Primary production sensitivity to phytoplankton light attenuation parameter increases with transient forcing

Abstract. Treatment of the underwater light field in ocean biogeochemical models has been attracting increasing interest, with some models moving towards more complex parameterisations. We conduct a simple sensitivity study of a typical, highly simplified parameterisation. In our study, we vary the phytoplankton light attenuation parameter over a range constrained by data during both pre-industrial equilibrated and future climate scenario RCP8.5. In equilibrium, lower light attenuation parameters (weaker self-shading) shift net primary production (NPP) towards the high latitudes, while higher values of light attenuation (stronger shelf-shading) shift NPP towards the low latitudes. Climate forcing magnifies this relationship through changes in the distribution of nutrients both within and between ocean regions. Where and how NPP responds to climate forcing can determine the magnitude and sign of global NPP trends in this high CO2 future scenario. Ocean oxygen is particularly sensitive to parameter choice. Under higher CO2 concentrations, two simulations establish a strong biogeochemical feedback between the Southern Ocean and low-latitude Pacific that highlights the potential for regional teleconnection. Our simulations serve as a reminder that shifts in fundamental properties (e.g. light attenuation by phytoplankton) over deep time have the potential to alter global biogeochemistry.

covarying factors, with the current default model value applied to both diazotrophs and the single general phytoplankton type (Eqn. 1). The Schmittner et al. (2008) k c value of 0.03 (m mmol N m −3 ) −1 is probably derived from Fasham et al. (1990), but was increased in Keller et al. (2012) to 0.047 (m mmol N m −3 ) −1 . Light attenuation parameters are measured based on chlorophyll (commonly Chlorophyll a) concentration but the model uses nitrogen units, necessitating the application of an assumed conversion factor also implicit to k c . Early tests of k c at steady-state (e.g., Fasham et al., 1990) demonstrated low model 5 biomass sensitivity to parameter value choice, and this has been the prevailing wisdom of biogeochemical modellers for over 20 years. Replacing the UVic ESCM default value with a different one might result in a k c of 0.014 m 2 (mg Chl a) −1 (generally applicable, Lorenzen, 1972), 0.041 m 2 (mg Chl a) −1 (Southern Ocean, Bracher and Tilzer, 2001), or a range anywhere from 0.006-0.015 m 2 (mg Chl a) −1 assuming all phytoplankton represent mixes of specific species of dinoflagellates, calcifiers, or diatoms (Falkowski et al., 1985). However, even the simple assumption that k c varies with chlorophyll concentration can be 10 considered highly simplistic (Siegel et al., 2005). In practice, any value assigned to k c is going to be highly model-dependent (e.g., 0.058 m 2 (mg Chl a) −1 in Wang et al., 2008). Conversion of these k c values to nitrogen units using a recent overview of the data (Dutkiewicz et al., 2015) yields a range of 0.008-0.054 (m mmol N m −3 ) −1 (though higher values in models exist-Evans and Parslow 1985 used a value of 0.12 (m mmol N m −3 ) −1 ). For our test, we employ eight separate simulations using k c = 0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, and 0.08 (m mmol N m −3 ) −1 . In the following analysis, they will be referred to 15 as 'K1-8', respectively.

Pre-industrial equilibrium
Equilibrium primary production in the UVic ESCM shows modest sensitivity to the range of k c values chosen. Figure 1 provides zonally averaged surface chlorophyll, calculated from model nitrogen units using a conversion factor of 1.59 following 20 Schmittner et al. (2005). SeaWiFS satellite chlorophyll data from 1997-2007 (NASA Goddard Space Flight Center, Ocean Biology Processing Group), regridded to the model grid, are also included. All simulations overestimate chlorophyll with respect to observations in the tropics and the southern hemisphere middle latitudes, and underestimate chlorophyll at high latitudes.
Model-observation RMSE reveals best agreement between SeaWiFS and K1 chlorophyll (RMSE 0.837) and worsening agreement with increasing k c (Fig. 1). The simulation spread is slightest between 20 and 40°, where phytoplankton biomass is low.

