Decoupling of 1O2/Ar and particulate organic carbon dynamics in nearshore surface ocean waters

We report results from two Lagrangian drifter surveys off the Oregon coast, using continuous shipboard sensors to estimate mixed-layer gross primary productivity (GPP), community respiration (CR), and net community production (NCP) from variations in biological oxygen saturation (1O2/Ar) and optically derived particulate organic carbon (POC). At the first drifter survey, conducted in a nearshore upwelling zone during the development of a microplankton bloom, net changes in1O2/Ar and [POC] were significantly decoupled. Differences in GPP and NCP derived from 1O2/Ar (NCPO2/Ar) and POC (NCPPOC) time series suggest the presence of large POC losses from the mixed layer. At this site, we utilized the discrepancy between NCPO2/Ar and NCPPOC, and additional constraints derived from surface water excess nitrous oxide (N2O), to evaluate POC loss through particle export, DOC production, and vertical mixing fluxes. At the second drifter survey, conducted in lower-productivity, density-stratified offshore waters, we also observed offsets between 1O2/Ar and POCderived GPP and CR rates. At this site, however, net [POC] and 1O2/Ar changes yielded closer agreement in NCP estimates, suggesting a tighter relationship between production and community respiration, as well as lower POC loss rates. These results provide insight into the possibilities and limitations of estimating productivity from continuous underway POC and 1O2/Ar data in contrasting oceanic waters. Our observations support the use of diel POC measurements to estimate NCP in lower-productivity waters with limited vertical carbon export and the potential utility of coupled O2 and optical measurements to estimate the fate of POC in highproductivity regions with significant POC export.

in POC. This diurnal variability results from the daytime accumulation of photosynthetically produced organic carbon, as well as nighttime loss of fixed carbon through community respiration, and can thus be used to infer NCP on daily timescales. The accuracy of this approach depends on the key assumption that variations in c p capture most of the variability in POC concentration, and it has been shown that beam attenuation is most sensitive to particles with a diameter range of 0.5-20 µm (Stramski and Kiefer 1991;Marra, 2002;Claustre et al., 2008). To date, most efforts to calculate daily NCP from c p variability have focused on lowproductivity offshore regions, where particle sizes are small and POC losses through particle export are limited (Claustre et al., 2008;White et al., 2017). These studies have reported good agreement between optically derived GPP estimates and independent estimates of NPP from 14 C incubations (White et al., 2017), suggesting a tight coupling between primary productivity and mixed-layer POC dynamics over daily timescales.
Another approach to NCP quantification is based on autonomous measurements of surface water dissolved-oxygento-argon ratios (O 2 /Ar). Argon normalization is used to correct for any physically induced changes in O 2 saturation, such that the derived saturation anomaly, O 2 /Ar, is a tracer of net biological O 2 production (Kaiser et al., 2005;Tortell, 2005;Cassar et al., 2009). At steady state, and in the absence of significant lateral advection and vertical mixing, the sea-air flux of excess biologically produced O 2 is equivalent to NCP. With the development of automated shipboard mass spectrometers, there has been a significant expansion of surface water O 2 /Ar measurements. These data have been used to examine O 2 variability resulting from diurnal variations of photosynthesis and respiration, as well as to infer NCP in a variety of oceanic ecosystems (Reuer et al., 2007;Stanley et al., 2010;Tortell et al., 2011Tortell et al., , 2014Hamme et al., 2012;Nicholson et al., 2015;Manning et al., 2017). Recent efforts have shown that NCP estimates from O 2 /Ar measurements can be corrected for vertical mixing using water column N 2 O measurements as a tracer (Cassar et al., 2014;Izett et al., 2018), but application of this methodology assumes that lateral advective fluxes of O 2 are negligible.
Combined measurement of mixed-layer POC and O 2 dynamics holds the potential to better constrain surface water carbon budgets in biogeochemically dynamic regions at high spatial and temporal resolution. In net autotrophic systems, an increase in O 2 /Ar reflects the accumulation of excess photosynthetic O 2 in the mixed layer but provides no direct insight into the fate of the resulting organic carbon. In the absence of particle export, grazing or DOC production, an increase in O 2 /Ar, corrected for air-sea exchange and vertical mixing, should be matched by a parallel increase in POC accumulation measured by optical sensors. By comparison, high POC export, DOC production, or grazing coupled to vertical migrations would act to decouple O 2 /Ar from optically derived POC measurements in the mixed layer.
Previous authors have used simultaneous O 2 and c p measurements on moorings to describe mixed-layer O 2 and POC dynamics in various marine environments (Stramska and Dickey, 1992;Kinkade et al., 1999;Dickey and Chang, 2002). However, few studies to date have compared estimates of primary productivity from simultaneous measurements on daily timescales. Briggs et al. (2018) and Alkire et al. (2012) were the first to explicitly combine concurrent measurements of O 2 and POC from in situ autonomous sensors to quantify mixed-layer productivity during a ∼ 2-month Lagrangian study of the 2008 North Atlantic spring bloom. Tracking daily changes in mixed-layer O 2 and POC concentrations, Alkire et al. (2012) constructed a detailed budget of surface ocean organic carbon throughout the course of the bloom, using the difference between O 2 -based NCP and net POC accumulation to assess the partitioning of NCP into different carbon pools (sinking particles, phytoplankton biomass, and DOC). Building on this work, Briggs et al. (2018) examined the role of respiration, particle export, and DOC production in decoupling O 2 and POC dynamics through different bloom stages, demonstrating significant differences between GPP estimates derived from O 2 , beam attenuation, and backscatter measurements. To our knowledge, such a detailed examination of O 2 and POC dynamics has not been reported for other marine systems.
Here, we present new results from a field study of diel variability in O 2 /Ar and optical properties in two contrasting nearshore regions of the subarctic North Pacific. Using shipboard automated sensors deployed along a Lagrangian drifter track, we resolved fine-scale temporal patterns in biological oxygen production and POC concentration in a high-productivity coastal upwelling zone over the continental slope and in lower-productivity stratified waters offshore. The biogeochemical differences between both sites provided a unique opportunity to compare GPP, CR, and NCP estimates derived from O 2 /Ar and POC in contrasting trophic regimes. We expected to observe significant differences between O 2 /Ar and POC-derived GPP, CR, and NCP estimates in the higher-productivity site, reflecting greater carbon export capacity and DOC production. By comparison, we hypothesized that discrepancies in these rates would be smaller at the lower-productivity site, reflecting a tighter coupling between O 2 and POC dynamics.
The results of this investigation extend the results from the 2008 North Atlantic bloom to a high-productivity coastal upwelling environment where vertical mixing fluxes significantly influence the surface water mass balance. These dynamic systems play a disproportionately important role in marine biogeochemical cycling, but they pose significant challenges for interpreting time series of ecosystem metabolism. Furthermore, our study results further expand applications of a recent field approach to correcting NCP for vertical mixing , suggesting that this approach has significant merit in reconstructing productivity estimates from a variety of mixed-layer tracers. We dis-cuss the implications of our coupled O 2 -POC measurements for understanding biological carbon cycling in marine waters and suggest some approaches to improve the utility of these measurements for evaluating the fate of marine primary productivity across marine trophic gradients.

Field site and Lagrangian surveys
Field studies were conducted on board the R/V Oceanus in August 2017, during a transect through the northeastern subarctic Pacific Ocean. Two Lagrangian drifters were deployed off the Oregon coast, allowing us to track diurnal patterns in phytoplankton productivity and particulate organic carbon cycling in two distinct water masses (Fig. 1). Underway temperature and salinity measurements, collected by a Sea-Bird SBE 45 thermosalinograph, as well as satellite-(Aqua MODIS) and ship-based chlorophyll-a (Chl-a) observations, were used to guide the specific location and timing of the drifter deployments. Drifter 1 was deployed on 20 August 2017 (∼ 09:30 PDT), ∼ 40 km from the Oregon coast (44.54 • N, 124.58 • W), in the vicinity of an upwelling feature detected based on low sea surface temperature (SST) and elevated salinity and [Chl-a]. The drifter, consisting of a beacon, GPS transmitter and 5 m drogue, was recovered at ∼ 18:30 on 23 August 2017 (44.40 • N, 124.55 • W) for a total deployment of 3 d and 9 h. Upon recovery, the drogue was missing, implying the potential for some erratic subsurface drifting (discussed below). Drifter 2 was deployed approximately 200 km from shore (43.75 • N, 126.50 • W) in a relatively warm and low-salinity water mass, with low Chl-a concentrations. This second drifter was deployed at ∼ 07:45 on 24 August 2017 and was recovered after 2 d and 6 h at ∼ 14:00 on 26 August 2017 at 43.80 • N, 126.99 • W. Because the R/V Oceanus lacks a dynamic positioning system, the ship was not always able to perfectly track the drifter locations. To correct for these positional offsets, we discarded any observations obtained when the ship was more than 1.5 km away from the drifter location. This filtered data set resulted in underway measurements (Sect. 2.2) every ∼ 15 min during the two drifter deployments, yielding 325 and 218 quality-controlled underway observations for drifters 1 and 2, respectively.

