We investigate the functioning of the ocean’s biological pump by analyzing the vertical transfer efficiency of particulate organic carbon (POC). Data evaluated include globally distributed time series of sediment trap POC flux, and remotely sensed estimates of net primary production (NPP) and sea surface temperature (SST). Mathematical techniques are developed to compare these temporally discordant time series using NPP and POC flux climatologies. The seasonal variation of NPP is mapped and shows regional- and basin-scale biogeographic patterns reflecting solar, climatic, and oceanographic controls. Patterns of flux are similar, with more high-frequency variability and a subtropical-subpolar pattern of maximum flux delayed by about 5 days per degree latitude increase, coherent across multiple sediment trap time series. Seasonal production-to-flux analyses indicate during intervals of bloom production, the sinking fraction of NPP is typically half that of other seasons. This globally synchronous pattern may result from seasonally varying biodegradability or multiseasonal retention of POC. The relationship between NPP variability and flux variability reverses with latitude, and may reflect dominance by the large-amplitude seasonal NPP signal at higher latitudes. We construct algorithms describing labile and refractory flux components as a function of remotely sensed NPP rates, NPP variability, and SST, which predict POC flux with accuracies greater than equations typically employed by global climate models. Globally mapped predictions of POC export, flux to depth, and sedimentation are supplied. Results indicate improved ocean carbon cycle forecasts may be obtained by combining satellite-based observations and more mechanistic representations taking into account factors such as mineral ballasting and ecosystem structure.
 The marine biosphere is a major component of the global carbon cycle, responsible for roughly half of the annual photosynthetic absorption of CO2 from the atmosphere [Field et al., 1998]. However, the influence of marine carbon cycle variability on atmospheric CO2 concentrations is poorly understood [Sarmiento and Le Quéré, 1996; Le Quéré et al., 2005]. In particular, there is little basis to predict how potential alterations in upper ocean ecosystems may influence the capacity of the ocean to store carbon [Sarmiento et al., 1998]. Significant variability in the vertical flux of particulate organic carbon between oceanographic regions [Lutz et al., 2002; Neuer et al., 2002] implies the potential for substantial biogeochemical consequences resulting from alterations to upper ocean dynamics. Changes in the efficiency of the marine organic carbon cycle may contribute to variability of atmospheric CO2 on glacial-interglacial timescales [Sigman and Boyle, 2000]. Understanding of the fate of biogenic carbon within the ocean is essential for improved forecasts of future atmospheric CO2 concentrations and rates of global warming, especially in the context of changing ocean circulation [Sarmiento et al., 1998; Schmittner, 2005].
 The transport of biogenic elements from surface waters to the deep ocean and sediments occurs through a variety of processes collectively known as the “biological pump” [Volk and Hoffert, 1985]. This biogeochemical and physical system may be conceptually divided into four interrelated components: production, export, flux to depth, and sedimentation. During primary production, phytoplankton incorporate dissolved nutrients and inorganic carbon into particulate organic matter (POM), lowering the partial pressure of CO2 in surface waters, thus enhancing the ocean’s ability to absorb CO2 from the atmosphere. Much POM is recycled within surface waters where associated nutrients support regenerated primary production [Eppley and Peterson, 1979]. A fraction this POM is exported beneath surface waters and fluxes to depth within the ocean’s interior where it is predominantly consumed and regenerated by subsurface heterotrophs. The minor portion of POM that reaches the seafloor is mostly remineralized in support of benthic biological activity. Flux to the seafloor escaping benthic regeneration is buried in the sediments.
 In general, the greater the depth at which sinking organic carbon is regenerated, the longer time it takes to return to the photic zone as dissolved CO2, where it may reenter atmospheric carbon cycle. Organic carbon reaching the deep ocean is entrained in water masses with longer flow pathways back to the surface and smaller advective velocities than in the upper ocean. Ventilation of subsurface waters occurs on timescales ranging from annual to hundreds of years in the upper ocean (∼100–1000 m) to thousands of years in the deep ocean (>1500 m). Carbon entering sediments may be geologically stored for hundreds of millions of years. Thus to describe the storage of biogenic carbon in the ocean, the depth of remineralization must be known.
 This paper explores how global-scale environmental parameters, sea surface temperature (SST) and the seasonality of net primary production (NPP), relate to the efficiency of the biological pump. The ability to infer SST and NPP from remote observations allows for their synoptic comparison to sediment trap data. This investigation introduces methodology to facilitate the synthesis of these time series. Our metadata analysis is used to describe the seasonal efficiency of production-to-flux relationships, evaluate hypotheses regarding the functioning of the biological pump, assemble new particle flux algorithms, and test parameterization accuracy.
2. Surface Temperature and Seasonality of NPP
 Recent evidence indicates water temperature may be a key environmental parameter influencing the proportion of NPP exported from surface waters to the deep ocean [Laws et al., 2000; Rivkin and Legendre, 2001]. This hypothesis suggests the rate of organic matter recycling is influenced by SST. Specifically, lower-water temperatures limit the activity of microbial heterotrophic decomposers more than the activity of autotrophic producers. This scenario implies that in colder regions of the ocean, a greater fraction of production is not remineralized and is therefore available for export. Furthermore, due to the lack of surface recycling where SST is colder, exported detritus may be more labile than where SST is warmer. Laws et al.  propose a euphotic zone food web model describing biogenic export relying heavily on this premise. In support of this rationale, the authors demonstrate a strong near-linear correlation between observed f ratios, the ratio of new to total production [Eppley and Peterson, 1979], and those predicted by their model. Water temperature alone explains most of the f ratio variability found in the model by Laws et al. .