25
The Southern Ocean and tropics are the two regions where chlorophyll concentrations are most sensitive to k c value (Figs. 1 and 2). In the Southern Ocean, K1 produces zonally averaged biomass concentrations more than 5 times larger than K8 because phytoplankton in K1 do not self-shade as strongly during the Austral summer, thereby allowing for a stronger seasonal cycle.
In the stratified tropics, the effect is opposite in that K8 yields zonally averaged biomass concentrations of up to double K1 because stronger self-shading inhibits deeper photosynthesis (Fig. 1), making more nutrients available at the surface (  Which k c value performs the "best" with respect to biogeochemical observations is not quantified here, but generally K4 and above perform better in the deep ocean than K1-K3. Observations included in Figure   otrophic conditions in the low latitudes and the northern hemisphere, where these simulations had relatively lower pre-industrial near-surface nutrient concentrations owing to deeper primary production. Relatively lower starting concentrations causes the biology in these simulations to be relatively more sensitive to an increase in stratification. Figure 5 repeats Figure 1 for years 2100 and 2300 and shows a decline in biomass in both of these regions for these simulations. These declines are not offset by increasing biomass in the Southern Ocean, which is driven by regional increasing temperature, wind-driven overturning, 30 and nutrient remineralisation (Kvale et al., 2015). The pre-2100 increase in global NPP in K8 is due primarily to increasing biomass north of 60°N and modest increases in biomass in the low latitudes (Fig. 5). Biomass in K7 and K8 is relatively less sensitive to increasing stratification because their high k c values limit primary production to the very near-surface, thereby After about year 2100, physical limitation of nutrients becomes a less important driver of changes in global NPP than temperature-enhanced biological processes (Kvale et al., 2015). Increasing global NPP in all simulations is dominated by increasing biomass in the Southern Ocean, though biomass also increases modestly in the low latitudes for K3-K8 (Fig. 5).

5
The drivers of change in NPP in the Southern Ocean are the same as those mentioned earlier, with alleviation of light and temperature limitations increasing production rates. Declines in northern hemisphere NPP do not offset the Southern Ocean increases, however the three simulations with the weakest self-shading (K1-K3) also show declining rates of global NPP increase due in part to declines in the northern hemisphere NPP.
Model spread in biomass and NPP response generally increases with radiative forcing. Change in global NPP differs by about 10 2.5 Pg C y −1 by 2100 (more than 100% of the total change in NPP at 2100 for all simulations) and k c parameter choice can determine the sign of the change. This is true even if only considering the k c parameters offering the better fits to pre-industrial nutrient and carbon observations (K4-K8). By 2300 this spread has increased to about 7 Pg C y −1 across all simulations, and 4 Pg C y −1 between K4 and K8. Differences in simulated response in Southern Ocean chlorophyll are the most remarkable, with K1 having an average chlorophyll concentration at 60°S of over 2 mg m −3 higher than K8 at 2100 (a concentration more 15 than four times higher than K8) that has increased to a difference of almost 2.5 mg m −3 higher by 2300. better reflect trends at hundred-year timescales (Fig. 7). For all biogeochemical quantities, simulated spread at the surface increases with time. K1 has generally lower pre-industrial NPP than K8 that also increases less over the transient simulation, and K8 Southern Ocean oxygen at 300 meters is up to 100 mmol m −3 by 2100, but by 2300 that difference has grown to a maximum of 160 mmol m −3 . All simulations experience a loss in oxygen due to warming and increasing remineralisation, but 30 K1 and K2 additionally experience denitrification in the Southern Ocean (not shown) as a result of very high primary production in the region and already lower oxygen concentrations at steady-state. This denitrification establishes a nutrient feedback with the low latitude Pacific that reduces Southern Ocean oxygen further (Fig. 8), thus producing a strong regional decline in oxygen despite K1 and K2 showing weaker global NPP trends than the other simulations. The feedback starts with increased stratification in the low-latitude Pacific, which limits nitrate availability for local primary production. As a result, more phos- phate begins to advect into the Southern Ocean, where it fertilizes phytoplankton growth. Warming seawater increases both primary production and remineralisation rates. Phytoplankton in K1 and K2 are only weakly inhibited by self-shading, and the resulting large increases in primary production leads to the consumption of enough oxygen that denitrification establishes in the Southern Ocean. Denitrification reduces the flow of nitrate in intermediate water back into the low latitude Pacific, which becomes even more nitrate-limited.

