Volume 123, Issue 5 p. 1447-1459
Research Article
Free Access

A New Approach to Estimating Coccolithophore Calcification Rates From Space

Jason Hopkins

Corresponding Author

Jason Hopkins

Bigelow Laboratory for Ocean Sciences, East Boothbay, ME, USA

Correspondence to: J. Hopkins,

[email protected]

Search for more papers by this author
William M. Balch

William M. Balch

Bigelow Laboratory for Ocean Sciences, East Boothbay, ME, USA

Search for more papers by this author
First published: 26 March 2018
Citations: 16

Abstract

The production and ultimate fate of calcium carbonate in the global ocean has implications for the efficiency of the biological carbon and alkalinity pumps. Historically, sediment trap flux data and/or mass balance equations have been used to estimate the rate of particulate inorganic carbon production in the ocean. More recently, satellite data have been used to provide a more comprehensive global overview of this important biogeochemical process based on relationships determined from multilinear regression of measured calcification rates against a number of measurable variables. Here we describe a simple model to estimate calcification rate based around elements of coccolithophore physiology that can be easily parametrized with satellite ocean color data. The model output conforms to our understanding of the spatial and temporal distribution of coccolithophores and performs relatively well at reproducing global rates that are of the correct order of magnitude, while capturing the variability in such a complex, natural process when compared to field calcification rate measurements (slope = 0.98; R2 = 0.28; p < 0.05; RMSE = 0.53 mg C · m−3 · day−1). Average, global, euphotic zone depth-integrated calcification rate is estimated to be 1.42 ± 1.69 Pg particulate inorganic carbon/year with the oceanic gyres contributing the greatest influence.

Key Points

  • Estimates of global calcification rate are essential for understanding the efficiency of both the alkalinity and biological carbon pumps
  • A simple model parameterized with remotely sensed data can be used to estimate global calcification rates
  • Average, global, coccolithophore calcification rate is estimated to be 1.42 Pg C/year

1 Introduction

There are two fundamental pump paradigms in biological oceanography. The biological carbon pump (BCP) is concerned with the photosynthetic fixation of inorganic carbon into organic carbon and its eventual fate—remineralization or export from the upper to the deep ocean. This latter process is particularly significant as it results in a net drawdown of atmospheric CO2 (Rost & Riebesell, 2004). The other is the carbonate counter pump, also referred to as the alkalinity pump (AP) (Heinze et al., 1991). This system is associated with the production and export of calcium carbonate (CaCO3), often referred to as calcite or particulate inorganic carbon (PIC). In contrast to the BCP, this process yields CO2 (equation 1) and has the potential to alter the air-sea flux of CO2 (Harlay et al., 2010; Shutler et al., 2013).
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0001(1)

The impact of any CO2 production during calcification may be offset somewhat by an increase in the efficiency of the BCP through the ballasting of organic carbon by biomineral particles. This process has been suggested to increase the flux of organic carbon from the upper ocean (Klaas & Archer, 2002; Sanders et al., 2010) although the significance may vary regionally (Le Moigne et al., 2014). The impact of ballasting has been questioned, however, with work suggesting that the export of organic carbon may actually control mineralogical fluxes (Passow & De La Rocha, 2006) or, alternatively, that ecosystem structure may play a greater role in regulating the efficiency of the BCP than ballasting (Henson et al., 2012). Given the apparent conflicting evidence for the significance of the equilibrium between the AP and BCP, the ability to estimate spatial and temporal variability in calcite production both globally and regionally will provide further insights into these fundamental pump paradigms and presents opportunities to better model the oceanic carbon cycle.

Coccolithophores—phytoplankton that produce an outer covering of intricate calcite plates (coccoliths) (Pienaar, 1994)—are considered to be the dominant oceanic calcite producers on the planet (Brownlee & Taylor, 2004). Of the approximately 200 extant species of coccolithophore (Probert & Houdan, 2004), Emiliania huxleyi is considered to be the most cosmopolitan and is recognized as forming extensive blooms (Holligan et al., 1993) across the majority of the world's oceans (Tyrrell & Merico, 2004). While the exact function of coccoliths remains unresolved, they contribute to the largest geological sink for carbon (Monteiro et al., 2016). The optical properties of E. huxleyi coccoliths (Balch et al., 1996) are such that they have enabled an algorithm to be developed that provides estimates of PIC concentration in the surface ocean from space (Balch et al., 2005; Gordon et al., 2001). Currently, long-term, synoptic satellite data records provide one of the best opportunities to better understand calcification and the carbonate counter pump on a global scale.