 Another proposed control on biogenic carbon export is variability in the efficiency of trophic transfer within the upper ocean food web [Berger et al., 1989; Berger and Wefer, 1990; Lampitt and Antia, 1997; Rühlemann et al., 1999; Fischer et al., 2000]. This hypothesis suggests that organic matter recycling is influenced by controls related to variability of upper ocean ecosystem structure. In particular, the fraction and biodegradability of production exported from more variable ecosystems may be greater than from stable ones. Where the supply of solar irradiation and nutrients are relatively constant, environments are stable throughout the year and activities of autotrophs and heterotrophs are balanced. Trophic transfer is efficient and recycling dominates. The fraction of production exported is minimal. Conversely, seasonal environments reflect time-varying physical conditions, solar irradiation, and nutrients, and are characterized by “leaky” food webs with temporally imbalanced components. New production dominates and rapid autotrophic growth outpaces consumer feeding and decompositional activity. Particulate matter has a greater opportunity to avoid surface heterotrophy and be exported in a more labile or “fresh” state. Thus in variable environments particulate matter has a greater opportunity to escape the euphotic zone and be more biodegradable than in stable environments.
Particulate organic carbon flux (Cflux(z); mg Corg m−2 d−1) is described as a function of the production of organic carbon in surface waters (Cnpp) or the export of organic carbon (Cexport) from the base of the euphotic zone (z0), scaled to depth below the sea-surface (z). The Suess  equation was determined from field-based production and sediment trap flux measurements, collected from the subtropical eastern Pacific and northwestern Atlantic at depths between 50 and 5400 m. The Martin et al.  normalized power function is a “best fit” based on sediment trap data collected in the low to midlatitude east Pacific from depths between 100 and 2000 m. Global carbon models including marine biology commonly describe flux to depth using the fixed parameters shown in equations (1) and (2).
 Our approach builds on the relationships of Suess  and Martin et al.  [equations (1) and (2)] by using the greater amount of recently available data to constrain more of the variability between production and flux to depth. In particular, we compare globally distributed annual and time series sediment trap flux measurements, and remotely sensed estimates of NPP and SST. Sediment traps and satellites allow for the characterization of marine biogeochemical variability on timescales longer than the data sets collected through traditional ship-based oceanography.
 The global sediment trap data set assembled for this analysis contains estimates of flux collected during the past 25 years. Satellite-based estimates of production are employed because field-based observations of production are not available at the same locations and durations as the sediment trap experiments. Annual climatologies are used to compare the seasonality of these temporally discordant time series. Interannual variability is not focused upon in this study.
4.1. Sediment Traps
 Moored sediment traps are funnel-shaped devices attached at varying depths to line that is fixed in location to the seafloor [Honjo, 1978]. Settling detritus intercepted by these funnels collects in sample bottles opened and closed at the funnel base. Each sediment trap measurement reflects the total mass of flux collected during each bottle open-close interval. Sampling frequency and interval duration typically varies within sediment trap experiments. Experiments often include interruptions produced during collection interval malfunctions and between trap deployments. The result of a sediment trap experiment is typically a discontinuous time series of average flux rates representing bottle open-close intervals of variable durations.
 The following criteria are used in assembling sediment trap data that best reflect surface production (Table 1 and Figures 1 and 2) :
 (1) Values are from depths greater than the local export depth (ze). The depth of export herein describes maximum of the euphotic zone and mixed layer depths during flux formation. Sediment traps located above this depth may not reflect the flux of particulate matter to the deep ocean. On the basis of the geographic and depth distribution of sediment trap observations, we apply a first-order approximation of the export zone depth as 100 m at equatorial latitudes (<35°) increasing linearly to 400 m at high-latitudes (>50°).
 (2) Observations are minimally influenced by resuspension of benthic surface sediment, typically those from depths above the nepheloid layer.
 (3) Terrestrial sources of detritus are minimal.
 (4) Observations must describe at least one entire year of flux to depth. This includes trap experiments of less than a year in duration if the seasonal cycle of flux may be well approximated. The sediment trap data set assembled for this study includes 244 annual flux estimates and 153 subannually resolved flux time series used to create flux climatologies.
 Flux climatology construction involves multiple steps depending on the temporal coverage of trap measurements. First, time series are made continuous for the entire duration of the experiment. Gaps produced during sampling bottle failures or between trap deployments are filled with the linear integral of temporally adjacent observations. Next, to account for timing inconstancies between trap experiments, time series are temporally standardized. Temporal standardization involves a method of describing sediment trap measurements that is a combination of the regularly applied techniques.