4 Discussion
Our results show choice of k c value for our model matters little for primary production in equilibrium tests within a range above 0.04 (m mmol N m −3 ) −1 . Primary production sensitivity increases with lower k c values, with reduced shading sensitivity allowing for a stronger seasonal cycle in the high latitudes, producing higher carbon and nutrient export. Equilibrium deep ocean oxygen is particularly sensitive to the application of k c values between 0.01 and 0.03 (m mmol N m −3 ) −1 .

10
Model spread increases in our transient simulations, and k c parameter choice can determine the sign as well as the magnitude of the global NPP response. Substantial differences in model behaviour occur even within the k c range shown insensitive in equilibrium tests, and within the range performing best with respect to pre-industrial biogeochemical observations. That this is true is a general reminder of the potential omission of important tuning information when focusing only on steady-state models. These differences in model behaviour have biogeochemical consequences below the surface, with oxygen again showing 15 particular sensitivity to parameter choice. A nutrient exchange feedback establishes in the two lowest k c value simulations, substantially reducing Southern Ocean oxygen concentrations. While these two simulations performed less favourably with respect to gridded nutrient observations in a pre-industrial comparison and might therefore be considered less reliable representations of the modern ocean, they performed best against SeaWiFS chlorophyll data, which might point to needed additional export or remineralisation parameter adjustments to tune the deep ocean. It is worth noting this feedback occurs because: 1) it 20 highlights the potential for strong biogeochemical teleconnection between the Southern Ocean and the low latitude Pacific, and 2) both light attenuation characteristics of dominant phytoplankton and ocean oxygen volatility have changed over geologic timescales (e.g., Katz et al., 2004;Lenton et al., 2014). A recent model study by Meyer et al. (2016) explored the sensitivity of oxygen to e-folding depth of remineralisation and total phosphate inventory and hypothesized an increase in remineralisation depth has occurred over the Phanerozoic alongside a stabilisation of ocean oxygen inventory. Our tests demonstrate another 25 potential mechanism (evolutionary increase in light attenuation characteristics by dominant phytoplankton) for the increase in ocean oxygen inventory in steady-state conditions as well as stabilisation of oxygen under rapid climate change.

Conclusions
The typical, highly simplistic parameterisations of underwater light availability used in climate and ocean models to calculate primary production and associated chemistry (alkalinity, DIC, nitrate, phosphate, and oxygen) contain substantial sensitivity 30 to light attenuation parameter. This applies both in steady-state and when using forced biogeochemistry for the range of val-7 Biogeosciences Discuss., doi:10.5194/bg-2017-118, 2017 Manuscript under review for journal Biogeosciences Discussion started: 13 April 2017 c Author(s) 2017. CC-BY 3.0 License. ues tested here. This sensitivity can grow with climate forcing as complex biogeochemical feedbacks develop, with primary production and ocean oxygen being especially susceptible to parameter choice. Our study highlights the need to assess biogeochemical models under transient as well as equilibrium conditions. In addition, the biogeochemical feedback we describe in two of our transient tests also serves as a reminder that even seemingly small events, like the emergence of shell-secreting phytoplankton, could have potentially large biogeochemical consequences just by altering the light field.