Prior to the advent of robust satellite PIC estimates, various methods were used in efforts to estimate global pelagic calcite production (Table 1). Using sediment trap data, accumulation rates, and average production rates, Milliman (1993) estimated world ocean annual calcium carbonate production to be approximately 5 billion tons (0.6 GT PIC/year). This estimate consists of calcification from various sources including reefs, banks, and enclosed seas. Calcification by coccolithophores and foraminifera—assumed to be the principal sources of open ocean calcium carbonate—account for approximately 0.26 GT C/year. This estimate of annual calcium carbonate production is comparable to the 0.7 GT C/year derived from inventories of alkalinity and water mass residence times (Milliman et al., 1999). In a similar approach, the PIC production estimate of Wollast (1994) of 1.0 GT C/year was derived by incorporating mean surface seawater carbonate chemistry conditions into the oceanic carbonate cycle. As part of an effort to determine global net primary production, Lee (2001) estimated global PIC production to be 1.1 GT C/year using a method that estimates CaCO3 concentration from salinity normalized total alkalinity. Moore, Doney, Glover, and Fung (2002) used a marine ecosystem model in which calcification by phytoplankton was parameterized as a variable fraction of primary production (Moore, Doney, Kleypas, et al., 2002) to come up with a PIC production estimate of 1.1 GT C/year. A summary of literature values for CaCO3 production (0.5–2.0 GT C/year) is included in a study of the ocean's carbonate budget (Iglesias-Rodríguez, Armstrong, et al., 2002). The advent of readily available, robust satellite-derived data provides an alternative way to estimate global PIC production from space. Using an empirical relationship derived from least squares multiple linear regression of measured calcification rate against several environmental parameters that could be simply predicted or measured from space, Balch et al. (2007) estimated annual global calcification rate to be 1.6 GT C/year, within the range of values reported by Iglesias-Rodríguez, Brown, et al. (2002).

Table 1. Examples of Previous Estimates of Global Calcification Rate
Author Technique Global CaCO3 production (GT C/year)
Balch et al. (2007) 14C measurements and remote sensing algorithm 1.6 ± 0.3
Moore, Doney, Glover, and Fung (2002) Global ecosystem model 1.1
Lee (2001) Salinity normalized total alkalinity 1.1 ± 0.3
Milliman et al. (1999) Alkalinity and residence times 0.7
Milliman and Droxler (1996) Carbonate mass balance; planktonic and benthic only 0.82
Wollast (1994) Carbonate system and mean seawater composition 1.0
Milliman (1993) Accumulation rates and sediment trap data 0.6
Morse and Mackenzie (1990) Geochemistry of carbonate sediments 1.0
  • Note. Adapted from Balch et al. (2007).

The use of remotely sensed data provides an effective means to estimate surface ocean processes over time and on a global scale. The purely empirical algorithm of Balch et al. (2007) provided some degree of justification that such estimates are possible; however, their calcification algorithm was applied to only 1 year of satellite data. With the increased volume of satellite-derived data now available, we revisit the global calcification rate problem and take advantage of data derived over the lifetime of the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor on board the AQUA platform (2002 to present). In addition, we are able to exploit a more extensive database of field calcification measurements than that available to Balch et al. (2005) although in contrast to their approach, we use these data to validate, rather than derive, our algorithm.

We choose to move away from a purely empirical approach to one that might be considered to be semi-analytical and adopt a framework similar to that used in the determination of net primary production using the carbon-based productivity model (CbPM) (Behrenfeld et al., 2005). The basis for this model is that net primary production (NPP) can simply be derived from carbon concentration (C), growth rate (μ), euphotic depth (Zeu), and an irradiance function (I0):
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0002(2)

While more complex biogeochemical models (although with their own built-in generalizations) have previously been used to estimate calcification rate (e.g., Moore, Doney, Kleypas, et al., 2002), we use the CbPM framework to propose a simplified calcification rate model based on elements of coccolithophore physiology. The aim of this study is to characterize this complex process as a “manageable model” (Levins, 1966), using readily accessible satellite data to improve our understanding of the global calcite cycle, the BCP, the AP and the transport of carbon within the global ocean.

