utilized in coastal waters by phytoplankton taxa with higher nitrogen uptake capacity, such as diatoms. Consistently, the spatial distribution of pigment markers in this study indicates that the relative concentration of diatoms increased most strongly near the island shelf (see Sect. ; Figs.  and ).

The data collected on board during the inbound transect to Rapa Nui show that the IME detection algorithm from satellite only captured the strongest [Chl a] increase near-shore, missing the marginal increase that is highlighted in the gray shaded area (Fig. ). The strong latitudinal SST front associated with the highest [Chl a] located south of Rapa Nui shifted north toward Rapa Nui at this period, masking the increase in [Chl a] due to IME, and caused the iterative IME detection algorithm to stop before the contour of [Chl a] was low enough to encompass the entire IME zone see video of time series in. Interestingly, this slight increase in [Chl a] in this zone was not associated with a detectable increase in cp660 see Appendix D Fig. D1 in. This suggests that despite evidence of iron enrichment (lower ΦSat and higher total iron concentrations) and reduced macronutrient stress (lower η′) compared to the rest of the inbound transect, there was no detectable biomass increase in this zone. As discussed previously, any increase in picophytoplankton biomass is rapidly consumed in this pico-sized dominated environment, often preventing any detectable biomass accumulation using in situ or remotely sensed bio-optical proxies, despite increased nutrient availability.

To investigate the impact of the different enrichment mechanisms on phytoplankton community composition, we focus on the transects of the Fiji-Tonga case study (Fig. ). Suspended particles were larger on average (i.e. lower γcp) and accessory pigment concentrations normalized by [Chl a] were higher in the IME of Fiji-Tonga compared to the BO zones. Both mean particle size and accessory pigment concentrations increased sharply in the inbound transect and gradually decreased in the outbound transect when crossing the IME zones located on the downstream side of the archipelago, suggesting that the strongest change in pigment-based phytoplankton community composition along this transect occurred close to the island shelf. The relative concentrations of accessory pigments did not all increase at the same distance from shore, suggesting a succession of dominant phytoplankton groups along the gradient from the background ocean to the island coast. Using a Lagrangian particle tracking model and MERCATOR Ocean daily surface current products, we estimated that the time of advection for a given water parcel from the coastal area around Fiji to the BO zone exceeds 30 d, a period sufficiently long for an ecological succession in the phytoplankton community to occur downstream of the island. Consistent with this interpretation, [PSC]/[Chla] was highest near shore and decreased relatively quickly offshore, suggesting a higher proportion of diatoms in coastal waters compared to the rest of the IME and BO zones sampled. The proportion of chlorophyll b and chlorophyll c relative to [Chl a] increased as soon as Tara entered Fiji's IME zone in the inbound transect and remained elevated across the IME zone sampled on the outbound transect relative to the BO, suggesting that the contribution of other red algae and green algae remained higher further south of Fiji within the IME zone. Similarly, the proportion of phycoerythrin, indicative of cyanobacteria, also remained higher further south of Fiji, consistent with the Trichodesmium spp. bloom observed in the outbound transect using automated microscopy . While pigment ratios provide limited information on phytoplankton community composition and are subject to biases, this analysis points to possible changes in community composition in association with the IME, a hypothesis that will need to be confirmed with more direct methods (e.g., imaging and metabarcoding), but beyond the scope of this study.

The IME detected around Fiji-Tonga, increased in surface and surface-area integrated [Chl a] from the end of February to early May 2017, when it covered more than one million square kilometers. The impacted area then decreased between May and the end of August 2017 (Fig. B). The difference in ΦSat between the IME and the BO (ΔΦSat) was consistently negative, indicating a lower phytoplankton physiological stress and a net enrichment in iron in the Fiji-Tonga IME relative to the BO, especially at the beginning of the time-series, during the expanding phase of the IME (i.e. February to May). ΦSat in the IME decreased overall during the studied period, suggesting an increase in iron enrichment in the IME. The difference in ϕsat between the IME and BO (i.e. ΔΦsat) remained statistically significant (|ΔΦsat|>SEMΔΦsatf), indicating persistent iron enrichment within the IME associated with Fiji-Tonga. However, |ΔΦsat| decreased over the six-month period, approaching but not reaching zero, implying that ΦSat values in the IME became more similar to those in the BO. Because ΦSat within the IME also decreased over this period, this trend suggests that ΦSat decreased more strongly in the BO than in the IME. This pattern is consistent with an increase in iron availability in the BO, which reduced the contrast in ΦSat between the IME and BO. These results indicate the presence of seasonal variability in iron stress within the BO and suggest that the processes controlling this variability differ from those governing iron enrichment within the IME. Phytoplankton communities in the IME were not experiencing stress due to macronutrient limitation during the expanding phase of the IME (i.e. February to May with η′<0.2) but started to experience a moderate stress three weeks before the end of the expansion phase of the bloom (i.e. 0.2<η′≤0.6). ΦSat and η′ were significantly lower in Fiji-Tonga's IME relative to the BO, even during bloom demise, suggesting that top-down controls, such as grazing, contributed to the bloom demise before the iron and macronutrients were fully depleted from the euphotic zone in the IME.

