Modelling Patch Dynamics During Ocean Fertilisation Andrew Crawford Submitted in partial fulfilment of the requirements for the Bachelor of Engineering (Environmental) degree with Honours at the University of Western Australia School of Water Research University of Western Australia November 2003
83
Embed
Modelling Patch Dynamics During Ocean Fertilisation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Modelling Patch Dynamics During Ocean Fertilisation
Andrew Crawford Submitted in partial fulfilment of the requirements for the Bachelor of Engineering (Environmental) degree with Honours at the University of Western Australia School of Water Research University of Western Australia November 2003
iii
Abstract In light of global warming, the efficiency of the biological pump has become an increasingly
interesting process in the sequestration of carbon from the atmosphere. Iron enrichment of
high nitrate, low chlorophyll oceanic regions have been shown capable of stimulating growth
and increasing the efficiency of the biological pump in these regions. This study uses a model
incorporating nutrient-limited phytoplankton growth, lateral diffusion and export by sinking
of aggregates to investigate the factors which enable maximum export of carbon from the
mixed layer. Export per unit area and total export increases with increasing maximum growth
rate, fertilising concentration and the length scale of the patch. Growth rate was found to be a
major determinant of the time scales involved in export, including the time at which nutrients
become depleted, and the time to first export. Increasing growth and fertilising concentration
increase maximum export flux, and multiple export peaks may occur when either growth or
fertilising concentration are sufficiently large to allow the critical concentration for export to
be reached before diffusion can dissipate concentrations. Successively fertilising the same
patch demonstrates that export increases as the inter-fertilising period increases. The time
series of export is complex and does not suggest a stable pattern of export in runs conducted
for 18 days. The scale of randomness in the initial phytoplankton distribution has little effect
on the export per unit area generated, however, introducing small scale subpatches within the
fertilising patch significantly increases export by allowing diffusion into areas vacated by
export. Knowing these characteristics in which export is maximized, and the parameters
which characterise the export behaviour, allows for the more efficient implementation and
management of iron enrichment to mitigate global climate change.
iv
Acknowledgements I’d like to thank my supervisors, Anya Waite, Greg Ivey and David Johnson for all their
guidance and support throughout the year.
Anya has aided me all the way through this project, not only in an academic sense but also by
giving me just enough space to go at my own pace, but keeping me on track when I asked for
it. Most importantly though, she has given me the encouragement needed to move forward
whenever my confidence was fading.
David has provided invaluable help in understanding the model and has been great for
bouncing some ideas off. His enthusiasm for the project was fantastic.
Greg has given fluid mechanics know-how and constructive criticism when presented with
‘rough and ready’ material.
Also a host of family and friends for helping me through the year. Mum, Dad, Lisa and Bec;
The Quaternity – Abi, Chris and Emma; All who have ever played soccer with me on Sunday;
and the final year engineering crew who have all suffered together with me.
Also others; some over-enthusiastic third years; Michelle Carey for running a club for which I
was meant to be president; and Luke Brown for quick and cheerful computer support.
v
Table of Contents
Table of Contents v List of Figures vi List of Tables viii Chapter 1 Introduction 1
1.1 Background 1 1.2 Direction of this study 2
Chapter 2 Literature Review 3 2.1 Climate change and the role of the oceans 3 2.2 In-situ iron fertilisation experiments 6
2.2.1 IronEx I 7 2.2.2 IronEx II 9 2.2.3 SOIREE 11
2.3 Modelling an iron-stimulated biological pump 13 2.3.1 Models of phytoplankton dynamics 13 2.3.2 Models of carbon export 14
Chapter 3 Methodology 21 3.1 Model Theory 21 3.2 Application of model theory 22 3.3 Investigation of parameters 23
3.3.1 Basic parameter set 24 3.3.2 Variation of growth and fertilising concentration 24 3.3.3 Variation of spatial distribution 25 3.3.4 Multiple fertilisation events 26
Chapter 4 Results 28 4.1 Variations of growth and fertilising concentration parameters 28
4.1.1 Time scales of export 29 4.1.2 Maximum growth rate 30 4.1.3 Fertilising concentration 31 4.1.4 Variation of the initial phytoplankton distribution length scale of randomness 33 4.1.5 Variation of the fertilising concentration length scale 36 4.1.6 Multiple fertilisation events 40 4.1.7 Multiple fertilisation export time series 41
Chapter 5 Discussion 48 5.1 Interpretation of results 48
5.1.1 Variations of maximum growth rate and fertilising concentration 48 5.1.2 Time scales of export 50 5.1.3 Spatial distributions 51 5.1.4 Multiple fertilisation events 52 5.1.5 Summary of interpretations 54
5.2 Practical implications 55 5.3 Issues concerning iron fertilisation for the mitigation of climate change 59 5.4 Recommendations 60
Figure 2.2 – Time series of mean export of a nutrient-limited growth-diffusion export model (from Waite and Johnson, 2003) 19
Figure 2.3 – Export simulated from a growth-diffusion-export model and its relationship with patch size and the non-dimensional parameter, Q (from Waite and Johnson, 2003) 19
Figure 3.1 – Example of initial phytoplankton distributions at the four different random length scales input. 26
Figure 3.2 – Nutrient distributions of the three fertilising patch variations used. 26
Figure 4.1 – The effect of variations of maximum growth rate, µmax and fertilising concentration for the three different patch lengths shown. The plots represent export per unit area in the left column, time to nutrient depletion, tN in the central column and tN x µmax in the right column. Derived from 10-run ensemble. 28
Figure 4.2 – Export per unit area per unit concentration of nutrient applied, generated by variations of maximum growth rate and fertilising concentration for three different patch lengths – 10k, 25 km and 50km. 29
Figure 4.3 –Time series of export showing variations with maximum growth rate. 30
Figure 4.4 – The variation of parameters characterising export time series behaviour with maximum growth rate. 31
Figure 4.5 – Time series of export showing variations with fertilising concentration. 32
Figure 4.6 - The variation of parameters characterising export time series behaviour with fertilising concentration. 33.
