As part of a campaign to study intraseasonal oscillations in monsoon weather, we deployed an array of three densely instrumented buoys and ocean-wave-powered Wirewalker profiling vehicles that gathered physical, bio-optical, and irradiance measurements in July 2019 in the Bay of Bengal. Each Drogued Buoy Air-Sea Interaction System (DBASIS) was deployed for ∼19 d (Fig. ): M1 on 9 July, M2 on 10 July, and M3 on 11 July, generally following the mesoscale flow. Each profiler had a metocean buoy on the surface, a vertically profiling Wirewalker over the upper 100 m , and six X-wings (1 m² drag elements) over 200 to 210 m depth so that the system drifted with the subsurface currents at 200 m. They collected approximately 200 profiles per day spanning the upper 100 m of the water column with sub-meter vertical resolution.

The buoy's meteorological package measured wind (2.9 m a.m.s.l. – above mean sea level), air temperature (2.6 m a.m.s.l.), humidity (2.6 m a.m.s.l.), precipitation (2.9 m a.m.s.l.), barometric pressure (2.3 m a.m.s.l.), sea surface temperature (0.3 and 1.5 m b.m.s.l. – below mean sea level), and downwelling solar and infrared radiation (3.2 m a.m.s.l.). The metocean buoys were equipped with Kipp and Zonen SMP21 shortwave radiation sensors that measure downwelling radiation over wavelengths of 285 to 2800 nm (50 % points). These measurements were used to estimate the surface photosynthetically available radiation (PAR).

We equipped the Wirewalker profilers with RBR Inc. conductivity, temperature, and depth (CTD) sensors augmented by SBE Wetlabs Ecopucks measuring Chl fluorescence (ChlF, a proxy for Chl biomass), optical backscatter at 532 nm, and chromophoric dissolved organic material (CDOM; not used in the analysis presented here). We measured subsurface downwelling irradiance onboard the Wirewalkers using a Satlantic OCR-504 multi-spectral radiometer at four bands, 380, 412, 490, and 532 nm. All parameters were collected continuously at 6 Hz and telemetered at that resolution in real-time via RUDICS Iridium modems on the surface buoy. Using the factory calibration, we convert the optical backscatter to Nephelometric Turbidity Unit, τ (NTU). The median time between profiles was 10 min at M2 and M3 and 20 min at M1. For all quantities, we only use measurements during the smooth upward ascent of the profiler.

To estimate the diffuse attenuation averaged by depth, Kd (m⁻¹), the raw measurements of downward irradiance were grouped in time and depth into 3 h and 4 m bins, respectively, and then least-squares fitted to an exponential curve. Irradiance was then interpolated back to the original O(10 min) time resolution (see Supplement for further details).

We also equipped the profilers with a downward-looking Nortek Signature with a frequency of 1 MHz, from which we show only the acoustic backscatter . For each profiler's upcast, we averaged the acoustic amplitudes of each of the four beams over the first 20 bins (2.5 m depth range) closest to the transducer. We then interpolated range-averaged beam amplitudes onto a 0.25 m uniform depth grid before averaging all four beams. The 1 MHz backscatter was similar to the R/V Sally Ride acoustic backscatter that had a frequency of 150 kHz, with differences between frequencies as expected from literature .

All three DBASIS floats had the same bio-optical instrumentation and generally collected comparable measurements. We present time series from only M3 here, which collected continuous (unlike M1) and high resolution (10 min) data, and show M1 and M2 observations in the Supplement. Our key results and findings were similar for all floats.

The primary objective of the field campaign was to investigate air-sea interactions and the underlying physical variability in the BoB. As such, while biogeochemical observations were collected autonomously, we did not sample water in order to field calibrate the chlorophyll sensors. Instead, using laboratory calibration carried out just prior to the experiment, we converted the observed ChlF in relative fluorescence units (RFU) to real units (µg L⁻¹) via ChlF=0.012×(RFU-50). This conversion has an unknown uncertainty that linearly scales for all values (i.e., percentage error). For context, the observed surface ChlF from the drifting systems and MODIS-Aqua surface Chl from July 2019 were 0.17 and 0.33 µg L⁻¹, respectively, with these points located an average of 54 km apart. The focus of our work is not on the absolute quantities of ChlF but rather on their response to surface and penetrative solar radiation, so we focus on the robust trends in co-variability rather than absolute values.