2 Materials and Methods

The approach that we adopt is to take the framework of the CbPM (equation 2) and modify it from a model that estimates primary production to one that estimates PIC production. In our revised model calcification rate by coccolithophores is proposed to be simply a function of PIC concentration, growth rate, irradiance, and depth:
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0003(3)
We make the following general assumptions in our model:
  1. PIC production is proportional to coccolithophore growth rate.
  2. Coccolithophore growth rate is a function of temperature and irradiance without an explicit nutrient term (see below).
  3. E. huxleyi calcite production is the major contributor to satellite-derived PIC concentration.
  4. Calcification rate decreases as a function of light availability through the water column.
This approach allows us to generalize the surface calcification rate process as follows:
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0004(4)
and the depth-integrated calcification rate process as follows:
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0005(5)
where μ is a temperature derived growth rate, h(I0) is a growth limiting irradiance function, and g(Zeu) is a depth dependency function.

2.1 Growth Rate

In the CbPM, growth rate is derived from a maximum growth rate moderated for the influence of nutrient and temperature stress and light limitation (Behrenfeld et al., 2005). We take an empirical approach to our determination of growth rate that allows us to include any temperature or irradiance effect on coccolithophore growth. No nutrient term was included for predicting calcification based on several lines of evidence. Previous observations of several ecologically important coccolithophore species have suggested that calcification rates are substrate limited by bicarbonate, not nutrients (temperature and light intensity are known to change calcification rates, whereas much less is known about the effect of nutrients) (Bach et al., 2015). Additionally, coccolithophores channel the most carbon into coccoliths in high-light, stratified, low-nutrient surface waters (Balch, Poulton, et al., 2011) and those growing in nutrient-limited conditions have been found to calcify more than exponentially growing cells in nutrient replete medium (Paasche, 1998; Paasche & Brubak, 1994).

Our coccolithophore-specific growth rate function follows the parameterization of Moisan et al. (2002), which also does not include a nutrient term. We modify their general growth equation, determined from model output of 100 phytoplankton groups (equation 9; Moisan et al., 2002), to best fit experimental E.huxleyi temperature—growth data from Paasche (2002) and Rosas-Navarro et al. (2016). This relationship, which follows the same form as that in Moisan et al. (2002), can be written as
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0006(6)

We have chosen Topt to be 20°C and have selected values for urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0007 and urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0008 that produce a curve that best fits experimental data (Figure 1). Observational data from the Southern Hemisphere suggest that E. huxleyi are absent in waters with a temperature of less than 2 °C (Holligan et al., 2010), and we have therefore chosen this temperature as the lower limit for growth.

Details are in the caption following the image
Coccolithophore temperature-growth rate relationship. Blue circles represent data from Paasche (2002). Red circles are data from Rosas-Navarro et al. (2016). Green line represents maximum growth rate determined from Eppley (1972). Black line is best fit to experimental data (see equation 6).

2.2 Irradiance Function

Light availability will impact coccolithophore growth rates (Paasche, 2002). We account for this with the inclusion of a generalized irradiance term based on the function reported in MacIntyre et al. (2002):
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0009(7)
where h(I0) is our irradiance function, E is satellite-derived photosynthetically available radiation (PAR) and KE is the average saturation parameter for E. huxleyi reported in MacIntyre et al. (2002).

2.3 Depth and Light Dependency

We integrate surface calcification rate data to euphotic depth (Zeu) using a Beer's law-type relationship. Here we assume that calcification rate is primarily a light-dependent process (Paasche, 2002), and thus, surface calcification rate will decay with depth at a rate similar to 490 nm light. The decay profile of calcification rate is therefore modeled as follows:
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0010(8)
where CRz is calcification rate at depth z, CR0 is surface calcification rate, and k is the attenuation coefficient at 490 nm (Kd 490). We use Kd 490 on the basis that calcification is influenced by light in the blue wavelengths (Paasche, 1966) due to its primary pigment being chlorophyll a and accessory pigment being 19′-hexanoyloxyfucoxanthin (Haxo, 1985).
We determine euphotic zone depth (Zeu) using satellite-derived PAR and assume that euphotic zone depth is the depth at which light availability diminishes to 1% of the surface light:
urn:x-wiley:21698953:media:jgrg21020:jgrg21020-math-0011(9)

2.4 Satellite Data

All satellite data used in this analysis were downloaded from the NASA Ocean Color website (https://oceancolor.gsfc.nasa.gov). We use 9 km resolution, level 3, monthly climatologies from the Aqua MODIS sensor (R2014.0 reprocessing). Specifically, we use the following products: (a) sea surface temperature (11 μm daytime, SST), used in the determination of growth rate; (b) PIC, used as the carbon variable in determining calcification rate; (c) PAR, used in the growth limiting irradiance function; and (d) diffuse attenuation coefficient at 490 nm (Kd 490), used to determine euphotic depth and depth-integrated calcification rate. Seasonal climatologies were derived from monthly composite model output (winter: December, January, and February; spring: March, April, and May; summer: June, July, and August; autumn: September, October, and November).