Figure 4.7 – Export time series for four length scales of randomness in initial phytoplankton distribution. Patch length 25km. 34
Figure 4.8 – Export time series for four length scales of randomness in initial phytoplankton distribution. Patch length 40km. 34
Figure 4.9 - Export time series for four length scales of randomness in initial phytoplankton distribution. Patch length 70km. 34
Figure 4.10 - Export time series for four length scales of randomness in initial phytoplankton distribution. Patch length 85km. 34
Figure 4.11 – Export per unit area generated over 10 days from a single fertilisation for different length scales of randomness of the initial phytoplankton distribution. 35
Figure 4.12 - Export time series for three length scales of fertilising subpatch distribution. Patch length 25km. 36
vii
Figure 4.13 - Export time series for three length scales of fertilising subpatch distribution. Patch length 40km. 36
Figure 4.14 - Export time series for three length scales of fertilising subpatch distribution. Patch length 70km. 37
Figure 4.15 - Export time series for three length scales of fertilising subpatch distribution. Patch length 85km. 37
Figure 4.16 – Export per unit area generated over 10 days from a single fertilisation for different length scales of fertilising subpatch. 37
Figure 4.17 - Maximum export flux occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L. 38
Figure 4.18 - Mean time of export weighted by export flux occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L. 38
Figure 4.19 - Timing of the first export occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L. 39
Figure 4.20 - Timing of the first export occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L. 39
Figure 4.21 – Export per unit area generated by three fertilising distributions – at length scales of L/10 and L/5 – and uniform. Q varied by patch length, L. 41
Figure 4.22 – Export per unit area generated by the uniform fertilising distribution. Q varied by µmax. Note that the range of Q, especially where the downturn occurs, is not shown in Figure 4.21. 41
Figure 4.23 - Ten-run ensemble of export time series for multiple fertilisations at different patch length scales corresponding to Q = 0.0431, 0.0271, 0.0207, 0.0171, 0.0147, 0.0130, 0.0118, 0.0108 respectively. Growth rate is 0.4 /day. The period between fertilisations is 1 day (upper left), 3 days (above) and 9 days (left) and is indicated by the dotted line. 42
Figure 4.24 - Ten-run ensemble of export time series for multiple fertilisations at different patch maximum growth rates corresponding to Q = 0.0936, 0.0624, 0.0468, 0.0374, 0.0312, 0.0267, 0.0234, 0.0208, 0.0187, 0.0170 and 0.0156 respectively. Patch length is 25km. The period between fertilisations is 1 day (upper left), 3 days (above) and 9 days (left) and is indicated by the dotted line. 43
Figure 4.25 - Power spectrum density of export time series of 180 day pass. Analysis of final 160 days. The 6 fertilising periods are shown. 44
Figure 4.26 - Power spectrum density of export time series of 180 day pass. Analysis of final 160 days. 8 patch lengths for a fertilising period of 6 days. 45
Figure 4.27 - Power spectrum density of export time series of 18 day pass of different spatial distributions. 44
viii
List of Tables
Table 3.1 - Inputs into MATLAB code, “diffgrow2.m” 31
Table 3.2 - Outputs from MATLAB model, “diffgrow2.m” 31
Table 4.1 – Number of export peaks in 10-run ensemble. 35
Table 4.2 - Distinguishable periods of export from spectral analysis of export time series. 45
most efficiently to apply iron in specific natural conditions, and how the resulting patch might
be suspected to behave.
1.2 Direction of this study Waite and Johnson (2003) have developed a simple model that considers three important
processes in phytoplankton dynamics under a nutrient-limited regime – nutrient-limited
growth, lateral diffusion and export by aggregation. They have also highlighted the spatial
and temporal complexity that exists in phytoplankton dynamics, even when simplifying
assumptions are used, and how this complexity leads to carbon export behaviour that is non-
trivial.
This study aims to investigate the parameters that characterise the carbon export behaviour of
the Waite and Johnson (2003) growth-diffusion-export model. In particular, it seeks to
determine the conditions and methodology for applying the fertilising nutrient that most
efficiently generates carbon export from the mixed layer.
Modelling Ocean Fertilisation Patch Dynamics Literature Review
3
Chapter 2 Literature Review
2.1 Climate change and the role of the oceans Since the industrial revolution, anthropogenic activity has unbalanced the process known as
the global carbon cycle, by which carbon is moved and partitioned over the earth. Changing
land use and the mining and consumption of fossil fuels have converted terrestrial carbon into
atmospheric carbon, in the form of the greenhouse gas, carbon dioxide (CO2), at a rate that
has not been seen in the previous 1000 years (Crowley, 2000). The CO2 present in the
atmosphere has increased by more than 25% since 1957 (Wigley and Schimel, 2000) and it
has been determined that CO2 contributes most significantly to the changing global climate
(Crowley, 2000). The effect of this increasing pool of carbon in the atmosphere is an increase
in global temperatures as famously reported from the Mauna Loa research station in 1985
(Bacastow et al., 1985).
Although the atmospheric carbon pool is increasing, it is by no means the largest carbon pool
in the global carbon cycle. The intermediate and deeper layers of the world’s oceans contain
98% (38,100 Gt) of the earth’s carbon (Wigley and Schimel, 2000). Consequently, it is the
oceans that offer the greatest capacity for buffering the atmospheric carbon change and global
temperature increase.