Underway measurements
Continuous underway measurements of surface seawater optical properties were collected using Sea-Bird (formerly WET Labs) ECO-BB3 and ac-s sensors, following the methods outlined in detail by . Water was collected from the ship's seawater supply system with a nominal intake of 5 m depth. Our instrument package included fully automated data collection and hourly filtered blanks (0.2 µm), which provided measurements of dissolved sea-water optical properties used to infer particulate absorption (a p ) and beam attenuation (c p ) at 82 wavelengths between 400 and ∼ 735 nm, as well as backscatter (b bp ) at 470, 532, and 650 nm. The BB3 and ac-s measurements were binned into 1 min intervals. Prior to binning, the absorption and beam attenuation data were first subsampled every 50 data acquisition cycles (∼ 12.5 s) to enable faster processing time. The optical measurements were accompanied by continuous surface photosynthetically active radiation (PAR) and wind speed data obtained from a Biospherical QSR-220 PAR sensor and Gill WindObserver II ultrasonic wind sensor mounted on the ship's bow.
Chlorophyll-a (Chl-a) concentrations were derived from the particulate absorption line height at 676 nm (a LH ) (Roesler and Barnard, 2013). Five-minute matchups between underway a LH and discrete filtered [Chl-a] measurements from the entire cruise transect (Sect. 2.4) were used to derive a best-fit coefficient for the linear relationship between a LH and [Chl-a] (r 2 = 0.87, n = 58, p < 0.01). Particulate organic carbon (POC) concentrations (µg L −1 ) were derived from particulate beam attenuation at 660 nm (c p,660 ), using the empirical model in Graff et al. (2015). Similarly, phytoplankton organic carbon (C ph ) concentrations were calculated, using an empirical relationship between particulate backscatter at 470 nm (b bp,470 ) and [C ph ] in micrograms per liter (µg L −1 ; Graff et al., 2015). We used a limited set of 5 m discrete measurements (n = 6; Sect. 2.4) to evaluate the relationship between POC concentrations and c p at 660 nm, as well as the applicability of the Graff et al. (2015) model to our observations. As shown in Fig. S1 in the Supplement, the POC measurements were significantly correlated to c p (r 2 = 0.88, p < 0.05), with a slope and intercept of 391.6 ± 201.6 and 36.7 ± 79.1, respectively. This slope was not significantly different from that of the Graff et al. algorithm (419.8) although our y intercept was higher. Notwithstanding the relatively small number of discrete POC samples, and some scatter around the regression line, the similarity of our POC-c p calibration to that reported by Graff et al. (2015) suggests that our optically derived POC estimates are reasonably robust.
To obtain information on the particle size spectrum, we derived the wavelength-dependent slope of particulate backscatter by fitting the three b bp coefficients (470, 532, 650 nm) to an exponential equation (Stramska et al., 2003;Loisel et al., 2006;Kostadinov et al., 2009). Finally, to assess interference of inorganic minerals on POC, and C ph variability, we calculated the wavelength-specific bulk refractive index (η p ) from backscatter / total scatter ratios b bp c p −a p and the wavelength-dependent c p slope, following the approach of Boss et al. (2001), Twardowski et al. (2001), and Sullivan et al. (2005).
In addition to optical measurements, the seawater biological oxygen saturation anomaly ( O 2 /Ar) was measured at a ∼ 20 s resolution using a membrane inlet mass spectrometer show a detailed view of the two drifter tracks (cross symbol), with the ship's track shown in a light gray line and open circles denoting times when the ship was < 1.5 km away from the drifter position. Only measurements taken at these crossover locations were used for analysis.
connected to the ship's seawater intake. The seawater ratio of dissolved O 2 and Ar was determined by diverting a continuous flow of water across a dimethylsilicone membrane interfaced with a Hiden Analytical HAL20 triple filter quadrupole mass spectrometer. The O 2 /Ar ratio of air-equilibrated standards ([O 2 /Ar] eq ), incubated at ambient sea surface temperature, was measured every 2 h. Values of O 2 /Ar were thus calculated as the percent deviation of seawater O 2 /Ar measurements from the air-equilibrated ratio, using O 2 /Ar = 100 % · ([O 2 /Ar] meas /[O 2 /Ar] eq − 1) (Tortell, 2005;Tortell et al., 2011).

Mixed-layer depth
Over the course of both drifter deployments, we conducted regular sampling casts (every 6 to 10 h) to examine depth profiles of seawater hydrography and biogeochemical variables. Temperature, salinity, dissolved O 2 concentrations, and Chla fluorescence profile data from the CTD casts were measured by a Sea-Bird SBE 38 temperature sensor, Sea-Bird SBE 4 conductivity sensor, SBE 43 dissolved O 2 sensor, and a Sea-Bird ECO fluorometer, respectively, and binned into 1 m intervals. Vertical profiles at the drifter 1 site showed relatively weak density stratification, likely as a result of recent upwelling. For this reason, we estimated mixed-layer depths (z mld ) based on visible inflection points in the dissolved [O 2 ], fluorescence, and density profiles, assuming that dissolved O 2 concentrations and fluorescence are relatively uniform in the mixed layer. Within a single CTD cast, mixed-layer depths varied by up to 28 % across all three profile measurements. The [Chl-a] fluorescence profiles had the most well-defined inflection points, and we thus used these data to estimate z mld at all casts. Excluding fluorescence profiles from the first day (Sect. 3.1) and two casts at 06:00 and midnight on second and third 24 h intervals, respectively, which displayed relatively noisy density profiles, an average z mld value (19 ± 2 m) was derived and applied to all subsequent analyses.
In comparison to the drifter 1 site, CTD cast profiles during drifter deployment 2 showed larger density gradients. We thus computed z mld using a density difference criterion of 0.25 kg m −3 ( Thomson and Fine, 2003;de Boyer Montégut et al., 2004) from median values within the uppermost 4-6 m of the profile. We found that this critical density criterion was necessary to capture the depth of inflection in O 2 and [Chla]. In all CTD casts except one, density difference-based z mld values were within 5 m of the values derived from the inflection points on density profiles. An average z mld value estimated from the density-difference approach (22 ± 5 m) was applied to all subsequent analyses.

Discrete samples
Concentrations of phosphate ), dissolved silica ([SiO 2 ]), and nitrate and nitrite ([NO − 3 + NO − 2 ]) were measured in seawater samples collected from daily Niskin bottle casts. Following collection, nutrient samples were filtered through 0.2 µm pore polycarbonate membranes and immediately frozen at −80 • C on board the ship. These samples were stored at −20 • C until subsequent colorimetric laboratory analyses (Murphy and Riley, 1962;Riley, 1977) with a Lachat QuikChem 8500 Series 2 Flow Injection Analysis System.
Concentrations of nitrous oxide (N 2 O) were measured in discrete samples collected in Niskin bottles during both drifter deployments (Fig. S2), following methods outlined in Capelle et al. (2015). These N 2 O measurements were used to correct NCP estimates for vertical mixing (see Sect. 2.6), following the approach described by Cassar et al. (2014) and Izett et al. (2018). Profile samples from the first day of drifter deployment 1 (20 August) were omitted from calculations, as underway surface temperature and salinity measurements indicated intrusion of an external water mass (further discussed in Sect. 3.1) (Fig. S3). Three profiles collected from 12:00 (PDT) CTD casts during the following 3 d of the deployment (21, 22, and 23 August) were applied to the NCP mixing correction at drifter station 1 (Sect. 2.6.1).
Surface (∼ 5 m) discrete seawater samples were collected either from Niskin bottles or from the ship's surface seawater intake system for high-performance liquid chromatography (HPLC) analysis of Chl-a concentrations and other phytoplankton pigments. Single or duplicate samples were filtered onto 25 mm GF/F filters, flash-frozen in liquid nitrogen, and stored at −80 • C until analysis, following the methodology described in Schuback et al. (2016). Additional samples were collected from the seawater intake for size-fractionated Chla analysis . These samples were filtered through stacked 47 mm filters (0.2, 2 and 20 µm pore size) separated by a mesh spacer. Filtered samples were extracted in 5 mL of 90 % acetone at 4 • C until analysis within 24-48 h using a Turner Trilogy fluorometer on board the ship.
Discrete samples for POC analysis were collected at two depths from several CTD casts. Surface samples were collected at both drifter sites from 5 m depth, while deeper samples were collected at near the base of the euphotic zone (∼ 1 % PAR), corresponding to 40-60 m at drifter site 1 and 100-120 m at drifter site 2. POC samples (∼ 1-4 L) were filtered through a precombusted (450 • C) Whatman GF/F filter (nominal pore size ∼ 0.7 µm) and stored at −80 • C until laboratory analysis. Prior to analysis, samples were thawed and dried at 50 • C overnight, fumigated with concentrated hydrochloric acid for 48 h, and dried again at 50 • C overnight. POC concentrations in samples (and blank combusted filters treated as described above) were quantified using an Elementar vario MICRO cube CHNS analyzer. Blank-corrected discrete POC concentrations were used to validate application of the [POC] model in Graff et al. (2015) to our underway c p data (Sect. 2.2; Fig. S1).