 Authors typically depict sediment trap experiments using linear connections between observation midpoints [Bathmann et al., 1990; Neuer et al., 1997; Honjo et al., 1999; Hargrave et al., 2002]. Flux is also frequently reported in bar-graph format as the average rate during the open-close intervals of each bottle [Newton et al., 1994; Honjo et al., 2000; Lampitt et al., 2001; Honda et al., 2002]. Our approach, demonstrated in Figure 3, is a smoothing of the linear style held equal in mass collected per bottle interval to the bar-graph style. Dashed lines connecting bottle midpoints (closed circles) and grey columns indicate the linear and bar-graph data representation styles. Open circles designate temporal intersections of the linear style with sample bottle openings (xo) and closings (xc). The solid thin line l connects these intersections. The curve D indicates the “stretching” of line l such that the integral of D equals the area represented by the bar-graph style for each bottle:
where denotes the average flux rate. The stretch function D is accomplished by multiplying line l by a daily cubic spline interpolation of normal distribution constrained so that the endpoints (open circles) remain constant. The direction line l stretches to become curve D, up, down, or not at all, depends on the flux rate measured in each bottle relative to the rates of neighboring bottles. Minimum daily flux rates estimated by the stretch function D are constrained not to be negative. Finally, sediment trap measurements including more than 1 year of observations are averaged into one climatological year.
4.2. NPP and SST
 NPP is estimated following Behrenfeld and Falkowski [1997a] using 7 years of 8-day satellite images available between 19 August 1997 and 24 June 2004. Data sets applied include the NOAA/NASA AVHRR Oceans Pathfinder SST (http://podaac.jpl.nasa.gov/products/product216.html), NASA SeaWiFs surface chlorophyll concentrations, and photosynthetically active radiation (http://daac.gsfc.nasa.gov/data/dataset/SEAWIFS/). Images applied display global coverage with an equal-area grid of 9-km resolution. NPP climatologies are constructed in manner identical to that of flux climatologies. During climatology construction, missing production estimates, for example, those obscured by cloud cover, are substituted with the linear interpolation of adjacent values.
4.3. Potential Error
 Errors associated with methods used to estimate marine biogeochemical processes are difficult to assess. Field-based oceanography is far from a controlled well-calibrated laboratory experiment. The ability of sediment traps to accurately measure flux has received much attention in the scientific literature. Possible trap biases include inadvertent “swimmer” capture [Lee et al., 1988], uncertain preservation of trapped particulate matter [Lee et al., 1992], hydrodynamic interactions with traps of various designs [Gust et al., 1994; Buesseler et al., 2000], and influences in brine and poison addition [Lee and Cronin, 1982; Knauer et al., 1984]. A standard trap design and methodological protocol is not adopted by the oceanographic community. The magnitude of these and other trap biases may be highly variable and as large as a factor of two or more [Gardner, 2000]. Calibration of sediment trap results have been attempted [Buesseler et al., 2000; Yu et al., 2001; Scholten et al., 2001]. In defense of sediment trap technology, we note that sediment trap results: (1) are interannually consistent with respect to timing, variability, and magnitude of particle fluxes [Deuser, 1986; Conte et al., 2001]; and (2) reflect euphotic zone biological variability [Deuser et al., 1981; Honjo, 1982; Deuser et al., 1990]. We use sediment trap data sets here because they represent the only available direct measurements of annual subsurface particle fluxes through the water column.
 To address potential error of sediment trap observations, estimates global flux to depth include the calibration rationale suggested by radionuclide studies. The validity of sediment traps to measure flux to the deep ocean (>1.5 km) is validated by 230Th and 231Pa calibration studies [Scholten et al., 2001; Yu et al., 2001]. Within the upper ocean (<1500 km) these radionuclide studies suggest sediment traps may often underestimate fluxes. With the exception of the California margin, Yu et al.  found a trapping efficiency of 40% is a typical minimum value for the pelagic upper ocean. To account for the potential undertrapping error, we report global estimates of flux to the upper ocean with (observed flux divided by 0.4) and without radiochemical calibration. This trapping efficiency estimate may be uncertain in part because of the variable incorporation of radionuclides on particles of different sizes [Buesseler et al., 1995; Yu et al., 2001].
 Satellite-based estimates of NPP also include significant uncertainties [Platt and Sathyendranath, 1993; Behrenfeld et al., 2002a]. These uncertainties largely stem from errors in relating surface chlorophyll biomass to phytoplankton carbon biomass [Behrenfeld et al., 2005]. Furthermore, empirical models used to estimate NPP from ocean color may not correctly simulate relevant phytoplankton physiological variability [Behrenfeld and Falkowski, 1997b; Behrenfeld et al., 2002b; Banse and Postel, 2003]. Ocean color is a property of the uppermost portion of euphotic zone and additional factors are needed to estimate deeper ocean production. Finally, field-based radiocarbon measurements of production used to calibrate satellite-derived algorithms include error [Sakshaug et al., 1997]. A recent performance analysis indicates various NPP algorithms yield significantly dissimilar results [Campbell et al., 2002]. We acknowledge algorithms describing production using satellite data may have significant systematic biases.
 Specific estimates of error regarding satellite-based NPP and sediment trap flux are not easily quantified and such an exercise is generally avoided in the literature. There is no a priori reason to believe errors associated with sediment trap observations covary with errors associated with satellite-based observations. Thus we infer that statistical significance in the results described below is found in spite of, and not because of, error in the data sets used.
