Carbon sequestration by multiple biological pump pathways in a coastal upwelling biome

Multiple processes transport carbon into the deep ocean as part of the biological carbon pump, leading to long-term carbon sequestration. However, our ability to predict future changes in these processes is hampered by the absence of studies that have simultaneously quantified all carbon pump pathways. Here, we quantify carbon export and sequestration in the California Current Ecosystem resulting from (1) sinking particles, (2) active transport by diel vertical migration, and (3) the physical pump (subduction + vertical mixing of particles). We find that sinking particles are the most important and export 9.0 mmol C m−2 d−1 across 100-m depth while sequestering 3.9 Pg C. The physical pump exports more carbon from the shallow ocean than active transport (3.8 vs. 2.9 mmol C m−2 d−1), although active transport sequesters more carbon (1.0 vs. 0.8 Pg C) because of deeper remineralization depths. We discuss the implications of these results for understanding biological carbon pump responses to climate change.


9
This document contains supplementary tables S1 -S3 followed by supplementary methods. 10 11 SUPPLEMENTARY TABLES S1 -S3 12 Quantifying carbon flux due to subduction and vertical mixing 23

Supplementary
The physical processes that transport organic carbon to depth (sometimes referred to as the large-scale physical 24 pump, the mixed layer pump, and the eddy subduction pump) act across multiple spatial scales from the 25 submesoscale to basin scale. Consequently, it is not possible to measure carbon transport resulting from all of these 26 interrelated processes (which hereafter we refer to simply as "subduction") directly from field measurements. 27 Hence, we quantified rates of particulate organic carbon subduction by combining a biogeochemical data 28 assimilation approach, a physical circulation data assimilation approach, and a Lagrangian particle tracking model 1 . 29 We outline each of those approaches here, but refer readers to ref 1 for additional details. 30 Biogeochemical data-assimilation submodule: The fundamental premise in our modeling approach is that 1) 31 particles are simultaneously transported by ocean circulation and their own gravitational settling and that 2) 32 gravitational settling rates should be modeled as a continuous spectrum of sinking speeds because a diverse suite of 33 sinking particles with highly variable settling rates exist in the ocean. We made the a priori assumption that the 34 spectrum of sinking particle speeds in the CCE would be best modeled by considering two classes of particles, both 35 of which have sinking speed distributions described by a log-normal distribution (i.e., most particles sink slowly but 36 some sink substantially faster) but with different mean sinking speeds for each particle class. These particle classes 37 were chosen to be representative of phytoplankton (which mostly have very slow settling rates) and fecal pellets 38 (which typically dominate sinking flux in the CCE 2 ). We thus define the production rate of particles as a function of 39 sinking speed (S) as: 40 Eq. S1 41 where  is the total particle production rate,  is the ratio of fecal pellet production to total particle production, and 42  1 and  2 are parameters related to the median sinking rate of phytoplankton and fecal pellets, respectively.  1 and 43  2 define the variability in the sinking rates of phytoplankton and fecal pellets, respectively and we assume a priori 44 that both of these parameters are equal to 1, because this generates realistic variability in particle sinking speeds. An 45 additional parameter used in the subduction model was the particle remineralization rate (λ). Because there is a limit 46 to the number of parameters that could be uniquely defined from the field data, we further chose to assume a fixed 47 ratio of 100 between the median sinking speeds of phytoplankton and fecal pellets. We thus assume that fecal pellet 48 settling rates are two orders of magnitude greater than those of phytoplankton, a difference in settling rates that was 49 chosen based on multiple studies showing phytoplankton settling velocities on the order of 1 m d -1 and fecal pellet 50 settling rates on the order of 100 m d -1 (e.g., 3-7 ). Thus µ 2 =ln(100×exp(µ 1 )). This left us with four unknown 51 parameters that needed to be fitted from our field data:   µ, and λ 52 To objectively define these four parameters, we used field data from Lagrangian experiments conducted during 53 three process cruises of the CCE LTER program (see main text for descriptions of these sampling programs). 54 During each of these 14 Lagrangian experiments (duration = 2 -5 days), we quantified NPP using H 14 CO 3uptake 55 daily at 6 -8 depths spanning the euphotic zone 2 , we measured POC concentrations at 8 depths in the euphotic zone, 56 we determined sinking particle flux using sediment traps or 238 U-234 Th disequilibrium 8,9 , and we quantified 57 mesozooplankton grazing rates using the gut pigment method 10 . Our data assimilation process began by defining the 58 flux of sinking particles as a function of speed (S) and depth (z): 59