2.5 Model Validation

We use an extensive database of in situ calcification rate measurements to test how well our model performs. These data consist of a mix of surface and depth profile measurements of calcification determined using the 14C microdiffusion technique (Balch et al., 2000) taken during a number of global research cruises from 2002 onward. For this validation exercise, we use 8-day satellite data that match, as closely as possible, the date the field sample was measured as inputs into our model. These data provide a trade-off between being close enough in time to the in situ measurement while minimizing data loss caused by cloud or overpass gaps. We assess the performance of our model through linear regression of modeled and measured estimates of calcification rate.

3 Results

The PIC algorithm is generally considered to be a Case I algorithm (Balch et al., 2005; Morel & Prieur, 1977). The optical properties of Case I waters are correlated with phytoplankton and their associated by-products, whereas in Case II waters they can be influenced by other constituents, such as suspended sediments. We have therefore chosen to exclude satellite-derived data obtained from water column depths of less than 200 m and focus our interpretation of the output from our model to the open ocean (i.e., Case I waters only).

3.1 Surface Calcification Rates

Seasonal, surface calcification rate data are presented in Figure 2. Areas with no remotely sensed data, water column depths <200 m, or regions, where SST is <2°C (our lower temperature limit for coccolithophore growth) are shown in white. In boreal spring (Figure 2a) calcification rate is high (> 5 mg C · m−3 · day−1), relative to the global boreal spring average, in a small region of the subtropical North Atlantic and subtropical North Pacific. There is also evidence of a band of higher than average calcification rates skirting the southern edge of the Indian Ocean. The Pacific and Atlantic gyre regions in both hemispheres are characterized by patches where calcification rates range between 0.1 and 5 mg C · m−3 · day−1. Regions within the high-latitude North Pacific (> ~ 45°N), high-latitude North Atlantic (> ~60°N), high-latitude Southern Ocean (>~60°S), the Indian Ocean, and Equatorial West Pacific have the lowest boreal spring calcification rates (generally <0.1 mg C · m−3 · day−1).

Details are in the caption following the image
Modeled seasonal surface calcification rate (mg C · m−3 · day−1). (a) Northern Hemisphere spring (Southern Hemisphere autumn); (b) Northern Hemisphere summer (Southern Hemisphere winter); (c) Northern Hemisphere autumn (Southern Hemisphere spring); (d) Northern Hemisphere winter (Southern Hemisphere summer). Areas in white represent pixels with no data, water column depths <200 m, or waters <2°C.

There is reduced data coverage in the high-latitude Southern Ocean in the boreal summer composite (Figure 2b) compared with boreal spring. The data available in this region indicate that the region of relatively low calcification rate (< 0.1 mg C · m−3 · day−1), identified in the boreal spring composite has advanced northward, forming a band that now encircles the Southern Ocean at approximately 50°S. The band of high calcification rate previously identified in this area has disappeared with the transition from austral autumn into winter. The region of relatively low calcification rate in the boreal spring high-latitude North Pacific has now been replaced by an area of relatively high (>5 mg C · m−3 · day−1) calcification rates that also appears in the North Atlantic (> ~40°N). In general, calcification rates in the Northern Hemisphere appear to be greater than those in the Southern Hemisphere, which is to be expected for this time of year (Hopkins et al., 2015).

The transition from boreal summer to autumn results in a decrease in calcification rate in the high-latitude North Atlantic and Pacific, although there is evidence of relatively high (> 5 mg C · m−3 · day−1) rates that persist in the Bering Sea and a band that extends across the North pacific at approximately 45°N (Figure 2c). In the Southern Hemisphere, the shift from austral winter into spring results in patches of increased calcification rate across much of the region. Our results show relatively high calcification rates associated with the area offshore of Chile and Peru and extending out from the coast of Namibia (South Africa). There is also evidence of increasing calcification rates (> 1 mg C · m−3 · day−1) in the region of the southern subtropical convergence (~ 45°S).