Net primary productivity
Daily integrated net primary productivity (NPP) was calculated in two ways. First, carbon uptake was determined from 24 h 14 C incubations with 5 m triplicate seawater samples collected from early morning CTD casts. Measurements were made on two different mornings during drifter deployment 1 and on one morning during drifter deployment 2. The measurements were conducted following the protocol outlined in Hoppe et al. (2017). Depth-integrated NPP was calculated by multiplying the derived 24 h volumetric carbon fixation rate by the average mixed-layer depth for the respective drifter period.
Second, daily integrated net primary productivity was also estimated as a product of [C ph ] values derived from b bp and phytoplankton growth rates according to the carbon-based production model (CbPM) (Behrenfeld et al., 2005;Westberry et al., 2008;Graff et al., 2016;. In these calculations, daily averaged [C ph ], [Chl-a]/[C ph ], and mixed-layer irradiance (E g ) calculated from the MODISderived surface PAR matched to drifter location were used to calculate growth rates and NPP every 24 h. Chlorophylla concentrations were derived from absorption line height, [C ph ] values from b bp (Sect. 2.2), and light extinction coefficients (K d ) obtained from [Chl-a] to calculate E g (Morel et al., 2007). An average mixed-layer depth for each drifter period was applied to estimate mixed-layer NPP (Sect. 2.3).

Quantification of GPP, CR, and NCP
Gross primary productivity (GPP), community respiration (CR), and net community production (NCP) rates were calculated based on linear regressions of O 2 /Ar and POC against time (dt in units of days) over subsequent day (d) and night (n) intervals during both drifter deployments. Daytime was defined as the period during which PAR levels exceeded 20 µmol quanta m −2 s −1 . The average length of the daytime period was 13.6 ± 0.14 h over the two drifter deployments. In the following sections, t d represents the day length normalized to 24 h, and t n analogously represents the fractional night length, equivalent to 1 − t d . All daily rates were integrated through the mixed layer using the average z mld for each drifter period, as described in Sect. 2.3.

O 2 /Ar-derived rates
Quantification of GPP O 2 /Ar , CR O 2 /Ar , and NCP O 2 /Ar rates from diurnal cycles in O 2 /Ar (Ferrón et al., 2015) requires corrections for gas exchange and, potentially, vertical mixing fluxes. For these calculations, we first computed the rate of change in O 2 /Ar (dO 2bio /dt) using linear regression analysis within successive day or night intervals. We then derived estimates for the air-sea gas exchange (J ex ) and vertical mix-3282 S. Z. Rosengard et al.: Decoupling of O 2 /Ar and particulate organic carbon dynamics ing fluxes (F mix ) over the respective time interval to isolate the NCP contribution to observed O 2 /Ar changes Tortell et al., 2014). A negative J ex indicates net transfer of O 2 from the atmosphere to the mixed layer, while a negative F mix indicates vertical transfer of O 2 /Ardepleted to the mixed layer, both in units of millimoles per square meter per day (mmol m −2 d −1 ). Gross O 2 production rates were converted into carbon units using a photosynthetic quotient (PQ) for new production of 1.4 for drifter-period-1 calculations and a PQ for regenerated production of 1.1 for drifter period 2 (Laws, 1991). Community respiration rates were converted into carbon units using the same PQ values and considered constant over each respective day length period (i.e., t d + t n ). This assumption of an equivalent respiratory quotient (RQ) and PQ within each drifter period is reasonable given the wide range of respiration ratios reported in prior studies across a range of oceanic environments (Anderson and Sarmiento, 1994;Robinson et al., , 2002Hedges et al., 2002;Lønborg et al., 2011;Daneri et al., 2012;Fernández-Urruzola et al., 2014). Moreover, Robinson and Williams (1999) estimated lower RQ values at lower-productivity stations in the Arabian Sea, suggesting that it is reasonable to assume a lower RQ value (equivalent to PQ = 1.1) at drifter site 2.
Equilibrium concentrations of O 2 and N 2 O ([O 2 ] eq and [N 2 O] eq ) were calculated using the salinity-and temperature-dependent equations of Garcia and Gordon (1992) and Weiss and Price (1980), respectively, and sea surface temperature and salinity from the ship's thermosalinograph. Estimates of surface excess N 2 O saturation, [N 2 O] bio , included a heat flux correction to account for solubility changes (Keeling and Shertz, 1992;Jin et al., 2007;Izett et al., 2018). Nonweighted piston velocities (k O 2 and k N 2 O ; units of m d −1 ) were calculated using the diffusive airsea gas flux and Schmidt number parameterizations of Wanninkhof (2014) and Raymond et al. (2012), as well as shipbased wind speed data 10 m above the sea surface. Daytime and nighttime estimates for the gas exchange term, J ex , were calculated using day or night average [O 2 ] eq , O 2 /Ar, and k O 2 values. Vertical gas gradients dN 2 O bio dz and dO 2bio dN 2 O bio were estimated from our discrete N 2 O measurements and rosette O 2 profiles over the upper 100 m of the water column, following Izett et al. (2018).
At drifter site 1, daily F mix values were calculated using daily [N 2 O] bio , daily vertical gradient, and daily average k N 2 O values and converted to carbon units using a PQ of 1.4. Denitrification should not have been a source of N 2 O within the upper 100 m of the water column because measured O 2 concentrations were consistently greater than the threshold value of ∼ 50 mmol m −3 (e.g., Hopkinson and Barbeau, 2007). Likewise, we assumed no lateral advection of N 2 O into drifter site 1, as there were little differences in the mixing ratio [O 2 ] bio /[N 2 O] bio across profile measurements (Fig. S2). While the 22 August CTD cast did exhibit a more anomalous [O 2 ] bio /[N 2 O] bio profile relative to the other two cast profiles, inclusion of these data had little impact on the vertical mixing correction. At drifter site 2, we assumed that vertical mixing was negligible due to the presence of strong density stratification and therefore did not calculate a mixing flux correction at this site. In any case, the presence of a subsurface O 2 maximum ( Fig. S2) at this site would limit the application of the N 2 O correction .

Optically derived rates
We used the approach of Claustre et al. (2008) and White et al. (2017) to calculate daily integrated GPP POC , CR POC , and NCP POC from daytime and nighttime changes in POC (dPOC/dt), derived from linear regressions of POC concentrations against time through day and night intervals. In certain ocean environments, NCP POC will not equate to NCP O 2 /Ar as a result of additional POC sinks, including export, grazing, and DOC production. Under these conditions, CR POC includes these loss terms, and therefore NCP POC more accurately reflects net POC accumulation, as will be discussed further in Sect. 4. Nonetheless, for consistency with previous studies, we use the term NCP POC to describe the quantities computed in Eq. (8).
The presence of significant upwelling at drifter site 1 provides additional complexity in the estimate of NCP from optically derived POC measurements. In particular, vertical transport of particle-deficient seawater from below the mixed layer into the surface could dilute the c p signal used to derive POC concentrations (Stramska and Dickey, 1994). To address this, we applied the vertical mixing term, k mix , derived from Eq. (6) to estimate the average daily dilution effect on mixed-layer POC concentrations through drifter period 1: In total, three sets of 24 h GPP, CR, and NCP values were calculated during the drifter 1 deployment from the three pairs of consecutive day and night intervals, starting with the first night interval and ending with the last day interval. We excluded the first daytime interval from our calculations, due to the erratic salinity values observed during the first day of this drifter deployment (Sect. 3.1; Fig. S3). Because the drifter period was terminated prior to sunset, the last day interval was 1.6 h shorter than the average daytime duration. For the second drifter deployment, two sets of GPP, R, and NCP values were calculated from consecutive day and night intervals, starting with the first daytime interval and ending with the last nighttime interval. The initiation of the drifter period occurred after sunrise, so the first day interval was 1.1 h shorter than the average daytime duration.