Chlorophyll pigment concentration, ChlF, and phytoplankton biomass (i.e., carbon content) are highly variable under changing light conditions as phytoplankton adapt to changing light environments by adjusting their light-harvesting pigments . Due to this, estimates of biomass from backscatter observations are typically a more robust estimate of carbon biomass than ChlF. However, in the situation encountered here, non-algal sources of backscatter at-times confounded measure of algal concentrations, for example in the high-sediment coastal waters we observed at M3 before 15 July (Fig. d). Focusing only on times when the backscatter signal was due to algal diel variability, ChlF and backscatter variability was similar. This analysis showed that ChlF was a good proxy for phytoplankton biomass at the depth of the subsurface chlorophyll maximum (SCM), which we discuss throughout the manuscript.

Near the surface, non-photochemical quenching (NPQ) results in a non-biomass-related diel reduction in fluorescence under high light conditions . To correct for NPQ in the ChlF measurements, we follow methods established for other rapidly profiling platforms . We least-squares fit irradiance at 490 nm to ChlF over the upper 20 m over 1 d windows, stepping in time every 0.1 d and averaging overlapping time steps. We set a maximum correlation coefficient of -0.1, ensuring we only modify ChlF if ChlF decreased when irradiance increased. Although we used only ChlF over the upper 20 m for the least-squares fitting, we corrected ChlF for all depths with non-zero irradiance. The NPQ correction to ChlF was small, with a 95th percentile of 0.016 µg L⁻¹. Further details are available in the Supplement.

At all profilers, the depth of maximum ChlF was identified within a narrow density range. Excluding the initial 2 d observed at M1 (see Supplement), the SCM and the 1021.5 kg m⁻³ isopycnal were significantly correlated (r2=0.85, p<0.001) with a mean absolute error of 1.65 m. By defining the SCM as the depth of the 1021.5 kg m⁻³ isopycnal ±1 m, the SCM can be matched to an along-isopycnal reference frame. Given the internal wave fluctuations (∼15 m) observed during the campaign, employing an along-isopycnal reference frame was convenient for isolating the impact of changing surface irradiance and diffuse attenuation on ChlF at the SCM (denoted ChlF_SCM). The 1021.5 kg m⁻³ isopycnal had an average depth of 39 m at M3 (Fig. ) that deepened to 50 m at M1.

ChlF and integrated ChlF reflects the standing stock of chlorophyll, and therefore integrates pigment concentration over days to weeks. To understand the rate at which ChlF is produced and lost due to the local conditions at the SCM, we applied the diel cycle method . We identify the rate of gross ChlF production (GCP_SCM) and ChlF loss (code available at https://github.com/duebi/dielFit, last access: February 2020) to the NPQ corrected ChlF_SCM. This loss term describes all processes resulting in a reduction of ChlF_SCM, including grazing, mortality, and downward export.

The diel cycle method linearly fits three terms via ordinary least-squares fitting to ChlF_SCM. This allows us to directly evaluate the response of SCM production to intraseasonally modulated PAR, rather than relying on changes in the accumulated concentration. We note that when deciding the depth region to vertically average ChlF_SCM over (2 m), averaging a larger depth range decreased the effectiveness of the diel cycle fit (i.e., the correlation coefficient decreased and mean absolute error increased). The three fitted terms in the diel cycle method included a constant, a constant loss rate for both day and night, and a variable in time GCP_SCM rate similar to the clear-skies PAR, with day-time growth maximum at noon and a zero rate of change at night.