The atmospheric and oceanic carbon pools interact via several pathways, including chemical
gradient processes, biological processes and physical processes such as subduction (Chisholm
1995). Sarmiento and Bender (1994) estimate, however, that 75% of the difference in
dissolved inorganic carbon concentrations between the surface and deep ocean is due to the
biologically-mediated pathway known as the ‘biological pump’. The biological pump is the
process by which CO2 fixed in photosynthesis is transferred to the deeper layers of the ocean,
resulting in temporary or permanent sequestration (storage) of carbon (Figure 2.1).
Atmospheric CO2 is initially fixed by autotrophs, such as phytoplankton, which then either
become senescent and sink out as aggregates or are consumed by herbivores that produce
sinking faecal pellets (Chisholm, 1995). Removing the biological pump, and allowing carbon
Modelling Ocean Fertilisation Patch Dynamics Literature Review
4
to be cycled in the ocean only by physical and chemical processes, would more than double
the CO2 concentration in the atmosphere through carbon release as the ocean equilibrated
(Chisholm, 1995).
The efficiency of the biological pump for carbon sequestration is not uniform across the
world’s oceans. High nitrate, low chlorophyll (HNLC) regions, namely the sub-arctic north-
east Pacific, the equatorial Pacific and the Southern Ocean (Chisholm et al., 2001), have long
provided a puzzling phenomenon for oceanographers because despite the favourable nutrient
conditions for production in these regions, productivity is extremely low (Gran, 1931). In
these regions productivity is not coupled to macronutrient concentrations, and even where
upwelling occurs, the resulting nutrients are not utilised (Martin, 1990). Considering that
these regions make up 25% of the world’s oceans (de Baar et al., 1999) they have attracted
considerable research interest into how they are sustained. This scientific interest has been
further accentuated in the last 30 years by the notion that one might use oceanic carbon
sequestration to mitigate the effects of global climate (Chisholm et al., 2001).
Figure 2.1 – Atmospheric and oceanic carbon interactions. The biological pump, with the fixation and removal of carbon to deep waters and recycling through grazing is shown on the left. Physical processes, subduction downwards and upwelling upwards, are represented on the right. (from Chisholm, 2000)
Modelling Ocean Fertilisation Patch Dynamics Literature Review
5
Several theories have been put forward to address the HNLC phenomenon since its discovery.
It seemed apparent given the adequate light availability in these regions that nutrient
limitation was occurring, but the nutrient of limitation was debated. As early as the mid
1930s, iron was invoked as the limiting nutrient responsible (Hart, 1934), however only
recently has the pervasive contamination that distorted early trace element measurements been
reduced so that such propositions could be accurately tested (Morel et al., 1991). One of the
reasons for the limitation of iron has been that it is poorly recycled in the oceans and must
therefore come from wind-transported dust from terrestrial sources. This process is known as
Aeolian deposition (Duce and Tindale, 2001). The HNLC regions have poor rates of Aeolian
deposition because they are many hundreds of kilometres from a terrestrial arid source (Duce
and Tindale, 2001). Some offshore upwelling regions, with high macronutrient
concentrations, do not exhibit HNLC conditions and these regions coincide with areas of
relatively high Aeolian deposition via long-range transport (Martin, 1990).
It is generally considered (Boyd, 2002, Chisholm et al., 2001., Coale et al., 1998, Cullen,
1995) that what is now termed Martin’s Iron Hypothesis (Martin, 1990) was a major
development in the case for iron limitation in HNLC areas. Martin’s hypothesis states that
oceanic productivity during glacial periods may have been higher due to increased upwelling
of nutrients and greater Aeolian deposition of iron from atmospheric dust loads 10-20 times
greater than at present. Martin (1990) hypothesised that these conditions maximised the
efficiency of the biological pump, transporting much more atmospheric carbon to the
intermediate and deeper oceans than currently occurs and resulting in the lower CO2 levels
observed in the paleogeological ice core record. The low atmospheric CO2 in turn had already
been suggested as a mechanism for preventing radiative heating and ultimately leading to the
freezing of the planet over the short time scales observed. Martin’s Iron Hypothesis framed
iron limitation in oceanic regions within the context of geological CO2 levels, thereby
providing circumstantial evidence and a testable theory for the role of iron in large-scale
oceanic processes.
Several elaborations of Martin’s hypothesis have emerged since it was first proposed as
successive bottle and in-situ iron enrichment experiments refined the knowledge of iron
limitation in the open ocean (Cullen, 1995). Cullen (1995) notes that much of the early focus
of Martin’s ideas was considered in the context of HNLC regions. The outcome from a formal
workshop on the topic in 1990 formulated what Cullen calls “the HNLC-iron hypothesis: “An
increase in the rate of supply of iron to the surface layer of the ocean will reduce to depletion
Modelling Ocean Fertilisation Patch Dynamics Literature Review
6
the unused macronutrients, nitrate and phosphate.” ” (Cullen, 1995, p1336). It was intended
that this hypothesis would provide a testable theory of iron limitation. When the rate of supply
of iron to the surface layer of the ocean was increased in the first in-situ iron enrichment
experiment, IronEx I, but the concentrations of nitrates and phosphates varied very little, there
was good evidence to reject the HNLC-iron hypothesis. Cullen (1995) suggests however, that
neither the test nor the hypothesis were perfect since the increase in the rate of supply of the
iron is not defined by the hypothesis and that the iron infusion applied in IronEx I was too
ephemeral to be considered an appropriate test of the hypothesis.
When later discussions progressed, those inclined towards the ecological implications of
Martin’s hypothesis, independently arrived at what is called the “ecumenical iron hypothesis”
(Cullen, 1995). The ecumenical iron hypothesis states that when iron is scarce, smaller cells,
with greater surface:volume ratios can grow more rapidly than larger cells. Therefore the
population of small cells is not limited by iron, but by microzooplankton grazers whose rapid
growth rates can keep phytoplankton under control. By contrast, large cells are not able to
achieve high growth rates when iron is scarce. Under enrichment, however, they can
assimilate macronutrients at a rate that is too great for mesozooplankton to respond, and their
size prevents them from being consumed by microzooplankton. Although this hypothesis
views the HNLC condition by the vague notion of a grazer controlled iron-limited system,
ultimately it predicts that during the enrichment of HNLC areas, we would expect larger
phytoplankton to dominate (Cullen, 1995).