Integration timescales
The approach to calculating NCP on the basis of linear regressions utilizes the high temporal resolution of our data set. We compared our results from Sect. 2.6.1 and 2.6.2 to NCP values calculated using several of other integration timescales. Following studies that have calculated daily NCP values from "instantaneous" rates of change (e.g., hourly rates in Hamme et al., 2012, andTortell et al., 2014), we divided our NCP calculations into shorter increments. Given that the average measurement interval was ∼ 15 min (after removing values where the ship was not sufficiently close to the drifter; Sect. 2.1), we calculated NCP within 3 h intervals: (1)-(9) were used to calculate NCP, and a PQ was used to convert O 2 to carbon units.

Error analysis
Errors for all estimates of net primary productivity (CbPM-NPP, 14 C-NPP) and net community production (NCP O 2 /Ar , NCP POC ) were propagated from uncertainties associated with all variables used for the computations. Error estimates for time-averaged variables were generally represented by the standard deviation, as we assumed that this significantly exceeded the error of the individual measurements prior to averaging. The uncertainty in z mld , derived from the standard deviation of mixed-layer depths across individual CTD casts, was 2 m for drifter site 1 and 5 m for drifter site 2 (Sect. 2.3). Small uncertainties in t d and t n were calculated as the standard deviations of all day or night lengths measured during both drifter periods (0.14 and 0.10 h, respectively was calculated as the confidence interval of the best-fit slope extracted from a linear regression of pooled drifter 1 [N 2 O] bio values against depth. In propagating the error associated with the dPOC/dz term in Eq. (10), we have included the standard deviation of the minimum transmissivity depth across daily CTD casts and the standard deviation of POC measured in multiple blank combusted filters (Sect. 2.4). Finally, to account for variability in the PQ and RQ, we assumed an uncertainty of 0.1, following the range reported in Laws (1991).

Water mass properties
Shipboard underway measurements revealed clear differences in hydrographic and biogeochemical characteristics between the water masses sampled by the two drifters. Surface water properties at drifter site 1 reflected the presence of a recently upwelled water mass that was relatively cold (11.8±0.4 • C), saline (32.6±0.04 g kg −1 ), and nutrientrich (Figs. 1, S3, S4). The Pacific Fisheries Environmental Laboratory's coastal upwelling index at 45 • N, 125 • W was positive throughout drifter period 1. In contrast, the water mass tracked by the second drifter deployment was warmer (17.5 ± 0.1 • C) and fresher (31.8 ± 0.05 g kg −1 ), with lower average mixed-layer nutrient concentrations.
Examination of surface water hydrographic properties during the two drifter deployments suggests that both drifters tracked a relatively homogenous water mass, excluding a period of salinity variability during the first day of drifter deployment 1, and several transient temperature and salinity excursions after the second night of this deployment (gray patches in Fig. S3). These features indicate potential intrusion of external water masses, possibly a result of loss of the drifter drogue (Sect. 2.1). Observations during these periods were thus removed from the data set prior to analysis. Outside of these intervals, variability in salinity (drifter 1: 32.5-32.7 g kg −1 ; drifter 2: 31.8-31.9 g kg −1 ) was small during both drifter deployments. Variability in sea surface temperature was also limited (drifter 1: 11.2-13.0 • C, drifter 2: 17.3-17.7 • C) and largely reflected a diurnal variation of warming and cooling, which was particularly evident for drifter period 2.
Temporal differences in CTD cast profiles point to some variation in mixed-layer depth (z mld ) during both drifter deployments. In general, there were no multiday trends or regular diurnal patterns in z mld through both periods, suggesting that transient shifts in water column turbulence likely contributed to changes in the shape of temperature, salinity, dissolved oxygen, and fluorescence profiles. Average z mld values, calculated over each drifter period, had relatively low relative standard deviations (< 25 %) and were applied to all subsequent calculations (Sect. 2.3). A sensitivity analysis, not shown, indicated that the choice of mixed-layer depth using different criteria (i.e., fluorescence profiles, density profiles, and the density difference criterion) and different timescales of integration (i.e., daytime or nighttime, 24 h, and multiday) did not significantly impact the results discussed below.
Average mixed-layer nutrient concentrations fluctuated during both drifter deployments but did not exhibit regular diurnal cycles (Fig. S4). At drifter site 1, concentrations ranged from 0.74 to 0.85 µM phosphate, 7.8 to 9.0 µM nitrate and nitrite, and 9.2 to 11.1 µM dissolved silica, excluding day 1 of the drifter deployment and anomalously high concentrations measured during a noisy CTD cast at midnight on the last day of the deployment. Excluding these outliers, a significant (p < 0.05) linear regression of each nutrient concentration against time revealed that phosphate concentrations decreased by ∼ 0. ] increased significantly (p < 0.05) by 0.015 and 0.48 µM, respectively, these changes were small compared to the nutrient drawdown observed during drifter period 1 and did not reflect Redfield ratio proportions. It is possible that intrusions of an external water mass with slightly elevated nutrient concentrations contributed to the small increase in [PO 3− 4 ] and [SiO 2 ] measured during these CTD casts, even though we assume that such effects on our derived productivity estimates are negligible based on inspection of underway temperature and salinity data (Fig. S3).

Biogeochemical comparisons between drifter sites
Elevated nutrient concentrations at the drifter 1 site supported high productivity and the accumulation of phytoplankton biomass, as indicated by elevated chlorophylla ([Chl-a] = 0.66-1.5 µg L −1 ), phytoplankton carbon ([C ph ] = 83-115 µg L −1 ), and particulate organic carbon concentrations ([POC] = 130-261 µg L −1 ) (Fig. 2a-c). We observed [C ph ]/[Chl-a] ratios ranging from 68 to 143 g g −1 , with a median value of 85 g g −1 (Fig. 2f). Using the carbonbased production model (CbPM; Sect. 2.5) and daily averaged mixed-layer PAR derived from satellite values matched to drifter location (within 5 km), these [C ph ]/[Chl-a] ratios translate into phytoplankton growth rates ranging from 0.75 to 0.94 d −1 . At the second drifter site, phytoplankton productivity and biomass were significantly lower in the nutrient-poor waters ([Chl-a] = 0.06-0.21 µg L −1 , [C ph ] = 11-17 µg L −1 , and [POC] = 25-38 µg L −1 ). Ratios of [C ph ] to [Chl-a] at site 2 were significantly higher (p < 0.05) than those observed at site 1, ranging from 69 to 203 g g −1 , with a median value of 108 g g −1 . The higher ratios may reflect reduced cellular [Chl-a] associated with greater nutrient limitation, higher daily integrated PAR, and proportionally more picoplankton than microplankton at drifter site 2 (Westberry et al., 2008;Hirata et al., 2011;Graff et al., 2016;. Median PAR levels were higher and less variable at site 2, in part contributing to lower variability in CbPM-based growth rates, which ranged from 0.81 to 0.85 d −1 . Several lines of evidence suggest that the phytoplankton assemblage at drifter site 1 was enriched in large-celled phytoplankton, as compared to drifter site 2. The wavelengthdependent slope of particulate backscatter (b bp ) was lower at site 1 (range: 1.4 to 1.6, median: 1.5) than at site 2 (range: 1.9 to 2.3, median: 2.1) (Fig. 2d), suggesting proportionally larger particle sizes (Stramska et al., 2003;Kostadinov et al., 2009). This observation is supported by size-fractionated Chl-a measurements. During the drifter 1 deployment, the > 20 µm size fraction (Sect. 2.4) increased from 21 % to 46 % of the total Chl-a pool, indicating the enrichment of large phytoplankton in the assemblage. Pigment-based estimates of phytoplankton taxonomic composition and size class (Hirata et al., 2011;Zeng et al., 2018) suggested that relative diatom and microplankton abundances exceeded 50 % on the final sampling time point. By comparison, size-fractionated [Chl-a] and HPLC analyses from drifter 2 indicated a lower proportion of large-celled phytoplankton, with 9 %-15 % of total [Chl-a] in the > 20 µm size fraction, and diatoms and microplankton estimated to account for 19 %-29 % of the phytoplankton assemblage. The proportion of picoplankton increased through time at drifter site 2 from 31 % to 50 % of total [Chl-a], alongside a slight increase in b bp slope, indicating accumulation of smaller particle sizes (Fig. S3d). Finally, median bulk refractive index values across three wavelengths (470, 532, 650 nm) were higher at site 1 (1.08-1.11) than at site 2 (1.02-1.04) (Fig. S3e), which is consistent with a greater proportion of diatom-derived silica in the particle pool (Lide, 1997;Twardowski et al., 2001).