Since we collected radiation measurements, we did not need to simulate the solar cycle. We modified the code to optionally include the observed PAR at the SCM (denoted PAR_SCM) to model growth (henceforward referred to as the PAR_SCM method). The PAR_SCM method is similar to the sinusoidal growth model but accounts for the impact of cloudiness, changes in diffuse attenuation, or perturbations of the SCM depth on growth. Before fitting to ChlF_SCM, we integrated each growth scheme in time. For all three profilers and days sampled, we fit each 26 h period (starting at 23:00 LT), overlapping each day by 1 h. We additionally modified the diel cycle methodology so that the fitted GCP_SCM and loss is always positive.

The performance of the fit was assessed by estimating the correlation coefficient and the mean absolute error, MAE=1/n∑||ChlFSCM-ChlFfit||, where ChlFSCM is the observed ChlF at the SCM, ChlFfit is the diel cycle fit, and n is the number of samples. Of the tested linear, sinusoidal, and PAR_SCM schemes for GCP_SCM, the PAR_SCM method consistently returned the smallest MAE and better predicted the timing of the observed ChlF_SCM maxima by 45 min on average across all floats (Table ). The sinusoid method also returned a good fit, with a larger MAE, but the linear method was relatively poor in describing the observed ChlF_SCM variability. The fitted mean absolute error doubled when fitting ChlF at a constant depth rather than along-isopycnal due to the vertical heaving of the SCM by passing internal waves (i.e., ChlF_SCM; Table ). The fitting routine was also effective at M1 and M2. At M1, the timing of peak afternoon ChlF_SCM was more variable, particularly when large-amplitude internal waves deepened the SCM during the day (see Supplement). We also applied the above diel cycle fitting methodology to turbidity to estimate gross turbidity production (GTP_SCM) and turbidity loss. We computed the confidence intervals of the model by bootstrapping the residuals with 200 iterations .

To contextualize our observations and investigate the frequency of light-limited productivity at the SCM in the bay, we analyzed 12 BGC-Argo floats deployed in the BoB, obtained via the OneArgo-Mat routine . These floats were equipped with a CTD and bio-optical sensors WET Labs ECO-FLBB AP2 or MCOMS, see measuring temperature, salinity, pressure, chlorophyll fluorescence (ChlF), optical backscatter coefficient at 700 nm (at ∼124°), as well as other sensors not used here. The data were quality controlled using standard Argo protocols and processed following . Quality-controlled data was used when available (quality flags 1, 2, 5 or 8), else visual inspection and simple outlier removal was performed (quality flags 0 or 3).

The raw signals were converted to ChlF (µg L⁻¹) and particulate backscattering coefficient (m⁻¹) following BGC-Argo procedures . To account for variations between sensors, we standardized the minimum ChlF and backscatter value by removing the median “dark” or background value at pressure > 600 dbar for each float . BGC-Argo floats vary in their temporal and vertical sampling frequencies, so following we interpolated the quality-controlled float data onto uniform temporal grids with time steps equal to the minimum temporal sampling rate, which is 10 d for most floats. We standardized the vertical grid from 4 to 1000 m. It had 2 m resolution in the upper 300 and 10 m resolution below that. This interpolation was performed using a hermite polynomial scheme (pchip in Matlab), but depths outside the observed range were excluded (i.e., no extrapolation). We applied a 3-point moving-median in the vertical to remove measurement noise and aggregates.

The BGC-Argo subsurface ChlF was also maximum near a consistent isopycnal, like the DBASIS observations, except ChlF was maximum near the 1022 kg m⁻³ isopycnal, so we defined this isopycnal as the SCM depth. Individual months could have enhanced ChlF near a slightly different isopycnal, like the 1021.5 kg m⁻³ isopycnal used when analyzing the DBASIS floats, but we simply use the same isopycnal year-round. To estimate the ChlF_SCM, we averaged ChlF over the 10 m above and below this depth. Since the floats profile once every 10 d, and there are too few floats to apply more advanced statistical techniques , we cannot estimate GCP_SCM from diel cycles for the BGC-Argo data. Instead, we examined the floats for any evidence of a correlation between ChlF_SCM and PAR at the surface and SCM (described below). We note the diel cycle in ChlF_SCM is prevalent in the Bay of Bengal and has the potential to alias multi-day trends in ChlF if measurements of ChlF are taken at different local times, as is the case for BGC-Argo measurements. To avoid this aliasing, in regions with diel cycles, ChlF and other biological or bio-optical variables should be measured at similar times of the day.