4.4. Seasonal Variation Index
 We use the seasonal variation index (SVI) to describe temporal variability of the time series. SVI is defined as the coefficient of variation, the standard derivation (σ(X)) normalized to the average (), of either the NPP or flux climatologies:
 SVI is dimensionless. This description of variability is statistically similar to the seasonality index, the number of months required to accumulate one half of the total annual primary production [Berger et al., 1989; Berger and Wefer, 1990], and the flux stability index, the minimum time taken for 50% of the annual flux to be collected [Lampitt and Antia, 1997].
4.5. Annual and Seasonal Production Ratios
 Previous research has compared e ratios, the ratio of flux to export production, to examine the transfer efficiency of POC [Laws et al., 2000; Francois et al., 2002]. A recent study shows transfer efficiency described by flux normalized to either production or export yield results of similar accuracy [Lutz et al., 2002]. Here following the approach of Suess , we compare the transfer efficiency between regions using annual production (p) ratios, defined as the annual particle flux at depth (z) normalized to annual NPP in the overlying surface waters:
This study also uses seasonal p ratios to examine subannual production-to-flux dynamics. Seasonal p ratio creation involves several steps and is conducted at locations where an entire year of flux time series is available. First, seasons of production are delineated by dividing NPP climatologies into autumn, winter, spring, and bloom intervals. The bloom production interval describes the average production during the 30 continuous days of an annual record with the maximum average production rates. Similarly, the winter production interval describes the average production during the 30 continuous days with the minimum average production rates. Autumn and spring production intervals describe average productions during the intervening time periods between bloom and winter seasons. The durations of autumn and spring production intervals varies depending on the timing of the bloom and winter seasons, and are constrained to be of at least one month in duration. By this constraint, 134 time series of production are included in the seasonal p ratio analysis.
 Next, the timings of corresponding sediment trap flux seasons are determined by adding a temporal lag to the timing of the production seasons. The temporal lag is defined by dividing the trap depth (m) by a sinking rate (m d−1). Here we use a sinking rate of 70 m d−1 based on the 1.5-month lag between surface events observed by satellite and arrival of the consequences of those events in sediment traps at 3.2 km shown by Deuser et al. . Finally, seasonal p ratios are determined by normalizing seasons of flux to corresponding seasons of production. We acknowledge this technique allows for a first-order description of production and flux dynamics, as sinking rates may differ between seasons, regions, and depths [Berelson, 2002]. Due to the lack of polar wintertime production, winter season p ratios are absent at latitudes >60°.
4.6. Flux Algorithm With Labile and Refractory Components
 To quantify relationships between satellite-based NPP and SST data, and sediment trap flux observations, annual production ratios are estimated using the following exponential algorithm:
 This algorithm was previously derived from an empirical fit between regional production and flux measurements [Lutz et al., 2002] and is applied here to determine if flux can be estimated using environmental parameters derived from satellite observations. This function describes the labile and refractory components of flux to depth below the export zone depth ze [depth (z) minus export zone depth]. The prd and rld coefficients measure flux that is available to decay and sinks more slowly. The rld coefficient describes the e-folding remineralization length/depth scale and measures the rate of sinking detritus degradation. The prr coefficient describes the more refractive and more rapidly sinking portions of flux. Larger prd and prr values indicate the export of a greater fraction of NPP. Larger rld values indicate the labile fraction of export sinks deeper before regeneration. For simplicity, this equation treats sinking and remineralization rates as constants, although we acknowledge these factors vary as particles are transformed during their descent through the ocean [Berelson, 2002]. For further derivation of equation (6) see Banse  and Lutz et al. .
 Annual p ratio estimates are used to determine the coefficients of equation (6) grouped by either SST or SVI. Coefficient approximation requires the p ratio data be divided into groups so that the entire water column is described for each SVI and SST value evaluated. To best characterize SVI and SST variability, groups used herein are of equal spacing on a logarithmic-scale for SVI and a linear-scale for SST. The prd, rld, and prr coefficients are determined for each group separately. Within each group the prr coefficient describes the median of the deepest 25% of p ratios, the depth of which vary slightly between groups. Thus for the purpose of this comparison it is assumed that on an annual basis flux to the deepest traps is refractory relative to upper ocean traps. The prd coefficient describes the maximum shallow p ratio minus the prr coefficient. The rld coefficient is determined by fitting equation (6) to the p ratio data with the previously determined prd and prr coefficients. For each coefficient the maximum number of groups is used that yields a statistically significant relationship to either SVI or SST. Hence the number of groups for the prd and prr parameters differs depending on the availability of the shallow and deep trap data within each group. Similarly the grouping for the rld parameter requires a sufficient number of observations to characterize the entire water column. Due to the infrequency of sediment trap observations near the local export zone depth, prd values are deemed minimal approximations.
5.1. Patterns of Remotely Sensed Parameters, NPP, and SST
 The results of our NPP climatology generally agree with those described by Behrenfeld and Falkowski [1997b]. Here we briefly describe the first-order characteristics of production temporal variability limited to those relating to patterns shown by our flux to depth climatology.