Diurnal variability and primary production
As shown in Fig. 3a, clear diurnal cycles in biological oxygen saturation ( O 2 /Ar) were observed during both drifter deployments. Generally, values of O 2 /Ar increased from dawn to dusk and decreased from dusk to dawn, yielding positive slopes of linear regressions of O 2 /Ar against time in the daytime and negative slopes at night. During drifter deployment 1, this diurnal cycle was superimposed on a longer-term increase in biological O 2 saturation as undersaturated values returned toward atmospheric equilibrium. At least part of this increase is attributable to gas exchange, which would act to erase O 2 undersaturation in the mixed layer caused by recent upwelling. However, calculation of the sea-air O 2 flux shows that, except for the first 24 h period, only a small amount of the daily increase in O 2 /Ar can be explained by gas exchange (absolute value of J ex < 10 mmol O 2 m −2 d −1 ) (Table 1). Thus, the temporal change in O 2 /Ar can be attributed to a primarily biological source. The magnitude of this increase is further underestimated because of vertical upwelling of deep oxygenpoor waters, which would act to dampen the increase in O 2 /Ar through time. After accounting for a mixing correction ranging between 22 and 97 mmol m −2 d −1 O 2 (equivalent to 16 to 70 mmol m −2 d −1 C when assuming a photosynthetic quotient of 1.4), daily integrated gross primary productivity (GPP O 2 /Ar ) ranged from 270 to 358 mmol C m −2 d −1 , and community respiration (CR O 2 /Ar ) rates ranged from 74 to 172 mmol C m −2 d −1 ( Table 1).
Examination of the diel variability in POC and Chl-a during drifter period 1 revealed significant differences in the behavior of these variables as compared to O 2 /Ar (Fig. 3b, c). In particular, while O 2 /Ar increased during the first drifter deployment, [POC] and [Chl-a] values decreased. We estimated that vertical mixing (F mix,POC ) accounted for 12 to 68 mmol m −2 d −1 C of these daily changes in [POC], similar to the magnitude of the mixing correction for O 2 /Ar variability (Table 1). After taking mixing into account, daily integrated GPP POC decreased from 242 mmol m −2 d −1 on day 1 to 98 mmol m −2 d −1 on day 3, while CR POC rates ranged from 77 to 147 mmol m −2 d −1 .
Calculated daily averaged net primary productivity (NPP) was lower than GPP O 2 /Ar . Rates derived from the CbPM model (Sect. 2.5) declined from 147 mmol C m −2 d −1 on day 1 of drifter deployment 1 to 112 mmol C m −2 d −1 on day 3 (Table 1), reflecting the trend in Chl-a concentrations used to derive NPP (Fig. 3c). The CbPM-derived NPP estimates were similar to that obtained in 14 C incubations (150±18 mmol C m −2 d −1 ) within the first 24 h of drifter deployment 1. However, 14 C-based NPP estimates on the third day of the deployment (49 ± 8 mmol C m −2 d −1 ) were about 2-fold lower than those obtained from CbPM calculations.
Dissolved oxygen and POC dynamics at drifter site 2 differed significantly from those observed at site 1. Compared to the drifter site 1, diel variability in O 2 /Ar and [POC] was more tightly coupled during the second drifter deployment (Fig. 3a, b). Both O 2 /Ar and [POC] displayed regular diurnal variations, increasing in the daytime to a maximum around dusk and decreasing at night to a minimum around dawn. Over the full drifter deployment, concentrations of Chl-a and, to a lesser extent, POC decreased, in contrast to O 2 /Ar, which remained relatively constant across days. Daily integrated GPP O 2 /Ar values ranged from 108 to 219 mmol C m −2 d −1 , and CR O 2 /Ar rates ranged from 83 to 186 m −2 d −1 . POC-derived values were considerably lower and less variable, from 41 to 38 for GPP POC and 36 to 44 for CR POC (Table 1). NPP derived from CbPM calculations was 22 mmol C m −2 d −1 on the first day of the drifter period and 18 mmol C m −2 d −1 on the second day, while NPP calculated from one 14 C bottle incubation during the first day of the drifter 2 deployment was 12 ± 4 mmol C m −2 d −1 , showing good agreement with the CbPM calculations.   Table 1. Daily integrated mixed-layer net primary production (NPP) and net community production (NCP), including all components used to calculate NCP from O 2 /Ar or POC time series, as indicated: gross primary productivity (GPP), respiration (CR), vertical mixing (Mix), and gas exchange (J ex ). All units here are in millimoles of carbon per square meter per day (mmol C m −2 d −1 ). Note that CbPM is the carbon-based production model (Behrenfeld et al., 2005;Westberry et al., 2008;Graff et al., 2016)