To understand the co-variance of light and BGC-Argo observed ChlF, we considered both MODIS Aqua PAR and the atmospheric reanalysis ECMWF Reanalysis v5 (ERA5) product . On average, the PAR estimated at M3 was less than the satellite derived estimate but similar to the ERA5 estimate. We thereby estimated surface PAR from the ERA5 shortwave radiation, by dividing shortwave radiation by 2.114 . Then, to estimate PAR at the SCM, we utilized the 490 nm diffuse attenuation (Kd490) from the monthly averaged MODIS-Aqua and converted to a PAR equivalent using the relation in . If this data was missing due to cloud cover, we took the monthly climatology value instead. For a short-term estimate of 490 nm diffuse attenuation (Kd490) at the start of our DBASIS time series, we analyzed the Ocean Color CCI (OC-CCI) merged satellite product, which had a spatial resolution of 4 km and was averaged over the 5 d period of 10 to 15 July 2019 .

The 2019 field campaign enabled a detailed investigation of the co-variability of subsurface physical and bio-optical properties on an hour-by-hour basis over a nearly 3-week deployment (Figs.  and ). Conditions at deployment were consistent with the southwest monsoon “break” period with weak winds (∼7.5 m s⁻¹), large net heat fluxes into the ocean (QN, daily average QN of 180 W m⁻²), and no precipitation (Fig. ).

The DBASIS array was deployed at the intersection of two eddies, as inferred from sea surface height (SSH, Fig. ). Low-SSH (SSH < 0.5 m) and high-SSH (SSH > 0.5 m) eddies were present to the northwest and southwest, respectively. Between these two eddies, a plume of elevated surface chlorophyll was advected more than 400 km from the coast toward the DBASIS array (Fig. a). This coastal plume exhibited elevated turbidity (τ) and surface chlorophyll a fluorescence (ChlF), both of which reduced light penetration into the upper ocean (Fig. ). Similar coastal plumes have been observed elsewhere at the intersection of cyclonic and anticyclonic eddies i.e., eddy dipole,.

Over the 19 d deployment, the array drifted southeastward and sampled two distinct regimes – one influenced by a coastal plume and one outside its influence. Surface turbidity and ChlF decreased after 5 d, following the profiler exiting the coastal plume. Below the surface, the pattern was reversed. ChlF at the subsurface chlorophyll maximum (SCM, ChlF_SCM) was initially relatively low and then approximately doubled after the drifting platforms left the plume (14 to 20 July). Meanwhile, turbidity near the SCM (τSCM) initially decreased and subsequently increased as ChlF_SCM increased (Fig. d and e). Outside the plume, turbidity is expected to primarily reflect algal biomass. The observed turbidity outside the plume, especially with its co-variability with ChlF near the SCM, indicates that variability in ChlF reflects changes in biomass and not just pigment concentrations.

Diffuse attenuation was strongly influenced by turbidity and was enhanced within the plume, limiting PAR_SCM (±1 m of 1021.5 kg m⁻³ isopycnal, Fig. b). After the drifting buoys left the plume, however, turbidity and diffuse attenuation decreased, leading to a doubling of maximum day-time PAR_SCM, even though surface PAR was relatively steady. This increase in PAR_SCM was coincident with increasing ChlF_SCM (Fig. a). From 18 to 23 July, wind speeds decreased to an average of 4.4 m s⁻¹, net heat fluxes were consistently large at a daily average of 139 W m⁻², and the total rainfall measured 4.4 mm (Fig. ). Subsequently, the atmospheric conditions transitioned from mostly sunny conditions to mostly cloudy conditions around 23 July with the onset of the “active” phase of the southwest Monsoon (Fig. a). From 23 to 28 July, surface-buoy-measured shortwave radiation and net heat fluxes decreased, with a daily average QN of -39 W m⁻², winds increased to an average of 8.8 m s⁻¹, and rainfall remained low, totaling 28 mm (Fig. ). During this time, PAR_SCM dropped significantly and ChlF_SCM decreased, even though near-surface turbidity and Kd were at a minimum.