The boreal winter (Figure 2d) is characterized by reduced data availability (due to low sun angle) and decreased calcification rates, in the high latitudes of the Northern Hemisphere (>~45°N), when compared to those observed in the autumn composite. In contrast, calcification rates across the Southern Hemisphere have increased. There is evidence of increased calcification (> 5 mg C · m−3 · d−1) along the Patagonian Shelf, the coast of Chile, and Peru, off the coast of Namibia and in a distinct band that all but encircles the sub-Antarctic. Calcification rates in the northern part of the Indian Ocean and Equatorial West Pacific remain consistently low (< 0.1 mg C · m−3 · d−1) across all seasons. Calcification rates are generally higher in the Southern Hemisphere compared to the Northern Hemisphere, which again is consistent with data on coccolithophore bloom timing (Hopkins et al., 2015).

3.2 Depth-Integrated Calcification Rates

For the sake of computational efficiency, we bin all satellite data into 1° × 1° averaged pixels prior to computing depth-integrated calcification rates. Seasonal, depth-integrated calcification rates over the euphotic zone are presented in Figure 3. While the broad temporal-spatial patterns remain (i.e., highest calcification rates in the Northern and Southern Hemisphere summers), the oceanic gyres are associated with elevated calcification rates across all seasons. Depth-integrated calcification rates in the boreal spring composite (Figure 3a) follow a broadly similar spatial pattern to surface calcification rates (Figure 2a) insomuch as the high latitudes (>~ 60°N and >~ 60°S) exhibit significantly lower rates (< 1 mg C · m−2 · d−1) compared to the rest of the globe. In contrast to surface calcification rates, relatively high depth-integrated calcification rates (> ~30 mg C · m−2 · d−1) are observed in the North and South Pacific and Atlantic gyres and the southern part of the Indian Ocean. Outside of these regions, depth-integrated calcification rate ranges between ~5 and 10 mg C · m−2 · d−1.

Details are in the caption following the image
Modeled seasonal depth-integrated calcification rate (mg C · m−2 · day−1). (a) Northern Hemisphere spring (Southern Hemisphere autumn); (b) Northern Hemisphere summer (Southern Hemisphere winter); (c) Northern Hemisphere autumn (Southern Hemisphere spring); (d) Northern Hemisphere winter (Southern Hemisphere summer). Areas in white represent pixels with no data, water column depths <200 m, or waters <2°C.

A similar pattern is observed in the boreal summer depth-integrated calcification rate composite (Figure 3b). The band of low depth-integrated calcification rates observed in the high latitude Southern Ocean has again pushed northward similar to observations in boreal spring surface calcification rate (Figure 2b).

While we again observe increased rates in the Northern Hemisphere compared with the Southern Hemisphere transitioning from spring to summer and relatively high depth-integrated calcification rates in similar locations as high boreal summer surface calcification rates (i.e., the high-latitude North Pacific and Atlantic), there is also a significant increase in depth-integrated calcification rates associated with the Northern Hemisphere gyres when compared to boreal summer surface calcification rates.

The higher depth-integrated calcification rates observed in the Northern Hemisphere gyres diminish in magnitude and extent in boreal autumn (Figure 3c). The Southern Hemisphere gyres and southern Indian Ocean now become regions of relatively high (>~30 mg C · m−2 · d−1) calcification rates although there is still evidence of some high calcification rates in the tropical Atlantic and Eastern North Pacific. There is a clear shift from Northern to Southern Hemisphere dominance of depth-integrated calcification rates during the onset of austral spring. The high depth-integrated calcification rates observed in the Northern Hemisphere gyres are significantly diminished in the boreal winter seasonal composite (Figure 3d). In the Southern Hemisphere, higher depth-integrated rates are observed not only in similar regions as high surface rates (i.e., in a band encircling the sub-Antarctic) but also across the southern oceanic gyres and southern Indian Ocean. The high-latitude Northern Hemisphere (>~45°N) is associated with relatively low (<1 mg C · m−2 · d−1) calcification rates in the boreal winter, similar to the pattern observed in the surface data (Figure 2d).

3.3 Validation of Model Using Measured Calcification Rate Data

By way of validating our model, we use 8-day satellite data composites as inputs to the calcification rate algorithm and compare the output with field calcification rates from a number of global research cruises (Figure 4). As previously mentioned, remotely sensed PIC data from Case II waters can be unduly influenced by other constituents within the water column, reducing confidence in the accuracy of data. We have therefore excluded data collected in the Gulf of Maine as part of the Gulf of Maine North Atlantic Time Series program from our validation process as these waters are reported as being Case II dissolved all the time and Case II particulate approximately half of the time (Balch et al., 2004). We plot measured calcification rate against predicted calcification rate, across all discrete euphotic depths, and assess model performance through linear regression. There is a statistically significant linear correlation (p < 0.05) between our modeled and measured calcification rates that has a slope of 0.98 and a coefficient of determination (R2) of 0.28.