Net community production
Daily net community production (NCP) rates were calculated using linear regressions of O 2 /Ar and POC over day and night intervals, corrected for gas exchange and vertical mixing (Sect. 2.6.1, 2.6.2). During drifter period 1, NCP O 2 /Ar and NCP POC exhibited contrasting trends, as NCP O 2 /Ar remained > 100 mmol C m −2 d −1 throughout, while NCP POC declined to negative values on the second and third days (Table 1; Fig. 4). The transition to negative NCP POC values over the course of the drifter 1 deployment primarily reflected diminishing daytime rates of POC accumulation (dPOC/dt term in Eq. 8). At drifter period 2, we observed closer agreement between NCP values. O 2 /Arderived NCP ranged from −12 to 33 mmol C m −2 d −1 over two consecutive 24 h periods, while NCP POC values ranged from −3 to 1 mmol C m −2 d −1 . These lower rates at drifter site 2 are consistent with the lower observed phytoplankton biomass and nutrient concentrations.
Additional constraints on NCP during drifter period 1 can be derived from examining nutrient drawdown. Because vertical upwelling of nutrient-replete waters would dampen the magnitude of observed nutrient drawdown over time (Sect. 3.1; Fig. S4), we used the derived k mix from Eq. (6) and a best-fit vertical gradient in nutrient concentrations between the mixed layer and 100 m (Sect. 2.4) to account for this mixing flux. This correction increases the cumulative 3 d nutrient drawdown by 2.1 to 2.6 times. Over the 3 d drifter deployment, surface Si, N, and P concentrations declined in a ratio of 17 : 13 : 1, which is consistent with the stoichiometry expected for organic matter produced by a diatomrich assemblage (Brzezinski et al., 1998;Turner et al., 1998;Figure 4. Daily net community production (NCP) during successive days of the two drifter deployments derived from diurnal variations of biological oxygen saturation ( O 2 /Ar) and particulate organic carbon (POC) concentration. Each set of bars is for one 24 h period, with approximate starting times on the x axis. Note that the negative NCP O 2 /Ar value for the first day of drifter period 2 was computed by integrating NCP O 2 /Ar values over eight consecutive 3 h increments (refer to Table 2). Brzezinski, 2004). Assuming that the observed decrease in SiO 2 concentrations over the 3 d is attributable to growth of diatoms in the mixed layer, and applying a stoichiometric ratio of 106 C : 16 Si, we estimate an average C production rate of ∼ 128 mmol C m −2 d −1 for the drifter period. This value is consistent with NCP O 2 /Ar rates, which were 137 mmol C m −2 d −1 on average over 3 d, but significantly greater than NCP POC estimates (7 mmol C m −2 d −1 on average) ( Table 2). Table 2 summarizes comparisons among NCP values calculated using day and night linear regressions of O 2 /Ar and POC against time, as well as other approaches described in Sect. 2.6.3. In general, the different calculation methods did not significantly alter the results. NCP values derived from one linear regression over each drifter period agreed well with the average of two (drifter 2) to three (drifter 1) daily NCP values calculated via the other approaches. Small differences between linear regression-based NCP values and both NCP values calculated from either 3 h increments or two time points are likely due to the effect of lower signal to noise in O 2 /Ar, [O 2 ] bio and [POC] values utilized in these latter two approaches. The following discussion thus focuses on productivity rates derived from day and night linear regressions (i.e., Eqs. 1 and 8), which utilize all data points while minimizing uncertainty in the derived rates of change. The exception is the NCP O 2 /Ar value calculated for day 1 of drifter period 2 using the daytime and nighttime linear regression method. By this approach, we calculated NCP O 2 /Ar as 26 mmol C m −2 d −1 , even though the time series in Fig. 3a Fig. 4, the NCP value derived from the integrated 3 h increments represents net community production during this particular interval.
Biogeochemical properties during the first Lagrangian survey suggested a dynamic, highly productive phytoplankton community, influenced by upwelling and elevated mixed-layer nutrient concentrations (Figs. 1, S4). Several lines of evidence imply the presence of a developing diatom bloom at this site (Sect. 3.2;Figs. 2,3). Increasing mixed-layer biological oxygen saturation ( O 2 /Ar) was contrasted by a general decrease in particulate organic carbon (POC) concentrations, suggesting a significant decoupling between O 2 and POC dynamics. This was reflected in significant differences between O 2 /Ar-derived gross primary productivity (GPP) and net community production (NCP) rates derived from O 2 /Ar and POC measurements (Figs. 4, 5; Table 1). In contrast, biogeochemical properties during the second drifter deployment were indicative of a lower-productivity, nutrient-limited phytoplankton assemblage, with near-zero O 2 /Ar values reflecting a close balance between water column photosynthesis and respiration (Fig. 3a). Relative to the drifter 1 site, diurnal variations in O 2 /Ar and POC were more closely coupled, while phytoplankton biomass (C ph ) and chlorophylla (Chl-a) concentrations (dominated by smaller cells) varied little through time. Contrary to our expectations, even though NCP O 2 /Ar and NCP POC rates agreed well, we also observed significant discrepancies between GPP O 2 /Ar and GPP POC and between CR O 2 /Ar and CR POC during drifter period 2. The contrasting properties between the two drifter deployments enable us to examine the coupling of O 2 and POC dynamics under different ecological states, with implications for the use of O 2 /Ar and POC measurements as proxies for GPP and NCP. In the absence of significant POC sinking and net loss to the dissolved organic carbon (DOC) pool, POC-based productivity rates should approximate O 2 /Ar-based rates (Claustre et al., 2008;White et al., 2017). However, at drifter station 1, both GPP O 2 /Ar and NCP O 2 /Ar greatly exceeded GPP POC and NCP POC , respectively (Figs. 4, 5a; Table 1). Over the three successive 24 h periods of drifter deployment 1, the absolute difference between GPP measures increased from 41 to 260 mmol C m −2 d −1 , while the absolute difference between NCP estimates increased from 42 to 193 mmol C m −2 d −1 . This discrepancy exceeded the propagated NCP uncertainties during the second and third days of the deployment and was apparent in all approaches used to calculate NCP (Sect. 2.6.3, Table 2). While mixed-layer O 2 /Ar primarily reflected O 2 accumulation from GPP and O 2 loss from CR, diurnal variability in [POC] was likely affected by several additional loss factors, which are discussed below. The variable difference between O 2 -based and POC-based NCP measured over 3 h increments (Eq. 11; Fig. S5) suggests that apparent POC loss rates were variable throughout the drifter period and lower at night relative to day. Thus, the higher NCP O 2 /Ar may be Table 2. Comparisons of NCP calculated using four different timescales of integration: (rows 1-2) day and night linear regressions, (rows 3-4) summed linear regressions over 3 h increments, (rows 5-6) the difference between two time points every 24 h, and (rows 7-8) a single linear regression over the entire drifter period. Refer to Sect. 2.6.3 for further details. For every calculation approach, "Export + DOC" is the average difference between NCP O 2 /Ar and NCP POC values ±1 SD or ± the propagated error. All units here are in millimoles of carbon per square meter per day (mmol C m −2 d −1 103 ± 56 13 ± 9 NCP POC (drifter trend) −21 ± 28 123 ± 62 −4 ± 2 17 ± 9 attributed more to differences in daytime accumulation of POC and O 2 rather than differential POC and O 2 losses at night. Indeed, we found that differences between CR O 2 /Ar and CR POC were smaller than differences in NCP or GPP throughout drifter period 1, and CR O 2 /Ar exceeded CR POC during two of three nights (Fig. 5b).
In the dynamic, high-productivity upwelling environment of drifter site 1, a number of processes can account for variable POC loss rates on various timescales (Gard-ner et al., 1993(Gard-ner et al., , 1999White et al., 2017;Briggs et al., 2018). During a diatom bloom, enhanced aggregation of large silica-rich particles and zooplankton fecal pellet production can stimulate POC export and diatom cells out of the mixed layer (Buesseler, 1998;Guidi et al., 2009;Brzezinski et al., 2015;Stukel et al., 2017), progressively decreasing NCP POC relative to NCP O 2 /Ar . The discrepancy we observed between NCP POC and NCP O 2 /Ar (up to 193 mmol C m −2 d −1 ) is in the upper range of prior export estimates from various oceanic regions, including the Southern Ocean (∼ 83 mmol C m −2 d −1 ), North Atlantic spring bloom (96 mmol C m −2 d −1 ), and southern California Current system (∼ 36 mmol C m −2 d −1 ) (Henson et al., 2012;Alkire et al., 2012;Stukel et al., 2017), suggesting that POC export fluxes could comprise a significant fraction of the inferred POC loss at drifter site 1. At the same time, subdaily changes in particle sinking velocities and size distributions could cause daytime export to exceed nighttime export (Durand and Olson, 1998;Waite and Nodder, 2001;Oubelkheir and Sciandra, 2008;Kheireddine and Antoine, 2014;Ribalet et al., 2015;Briggs et al., 2018), leading to greater differences between GPP O 2 /Ar and GPP POC than between CR O 2 /Ar and CR POC , as we observed.
Another likely POC loss is DOC production through cellular exudation, viral lysis, and/or grazing (Karl et al., 1998;Lochte et al., 1993;Claustre et al., 2008;Dall'Olmo et al., 2011;Briggs et al., 2018). On daily timescales, this loss term would lower NCP POC relative to NCP O 2 /Ar , provided that DOC production exceeds DOC respiration. Further, higher daytime net DOC production would cause GPP O 2 /Ar to increase more than GPP POC in the daytime, while a decrease at night would cause CR O 2 /Ar to exceed CR POC (Karl et al., 1998). Light-and productivity-dependent increases in DOC production in the daytime could result, for example, from the effects of photorespiration and other mechanisms of dissipating excess light energy . While we did not conduct direct measurements of DOC concentrations during the cruise, previous work in a variety of ocean environments has shown that DOC production can account for 3 %-37 % of NCP in the Ross Sea, up to 10 %-40 % in the equatorial Pacific Ocean, up to 66 % in the Sargasso Sea during the seasonal phytoplankton bloom, and 22 %-40 % during the North Atlantic bloom (Hansell and Carlson, 1998;Alkire et al., 2012). In the eastern subarctic Pacific, Bif and Hansell (2019) estimated springtime DOC / NCP ratios of 0.05-0.54 and summertime ratios of 0-0.28 along the Line P transect (130-152 • W).
In addition, assuming that DOC exudation from phytoplankton cells is positively related to growth in heterotrophic biomass (Fuhrman et al., 1985;Kuipers et al., 2000;Church et al., 2004), variations in total bacterial biomass may have impacted c p measurements at drifter site 1 (Oubelkheir and Sciandra, 2008;Gernez et al., 2011;Barnes and Antoine, 2014). If detected by the ac-s sensor, bacteria could potentially account for some of the discrepancy between diel POC and O 2 -derived variability. In particular, c p decreases from phytoplankton exudation would counter c p increases from heterotrophic growth. At night, this would decrease CR rates derived from c p -based [POC], relative to O 2 -derived CR rates.
A final consideration involves diurnal variation of zooplankton abundances and grazing rates, which could enhance POC loss without depleting O 2 /Ar (Dall'Olmo et al., 2011;White et al., 2017;Briggs et al., 2018), assuming that biomass accumulation rates from grazing surpass grazer respiration rates (Dagg et al., 1982). Further, once POC is assimilated into the body of a grazer, it joins a larger particle size class that likely exceeds the size-dependent detection limits of the beam attenuation coefficient (Stramski and Kiefer, 1991;Marra, 2002;Claustre et al., 2008), decreasing the c p signal used to derive POC. During our expedition, we observed a strong signature of diel migrating zooplankton based on increased nighttime signal spikes in surface optical backscatter measurements . These effects would enhance CR POC relative to CR O 2 /Ar , contrary to what we observed. We thus assume that grazing at drifter site 1 is minimal relative to the effects of particle export and DOC production on GPP, CR, and NCP.