The rapid vertical profiling of the Wirewalkers allowed the assessment of bio-optical variability from a time-varying, along-isopycnal frame of reference (Fig. d vs. Fig. a). We use the observed density to convert ChlF to an along-isopycnal reference frame. This approach moderates the effect of passing internal waves that confound measurements taken at discrete depths and highlighted the variability of ChlF at diel timescales (Fig. ). Due to high variability in surface densities, we omit the upper 20 m in the along-isopycnal figure (Fig. a). We also note that due to the relatively dense surface densities from 12 to 18 July, the along-isopycnal figure has missing data (black regions) to up to 30 m depth (≤1021.3 kg m⁻³ isopycnal), as no ChlF observations were made at these densities. The diel cycle at the SCM showed peak concentrations around 3 h after local noon and minimum concentrations at dawn (Fig. a and c). The noon maximum in PAR_SCM (Fig. b teal) coincided with a rapid increase in ChlF_SCM, as found previously in the bay . All records were corrected for NPQ, which had a fairly small effect on ChlF variability.

The consistent timing of irradiance and the diel cycle (e.g., maximum in ChlF rate of change at noon, not shown) has been used to estimate gross production and loss from dissolved oxygen and other biological variables . Here we adapt these diel models to estimate the gross ChlF production (GCP) and the rate of ChlF loss from the on-isopycnal variability of ChlF and co-located PAR at the SCM. By modeling growth from the co-located PAR (PAR_SCM), the impact of variable cloud cover, changes in ocean clarity, or passing internal waves on growth can be accounted for, improving the performance of the model.

The observed diel periodicity and the multi-day trend in ChlF_SCM were effectively emulated by the model for each DBASIS platform (Fig. c and Supplement). The fitted rate of ChlF_SCM growth was closely coupled to ChlF_SCM loss (Fig. d), reflecting strong recycling in this oligotrophic setting. When the rate of ChlF_SCM loss exceeded GCP_SCM, daily-averaged ChlF_SCM decreased even when GCP_SCM and the diel variability in ChlF_SCM remained large (from 21 July, Fig. c and d).

The observed GCP_SCM increased and decreased following respective changes in PAR_SCM. PAR_SCM was a function of shortwave radiation at the surface (largest effect, e.g., 22 July onwards, Fig. ), diffuse attenuation (Kd) in the water column (e.g., 12 to 15 July), and the SCM depth (e.g., 20 to 21 July at M1, see Supplement). Over the deployment, GCP_SCM linearly scaled with the daily averaged shortwave radiation and PAR_SCM (Fig. a and c), with a significant positive correlation (r2≥0.34, p<0.001, and n=40). From this linear regression, as shortwave radiation decreased by 80 % following the transition from the break to active monsoon conditions (21 to 26 July), GCP_SCM decreased by 82 %. Similarly, as the drifting systems left the coastal plume, PAR_SCM and the regression-estimated GCP_SCM increased by 58 % (12 to 21 July). The fit with in situ PAR_SCM was superior to the fit with shortwave radiation due to the influence of the time-variable diffuse attenuation reflected in that measurement, with days with Kd-1≤15 m decreasing the performance of the shortwave radiation fit. These findings suggest that spatial variability in diffuse attenuation, due to ocean clarity, and surface irradiance, due to intraseasonal oscillations in monsoon weather, act together to influence subsurface productivity in the northern Bay of Bengal.

