ISW signatures were identified in two illustrative images acquired on 28 September 2007 and 12 October 2018, over the Amazon shelf in the region of sunglint. These images consist of Level 1B data from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor onboard the TERRA satellite, using band 6 (centered at 1640 nm) with a spatial resolution of 500 m. The MODIS-Terra images were obtained from NASA's Earth Science Data System (ESDS) (10.5067/MODIS/MOD02QKM.NRT.061 ).

CHL data were derived from daily Level 1A acquisitions of the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Aqua satellite, covering the period from 1 January 2005 to 31 December 2021, at a spatial resolution of 1 km. The Level 1A MODIS-Aqua images were obtained from NASA's Ocean Color website (https://oceancolor.gsfc.nasa.gov/, last access: 13 December 2023) and processed to Level 1C using the SeaDAS 8.1.0 software package (https://seadas.gsfc.nasa.gov/, last access: 1 December 2022). Remote sensing reflectance (Rrs) was computed from the Level 1C MODIS data using POLYMER 4.13 (https://www.hygeos.com/polymer, last access: 1 December 2022) for atmospheric correction. POLYMER is specifically designed to recover marine reflectance and is particularly effective in correcting for effects of sunglint . Rrs values were then used to estimate near-surface CHL concentrations through NASA's standard algorithm implemented in SeaDAS. This algorithm combines an updated version of the OC3M band ratio algorithm with the color index (CI) approach .

Since the study area is highly affected by clouds throughout the year, we have used as well the daily L4 CHL data from the global multi-sensor Copernicus-GlobColour processor, which is a blend between OC5 and CI of . The GlobColour product was obtained from the Copernicus Marine Service (10.48670/moi-00281 ) for the period from 1 January 2005 to 31 December 2021, at a spatial resolution of 4 km. This product is a CHL daily composite obtained by merging multiple ocean satellite sensor acquisitions (SeaWiFS, MODIS, MERIS, VIIRS-SNPP, JPSS1, OLCI-S3A, and S3B) and by applying temporal averaging and interpolation methods to fill the missing data values.

Models based on band ratios in the visible spectrum offer reliable chlorophyll estimates for clear to moderately turbid waters but show limitations in highly turbid environments . Considering the presence of highly turbid waters in our study area, associated with the Amazon River plume, we applied the methodology developed by to exclude information from highly turbid locations. Specifically, if a given location (longitude, latitude) was classified as turbid waters (optical water types 4 and/or 5) in more than 20 % of the time series, all data from that location were excluded from the analysis. This approach helped avoid bias in the time series by removing pixels that were permanently or episodically associated with turbid environments. Overall, the excluded pixels are mainly confined to areas shallower than 100 m, where there is no evidence the ISWs discussed in this manuscript.

Gap-free maps of daily L4 foundation sea surface temperature (SST) data from the OSTIA were acquired from 1 January 2007 to 31 December 2021 in the Copernicus Marine Service (10.48670/moi-00165). This product uses in-situ data and merged satellite observations (AMSR2-GCOM-W, AVHRR-MetOp-B, SEVIRI-MSG, VIIRS-SNPP, NOAA-20, SLSTR-S3A, and S3B) and provides SST maps at 0.05°×0.05°. horizontal grid resolution.

In regions where M2 and S2 tidal constituents dominate, their nonlinear interaction generates the MSf (Lunisolar Synodic Fortnightly) oscillation, with a period of approximately 14.7 d. The MSf corresponds to the neap–spring tidal cycle, a phenomenon of great importance in tidal dynamics and a major physical factor influencing coastal and marine environments. Wavelet analysis was therefore applied to identify and quantify this fortnightly variability in the CHL and SST time series. This approach is a powerful tool for analyzing ecological systems since the methodology performs a local time-scale decomposition of the signal overcoming the problem of non-stationarity in time series . The 1D continuous wavelet spectral analysis was carried out on daily CHL GlobColour and OSTIA products. The continuous Morlet wavelet transform was applied to the CHL and SST anomaly time series, which was performed using Python 3.8 and the free Python package PyCWT (https://pypi.org/project/pycwt/, last access: 24 January 2024) based on the method developed in . The power spectra significance test was made according . applied a similar wavelet approach to investigate CHL variability associated with spring–neap tidal cycles.