The testing of these hypotheses formed the impetus behind more than 4 in-situ iron
fertilisation experiments including IronEx I, IronEx II, SOIREE and EisenEx, conducted so
far.
2.2 In-situ iron fertilisation experiments
Martin and his colleagues realised that in vitro bottle experiments that had been executed to
date were inadequate in testing the ecosystem and carbon export consequences of their iron-
limitation hypotheses, in particular the ecumenical iron hypothesis (Coale et al., 1996).
Martin therefore proposed “perform[ing] realistic large-scale Fe enrichment experiments in
which phytoplankton species composition, elemental ratios of C:N:P and Si, δ13C ratios, etc
… can be determined, as well as [investigating] the effects of grazing and associated fecal
Modelling Ocean Fertilisation Patch Dynamics Literature Review
7
The greatest hindrances to such an experiment lay in the lateral and vertical exchanges
between enriched and ambient waters due to advection and turbulent diffusion (Frost, 1996).
The logistics of the number and accuracy of measurements were also significant, however
more than four in-situ experiments have now been carried out, demonstrating that the
technology and the methodology for successful testing does exist.
2.2.1 IronEx I
Conducted in 1993 in the Equatorial Pacific, this is considered the first in-situ testing of
Martin’s hypothesis. Coale et al. (1998) implicitly refer to the hypothesis tested during this
experiment as “determin[ing] whether iron enrichment in the presence of the entire
community results in an increase in the net new production” (p921). The site near the
Galapagos Islands was chosen as the most favourable location for an initial iron experiment
for a number of reasons (Martin and Chisholm, 1992). Primarily these were its high light
intensities and warm temperatures (~25ºC) which would enable high phytoplankton growth
rates, and the vast oceanographic and biological data already available for the area. In
particular, National Oceanic and Atmospheric Administration (NOAA) drifters suggested that
surface flow paths with eddies were rarely observed. This was important since the physical
coherence of the patch was deemed to be of particular concern for the experiment (Stanton et
al., 1998). The NOAA data suggested that the problem of spreading and streaking of the patch
by turbulent diffusive processes at eddy or frontal boundaries (Garret, 1983) could be avoided
at this site.
Pre-fertilisation testing demonstrated that the site was typical of the equatorial Pacific HNLC
area (Coale et al., 1998). Concentrations of nitrate and Chlorophyll a were 10.8 µM and 0.24
µgL-1 respectively and there was a strong pycnocline, well mixed surface layer to 30m and
low horizontal gradients. Initial concentrations of dissolved and particulate iron were
measured at depleted levels of 0.07 and 0.22 nM respectively (Gordon et al., 1998).
443kg of iron as pharmaceutical grade Fe(II) sulphate was deployed once to a large square
patch approximately 8km × 8km. The choice of spatial distribution was due to sampling
considerations which ruled out other methods such as ‘point source’ (Martin and Chisholm,
1992) or ‘streak’ (Watson et al., 1991) applications.
Modelling Ocean Fertilisation Patch Dynamics Literature Review
8
The quantity of iron released was intended to achieve an ocean concentration of 4nM. This
was double the concentration necessary to achieve maximal phytoplankton growth in
laboratory bottle experiments but was considered necessary to account for possible iron
removal processes in the experiment. It was assumed that the concentration of iron when
released would be mixed through to the depth of the mixed layer within 24 h (Martin and
Chisholm, 1992). The root mean square distance for horizontal diffusion (square root of 2kxt)
in this period was calculated to be ~415m assuming a horizontal eddy diffusion coefficient of
kx ~ 10 km2d-1 (Coale et al., 1998). Therefore fertilisation tracks were separated by 450m.
These factors combined were intended to achieve the 4nM concentrations throughout the
patch after 1 day.
Iron behaviour was as expected. One day after iron release, maximum values in the patch
were 3.6 nM due to horizontal eddy diffusion and convective overturn (Coale et al., 1998).
The concentration of dissolved iron (DFe) decreased rapidly in the core of the patch over the
first four days of the experiment to 0.25nM Fe.
The biological response to fertilisation was dramatic. Productivity increased 3 to 4-fold in all
size fractions. Primary production increased monotonically from 10-15 mg C L-1 d-1 to 48 mg
C L-1 d-1 over three days and chlorophyll increased nearly 3-fold to 0.65 mg L-1. The chemical
response, however, was not correlated to this biological response. Although the biological
changes were sizeable, the magnitude of macronutrient drawdown was less than expected. In
contrast to bottle enrichment experiments in which there is complete drawdown (Coale et al.,
1998), nitrate drawdown was undetectable (<0.2 µM) and carbon dioxide fugacity was only
reduced by 10 µatm (Coale et al., 1996).
Approximately five days after the infusion of iron to the system, the core of the patch was
subducted to a depth of 30-35m beneath a low salinity front. At this depth it was confined to a
5-10m layer just above the thermocline. Although the patch could no longer be sampled by
the ship’s flow through system, the presence of the patch was still detectable from the SF6
signal, its distinct salinity and low light transmission (Coale et al., 1998). Since the SF6 signal
remained constant it is likely that the unfertilised waters did not penetrate and dilute the
subducted patch core (Coale et al., 1998).