We also note that due to the SCM depth being approximately 10 m deeper at M1 compared to the other profilers, PAR_SCM was lowest, on average, at this profiler. We can slightly improve the fit (MAE decreased by 12 %) of GCP_SCM to PAR_SCM (Fig. b) by accounting for phytoplankton light adaption by relating GCP_SCM to PARSCM/PARSCM‾ instead of PAR_SCM, where PARSCM‾ is the along-isopycnal average of the PAR_SCM at each profiler. This suggests phytoplankton were adapted to a lower irradiance (i.e., isolume) at M1 than the other profilers, but regardless, GCP_SCM similarly increased for larger PAR_SCM.

Despite the occasional non-algal sources of high turbidity, especially at M3, we estimated a good diel cycle fit (r2≥0.4) in turbidity for 28 d in total across the three profilers. The estimated gross turbidity production at the SCM (GTP_SCM) linearly varied with PAR at the surface and SCM, similar to GCP_SCM variability (Fig. b and d). This indicates the co-variability of ChlF_SCM and light availability is not limited to pigment concentration (e.g., photo-acclimation) but also corresponds to changes in biomass.

Our findings of light-limited ChlF could additionally be found by analyzing the instantaneous ChlF_SCM and depth-integrated ChlF, but the relationships were much weaker. We compared depth-integrated ChlF over the upper 80 m to depth-integrated PAR, and ChlF_SCM to PAR_SCM. Both linear regressions returned weak correlations (r2≤0.13) that remained significant (p<0.001, not shown). The instantaneous and depth-integrated ChlF was additionally influenced by ChlF loss and represents the accumulated biomass rather than just growth. Despite this, the strong influence of light availability on ChlF variability remained evident in both instantaneous ChlF_SCM and GCP_SCM.

In the southern BoB, SCM productivity was linked to barrier layer thickness, defined as the depth-region between the mixed layer depth (MLD) base and the isothermal layer ITL,. Here, we define the ITL depth as the shallowest depth where the temperature was at least 1 °C cooler than the sea surface temperature (ΔT≤1 °C), and the MLD as the shallowest depth where the density increase from the surface value corresponds to a temperature decrease of 1 °C. The ITL was on-average within 1.5 m of the SCM depth at M3 and the barrier layer thickness was highly dependent on salinity variations (Fig. a–c).

While M3 was within the coastal plume (before 15 July), the MLD approximately equaled the SCM depth, which continued till 17 July (Fig. a and b). While within the plume, water temperature and salinity at the surface and MLD were different, indicating a very thin barrier layer was present, but ChlF was the same (Fig. d–f). A passing storm resulted in a shallow rain-layer forming around 17 July, the rapid shoaling of the MLD, diverging surface and MLD temperature and salinity see, and a rapid decrease in ChlF at the MLD base (Fig. ). Over the remaining observation period, the MLD deepened until it reached the SCM around July 25. ChlF at the MLD and SCM then converged, and surface ChlF increased despite the transition to active monsoon conditions and decrease in light availability. The thick barrier layer that formed from the rain event should theoretically isolate upper ocean mixing from the SCM. The observed increase in ChlF_SCM and GCP_SCM from 18 July may therefore be due to both the sustained sunny break conditions and a decrease in SCM mixing from the barrier layer formation (Fig. d).

To determine whether our observations of co-varying PAR_SCM and GCP_SCM in the central Bay of Bengal were representative of the region and season, we analyzed the BGC-Argo subsurface ChlF, ERA5 estimated PAR and MODIS-Aqua diffuse attenuation climatology (see Sect.  and ). To illustrate, we show one example of positively correlated PAR_SCM and ChlF_SCM over 12 months (WMO ID: 2902193; r2=0.58, p<0.001, and n=70; Fig. ). At the seasonal time scale, trends in PAR_SCM were more similar to SCM depth than the incident surface PAR, where PAR exponentially decays with depth and so SCM depth can largely control PAR_SCM. In contrast, during the monsoon months of June to September, temporal variability was comparable between surface PAR and ChlF_SCM. This intraseasonal signal was sufficiently strong such that during the southwest monsoon period, surface PAR and subsurface ChlF (averaged over the upper 140 m) were significantly positively correlated (r2=0.26, p<0.01, and n=27), even without accounting for changes in SCM depth.