We developed an alternative methodology to investigate the fortnightly tidal modulation of CHL by constructing spring–neap tidal cycle composite maps. Assuming the IT signal is coherent (i.e., coherent ITs), these composites can resolve the broader MSf frequency. Unlike wavelet analysis or other commonly applied methods to examine IT-related frequency content (e.g., least-squares harmonic analysis; ), our approach also preserves the sign of spring–neap CHL variability. This enables us to determine whether tidal influences at the MSf frequency drive increases or decreases in CHL concentration within the affected region. If the IT signal is incoherent, however, this method is unlikely to extract the MSf frequency reliably.

Tidal elevation at the Amazon shelf break (45.5° W, 1° N) was analyzed using the Tidal Toolbox, available within the COMODO tools developed and maintained by the SIROCCO (SImulation Réaliste de l'OCéan COtier) national service (https://sirocco.obs-mip.fr/other-tools/prepost-processing/comodo-tools/, last access: 6 July 2023).

Let S and N denote the CHL concentrations at the maximum and minimum tidal range elevation, respectively, spring and neap tidal cycles during the time series period (please, see Fig. ). The composite maps, i.e., the weighted spring-minus-neap CHL difference for pixel p, fp(S,N), averaged over all cycles, can be calculated as follows: 3fp(S,N)=1n∑i=1n34Sp,i(w)+14Sp,i+1(w)-14Np,i(w)+34Np,i+1(w)CHLp,i‾ where 4Sp,i(w)=13∑j=-1+1Sp,tS,i+j,Np,i(w)=13∑j=-1+1Np,tN,i+j Sp,tS,i+j (Np,tN,i+j) is the CHL value at pixel p on day tS,i+j (tN,i+j), where tS,i (tN,i) is the nominal day of the ith spring (neap) tide event, i.e., and j=0, jϵ{-1,0,+1} days from spring tides. CHL_p,i represents the mean CHL concentration used to normalize each spring-minus-neap CHL difference in the series. Specifically, for each index i, CHL_p,i is computed as the average of all CHL values included in the calculation of the weighted difference, considering both the spring and neap tide observations within the local 3-pixel window. This normalization allows each difference to be expressed relative to the local CHL magnitude, ensuring comparability across the time series. In addition to the composite map centered on the spring and neap tide events, we also computed alternative composites by introducing temporal lags. This approach accounts for the phase lag between the maximum tidal potential and the maximum tidal elevation (the so-called age of the tide), when mixing intensity typically peaks. Specifically, we considered delayed versions of spring and neap tides, defined as S={Sp,i+delay} and S={Sp,i+delay}, where delay={0,1,2,3,4,5} days.

We investigated daily CHL data in the BEM derived from MODIS-Aqua and GlobColour data for days when ISWs signatures were identified in MODIS-Terra images acquired under sun glint conditions . The presence of ISWs is used as a proxy for the presence of ITs. We selected two illustrative cases to demonstrate how the passage of ITs can influence chlorophyll concentration, thereby justifying the subsequent analysis that extrapolates this effect to a 17-year time series. It is important to note that, although these cases exhibit chlorophyll modulation patterns previously associated with ITs in other studies , they are not necessarily representative of the typical conditions in our study area, particularly given the lack of data during months with high cloud coverage (from January to July). Furthermore, an examination of the background circulation and sea level anomaly (see Fig. ) for the days corresponding to the illustrative cases reveals that eddy activity is more intense during illustrative case I. However, in both periods, the ISW location remains outside and to the south of the region directly influenced by the mesoscale processes.

In both illustrative cases, the MODIS-Aqua images were captured 2.55 h after the MODIS-Terra ones, corresponding to a northeastward displacement of approximately 25 km in the ISW location. Figures  and  present these illustrative cases, with the leading waves of the ISW indicated by dotted-dashed black lines. The wave positions depicted have not been adjusted for their displacement between the two remote sensing image acquisitions.