 The NPP climatology, as described by the SVI (Figure 4), displays regional- and basin-scale geographic patterns. Temporal trends of the NPP climatology (Figure 5) are similar to those found in the literature [Yoder et al., 1993; Longhurst, 1995] and include the well-known open ocean equatorial near-constant low productivity, the multiple blooms of temperate and monsoonal regions, and the intense summer bloom of polar regions. A primary component of global distribution of NPP SVI is the increase of values poleward of roughly 45°S and N, reflecting the seasonal availability of photosynthetically available radiation and distribution of sea ice.
 The heterogeneous distribution of SVI values at lower-latitudes indicates provinces where interactions of annual climatic phenomenon with the density structure of upper ocean waters permit seasonal upwelling. In open ocean waters, diminished SVI values characterize subtropical regions of year-round high surface atmospheric pressure and the seasonal limits of the intertropical convergence zone. Reduced SVI values characterize the oligotrophic subtropical ocean gyres and equatorial western Pacific and Atlantic Oceans, where production is relatively constant year-round. Narrow longitudinal bands of moderate SVI values resulting from seasonal water mass divergence characterize the equatorial upwelling regions of the Pacific and Atlantic Oceans. Elevated SVI values approximate the influence of the circumpolar westerlies. Monsoonally influenced regions of the western Indian Ocean exhibit enhanced SVI values. Larger SVI values typically characterize the centers of eastern boundary seasonal coastal upwelling and the associated offshore areas where upwelled nutrients are advected. Additionally, SVI values reflect the seasonal input of riverine and aeolian nutrients, as well as limitations to production, such as mixing of surface waters beneath the photic zone, phytoplankton self-shading, and herbivore activity [Parsons et al., 1977; Broecker and Peng, 1982; Chester, 1990; Duce and Tindale, 1991].
 The Southern Hemisphere of the Pacific Ocean indicated by transect A of Figures 4 and 5 epitomizes the transition from low-amplitude seasonality and diminished production of equatorial regions to high-amplitude seasonality and strong summer blooms of polar regions. Heading south from the equator, between 5° and 15°S, SVI values decrease as the strength of equatorial upwelling diminishes. Continuing south, a latitudinal ribbon of elevated SVI values, between 25° and 35°S, indicates where mixing of subsurface waters increases surface nutrient levels sufficient to generate a summer bloom. Adjacently poleward, a narrow latitudinal band of minimum SVI values, between 35° and 40°S, indicates where upper water column mixing sustains elevated production year-round, without distinct spring and fall blooms. Farther poleward SVI values increase as bloom intensity amplifies. The latitude-dependent SVI fluctuations and the northeast trending “sustained-bloom” band of diminished SVI values distinguish the influence of the westerlies, and are less distinctly repeated at similar latitudes in the western and eastern portions of the Northern and Southern Hemispheres, respectively.
 For much of the global ocean the SVI of production and SST show similar trends (Figures 6 and 7) . Poleward of 35° latitude, where waters are cooler than approximately 18°C, SVI, and SST generally covary. At lower-latitudes the distribution of SVI and SST varies depending on regional oceanographic circumstances and differs significantly between ocean basins. Globally, neither variability of production nor SST correlates significantly with rates of NPP.
5.2. Patterns of Sediment Trap Flux
 The results of the flux to depth climatology are shown in Figure 8. Here we focus on regional- and basin-scale patterns of flux rather than restate findings of original sources. Global sediment trap sampling is heterogeneous in depth and geographic distribution. Often, the water column is not entirely measured, leaving either shallow, intermediate, or deep waters uncharacterized. Sediment trap observations are heavily weighted, by 70%, toward the Northern Hemisphere. Major portions of the Southern Hemisphere are uncharacterized, including the deep Antarctic circumpolar waters of Atlantic and Indian Oceans. The Indian Ocean has received particularly little attention, with less than an eighth of all observations. Large segments of continental shelf regions are unknown, such as surrounding South America and much of Antarctica, and bordering the Indian Ocean. Meagerly characterized open ocean regions include the equatorial upwelling of the Atlantic and Indian Oceans, and the moderately productive Southern Hemisphere subtropical convergence. Regions of mesotrophic productivity have received the majority of attention. The most productive centers of upwelling regions are not well characterized. The modestly productive oligotrophic central ocean gyres are similarly not frequently sampled. Hence regions characterizing minimum and maximum limits of global productivity rates largely await description.
 Flux seasonality of subtropical-subpolar open ocean waters is characterized by a latitude trending pattern, whereby maximum fluxes occur later in the year at higher-latitudes. This pattern is coherent across multiple sediment trap time series, recognizable in the Atlantic Ocean in intermediate [e.g., Figure 8: 40-i, 41-i, and 42-i, Honjo and Manganini, 1993] and deep waters [e.g., Figure 8: 36-d, Honjo et al., 1987; 38-d and 39-d, Lampitt et al., 2001; 40-d, Honjo and Manganini, 1993; 41-d, Deuser et al., 1981; Conte et al., 2001; 42-d, Neuer et al., 1997]. This pattern is also apparent in deep waters of the Pacific Ocean [e.g., Figure 8: 9-d and 10-d, Honda et al., 2002; 12-d, 15-d, and 16-d, Kawahata et al., 2002; 13-d and 14-d, Mohiuddin et al., 2004]. Overall, flux peaks are delayed by approximately five days per degree latitude increase. Maximum flux timings are generally synchronous with the timing of temperate spring blooms of lower latitudes and summertime polar blooms of higher latitudes, with a production-to-flux lag typically between 40 and 80 days. The range of flux rates between season’s maximum and minimum flux is roughly a factor of six.