Samoa's IME was characterized by two important phytoplankton biomass accumulation phases (i.e. blooms) over the period studied, highlighted by two distinct IME-surface-area integrated [Chl a] peaks around 9 November and 19 December 2016 (Fig. ). Interestingly, in Samoa's IME, iron and macronutrient stress did not decrease before and during the initiation of the first of these two blooms and were consistently lower in the IME compared to the BO along the entire time series, which suggests that the occurrence of the first bloom in Samoa's IME was not only triggered by bottom-up processes such as macronutrient and iron enrichment. This first bloom only initiated after being advected offshore and detached from the coastal IME of Samoa, which coincided with the dilution of coastal waters into the BO. This dilution may have decreased the encounter rate between grazers and their prey, reducing the grazing pressure, while the high levels of macronutrients and iron advected within this water mass maintained phytoplankton growth rate, allowing for positive accumulation of phytoplankton biomass such as hypothesized in. This hypothesis is also supported by the similarities in physical properties of Samoa's advected IME and the ones of its coastal IME (i.e. warmer and fresher than the BO south of Samoa; Figs.  and ), which suggest that the water mass advected offshore had a coastal origin (see Sect.  and Appendix ).

While these results suggest continuous enrichments in iron and macronutrients in IME zones in the western South Pacific Ocean, the results are more contrasted in the eastern South Pacific Ocean where ΦSat and η′ in IME areas are often not significantly lower than in the BO (illustrated by ΔΦSat+SEMΔΦSatf>0 or Δη′+SEMΔη′f>0; Appendix , Figs.  and ).

Although the four cases studied were sampled during different seasons, we detected a longitudinal gradient in the strength of the IME and its associated iron enrichment (i.e. from Rapa Nui to Fiji) in the SPSG. However, this longitudinal gradient should be interpreted with caution, as the size of the IME area is correlated with the size of islands (Fig. A), highlighting that island size can be an important driver of IME magnitude.

The Σ(30 m isobath area) effectively predicts the potential minimum IME area using a second-order polynomial model (dashed blue line in Fig. A): 7log⁡10(IMEareaminpredicted)=1.114+0.281X+0.105X2 where X=log⁡10(Σ(30misobatharea)). However, it is insufficient to accurately predict the mean strength and spatial extent of the IME, as evidenced by the large variability in IME area observed for a given Σ(30 m isobath area) (Fig. A).

Other factors associated with nutrient supply mechanisms that are independent of island size – such as the stoichiometry of macronutrients and trace metals arising from different sources, phytoplankton community composition, and grazing pressure – also play an important role in determining the magnitude of the IME.

The rank in the magnitude of the IME observed in Figs.  and (Fiji-Tonga > Samoa > Society Islands > Rapa Nui) does not correspond to the rank of the island/reef area contributing to the IME of each region (Fig. D). For example, the total 30 m isobath areas of all islands and reefs that contributed to the IMEs in the Society Island region are larger on average than the total 30 m isobath areas of Samoa and Fiji-Tonga (Fig. D). Island/reef size is represented here as the sum of all 30 m isobaths associated with an IME, which in the case of the Society Islands region, encompasses the Society Islands themselves, but also the Tuamotu archipelago, which is composed of 76 islands and atolls, some of which are among the largest atolls on the planet (e.g. Rangiroa covering 1640 km²). Their land area relative to their 30 m isobath area is very small, therefore, the terrigenous inputs of these coral reef atolls are likely smaller in comparison to those of the large high islands of Viti Levu and Vanua Levu of Fiji's archipelago. Moreover, islands of different geomorphic types may have different stoichiometries of macronutrient and iron enrichment. For example, volcanic islands are known to be sources of iron in the water column , thus their IMEs may have, on average, higher ΔΦSatIME-BO than the IMEs located around coral atolls. Our results, however, do not show clear trends between the geomorphic types of islands and their associated IME area or nutrient stress (Fig. ). The lack of a clear trend may result from the multiple sources of variability inherent to this study (e.g., seasonal variability, regional variability in oceanographic conditions, variability in responses of different phytoplankton assemblages to macronutrient and iron enrichment), which could mask the signal. Although characterized by weak correlations, the iron and nutrient enrichments associated with IMEs increased with larger reef/islands, which is illustrated here by the negative correlation between Σ(30 m isobath area) and ΔΦSatIME-BO and ΔηIME-BO′.