For illustrative case I, shown in Fig. a and b, daily CHL levels retrieved from MODIS-Aqua and GlobColour are displayed, respectively. On the left side of the IT pathway A (indicated by the red dashed line), two bands of elevated CHL concentration are visible, extending parallel to the shelf break between IT generation sites A and D. The bands are separated by approximately 100 km, typical mode-1 IT wavelengths (150–100 km). In the location of the mapped ISW leading wave, CHL concentration is similar to the background levels. The offshore band of enhanced CHL (around 2.5° N, 45.5° W) is flanked by two ISW signatures. On the right side of IT pathway A, we also observe bands of elevated CHL concentration separated by about 68 km, typical mode-2 IT wavelengths. In illustrative case II (Fig. a and b), a similar modulation of CHL concentration is evident, with the offshore band of elevated CHL also being flanked by ISW signatures. The positions of the enhanced chlorophyll bands in illustrative case II closely match those in illustrative case I. Furthermore, although the CHL concentration is generally higher in the GlobColours product for both illustrative cases, the spatial patterns in both products show agreement.

For illustrative cases I and II, CHL concentration values along IT pathway A, derived from MODIS-Aqua and GlobColour data, are shown in Figs. a, c and b, d, respectively. When ISW signatures cross the IT pathway, their positions are marked as vertical dashed magenta lines on the graph. Accounting for the displacement of the waves from different scene acquisitions, the recalculated wave position is indicated by a vertical dashed cyan line for the MODIS-Aqua CHL data. The recalculated position is not plotted for GlobColour because its exact acquisition time is unknown, as it is a merged product (without a defined acquisition time). In both illustrative cases, a distinct peak in the CHL concentration profile is observed at 262 km (202 km from the IT generation point A), with the peak being more pronounced in illustrative case I when considering the MODIS-Aqua data.

The maps of CHL relative difference are shown for illustrative cases I and II in Figs. c and c, respectively. In both illustrative cases, the bands of enhanced CHL correspond to concentrations higher than the 15 d average. Figure e and f depicts the chlorophyll relative differences along IT pathway A for both illustrative cases. In both instances, a pronounced CHL relative difference peak of up to 45 % is observed at the IT generation site A. For illustrative case I, two peaks in CHL difference along path A are observed at approximately 129 and 262 km (respectively, 69 and 202 km from IT generation point A, see red arrows), with the first peak reaching 19 % and the second up to 31 %. For illustrative case II, two peaks are found at 122 km (8 %) and 262 km (7 %), respectively, 62 and 202 km from the IT generation site A. A 1D horizontal band-pass filter for mode-1 and mode-2 wavelength (respectively, 100–150 and 50–100 km, see red and blue dashed lines in Fig. e and f) was applied to highlight the impact of the IT. For each transect, the mean spectral coherence between the original CHL relative difference signal and the band-pass filter components (mode-1 and mode-2) was computed to quantify the relative contribution of each internal tide mode. For illustrative case I, the mean spectral coherence values for mode-1 and mode-2 are 0.11 and 0.23, respectively, suggesting that mode-1 and mode-2 tides contribute approximately 11 % and 23 % of the observed signal. For case II, the corresponding values are 0.17 and 0.26, indicating contributions of about 17 % and 26 %, respectively. Modulations are stronger and more coherent with the original signal when considering the signal filtered for mode-2 wavelength in both cases, with the offshore signal weakening beyond the first 300–400 km.

Figure a shows the spatial variability of the fortnightly signal in the gap-free GlobColour CHL product, extracted using the mean Morlet wavelet power within the 14.2 to 15.2 d period where the power was statistically significant above the 95 % confidence for red noise. The SST fortnightly signal was calculated using the same methodology (see Fig. b). The depth-integrated barotropic and baroclinic energy flux (black arrows) and dissipation from , computed using the Nucleus for European Modeling of the Ocean (NEMO v4.0.2, ) with the AMAZON36 configuration, are shown in Fig. c and d, respectively. For both CHL and SST, the fortnightly signal is more intense in the northwest part of the shelf region around 46.5–52° W and 0–6° N, where barotropic tides dissipate through bottom friction, see Fig. c. Offshore, the highest values of the CHL fortnightly signal are mostly concentrated in the northwest part while, for SST, the fortnightly signal considerably reduces. Areas with high CHL fortnightly power can also be seen aligned with the IT pathways A, B, and C (see the gray dashed rectangles in Fig. a). The areas of high power along IT pathways A and B align well with the lower values of depth-integrated S2 baroclinic energy dissipation predicted by the model.