Several theories were put forward to explain the subdued geochemical response observed in
IronEx I. These were that (1) iron was lost form the patch, (2) the subduction of the patch to
Modelling Ocean Fertilisation Patch Dynamics Literature Review
9
lower light levels minimised the photodissolution of iron colloids and decreased rates of
bioavailable iron production, (3) zooplankton quickly cropped the increase in phytoplankton
biomass, and (4) another nutrient, such as zinc or silicate, became limiting thus preventing
further growth (Coale et al., 1996)
Although the experimental results were confounded by a subduction event, the results were
heartening. This initial experiment heralded the start of a new wave of oceanographic research
by demonstrating that that oceanographers were no longer restricted to observation but that
the problems associated with large scale in-situ experiments could be overcome.
2.2.2 IronEx II
The second iron fertilisation experiment which occurred in 1995 near 3.5°S, 104°W in the
Pacific, followed very closely the methodologies of IronEx I but tried to address the
hypotheses put forward for the unexpectedly low geochemical response to iron enrichment. It
was important that this experiment recreated the biological responses of IronEx I without the
confounding subduction event (Coale et al., 1996).
Prior to fertilisation, the concentrations of nitrate were typical of the HNLC region (~ 10µM)
and initial iron concentrations were expectedly low, recorded at 0.05 nM (Cavender-Bares et
al., 1999).
At day 0 (29th May), iron, in the form of FeSO4 and a SF6 tracer were applied over a 72km2
rectangular deployment area. This was done by creating streaks 400m apart, which were noted
to merge within 1 day. Mixed layer depth measured over the infusion period was averaged at
25m (Coale et al., 1996). Consequently, the day 1 concentration of iron was 2nM. Subsequent
infusions in days 3 and 7 maintained the iron concentration in the infused patch at
approximately 1nM (Landry et al., 2000).
The mixed layer of the patch increased to 50m by day 11 due to small mixing events. Periodic
increases in nitrate concentrations suggested these mixing events introduced nutrient-rich
waters from below into the patch. The patch also expanded horizontally with time from the
initial 72 km2 to 120 km2, however it retained cohesion throughout the experiment (Coale et
al., 1996).
Modelling Ocean Fertilisation Patch Dynamics Literature Review
10
Chlorophyll a concentrations demonstrated a rapid and monotonic increase from 0.15-0.2
µgL-1 initially to almost 4 µgL-1 on day 9, two days after the last infusion of iron. Following
this peak, concentrations decreased to 0.30 µgL-1 on day 17 (Coale et al., 1996).
The biogeochemical response was significantly more developed than that observed in IronEx
I. Nitrate drawdown was approximately 5 µM, however this may be conservative considering
that some nitrate was probably mixed through from below during mixing events on days 11
and 14. After two days, nitrate drawdown tracked silicate drawdown suggesting that diatom
growth was responsible for most of the nitrate uptake (Coale et al., 1996).
Carbon dioxide drawdown also paralleled nitrate drawdown. Maximum depletion occurred on
day 9 in conjunction with the maxima of most of the other biological and chemical indicators
of growth. The south equatorial Pacific near the site is recognised as a strong source of CO2 to
the atmosphere (fco2 in seawater, 526 ppm; fco2 in the atmosphere, 360ppm), however, iron
enhanced growth enabled a drawdown of about 90 µatm, which significantly reduced
outgassing of CO2 from these waters. However, it is not believed that drawdown was so
severe as to limit carbon (Coale et al., 1996).
Iron was rapidly taken up or removed following each of the infusions (initial and day 3 and 7
‘top ups’). It was noted that as the biomass increased in the patch due to iron addition, the rate
of iron removal also increased (Coale et al., 1996).
The community response to the iron enrichment was a shift towards larger organisms,
particularly diatoms. Diatom biomass increased over 85 times to dominate over the naturally-
abundant small (<5µm) phytoplankton which only doubled in size. Total phytoplankton
abundance increased dramatically since they were able to grow faster than predators could
consume them. This led to an imbalance in the early phase of the iron-induced bloom. The
modest picoplankton biomass increase demonstrates that these were most controlled by
zooplankton grazing. However, diatoms increased because they were too large to be
consumed by the fast-growing microzooplankton and too fast-growing to be controlled by the
slower-growing mesozooplankton (Landry et al., 2000).
Estimates of carbon new production suggest that between 5 and 12 µM C was exported from
the surface layer (Coale et al., 1996). Since community analysis showed a lack of larger
mesozooplankton grazers, which are commonly responsible for producing rapidly sinking
Modelling Ocean Fertilisation Patch Dynamics Literature Review
11
fecal pellets that transport carbon below the mixed layer, Coale et al. (1996) suggests that
grazing export did not remove the surface carbon. More likely is that export occurred by
vertical mixing and sinking of diatom aggregates. This is supported by the removal of the SF6
tracer to erosion at the base of the mixed layer by exchange with waters moving relative to the
advection of the patch and thus spread horizontally within the mixed layer (Coale et al.,
1996).
2.2.3 SOIREE
After the success of the iron enrichment experiments in the HNLC areas of the Pacific, the
next step was to trial the meso-scale perturbation experiment in the Southern Ocean. The
conditions of the Southern Ocean are significantly different to those of the equatorial Pacific,
in its physics, its geochemistry and its ecology/biology (Boyd, 2002). More importantly, an
experiment in the Southern Ocean is the ultimate test of the Iron Hypothesis because the
Southern Ocean has the greatest potential for carbon drawdown. This potential is due to the
size of the Southern Ocean, its vast quantities of unused nutrients, except for iron, and the
intermediate and deep water formation delivering surface waters (Chisholm, 2000). It is also
where the coherence between paleoclimate iron flux and carbon export has been observed
most strongly (Coale et al., 1996). Regardless of the iron hypothesis implications, its potential
for carbon drawdown also makes the Southern Ocean the most important place to implement
large-scale iron enrichment to further the understanding of the Southern Ocean’s ability to
mitigate climate change (Chisholm et al., 2001).