The BGC-Argo float observed the largest change in ChlF_SCM during the southwest monsoon between the end of July and start of August. The minimum and maximum PAR_SCM around this time were on 23 July and 6 August, with observed ChlF_SCM on these days of 0.38 and 1.65 µg L⁻¹, respectively (Fig. ). The time of observation will also impact ChlF_SCM in this region of strong diel cycles. We note the time of sampling of the minimum and maximum PAR_SCM was at 16:12 and 03:08 LT, respectively, where the DBASIS float observed daily maximum ChlF_SCM at 15:00 LT and minimum at 07:00 LT, on average (Fig. c). The over 300 % increase in ChlF_SCM observed by BGC-Argo was not only due to the transition from active to break monsoon conditions (Fig. ), but also the shoaling of the SCM that further increased light availability.

If we contrast to the DBASIS platform observations, at M3 ChlF_SCM decreased from 0.96 to 0.58 µg L⁻¹ on 21 to 25 July, and at the same local time as the minimum and maximum BGC-Argo sample (Fig. c). This equates to a 40 % difference at M3, compared to the 77 % difference in the BGC-Argo float. We note the change in gross ChlF production at the SCM at the DBASIS platform was double the ChlF_SCM change, at an ≥80 % decrease, where this number is independent of sampling time (Fig. e). This may suggest changes in gross ChlF production due to intraseasonal oscillations at the BGC-Argo exceeded what we observed at M3, given variability in ChlF_SCM was larger at the BGC-Argo float than at M3.

During the 2019 monsoon in the central Bay of Bengal and away from the coastal plume, ChlF and turbidity had a subsurface maximum, denoted the subsurface chlorophyll maximum (SCM, Fig. ), and the SCM almost always coincided with the 1021.5 kg m⁻³ isopycnal. Both ChlF and turbidity also had a diel periodicity at the SCM, further suggesting that variations in ChlF at the SCM represent similar changes in biomass, as has been previously shown in the Indian Ocean waters . The diel periodicity in ChlF has been observed elsewhere in the BoB , and highlights a tight coupling between phytoplankton growth and loss in these oligotrophic waters. At the SCM, the gross production derived from both ChlF and turbidity varied with PAR_SCM (Fig. ). Estimated gross ChlF production at the SCM (GCP_SCM) responded to variations in light from passing monsoon storms and changing ocean clarity (Fig. b and Supplement). GCP_SCM, estimated from the diel cycle in ChlF_SCM, linearly scaled with surface shortwave radiation at each DBASIS system, but the fit was sensitive to changing ocean color and turbidity. By accounting for variations in diffuse attenuation by linearly scaling GCP_SCM with PAR_SCM rather than PAR at the surface (PAR₀), we improved the linear fit (Fig. ). We conclude we observed light limited SCM growth, contrary to prior studies showing the subtropical oligotrophic waters were typically nutrient limited e.g.,.

The estimated GCP_SCM rapidly decreased following the transition from sunny “break” to cloudy “active” monsoon conditions by ≥80 % in the northern BoB (Fig. d and e). This intraseasonal signal in ChlF gross production was also evident in BGC-Argo ChlF measurements from central BoB, with a similar difference in ChlF_SCM levels during the active vs. break periods as that observed by our measurements (e.g., 24 July to 8 August in Fig. a). Due to the southwest monsoon, surface PAR fluctuates on timescales of days to months over the entire bay . Average surface PAR increased from 20 to 8° N, while the monthly standard deviation in surface PAR decreases from north to south (Fig. ). This indicates that the southern BoB waters have higher and more consistent surface PAR, and towards the north, intraseasonal oscillations have a larger impact on surface PAR. We hence predict our observed ≥80 % difference in SCM productivity due to intraseasonal oscillations was representative of northern and perhaps central BoB variability during the southwest monsoon, confirmed by the BGC-Argo observations from central BoB (Fig. ). From the surface PAR fluctuations, we expect weaker but still prevalent oscillations in productivity to occur in more southern waters.