Figure  presents the CHL spring–neap tidal cycle composites computed from Globcolour data using Eq. (), considering various delays (0–5 d) from spring and neap tides. Across all delays, CHL exhibits significantly higher shelf and shelf break variability than the offshore region, which is consistent with the strong shelf depth-integrated barotropic dissipation, as illustrated in Fig. c. On the shelf area, a dipole pattern is evident: in the northwest region (around 47–52° W and 2–6° N), a negative anomaly in CHL is observed, with minimum values reaching approximately -50 % occurring 2–3 d from the spring–neap tides. Conversely, in the northeast region (around 40–47° W and 3° S–2° N), a positive anomaly is seen, with maximum values of about 30 % occurring with a 1–2 d delay.

In the offshore region between IT generation points A and B, a positive spring–neap tidal cycle difference is observed, following a wave-like pattern in the horizontal structure of CHL. This wave pattern is most pronounced with a 1 d delay from the spring–neap tides, gradually diminishing in extent and coherence with longer delays. Referring to the spring–neap tidal cycle composites with a 1 d delay as a benchmark, Fig. b illustrates at least three peaks of positive CHL spring–neap tidal cycle difference. Figure a shows a close-up view highlighting the wave-like pattern in the CHL composite with a 1 d delay. The profiles along the IT pathways A, B, and C for spring–neap tidal cycle composites are shown in Fig. . Taking as reference the delay of 1 d, an average of 63 km separates the two first peaks and the second and third peaks are 83 km apart. The distances of the three peaks from IT generation points A, B, and C are detailed in Table . In Fig. , the black stars indicate areas with frequent ISW occurrences based on findings by . The third peak of positive CHL difference along pathway A is flanked by the regions of high ISW occurrence.

In Fig. , focusing on IT pathway A and a 1 d delay from spring–neap tides, the positions of the first and third peaks correspond closely to the peaks of positive CHL relative difference observed in illustrative cases I and II (see Fig. e and f). Additionally, the locations of the second peak of positive spring–neap tidal cycle CHL difference along IT pathways A and C, and the third peak along IT pathway B, correspond closely with areas of high Morlet wavelet power (indicated by gray dashed rectangles in Fig. b). The positive peaks of CHL differences during the spring–neap tidal cycle, with a 1 d delay, average around 2.1 % across all IT pathways. The modulation of spring–neap tidal cycle differences filtered for mode-1 and mode-2 IT wavelength diminishes further offshore after the initial 300–400 km. The mean spectral coherence between spring–neap tidal cycle signal and the band-pass filter component is higher for mode-2 than mode-1, considering all delays (see Table )

When considering the signal filtered for mode-1 wavelength, the modulations in the spring–neap tidal cycle composite map are less coherent along IT pathway B. However, the peaks of positive spring–neap tidal cycle CHL differences reach higher values along IT pathway B. This finding aligns with the Morlet wavelet analysis, which indicates higher fortnightly power along IT pathway B than other pathways. In contrast, lower values are observed along pathway A (see values in Table ) compared to other IT pathways. Note that, when considering a profile outside the influence of ITs, the signal filtered for mode-1 and mode-2 IT wavelengths are very low (see Fig. ).

Because of the temporal averaging and interpolation methods involved in deriving CHL concentration from the GlobColour product, we also generated spring–neap tidal cycle composites using daily MODIS-Aqua imagery, as shown in Fig. . Due to the substantial cloud cover in our study area, these composites are noisier than the GlobColour ones. Nevertheless, both datasets reproduce similar spatial patterns, particularly on the shelf region, where a comparable dipole pattern is evident in MODIS-Aqua as in GlobColour composites. In the northwest part of the shelf, MODIS-Aqua composites show maximum negative CHL differences (approximately -84 %) occurring 2–3 d after spring–neap tides, whereas in the northeast region, maximum positive CHL anomalies of about 70 % are observed under similar timing. Overall, CHL differences between spring and neap tides tend to be higher in MODIS-Aqua compared to GlobColour composites.