The Southern Ocean study site was chosen to be representative of a broad region of
circumpolar HNLC waters but have small current shear stresses in order to maximise the
timescales for tracking the fertilised patch. It was also necessary to try and balance a
regionally representative depth of the mixed layer, whilst making sure that the mixed layer
depth was not too deep so that phytoplankton would become dually iron and light co-limited,
and also so that the iron/sulphur hexaflouride (SF6) would be overly diluted (Boyd and Law,
2001). Whether a true balance was made between a regionally representative site, in the
horizontal and vertical sense, and one which fitted the criteria for successful experimental
design is difficult to say. The up- and down-welling behaviour of the Southern Ocean as well
as other physical behaviours are highly seasonal and also spatially variable (Trull et al.,
2001). Trull et al. (2001), also members of the SOIREE science team, contend that “less than
half of the Southern Ocean is likely to exhibit a response similar to that which occurred
during SOIREE (because only waters well south of the Polar Front are silica-rich throughout
Modelling Ocean Fertilisation Patch Dynamics Literature Review
12
the year), and that carbon export by bloom subduction is unlikely” (p2440). This is unless
there are significant changes in the community structure or algal physiology (Trull et al.,
2001). Given this, the site chosen is relatively representative of the summer period in which
the experiment was conducted, but will not represent any response seen in the winter at the
same site (Trull et al., 2001). The average mixed layer depth for the chosen site was 65m.
Pre-fertilisation testing showed that at the site, mixed layer nitrate and phosphate, at ~25 ±
1µM and ~1.5 ± 0.2µM respectively, were relatively high. Dissolved iron levels were low at
~0.08 ± 0.03 nM for polar waters and chlorophyll a concentration was 0.25 mg m-3 (Boyd et
al., 2000). These concentrations are indicative of the Southern Ocean.
The methods for conducting a mesoscale iron enrichment experiment were quite well
established at SOIREE, following both IronEx I and II. As in the case of both the earlier iron
enrichment experiments, SF6 was used as a tracer for iron added as acidified FeSO4.7H2O.
Testing was conducted in a Lagrangian framework. The study site was infused with iron to
3.8nM in a patch ~50km2 at day 0 (9th February 1999). Subsequent infusions occurred on day
3, 5 and 7 to ~2.6nM over areas of 32, 33.8 and 38.5 km2 respectively (Boyd and Law, 2001).
The initial patch had mixed through within two days to 100 km2 and within 13 days to ~200
km2. No physical structural changes were observed within the first four days of the
experiment, however, calm days caused the generation of transient and temporary thermal
strata on days 5/6, 8/9 and 13. (Boyd et al., 2000)
Dissolved iron was measured as a criterion for subsequent iron infusions. After the first
infusion, levels were initially increased similar to those in Iron Ex I and II and iron rich areas
of the Atlantic polar front (Boyd et al., 2000). By day 2, however, levels of iron had
decreased rapidly, presumably due to the effects of the patch spreading and conversion to
particulate iron. Total (unfiltered) iron remained at ~2nM. Later infusions followed when
dissolved iron concentrations approached background levels. After the fourth and final
infusion iron levels decreased, but thereafter remained relatively stable at ~1nM until the end
of the study site occupation (Boyd et al., 2000).
The geochemical response was expectedly large. Nitrate drawdown was 3µM and pCO2
drawdown was 35µatm, which, although less than the equatorial pacific equivalents, are still
sizeable (Boyd, 2002). The biological and physiological changes that were observed in both
IronEx I and II were also observed in SOIREE. Ratios of carbon to chlorophyll a halved by
Modelling Ocean Fertilisation Patch Dynamics Literature Review
13
day 13 to ~45 within the patch compared to just prior to the first infusion. The community
structure also changed. There was a floristic shift over the period of the experiment with
initial chlorophyll a increases attributed to pico-eukaryotes, to autotrophic flagellates between
days 2 and 8, and finally to large diatoms (particularly Fragilariopsis kerguelensis) from day
6. These large diatoms, of between 30-50 µm cell length and growing at 4.4 x 104 cells L-1 by
day 12, also increased the number of cells in their diatom chains twofold to 14 by day 12. It is
believed that this floristic shift is responsible for mediating changes in the concentration of
climate-reactive gases in surface waters (Boyd et al., 2000). The dominance by F.
kerguelensis is also believed to be responsible for the low rates of diatom herbivory as it is
morphologically adapted, with a highly silicified skeleton, to minimise grazing. Diatoms
accounted for 75% of production (Boyd et al., 2000)
Although the SOIREE study supported the first tenant of Martin’s iron hypothesis by
demonstrating a significant biological response to iron addition, it did not support the second
tenant in that there was no evidence of increased particle export (Boyd et al., 2000). It is
likely that the lateral diffusion, which amounted to 25% of the algal growth rate, prevented a
single large export event (Waite and Nodder, 2001). Furthermore, sediment traps suggest that
sinking rates and aggregation characteristics did change over the course of the experiment.
This is consistent with Muggli et al. (1996) who have shown that sinking rates are higher
under iron stress. The phenomenon of aggregate formation was increased during iron
enrichment (Waite and Nodder, 2001).
2.3 Modelling an iron-stimulated biological pump
Modelling the biological pump is important in understanding the role that the oceans play in
the global carbon cycle. Accurate modelling requires consideration of aspects of
phytoplankton growth and distribution, and how the carbon fixed by phytoplankton is
conveyed to the intermediate and deep ocean.