6.1. Limitations Using Seasonal Production Ratios
 Sediment trap measurements of flux to depth reflect variability of surface water production whereby periods of enhanced production generally correspond to periods of enhanced flux to depth [Deuser et al., 1981; Honjo, 1982; Deuser et al., 1990]. We use seasonal production ratios (section 4.5) to characterize global-scale subannual variability of particle transport efficiency. Inferences using seasonal production ratios are limited due to methodological and natural factors. The seasonal interval technique does not completely describe subannual variability of production and flux to depth. For example, dual production blooms and flux events of monsoonal regions [Honjo et al., 1999] are not completely characterized. Furthermore, the technique is limited by interannual differences between the timing of production and flux to depth time series used to compile the climatologies. For example, where production and flux are influenced by long-term climate phenomenon, such as ENSO [Kawahata and Gupta, 2004], some bloom production and flux intervals may be mismatched. Differences in particulate matter sinking rates between seasons and depths may further obscure interpretations. Furthermore, the potential inaccuracy of shallow sediment traps may limit the interpretation of shallow p ratio estimates. This analysis represents an initial step toward characterizing subannual production-to-flux variability.
6.2. The Seasonal Sinking Fraction of NPP
 Our results indicate the sinking fraction of production is not seasonally constant. The global range of seasonal production ratios (Figures 9a and 10a) spans four orders of magnitude, from 0.0003 to 0.93, with much overlap between seasons. This seasonal range is almost two orders of magnitude greater than the range of annual p ratios between different ocean regions (from about 0.001 to 0.1) [Lutz et al., 2002]. A trend toward lower p ratios during the bloom seasonal interval is indicated; however, an overlap between seasonal estimates makes differentiation between seasons difficult. Much of the global variability in seasonal p ratios reflects differences between regions that exhibit different ranges of NPP and different efficiencies of the biological pump [Lutz et al., 2002; Neuer et al., 2002]. In order to reduce the influence of variability between locations and changes between depths, seasonal p ratios are normalized to local annual p ratios (Figures 9b and 10b). Normalized p ratios during bloom intervals are on average half of those during the rest of the year. The greatest range of normalized p ratios and the bulk of the lowest normalized p ratios occur during the bloom season at <1 km.
 The production ratio results indicate a global-scale synchronicity in the seasonal functioning of the biological pump. This pattern may reflect a seasonal change in the biodegradability of sinking particulate matter. Diminished bloom p ratios are consistent with a greater proportion of particulate matter remineralized during enhanced production. Increases in flux lability in response to increased production is suggested by previous flux studies involving sediment traps [Lee and Cronin, 1984; Lohrenz et al., 1992; Haake et al., 1993; Newton et al., 1994; Lampitt and Antia, 1997]. Changes of biodegradability may be due to the seasonal manufacture of skeletal material by phytoplankton. For example, growth forms of many Antarctic diatoms progress seasonally from long, lightly silicified chains (more labile) in the austral spring, to heavily silicified valves (more dense and refractory) in the autumn and winter [El-Sayed and Fryxell, 1993].
 The pattern of seasonal production ratio may additionally reflect particulate matter retention or the delay of sinking particulate matter to reach deeper waters. Diminished p ratios during bloom production relative to other seasons is consistent with a portion of bloom-derived particulate matter being retained to be recycled or fluxed during latter seasons. Multiseasonal retention and recycling of detritus within the deep ocean is suggested to explain seasonal patterns of suspended particulate matter concentrations in the northeast Pacific [Bishop et al., 1999].
6.3. Seasonality of NPP and Flux to Depth
 The balance between seasonality of production and seasonality of flux reverses with latitude (Figure 11). At higher-latitudes seasonality of production is generally greater than seasonality of flux. At lower-latitudes seasonality of production is generally less than seasonality of flux. To account for this latitudinal difference we suggest that processes influence flux seasonality discernable at lower-latitudes, where variability of production is diminished, are masked by the larger-amplitude production seasonality signal of higher latitudes. Processes that may enhance low-latitude flux seasonality relative to production include phytoplankton mass sedimentation events [Kemp et al., 2000] and transient meteorological forcing of pulsed export [Conte et al., 2003]. Processes that may attenuate seasonality between production and flux at high-latitudes include the regeneration of detritus retained within surface waters, delayed herbivore growth, and seasonal mixed-layer deepening associated with polar winter onset. Overall, the production-to-flux seasonality signal may be influenced by horizontal water mass advection [Siegel and Deuser, 1997], which may covary with rates of production [Lampitt and Antia, 1997].