Observations from the southern BoB have highlighted the importance of the barrier layer in supporting SCM productivity by impeding the mixing of the SCM into the mixed layer . Our observations deviated from the southern BoB observation but still confirmed the importance of the barrier layer. The formation of a shallow rain-layer rapidly shoaled the mixed layer, increasing the barrier layer thickness, and isolating the SCM from being eroded by upper ocean mixing (Fig. ). Had the sunny break conditions not followed after a storm event, we expect observed intraseasonal oscillations in ChlF_SCM and GCP_SCM would have decreased. At M3, the erosion of the barrier layer co-occurred with minimum ChlF_SCM rather than when ChlF_SCM was enhanced, like observations by . The erosion of the barrier hence had minimal impact on the already decreased SCM productivity, but likely contributed to the observed decrease in ChlF_SCM. We also note that at M3, the bulk of the SCM productivity was below the barrier layer, while observed enhanced ChlF within the barrier layer when it was present. As the SCM is typically located at the deepest depth where PAR is sufficient for growth and the shallowest depth where nutrients are available, our observations may suggest a relatively deeper nutricline than in the southern BoB. The barrier layer still separated the SCM from upper ocean mixing, and hence will influence SCM productivity, as found in the southern BoB .

Both the DBASIS and BGC-Argo observations indicate the intraseasonal variability in surface light project onto the SCM primary productivity. These observations suggest propagating coupled ocean-atmosphere intraseasonal weather patterns may leave a biological footprint of time- and space-variable primary productivity within the Bay of Bengal. Barrier layer formation is additionally impacted by intraseasonal oscillations , and will impact ChlF vertical distributions and hence SCM productivity. Other biogeochemical properties may also be dependent on intraseasonal weather patterns, since primary productivity modulates dissolved oxygen, particulate organic carbon, and dissolved inorganic carbon concentrations . Similar patterns could also be evident in other tropical and subtropical oceans subject to intraseasonal atmospheric variability.

Although we focus on the light limitation on primary productivity in this manuscript, grazing by higher trophic levels can also constrain phytoplankton variability. To provide some information on this component of the food web, we used acoustic backscatter measurements from 1 MHz Acoustic Doppler Current Profilers onboard the Wirewalkers Fig. ,. The backscatter measurements showed enhanced values at night in shallow waters with depths less than 50 m and density of 1022 kg m⁻³ or less (Fig. ). Then during the day, backscatter measurements were enhanced in deeper waters at around 60 m depth. These observations are consistent with a shallow diel migration of zooplankton between around 60 m at daytime to near the SCM at nighttime. Although we do not have observations to confirm the size and species present, as highlighted in the “Introduction”, previous analysis with a 1 MHz acoustic backscatter confirms its sensitivity to smaller zooplankton species, including grazers . These smaller species tend to have slower and shallower vertical migrations than larger species , so the shallow diel migration in Fig. b is consistent with a small zooplankton species.

The acoustic backscatter strength near the SCM increased over time and reached a maximum on 24 July (Fig. b). During this period of increasing acoustic backscatter, the rate of ChlF_SCM loss began to exceed that of GCP_SCM (Fig. d), leading to a decrease in Chl_SCM from 21 July. This suggests that grazing could have contributed to the increased loss rate and might provide a link between intraseasonal variability in primary production and higher trophic levels. highlighted that a better understanding of barrier layer formation under variable phytoplankton growth rates and zooplankton grazing is required. Here, we reveal important linkages between these processes and additionally highlight the importance of the intraseasonal oscillations in imparting periodicity to phytoplankton growth at the SCM.