2.3.1 Models of phytoplankton dynamics
The early work of Kierstead and Slobodkin (1953) developed a classical model for oceanic
phytoplankton dynamics by balancing horizontal diffusion with growth of phytoplankton in
the mixed layer. They considered a body of water that was favourable to growth bounded by,
and mixing at the edges with, water that is unsuitable. The unsuitable water could be
Modelling Ocean Fertilisation Patch Dynamics Literature Review
14
characterised by any of several parameters including salinity, temperature or nutrients. In the
case of iron enrichment the suitable waters are no longer limited by iron whereas unsuitable
water is still iron-constrained and therefore has relatively small growth rates. By assuming
that diffusive losses were large, or that the water was otherwise unsuitable for concentrations
of phytoplankton outside a particular area of patch, the Kierstead and Slobodkin model
delivers a critical minimum patch size able to be sustained. Subsequent elaborations of this
model incorporated the effect of grazing by zooplankton (Wroblewski et al., 1975; Platt,
1975), the scale dependence of the diffusion coefficient (Platt and Denman, 1975), nutrient
limitation and light periodicity (Wroblewski and O’Brien, 1976).
In its simplest form, considering only horizontal diffusion with spatially constant diffusion
coefficients and by neglecting advection, phytoplankton dynamics within the patch can be
represented by:
QCSCdx
CdKdtdC
h −+= max2
2
µ
where C is the concentration of phytoplankton
Kh is the horizontal diffusion coefficient
µmax is the maximum growth rate
S is the function that details the growth formulation
and; Q is the rate of collective removal of phytoplankton from predation by herbivores,
extracellular release, and sedimentation.
(Wroblewski and O’Brien, 1976)
2.3.2 Models of carbon export
Below the surface or mixed layer it can be assumed that oceanic waters are quiet and that
there is little return of particles to the surface (O’Brien et al., 2003). In modelling the
biological pump one therefore needs only to consider carbon export from this mixed layer.
Sedimentation of particulate organic carbon (POC) is the primary biologically-mediated
method of carbon export from the mixed layer in marine systems. Due to Stokes’ Law larger
particles sink faster than smaller particles or cells and therefore have a significant effect on
the fate of organic matter (Jackson and Lochmann, 1992). They contribute disproportionately
to the carbon flux into deeper waters. Consequently, carbon export models have focussed on
these larger particles and the processes which form them.
Modelling Ocean Fertilisation Patch Dynamics Literature Review
15
Large particles may exist as large phytoplankton, especially diatoms, or as accumulations of
smaller particles, especially smaller phytoplankton (nanoplankton or picoplankton).
Furthermore, these accumulations may be aggregates formed from the collisions between
particles, or as non-aggregate particles formed through mechanisms including algal division
and diatom chain formation.
Aggregates of organic matter (also known as flocs or “marine snow”) are a highly visible
phenomena observed in the wake of an algal bloom (Jackson, 1990). The collision processes
by which they form have been artificially replicated in the laboratory with naturally occurring
organic matter (Waite et al., 1997). Kinetic coagulation theories developed in understanding
particle dynamics of lakes, conclude that the rates of algal losses to aggregation (coagulation)
can be comparable to losses caused by zooplankton grazing. This is supported by algal losses
in in-situ iron enrichment experiments in oceanic systems (Boyd et al., 2002; Coale et al.,
1996) and the ecumenical iron hypothesis (Cullen, 1995) which both demonstrate that when
iron is abundant sedimentation losses are at least as important as herbivory losses.
In coagulation theory, an aggregate is a particle formed by collision of two smaller particles,
the largest aggregates being the product of repetitive collisions and coalescence of smaller
ones. Up until Jackson (1990) the concepts of coagulation theory had been applied primarily
to freshwater environments to explain mass fluxes. Jackson applied these principles to a
marine system and expanded the focus of the theory to explain the effect of coagulation and
carbon export on phytoplankton dynamics. In the Jackson model a number of different size
classes were considered. Particles were removed from a particular size class either into a
larger size class following a collision and coagulation, or as they were exported below the
mixed layer. The likelihood of incorporation into larger and larger sized particles, the
collision rate, was given by a function of sizes of colliding particles, their concentrations and
environmental and physiological parameters. Three collision mechanisms were used to
describe particle collisions – Brownian motion, shear (either laminar or turbulent) and
differential sedimentation (in which a larger particle falling faster than a smaller one ‘runs
into the back’ of the smaller one).
The Jackson model used the criteria that the fundamental size class was a solitary cell and the
size of each particle is given by the number of cells it contains. The source of particulate
Modelling Ocean Fertilisation Patch Dynamics Literature Review
16
matter for the model was cell growth, increasing the size of aggregates and increasing the
number of separate solitary cells. The ultimate source is removal by sinking.
An important conclusion of Jackson’s model is that algal systems have a two-state nature,
either coagulation is not important or it is dominant. The distinction between these two states
occurs at some critical cell concentration (Ccr) (Jackson and Lochmann, 1992). This provides
a parameter to compare the likelihood of continued growth (<Ccr) or rapid loss of particles to
coagulation and decrease in particle concentration (>Ccr). The value of this Ccr therefore
becomes important when we are considering the aggregation response of a system and the
relationship between growth as a source and export as a sink. Jackson and Lochmann (1992)
further developed this model by incorporating light and nutrient limitations, which
constrained the growth dynamics of the phytoplankton.
Since coagulating particles can occur when growth occurs even at reasonably constant rates,
there is likely to be a cap placed on phytoplankton concentrations in natural systems because
export from the mixed layer is ongoing (Jackson and Lochmann, 1992). The importance of
coagulation in controlling biomass is not likely to be great in areas where shear is low or
grazing is high. In natural systems, this value of Ccr will roughly correlate with the maximum
phytoplankton concentrations that exist (Jackson and Lochmann, 1992).