Drifter 2
Relative to the drifter 1 site, drifter site 2 exhibited similar discrepancies between GPP O 2 /Ar and GPP POC , as well as greater discrepancies between CR O 2 /Ar and CR POC (Fig. 5ab; Table 1). Irrespective of the time of day, the rate of O 2 /Ar change computed over 3 h intervals (Eq. 11) consistently exceeded POC-derived changes throughout the drifter period (Fig. S5). The strong, positive relationship between these two 3 h measures (p < 0.05, r 2 = 0.64), compared to the weaker correlation at drifter site 1 (p < 0.05, r 2 = 0.39) ( Fig. 5c-d), suggests that, despite large differences in the magnitude of O 2 /Ar-derived and POC-derived GPP and CR rates, POC-based changes were a good relative indicator of O 2 -derived productivity rates at drifter site 2.
Because daytime increases in both O 2 /Ar and [POC] were balanced by nighttime decreases, absolute differences in NCP O 2 /Ar and NCP POC were smaller than at drifter site 1. This result suggests a closer coupling between primary production and heterotrophic consumption, as expected for this more oligotrophic ecosystem (Claustre et al., 2008;White et al., 2017). While the NCP discrepancy was negligible over the first 24 h period, it increased to 32 mmol C m −2 d −1 over the 24 h period (Table 1; Fig. 4), exceeding the uncertainty of both NCP calculations. This suggests low, but nonnegligible, rates of particle export, grazing, and/or net DOC production at drifter site 2. Although we lack direct DOC measurements, this result is consistent with several previous observations of low net DOC production in oligotrophic waters (Bif et al., 2018;Hansell and Carlson, 1998), with values approaching ∼ 30 % of NCP in low-productivity offshore waters of the subarctic Pacific (Bif and Hansell, 2019). Low particle sinking rates could also explain the smaller absolute discrepancy between NCP O 2 /Ar and NCP POC at drifter site 2. Low particle export is generally expected from phytoplankton assemblages dominated by small particle sizes < 20 µm, as evident in higher b bp slope values and Chl-a size fractionation measurements at drifter site 2 (Sect. 3.2; Fig. 2) (Fowler and Knauer, 1986;Guidi et al., 2008).
Prior studies have observed that the amplitude of diurnal variability in O 2 /Ar exceeds the amplitude of diurnal variability in c p -based [POC], as we observed at drifter site 2 (Kinkade et al., 1999;Hamme et al., 2012;Briggs et al., 2018). For example, Briggs et al. (2018) observed higher amplitude variations in O 2 relative to c p -derived [POC] during the North Atlantic bloom, leading to higher absolute O 2derived respiration and gross oxygen production (GOP) rates compared to c p -derived rates. In the Southern Ocean, Hamme et al. (2012) also observed high ratios of underway O 2 /Arderived gross oxygen production to gross carbon production (i.e., GPP) based on photosynthesis-irradiance incubations. As discussed above for drifter site 1, these offsets between O 2 /Ar and POC-based measures might result from the effects of bacteria on c p measurements, especially at a relatively low productivity site like drifter site 2 (Table 1; Fig. 2) (Claustre et al., 2008;Oubelkheir and Sciandra, 2008;Barnes and Antoine, 2014). Bacterial c p variability would act to counter phytoplankton c p variability, decreasing the magnitude of CR POC relative to the magnitude of CR O 2 /Ar . Indeed, the positive CR O 2 /Ar − CR POC discrepancy at drifter site 2 contributed to 58 %-82 % of the differences between O 2 /Ar and POC-derived GPP rates. The remaining difference may be attributed to greater daytime POC losses to the DOC pool and through particle export.

Other factors driving variability in NCP
In interpreting our results, it is important to consider a number of methodological caveats that could contribute to the apparent difference between NCP O 2 /Ar and NCP POC . One important variable in all of our comparisons of productivity rates is the O 2 -to-POC conversion factor, represented by the photosynthetic (PQ) and respiratory quotient (RQ). Given the relatively narrow range of possible PQ values applicable to our study sites (∼ 1.1-1.4) (Laws, 1991), variability in this term cannot account for the total discrepancy observed between O 2 /Ar and POC-derived GPP, CR, and NCP rates. By contrast, RQ values in the ocean are more variable than PQ Hedges et al., 2002). Therefore, variability in RQ values at both drifter sites could introduce considerable uncertainty into GPP O 2 /Ar , CR O 2 /Ar , and NCP O 2 /Ar calculations unaccounted for in our error propagations (Sect. 2.7). However, we found that use of RQ values ranging between 1.0 and 1.4 (Anderson and Sarmiento, 1994;Hedges et al., 2002;Daneri et al., 2012) did not greatly change calculated GPP O 2 /Ar , CR O 2 /Ar , and NCP O 2 /Ar relative to GPP POC , CR POC , and NCP POC . Therefore, it is unlikely that our selected RQ values, 1.4 and 1.1 for drifter sites 1 and 2, respectively, biased our main interpretations.
In our analysis, we interpret variations in particulate backscatter (b bp ) and beam attenuation (c p ) in terms of phytoplankton and total particulate organic carbon concentrations, assuming a negligible influence of inorganic suspended minerals from various sources, including sediment resuspension and transport by the Columbia River plume (Thomas and Weatherbee, 2006). This assumption is supported by the salinity of waters we sampled at both drifter sites, which was significantly higher than that expected for river-influenced regions (below 30 g kg −1 ; Hickey et al., 1998). At the same time, the observed bulk refractive index of particles (η p ) at drifter site 1 does not preclude the presence of mixing between POC and a small fraction of shelf-derived inorganic particles. Estimates of η p were generally below 1.12 for this nearshore site (Sect. 2.2; Fig. S3e), as compared to values as high as 1.26 for inorganic minerals in seawater (Lide, 1997;Twardowski et al., 2001). By comparison, calculated η p values during the drifter 2 deployment were below 1.08, which is much closer to values expected for water-containing predominantly nondiatom phytoplankton organic carbon.
Additional uncertainty in our analysis derives from the algorithms used to estimate POC and phytoplankton carbon C ph from optical measurements (Sect. 2.2). Because of particle size limitations in the optical measurements, they may not fully capture all significant size classes of the particulate pool, such as larger microplankton and zooplankton. Such a size bias in the c p signal at 660 nm, used to derive [POC], would cause an underestimate of larger POC particles measured by beam attenuation (Claustre et al., 2008;Marra, 2002;Stramski and Kiefer, 1991) and thereby contribute to the apparent discrepancy between diel changes in [POC] and diel changes in O 2 /Ar. Despite these potential caveats, recent work (Graff et al., 2016;Briggs et al., 2018; has demonstrated that c p -and b bp -based derivations of [POC] and [C ph ] can indeed be robust in highbiomass ocean regions, where productivity and the proportion of large-celled phytoplankton is significant. Equally important, changes in the c p -to-[POC] relationship through time could also drive apparent variability in optical [POC] estimates. The linear regression of [POC] against c p at 660 nm measured across diverse marine environments is defined over a range of POC concentrations from ∼ 5 to ∼ 175 µg L −1 (Graff et al., 2015). At drifter site 2, POC concentrations fell within the range of this fit, and particle properties that may influence POC/c p values (i.e., b bp slope values, phytoplankton community composition, particle size, and bulk refractive index) were relatively constant through time (Fig. S3d, e). By comparison, POC concentrations at drifter station 1 were 25 % higher than the empirical limits of the c p -based algorithm in Graff et al. (2015), requiring extrapolation of the POC/c p relationship beyond its calibration range. In a limited comparison with discrete POC samples, we found a POC-c p slope that was similar to that of Graff et al. (2015, albeit with a different y intercept) (Fig. S1). Nonetheless, we cannot rule out changes in the c p660 -[POC] relationship due to shifts in cell size and, to a lesser extent, bulk refractive index resulting from diatom accumulation (Kheireddine and Antoine, 2014;Stramski and Reynolds, 1993) (Fig. S3d-e). Indeed, Briggs et al. (2018) observed that the ratio of [POC] to c p decreased by ∼ 20 % during the rise of the North Atlantic bloom, while values increased by ∼ 60 % during the bloom decline. If we assume a 20 % decrease in POC/c p660 values (from ∼ 420 to ∼ 340 mg m −2 ) associated with diatom growth (Briggs et al., 2018), our daily NCP POC estimates would be less positive during day 1 and less negative during days 2-3. This, in turn, would increase the apparent decoupling between NCP POC and NCP O 2 /Ar on day 1 (∼ 27 %) and day 3 (∼ 1 %) and bring the values slightly closer on day 2 (∼ 8 %). Overall, the value of these potential changes is small relative to the differences we observed between NCP O 2 /Ar and NCP POC , and we thus conclude that variable POC/c p660 ratios cannot explain the observed decoupling between POC, C ph , and dissolved O 2 dynamics at the drifter 1 site.
There are a number of other potential caveats in our analysis of phytoplankton carbon from b bp and particle size distribution from b bp slope. Previous studies have reported that daily variations in b bp do not always track daily variations in c p , suggesting that b bp dynamics do not reflect phytoplankton carbon dynamics on diel timescales (Kheireddine and Antoine, 2014;Briggs et al., 2018). We observed a similar decoupling between b bp and c p in this study; for example, while c p values at 660 nm steadily declined in the last 24 h of drifter period 1, b bp at 470 nm stayed relatively constant. Nonetheless, [C ph ] estimates from b bp (Fig. 2) remain useful for comparisons between drifter sites, and differences in apparent phytoplankton biomass concentration were consistent with a number of the other biogeochemical differences measured between the two trophic regimes. Similarly, the relationship between b bp slope and particle size distribution has been challenged in previous literature (e.g., Zeng et al., 2018). While this limits our interpretation of daily b bp slope dynamics, we did find independent evidence for larger particle sizes at drifter site 1 (as predicted by the b bp slope), from size fractionated [Chl-a] measurements and pigment analysis showing a greater fraction of microplankton (Sect. 3.2).