Details are in the caption following the image
Validation of the calcification algorithm. X axis is measured calcification rates (surface and depth). Y axis is modeled calcification rates using 8-day satellite data composites as inputs. Black dashed line is the 1:1 line. Red line is the best fit line (y = 0.98x + 0.08; R2 = 0.28; p < 0.05; SE = 0.09 mg C · m−3 · day−1; DF = 321; F = 0.29; Fcrit = 0.83; root-mean-square error (RMSE) = 0.53 mg C · m−3 · day−1). Error bars indicate one standard deviation of triplicate calcification rate measurements.

4 Discussion

The aim of this study was to develop new global estimates of calcification rate using remotely sensed data. Here we propose a simple algorithm that takes satellite-derived SST, PIC, Kd 490, and PAR data and applies a set of basic physiological assumptions to generate estimates of coccolithophore-driven calcification rate and demonstrate that this complex process can be simply modeled. The ability to resolve calcification rate across time and at global and/or regional scales provides a unique opportunity to improve our understanding of the calcium carbonate cycle and acts as a foundation for further studies into the role that mineral ballasting may play in the export of carbon from the upper ocean (Klaas & Archer, 2002; Sanders et al., 2010), which influences the efficiency of the BCP.

Regression of our model output against measured calcification rates shows that, to a first order, our algorithm provides a statistically significant, albeit modest, correlation between observed and predicted calcification (Figure 4). The gradient of the best fit line is close to 1; however, the modest R2 value highlights the difficulties involved with estimating such highly variable natural processes from space. Our coefficient of determination for the best fit line is slightly better than that calculated for the previous study (R2 = 0.274; Balch et al., 2005); however, it is not as good as that calculated for remotely sensed primary production estimates such as the those determined using the Vertically Generalized Production Model (R2 = 0.58; Behrenfeld & Falkowski, 1997b) for a number of possible reasons. First, there is a distinct paucity of in situ calcification rate measurements compared with measurements of primary productivity, which reduces the dynamic range available to validate a multiseasonal, global model such as this. Indeed, calcification measurements tend to be heavily biased toward the lower end of the rate measurement spectrum. It should be noted that when the same analysis is undertaken using only low concentration in situ data (<0.5 mg · m−3 · day−1), the slope of the fit reduces to 0.24, with an R2 of 0.08. This suggests that, as with most ocean color algorithms, this algorithm performs best when applied globally, across the full dynamic range of properties. If it is only to be applied in oligotrophic regions with low calcification rates, its performance will degrade. Second, the signal-to-noise ratio for calcification measurements made using the 14C microdiffusion technique are significantly lower than those for similar primary productivity measurements. Finally, calcification rate measurements made in vitro using the microdiffusion technique measure bulk calcification within a sample. Our model is based on the physiology of coccolithophores, specifically E. huxleyi. We may therefore be comparing modeled calcification rates based on E. huxleyi data with measured calcification rate data that could include calcification by other species of coccolithophores or calcifying organisms. A recent study has suggested that saturation of the MODIS bands may result in high concentration coccolithophore blooms being masked and thus potentially excluded from this analysis (Land et al., 2017). The identical issue of high PIC concentrations being masked has also been identified by the personnel responsible for PIC algorithm maintenance. This phenomenon was associated with there being no switching between the two-band (used in low-concentration PIC regions) and the three-band algorithm (used in high-PIC concentration areas) as implemented by NASA in SeaDAS version 7.3 (released 23 December 2015). It has been estimated that the three-band algorithm should have been used in <0.05% of the total global pixels (B. C. Bowler, personal communication, 2018). Given that this is a global study and the fact that we have excluded data from Case II waters (typically associated with high reflectance, more likely to trigger the switch to the three-band algorithm), this will not appreciably affect the global analysis presented here. Should this older release for the PIC data be applied only to high-reflectance coastal waters nearshore, caution should be taken in interpreting PIC concentration in high-reflectance waters. This problem will be corrected by NASA in a future reprocessing.