Eq. S2 60
where λ is again the remineralization rate (units of d -1 ) and d prod is the mean depth of particle production (which we 61 calculated from our H 14 CO 3uptake profiles). We can then define the following relationships between our 62 measurements and Equations S1 and S2:

Eq. S5 66
Because the flux of sinking particulate carbon (at a specific depth) with any particle sinking speed will simply 67 be equal to the carbon content of particles with that sinking speed (at that same depth) times the sinking speed, we 68 can define the carbon content of particles as a function of sinking speed and depth as C(z,S) and relate this to our 69 measured vertically integrated POC concentrations in the water column as: 70 where d POC is the maximum depth at which POC measurements are made (if shallower than the depth of the export 72 measurement). 73 We can further relate  (the ratio of fecal pellet production to total particle production) through the equation: 74

Eq. S7 75
where Graz is the measured mesozooplankton grazing rate (units of mg C m -2 d -1 ) and EE is the egestion efficiency 76 of zooplankton which was assumed to be 0.3 following ref 11 . 77 This leaves us with 5 equations (S3 -S7) that can be used to define 5 parameters (  µ, λ, and d prod ). To 78 solve this system of equations we used a grid search approach while assuming that sinking speeds varied from 1 mm 79 d -1 to 1 km d -1 . The resultant parameters used for the particle subduction model are given in Table S4: 80 81 82 83 as well as in situ physical measurements (temperature, salinity) from our cruises 13,14 . These state estimates were 93 extended 30 days into the future (post-cruise and past the assimilated time frame) with a free-running model 94 estimate, to ensure that dynamically consistent fields were available to force advection of simulated particles for 30 95 days following each Lagrangian experiment conducted in the field. State estimates were saved with 8-hourly 96 temporal resolution. 97 Lagrangian particle simulations: To simulate the four-dimensional particle trajectories that result from a 98 combination of sinking and physical circulation we used a modified version of the LTRANS Fortran package 15 . 99 LTRANS advects particles using ROMS velocity and eddy diffusivity fields as well as "behavior", which here refers 100 to particle sinking speeds. The modified package allows each individual particle to have a sinking speed selected 101 randomly from the particle sinking speed distributions defined in Eq. S1 using the parameters in the table above. 102 For each of the 14 Lagrangian experiments conducted at sea, we released 10,000 simulated particles. Particles were 103 released at random locations along the in situ Lagrangian experiment trajectories and at depths proportional to the 104 measured vertical profiles of NPP. Particles were tracked for a period of 30 days (with one-hour time step) and 105 remineralized according to the remineralization constants (λ) listed above. Importantly, all particles sank (but at 106 very different speeds) and were transported by physical circulation. Particles were determined to have been 107 "subducted" across a particular depth horizon if they were transported across that depth horizon during the physical 108 time step in LTRANS (rather than the biological/sinking time step). Conversely, particles that were transported 109 across that depth horizon during the biological/sinking time step were considered to have sunk across that depth 110 horizon. Note that this explicitly allows particles to contribute to both of these BCP pathways at different depths, as 111 they can be subducted across one depth horizon and then sink past another horizon or vice versa. Carbon flux 112 associated with "subduction" was determined based on the carbon existing in the particle at the time of subduction. 113 We note however that this "subduction" really refers to downward physical transport associated with subduction and 114 vertical mixing generated by multiple physical mechanisms (e.g., it combines the "eddy-subduction pump", "mixed-115 layer pump", and "large-scale physical pump" as defined in ref 16 ). 116 117 SUPPLEMENTARY REFERENCES 118