6.4. Parameterization of the Annual Sinking Fraction of NPP
 Our analysis of SST- and SVI-associated hypotheses includes determining relationships between the remotely sensed parameters and production ratio data, and comparing the ability of algorithms derived to predict flux. Flux predictions are based on algorithms that describe the coefficients of equation (6) as a function of the remotely sensed parameters, as outlined in section 4.6. Coefficient parameterization involves testing different equation functional forms (polynomial and exponential), various orders of equations, and different numbers of data groups to find the most accurate predictions. Algorithm curve fits were constrained to not allow negative coefficient parameterizations. Figure 12 and Table 2 show the data groupings and coefficient algorithms that produced the most accurate flux predictions. Curve fits shown do not imply complex relationships, but rather show the simplest and most accurate coefficient algorithms attained.
Table 2. Curve Fit Algorithms Describing Coefficients Used to Estimate Annual Particulate Organic Carbon Flux to Depth Normalized to Overlying Production (p ratio(ze)) as a Function of Satellite-derived Parameters, the Seasonal Variation Index of Production (SVI; Annual Standard Deviation Divided by Average) and Sea-surface Temperature (SST; °C)
Coefficients describe flux using equation (6)a: the labile fraction of export (prd; Figure 12, A and B), remineralization length scale (rld; Figure 12, C and D), and more refractory and rapidly sinking fraction of export (prr; Figure 12, E and F).
 Comparison of SST and p ratio data groups shown in Figure 12 indicates little significant variation in p ratios for much of the range of SST. Dissimilar coefficient behavior is generally confined to minimal SST values. This discontinuous distribution may imply that SST influences remineralization in a stepwise manner with little influence until a certain low threshold temperature is reached, where flux is more labile as indicated by the prd and rld coefficients. Alternately, this behavior may reflect SST and SVI covariance, characteristic of lower temperatures (Figure 7). SVI coefficients and p ratio data groups show somewhat more continuous distributions and include a larger range of rld and prr coefficient values. SVI coefficient distributions are consistent with the suggestion that where production is more variable, sinking PM is more labile and decays more rapidly.
 Flux to depth forecast performance abilities of the SST- and SVI-associated parameterizations and the relationship presented by Suess  are compared in Figure 13. The most accurate forecast performance is found using the SVI-related algorithm (Figure 13). All equations overestimate flux to depth and p ratios in the upper ocean (≤1500 m), although to degrees varying depending on latitude range evaluated. Overestimation of flux and p ratios is largest using the equation of Suess , especially at shallow depths. The SVI- and SST-algorithms perform similarly in the deep ocean (≥1500 m). A portion of the improved accuracy derived using equation (6) may be due to our incorporation of a larger range flux and production data available to Suess . In the upper ocean the SST algorithm overestimates high-latitude flux and p ratios at all latitudes more than the SVI algorithm. This depth dependant discrepancy suggests the exponential form of equation (6) and the depth scaling of the equation of Suess  do not always account for the most rapid degradation of detritus in the upper water column.
6.5. Predictions of Annual Flux to Depth
 Coupling between surface and subsurface biogeochemical process has been proposed by comparing satellite-derived estimates of production and sediment trap-derived flux to depth [Lampitt and Antia, 1997; Fischer et al., 2000; Antia et al., 2001; Müller-Karger et al., 2005]. We build on this insight by using parameters in addition to the rate of NPP to further constrain flux to depth. Algorithms developed using the remotely sensed parameters, the SVI of production and SST, and equation (6) allow for predictions of flux and p ratios and the assessment of forecast accuracy relative to the equation of Suess .
 The accuracy of SVI-associated annual flux and p ratio predictions allows for prediction of global particulate organic carbon flux using satellite-derived NPP (Figure 14 and Tables 3 and 4). For much of the global ocean, the geographic pattern of flux is similar at different depths. Enhanced export and flux characterize regions where rates of production and production seasonality are enhanced. However, in the central northern Atlantic and Pacific Oceans and in the Southern Ocean where export is enhanced, flux to the seafloor is diminished. Flux to the seafloor is greatest on continental shelves and within centers of coastal upwelling. At depths greater than the continental shelves the geographic distribution of flux to the seafloor is similar to that of flux within the water column. While absolute flux rates differ, the geographic patterns of flux and export predicted by the SVI algorithm are generally similar to other satellite-derived global estimates [Falkowski et al., 1998; Laws et al., 2000; Müller-Karger et al., 2005], with proportionally larger fluxes in regions of enhanced seasonality (for example, higher latitudes and regions of seasonal upwelling).
Table 3. Annual Average Particulate Organic Carbon Export, Flux to Depth, and Flux to the Seafloor in the Global Ocean
Estimates are derived as a function of a satellite-derived net primary production (NPP) and the Index of Seasonal Variation (annual standard deviation divided by average) of NPP using equation (6).
Ocean basins are divided such that: the Southern Ocean is delineated south of 40°S; Arctic waters are delineated north of 70°N; the Indian Ocean is delineated east of the Cape of Good Hope (20°E) and west of Sumatra and Australia from Cape Londonderry (127°E) to Melbourne (147°E); the Atlantic Ocean includes the Mediterranean Sea, which has minor a contribution to global flux.