We make a number of broad assumptions in the design of our model. The first is that PIC production is proportional to growth rate in E. huxleyi. In simple terms, the faster coccolithophores grow, the more PIC that is produced. This basic assumption is supported by measurements of temperature manipulated growth rate and cellular PIC production (Figure 5; Rosas-Navarro et al., 2016). As previously mentioned, however, such an assumption may lead to our model overlooking the contributions from other coccolithophore species and calcifiers to global calcification rates, which may indeed account for over half of the global production of calcium carbonate (Milliman, 1993).

Details are in the caption following the image
Increase in particulate inorganic carbon production with temperature-manipulated growth rate. Data from Table 1 of Rosas-Navarro et al. (2016). Red dashed line is the best fit line (y = 10.415x + 0.4522; R2 = 0.878; p < 0.05; SE = 1.294; n = 11).

Our second assumption—that E. huxleyi growth rate can be modeled as a simple function of temperature and irradiance—disregards the influence that nutrient availability may have on coccolithophore physiology. While the impact of nutrient stress is included in the CbPM, we have chosen to omit it on the basis that coccolithophores, and in particular E. huxleyi, are recognized as being relatively tolerant to low-nutrient concentrations (Brand, 1994). It has been reported that E. huxleyi has both the lowest half saturation constant for nitrate and ammonium uptake (Eppley et al., 1969) and the ability to utilize alternative forms of phosphate (Riegman et al., 2000). Under phosphate and nitrate limitation, E. huxleyi still calcifies, however, the PIC:POC ratio increases mainly due to a cessation of division or reduction in protein synthesis (Raven & Crawfurd, 2012), suggesting that nutrient availability may not be a significant limiting factor for calcification in coccolithophores as it is for their organic growth (Monteiro et al., 2016).

The assumption that satellite-derived PIC concentration is related to calcite production by coccolithophores (and not other biocalcifiers like foraminifera) is fundamental in the derivation of the two-band algorithm that is used to estimate the majority of the ocean PIC concentration from space (Balch et al., 2005). This is because the algorithm is specifically designed to include the optical characteristics of E. huxleyi (Balch et al., 2005; Gordon et al., 1988) due in part to the wide-spread nature of this species (it has been found across the majority of the global ocean, Tyrrell & Merico, 2004) and also the favorable scattering cross section for E. huxleyi coccoliths that is orders of magnitude greater than other size classes of calcite particles (Balch et al., 1996). The algorithm, which has been validated against field measurements of discrete and optically derived estimates of PIC (Balch et al., 2016), as mentioned before, performs best in Case I waters meaning that our assumption will likely not hold true in coastal waters where other particulate matter, such as sediments, may contribute to the optical signal of the surface waters.

Our final assumption, that calcification rate decreases with the exponential decrease in light, provides us with the means to simply model calcification rate down to euphotic zone depth and thus estimate integrated PIC production. Calcification is an energetic process (Anning et al., 1996; Monteiro et al., 2016) that is considered to be predominantly light dependent (Paasche, 2002), and we have therefore chosen to model calcification rate through the water column as a simple Beer's law-type decay (although see Tyrrell et al., 1999, for a more comprehensive analysis of the impact coccolithophores have on the vertical light field). There is, however, evidence that calcification may be less light-dependent than photosynthesis (Zondervan, 2007) with increased calcification to primary production ratios measured deep in the water column (Balch, Poulton, et al., 2011; Poulton et al., 2010) and measurable calcification taking place in the dark, likely fueled by respiration (Balch, Holligan, et al., 1992). We tested our assumption by comparing our modeled depth dependent calcification rate with in situ field measurements of calcification rate at depth (Figure 6) and found that generally this relationship predicts depth dependent calcification rate reasonably well.

Details are in the caption following the image
Modeled calcification rate (green line) compared with measured calcification rate (blue dots) at (a) 2.0°S, 110°W in Dec 2004; (b) 3.0°N, 110°W in Dec 2004; (c) 50.4°S, 10.8°W in Feb 2011; and (d) 50.2°S, 56.3°W in Jan 2011. Error bars indicate one standard deviation of triplicate calcification rate measurements.