Offshore, a distinct wave pattern emerges with a 2 d delay, characterized by at least three peaks of positive CHL spring–neap tidal cycle differences along pathway A. A close-up view highlighting the wave-like pattern in the CHL composite, with 2 d delay is shown in Fig. b. A similar pattern is partially visible along pathway C, which exhibits two peaks of positive CHL spring–neap tidal cycle differences. Along pathway A, the third peak of positive CHL difference is in between regions of frequent ISW occurrence , aligning with the patterns observed in GlobColour composites. Figure  illustrates the CHL difference profiles along the IT pathways for MODIS-Aqua composites. For a 2 d delay, along pathways A and C, the CHL spring–neap tidal cycle difference peaks average around 6.8 %, more than double the signal detected in GlobColour composites. The wave-like structure begins to lose coherence after a 3 d delay from the spring–neap tides. The spatial distribution of the CHL concentration peaks closely aligns with the wave structures observed in GlobColour composites along IT pathways A and C (see Table ). In contrast, no significant signal is detected along pathway B in the MODIS-Aqua composites. In both pathways A and C, the signal filtered for mode-2 IT wavelength is more coherent than the signal filtered for mode-1 (see Table ).

The offshore band of enhanced CHL in illustrative cases I and II (around 2.5° N, 45.5° W, see Figs.  and ) likely corresponds to an IT crest, as two ISW signatures flank it. Similar bands of elevated CHL levels, flanked by ISW signatures, were observed by and in the Bay of Biscay. They attributed the enhanced CHL concentration to the uplifting of the DCM caused by the passage of internal tidal crests.

The CHL concentration bands associated with the IT in our illustrative cases are 7 % to 31 % higher than the 15 d average, aligning well with the findings of . Their study, based on data from a Slocum G2 glider equipped with an optical fluorescence sensor deployed off the Amazon shelf, reported that ITs increase total chlorophyll concentration by 14 % to 32 %.

The findings indicate that the highest CHL and SST fortnightly mean Morlet wavelet power occurs on the shallow shelf, where barotropic tides dissipate through bottom friction (see Fig. c). Based on these surface measurements, we infer that this dissipation induces mixing, likely impacting water properties.

The fortnightly signal in SST is weaker than that observed in CHL and becomes even less pronounced offshore. Using two regional simulations (with and without tides), estimated sea temperature anomalies along IT pathway A and reported offshore surface anomalies of approximately -0.2 °C. The study also showed that ITs modulate upper ocean–atmosphere interactions by enhancing the net heat flux at the air–sea interface. This enhanced flux from the atmosphere to the ocean tends to damp the IT-induced SST cooling and contributes to restoring surface temperatures . Such restoration could explain the weaker fortnightly signal in SST compared to CHL. Nevertheless, further studies are needed to understand these differences better.

On the shelf, using 1 and 2 d delays as benchmarks for GlobColour and MODIS-Aqua CHL composites, respectively (see Figs. b and c), based on surface measurements, we infer that the northwestern region, influenced by the turbid Amazon River plume, may experience higher sediment resuspension and mixing during spring tides, which in turn may inhibit primary production due to light limitation. In shallow, permanently mixed areas of the northwest European shelf, tidal mixing and resuspension of suspended matter hinder primary production, as noted. According to , near the Amazon River mouth, energetic spring tides resuspend muddy sediment in the water column, making fluid mud less dense and well-mixed but, during neap tides, the tides are less energetic and denser fluid mud consolidates on the seabed. Another possible explanation for the negative spring–neap tidal cycle CHL anomaly observed in this region is the influence of tidal variability on the Amazon River plume dynamics. During spring tides, strong vertical turbulence shifts the northeastward deflection of the plume waters further offshore. In contrast, during neap tides, the plume is deflected right upon entering the ocean remaining closer to the shore . In neap tide reduced mixing results in a well-stratified river plume, limiting nutrient dispersion and leading to higher nutrient concentrations within the plume. Under these conditions, phytoplankton likely can access these nutrients, considering that the light availability is sufficient. In the northeastern shelf region, where lower turbid waters are present (non-plume waters), we can infer that the dissipation of barotropic tides likely promotes mixing, which may contribute to enhancing CHL concentrations. As highlighted, tidal fronts in the northwest European shelf promote vertical mixing of nutrients, enhancing primary production, with approximately 16 % of annual mean primary production attributed to tidal forcing in these regions.