 Sediment trap-derived estimates of global export are significantly less than those developed using other methods. Multiple causes may influence this discrepancy. Accuracy of sediment trap-derived estimates of export is limited by the lack of measurements characterizing the upper mesopelagic where greatest rates of subsurface recycling occur. As noted in section 4.3, underestimation of upper ocean fluxes by sediment traps is suspected for a variety of reasons. While radiogenic calibration narrows the discrepancy between estimates, the difference remains significant. Variability associated with the radiogenic technique may be responsible [Yu et al., 2001]. Finally, a portion of this discrepancy may be due to export of dissolved organic carbon [Carlson et al., 1994] not measurable using sediment traps.
 Sediment trap estimates of flux to the seafloor, although not significantly dissimilar from one another, are significantly greater than flux estimated using apparent oxygen utilization (AOU) in surface sediments [Jahnke, 1996]. This discrepancy may be due to errors associated with sediment trap and AOU-associated flux methods. The accuracy of traps within the deep ocean may be influenced by some of the hydrodynamic limitations associated with traps in shallow waters. Resuspension of benthic detritus into deep-water traps may be greater than anticipated. Multiple factors may limit the accuracy of the AOU-associated flux estimations. The flux and subsequent accumulation of settled detritus on the seafloor may be geographically heterogeneous (for example, favoring topographic low points) and thus not well characterized by spatial distribution of coring techniques typically used to collect benthic surface sediments. Benthic biological activity and bioturbation may further enhance the geographic heterogeneity of benthic remineralization [Tedesco and Wanless, 1991; Meadows and Meadows, 1994]. Finally, the remineralization of flux arriving on the seafloor may occur on timescales not assessed by the methods reported in the work of Jahnke .
 Theoretical connections have been proposed between variability of upper ocean dynamics and pelagic biogeography [Longhurst, 1995], and between biogeography and biogeochemical cycling within the ocean [Longhurst and Harrison, 1989; Lampitt and Antia, 1997]. In particular, Lampitt and Antia  suggest biogeography, as described by the plankton climate categories of Longhurst , influences marine biogeochemical cycling. Results presented in this study further establish these findings. Our analysis suggests variability of production, as characterized by the seasonal variation index, reflects ecosystem-scale biogeochemical processes. This connection may be because, by measuring environmental change, SVI of production reflects pelagic ecosystem structure [Longhurst, 1995].
 The following conclusions are based on our study of the biological pump: The NPP climatology displays seasonal patterns coherent over large geographic provinces, indicating the relative dominance of solar, climatic, and oceanographic controls on the annual variability of NPP. In general, seasonal patterns of flux reflect those of production. Prominent patterns of flux include polar, equatorial upwelling, coastal upwelling, monsoonal, and within subtropical-subpolar regions. Notably, within subtropical-subpolar open ocean regions the timing of maximum flux is delayed by approximately 5 days per degree latitude increase. Coherence of flux patterns between multiple widely dispersed locations demonstrates the ability of sediment traps to characterize sinking particulate matter. Seasonal production-to-flux analyses indicate POC vertical transfer efficiency is significantly seasonally variable. In particular, the ratio of flux to production during bloom production is typically half that of the remaining year. Comparison of production and flux variability shows a latitudinal dependant relationship. At lower-latitudes seasonality of flux is typically greater than that of production, while at higher latitudes, seasonality of production is typically greater than that of flux. This reversal of variability may describe a biogeographic distinction in the controls of production-to-flux relationships.
 By analyzing a globally distributed data set of NPP and flux to depth, we show the accuracy of algorithms describing flux is improved by incorporating SVI- and SST-related controls and describing labile and refractory sinking fractions of production. The use of the Behrenfeld and Falkowski [1997a] global NPP model with equation (6), parameterized as a function of the SVI of production and SST, yields annual satellite-based estimates of deep-sea particulate fluxes with greater skill than the commonly applied equation of Suess . Global-scale coherence between seasonal patterns of production and flux to depth demonstrate the influence of annual climatic variability on the marine carbon cycle. Results suggest atmospheric CO2 variability is influenced by changes in ecosystem structure as well as the rate of production.
 Differentiation between SVI- and SST-associated controls is limited because of their covariance. Correlations found do not prove causality. Furthermore, the potential for systematic error in either the NPP or sediment trap flux data may influence the accuracy of predictions. Greater confidence in flux to depth forecasts may require consideration of additional NPP model estimates [Campbell et al., 2002], flux to depth techniques [Buesseler et al., 2000], or other marine carbon cycle methodologies [e.g., Aydin et al., 2004; Bishop et al., 2004]. This study is an initial step toward characterizing the efficiency of the biological pump.
 We thank the following researchers for their inspiration and advice: James Bishop (Laurence Berkeley National Laboratory), Kevin Arrigo, Gert van Dijken (for advice on satellite image processing), Pamela Matson, Alexandria Boehm, Alessandro Tagliabue, Rochelle Labiosa, and Tasha Reddy (Stanford University). This research was supported by a number of sources, including the NSF ROAVERRS program, Lawrence Livermore National Laboratory DOE Center for Research on Ocean Carbon Sequestration, Stanford University McGee and A. W. Mellon Foundations, and International JGOFS Program (Ocean Biogeochemical Modeling Course, Bangalore, India). Finally, we thank the J. Geophys. Res. reviewers for their recommendations.