The spatial and temporal patterns observed in our modeled global surface calcification rates are in line with expectations based on our understanding of the geographical distribution (Balch et al., 2005; Brown & Yoder, 1994; Iglesias-Rodríguez, Brown, et al., 2002) and phenological characteristics (Hopkins et al., 2015) of coccolithophores. For example, the North Atlantic (Holligan et al., 1993), Patagonian Shelf (Poulton et al., 2013), the Bering Sea (Merico et al., 2003), the Iceland Basin (Raitsos et al., 2006), the Barents Sea (Smyth et al., 2004), and the Southern Ocean Great Calcite Belt (Balch, Drapeau, et al., 2011; Balch et al., 2016) are all areas where extensive blooms of coccolithophores have previously been observed. Our model predicts high rates of surface calcification in these regions that coincide with the timings of the peaks in these blooms (Hopkins et al., 2015). This provides some confidence that the design of this model is able to capture the expected spatial and temporal patterns.

When calcification rate is integrated over the depth of the euphotic zone using Beer's law with an exponential decay model, relatively high calcification rates are consistently observed in the oceanic gyres (Figure 3). This may appear to be counterintuitive to some as these regions are often considered to be areas of low productivity (Ryther, 1969). There is evidence however that coccolithophores are abundant in these oligotrophic environments (Poulton et al., 2017) and can continue to calcify in the deep euphotic layers that are associated with these regions (Beaufort et al., 2008).

We use these data to estimate average global, annual PIC production to be 1.42 ± 1.69 Pg C/year. This is in line with previous estimates (Table 1) but lower than a previous satellite data based method (Balch et al., 2007), which may be in part due to the assumptions we have incorporated into this model (e.g., the masking of data from Case I waters) or more fundamentally, through the use of satellite-derived PIC data, estimated using the merged two-band and three-band algorithms (Balch et al., 2005; Gordon et al., 2001), as the carbon source in our model. The PIC algorithm estimates bulk PIC concentration and does not differentiate between contributions from coccolithophores, shed coccoliths, or other calcite particles. Our model associates PIC production with coccolithophore growth, however, shed coccoliths may remain in the water column for extended periods of time due to their slow sinking rate (Tyrrell & Merico, 2004), even when the coccolithophore population is diminishing. We therefore may be overestimating calcification rate in some regions (i.e., where coccolith concentration is high but coccolithophore concentration is low). Despite these limitations, and the assumptions incorporated into this model, it manages to capture the magnitude and range of calcification rates observed in measured calcification rates, for example, measured PIC production in the Iceland Basin in July–August 2007 ranged between 0.12 and 3 mg C · m−3 · day−1 (Poulton et al., 2010). Our model estimates mean calcification rate in the climatological July and August in this region to range between 3.2 and 6.8 mg C · m−3 · day−1.

Previous estimates of global calcification rates using remotely sensed data (Balch et al., 2007) were developed from an algorithm based on relationships identified from least squares empirical multiple linear regression of measured calcification rate against temperature, depth, day length, PIC concentration, and chlorophyll concentration. In this analysis we have chosen to move away from a purely empirical approach to a more physiologically driven algorithm, similar to the framework used in a productivity model (Behrenfeld et al., 2005). The results of our validation exercise with field measurements of surface and water column calcification rates suggest that the basis for this approach is reasonable. Measuring total phytoplankton productivity in the ocean from space is associated with notoriously large errors, with values predicted within a factor of 2 (Balch, Evans, et al., 1992; Behrenfeld & Falkowski, 1997a; Campbell et al., 2002). Estimating calcification from space would be expected to have even larger errors given that the signal to noise can typically be as low as 5% of that for primary productivity. However, our model shows that it is possible to get a realistic first-order estimate of mean calcification rate in the global open ocean using this physiological-based algorithm. This approach is an important first step for modeling the influence that coccolithophore calcification may play in the BCP and AP, and it provides a means of estimating calcification rates on both global and regional scales. In addition, the longevity (and continuity) of ocean color measurements from space provides the opportunity to explore long-term changes on calcite production by coccolithophores in the global ocean.

Acknowledgments

We gratefully acknowledge the comments of an anonymous reviewer and J. Shutler that have helped improve this paper. The authors wish to thank the NASA Ocean Biology Processing Group for the production and access to the ocean color data used in this analysis (http://oceancolor.gsfc.nasa.gov). We would also like to thank A. J. Poulton and C. J. Daniels for discussions and input on this paper. J. H. was supported by NASA grant NNX14AL92G to W. M. B. W. M. B. was supported by NASA grants NNX14AM77G, NNX14AQ43A, NNX14AL92G, NNX17AI77G, and NNX14AQ41G. Satellite data used in this paper are available from https://oceancolor.gsfc.nasa.gov. In situ calcification rate data are available from https://seabass.gsfc.nasa.gov.