Reconciling NCP and NPP
During both drifter surveys, we estimated daily integrated net primary productivity (NPP) values using carbon-based production model (CbPM) calculations and 14 C bottle incubations (Sect. 2.5). On several days, these two estimates of NPP were consistently lower than NCP O 2 /Ar integrated over the same timescales and mixed-layer depths (Table 1). Similarly, Briggs et al. (2018) and Alkire et al. (2012) also reported NCP values that were equal to or greater than NPP values obtained from different methodologies during their Lagrangian study of the North Atlantic bloom.
In theory, NCP cannot exceed NPP, as NCP includes additional respiration terms not included in NPP and must always be equal to or (more realistically) lower than NPP. Recent work in the northeastern Pacific Ocean has reported mean NCP/NPP ratios, based on O 2 /Ar measurements and CbPM calculations, ranging from 0.16 to 0.26 in offshore and coastal waters . These values, determined from continuous observations along a moving ship track, are consistent with theoretical expectations. The observed high (> 1) apparent NCP/NPP values observed in our study and that of Briggs et al. (2018) and Alkire et al. (2012) highlight a number of methodological limitations that could depress NPP estimates.
One possibility, which has been discussed at length by various authors (Gieskes et al., 1979;Fogg and Calvario-Martinez, 1989;Marra, 2009), is that bottle containment effects limit accurate estimates of 14 C uptake. This effect would have caused underestimates of 14 C-NPP during both drifter surveys, relative to CbPM-NPP and NCP O 2 /Ar , which do not require discrete sample incubations. In addition, during the last 14 C-uptake experiment of drifter survey 2, the incubator warmed (as the ship passed through warm SST water used to cool the tanks), potentially creating heat stress on phytoplankton and depressing 14 C-NPP values.
A number of factors may also influence CbPM-based NPP estimates. While the model applies a satellite-based relationship between [Chl-a] / [C ph ] and daily mixed-layer irradiance (E g ) to calculate growth rate, these E g values may not fully parametrize phytoplankton physiology for mixed assemblages in the ocean (Westberry et al., 2008). Indeed, phytoplankton photophysiology varies with other environmental conditions and phytoplankton composition (Cloern et al., 1995;Geider et al., 1998;MacIntyre et al., 2002;Westberry et al., 2008). In addition, the CbPM does not allow calculated growth rates to exceed 2 d −1 , which may not apply to all ocean environments (Graff et al., 2016). These uncertainties could potentially impact the applicability of the CbPM parameters to the specific ocean conditions at drifter sites 1 and 2. In addition, a vertical mixing correction for ac-s and backscatter-derived [Chl-a] and [C ph ], respectively, not feasible in the present data set, may improve CbPM-based estimates of NPP.

Conclusions
In the current study, biological oxygen saturation ( O 2 /Ar) and optically derived particulate organic carbon (POC) were measured continuously and simultaneously during two Lagrangian drifter deployments. This dual measurement approach facilitated direct comparison of O 2 /Arand POCderived measures of gross primary productivity (GPP), com-munity respiration (CR), and net community production (NCP), from a mesotrophic upwelling-influenced system and a more oligotrophic system further offshore. As hypothesized, the results show that O 2 -and POC-based measures of GPP and NCP diverge in mid-to-high-productivity phytoplankton communities, where daily fluctuations in O 2 /Ar are decoupled from POC cycling. Interestingly, oxygenbased GPP and CR also exceeded POC-based GPP and CR rates at the lower-productivity site, though we found that net changes in POC scaled with O 2 /Ar-based productivity estimates, suggesting a tighter coupling between O 2 and POC cycles.
These findings are generally consistent with current understanding of productivity dynamics and mixed-layer POC cycling in these two coastal Pacific environments and complement only one prior comparison of daily GPP and NCP estimates from simultaneous, autonomous measurements of c p and O 2 in the North Atlantic mixed layer (Alkire et al., 2012;Briggs et al., 2018). Importantly, however, our results differ from earlier studies by providing two examples of significant disagreement between GPP O 2 /Ar and GPP POC , as well as CR O 2 /Ar and CR POC , likely resulting from subdaily variations in particle export, net DOC production, and bacterial growth over respiration. In such cases, assuming constant daily respiration rates by extrapolating nighttime rates of change may pose challenges for comparing O 2 /Ar and POC-based GPP and CR. We have further shown that for upwelling regions like drifter site 1, it is important to account for vertical mixing of subsurface waters into the mixed layer and its effect on not only NCP O 2 /Ar calculations , but also on NCP POC estimates through dilution of the surface POC signature. Our study thus illustrates an application of the vertical mixing coefficient, k mix , derived from [N 2 O] profiles to more accurately estimate net changes in POC and nutrient concentration in such environments.
Moving forward, the disparity between POC-and O 2based NCP estimates offers an opportunity to continuously track cumulative POC losses in the mixed layer using autonomous shipboard or in situ sensors. As it is labor-intensive to measure POC export on short timescales with sediment traps and the 234 Th-238 U disequilibrium method Savoye et al., 2006), simultaneous underway measurements of dissolved O 2 , particulate beam attenuation and colored dissolved organic matter (CDOM) absorption and spectral slope over a range of wavelengths < 400 nm (Del Vecchio and Blough, 2004;Grunert et al., 2018) may provide a valuable, first-order approximation of POC partitioning among living phytoplankton biomass, particle export, and dissolved organic carbon (DOC) in the surface ocean on short timescales. At drifter site 1, for example, taking an upper bound of 40 % of NCP as DOC production (close to the fraction estimated by Alkire et al., 2012, dur-ing the North Atlantic spring bloom) yields a 3 d average DOC flux of 55 mmol C m −2 d −1 and residual export flux of 76 mmol C m −2 d −1 . Being able to estimate such quanti-ties with this approach is especially important in the California coastal upwelling regime and other similar ecosystems with high NCP and significant potential for carbon transfer to higher trophic levels.
For future work, we recommend a number of approaches to facilitate estimation of POC export from coupled O 2 , POC, and DOC dynamics. First, it will be valuable to constrain particle size and partition POC into detrital and living (phytoplankton and heterotrophic bacteria) components to properly assess the size range captured by optically derived POC and C ph measurements. Second, independent estimates of POC export and DOC concentrations during each drifter deployment could validate POC export fluxes derived from coupled O 2 and POC measurements. Relatedly, depth-resolved backscatter profiles (Briggs et al., 2013(Briggs et al., , 2018 could be used as another autonomous approach to calculating export fluxes, as an independent check on surface-based estimates. Going forward, there is significant future potential to exploit coupled O 2 and c p measurements on autonomous platforms, including various ocean moorings (e.g., the Optical Dynamics Experiment, the Biowatt II program, and the Bermuda Testbed Mooring program), and biogeochemical floats and gliders to resolve opportunistic, high-resolution POC export time series (Stramska and Dickey, 1992;Kinkade et al., 1999;Dickey and Chang, 2002). Deployment of such autonomous measurement systems across a range of oceanic regions will help to constrain POC and productivity dynamics on global scales.
Author contributions. SZR, PDT, and NS collected the data in the field. RWI processed the CTD cast data and measured nitrous oxide concentrations in discrete samples. SZR wrote the manuscript with significant input from the coauthors.
Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. Special thanks to Jessie Gwinn, Ross Mc-Culloch, Chen Zeng, Melissa Beaulac, Chris Payne, and Maureen Soon for assistance in field collection and analysis of samples, as well as to two anonymous reviewers for insightful suggestions on earlier versions of this paper.
Financial support. This research has been supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) and the US National Science Foundation (NSF project number 1436344).
Review statement. This paper was edited by Emilio Marañón and reviewed by two anonymous referees.