Results from SOIREE suggest that one of the primary algal losses from the iron-enriched
bloom was due to diffusive losses at the edges of the patch. The entrainment of surrounding
HNLC waters and subsequent dilution of phytoplankton stocks in the labelled patch was
given as a reason for its longevity. Abraham et al. (2000) report losses due to lateral diffusion
to be 0.1d-1, 75% of net growth rate. Boyd et al. (2001) call this a “physical artefact” in as far
as it obscures the biological processes underlying Martin’s hypothesis. Boyd et al. (2002)
investigate these “physical artefacts” in an aggregation model (Jackson and Lochmann, 1992)
modified by imposing a constant specific algal growth rate representative of both the SOIREE
and IronEx II mesoscale experiments. Two cases were modelled – a standard run,
incorporating best guesses of parameters from the respective enrichment experiments; and a
run with a higher net growth rate. The purpose of the higher net growth rate was in order to
mimic a larger scale enrichment experiment (~100km) where it was assumed that dilution of
bloom stocks via horizontal diffusion is negligible. An algal monoculture was assumed
because of its simplicity and the dominance of a single species in both the IronEx II and
SOIREE studies.
Modelling Ocean Fertilisation Patch Dynamics Literature Review
17
By using a higher net growth rate to represent lower diffusive losses the consideration of
diffusive losses by Boyd et al. (2002) is a simple one. Although the model is based upon a
thorough aggregation model (Jackson and Lochmann, 1992) it fails to consider classical
diffusion-growth theory (Kierstead and Slobodkin, 1953). Boyd et al. (2002) and Boyd et al.
(2001) mistakenly refer to lateral losses of phytoplankton bloom stocks as “lateral advective
losses” (eg – Boyd et al., 2002. pp 36-4) and the authors refer to Abraham et al. (2000) to
provide an estimate of this lateral loss. What these authors term “lateral advective losses” are,
however, more accurately referred to as “horizontal diffusive losses”. Indeed, this is the actual
context of Abraham et al. (2000) estimates quoted in Boyd et al. (2002) and Boyd et al.
(2001). In contrast to some of the assertions of Boyd et al. (2001), the concept of diffusive
losses is not uncommon in phytoplankton patch modelling (see Kierstead and Slobodkin,
1953). This paper will refer accordingly to lateral losses as horizontal diffusive losses, in the
more accurate context used by Abraham et al. (2000) and followed by Waite and Johnson
(2003).
The results of Boyd et al. (2002) do accurately reflect the timing of downward particle flux
measurements of both IronEx II and SOIREE. They do not, however, simulate the magnitude
of the changes in particle concentrations and downward flux from observations.
In modelling the role of diffusive losses in mesoscale enrichment experiments, applying a loss
term that is dependent upon length scale provides a more realistic consideration of the process
than merely using a higher net growth rate. This is the approach taken by Waite and Johnson
(2003). A simple horizontal diffusion-growth model is employed for phytoplankton dynamics
(Kierstead and Slobodkin, 1953; Wroblewski and O’Brien, 1976). Horizontal diffusion is
made to be length-scale dependent (Okubo, 1971) and spatially uniform.
A 2-dimensional model is justified by considering only phytoplankton which are uniformly
distributed vertically in the mixed layer. This is justified since sinking rates of unaggregated
phytoplankton are low (O’Brien et al., 2002). Vertical diffusion out of the mixed layer is
assumed to be small also. Vertical export from the mixed layer occurs only through
aggregation and sinking which is modelled as a two-state function (Jackson, 1990)
E= -1Ccr C≥Ccr
0 C<Ccr
Modelling Ocean Fertilisation Patch Dynamics Literature Review
18
This means that once concentration reaches a certain critical threshold there is an
instantaneous export of this amount out of the mixed layer. The rate of sinking of single cells
is considered to be negligible and it is assumed that the timescale of export by aggregation
and sinking is shorter than the timescale of growth (Jackson and Lochmann, 1992). Thus the
phytoplankton dynamics can be modelled according to the following relationship:
ECCKdtdC
h −+∇= µ2
Waite and Johnson (2003) investigate both the non-limited and nutrient-limited cases of
phytoplankton growth by developing the non-dimensional parameter, Q. Q represents the
ratio between diffusion growth and patch length scale.
2LKQµ
=
In both cases total mean export is dependent upon Q. The non-limited case demonstrates
characteristics chaotic patterns in surface patch structure as holes form from sinking in
concentrated areas.
In the nutrient-limited case the growth term, µ, is made to be dependent upon nutrient
concentration via the Monod equation (McCarthy, 1981). Nutrients are modelled in a similar
diffusion model to phytoplankton, diffusing horizontally but not vertically. Nutrient removal
is via uptake by phytoplankton. As phytoplankton sink out it was assumed that they take the
associated nutrients with them so there is no assumed recycling by grazing. Coagulation
theory also suggests that rapid removal of particles moves organic matter to the deep ocean
faster and makes it more likely they will fall rather than be eaten (Jackson and Lochmann,
1992).
In contrast to the non-limited case, nutrient limitation makes µ and therefore Q, time
dependent. Therefore Q is considered to be dependant upon µmax rather than µ in the nutrient-
limited case. Due to the time dependence, export occurs in events (Figure 2.2) with a primary
sedimentation event followed by smaller secondary events as nutrients are removed and
become increasingly limited (Waite and Johnson, 2003).
Modelling Ocean Fertilisation Patch Dynamics Literature Review
19
Figure 2.3 – Export simulated from a growth-diffusion-export model and its relationship with patch size and the non-dimensional parameter, Q. Note that export only occurs after Q ~ 0.7 (from Waite and Johnson, 2003).
Figure 2.2 – Time series of mean export of a nutrient-limited growth-diffusion-export model over 10 passes. At 10 km there is no export generated as the patch is too small. Duration and intensity increases as patch size increases. (from Waite and Johnson, 2003).
Modelling Ocean Fertilisation Patch Dynamics Literature Review
20
The total export increases with decreasing Q and therefore with increasing patch size (for
constant µmax) (Figure 2.3). The critical Q at which export begins to occur is 0.07 however the
maximum export occurs far beyond the assumptions of this model. At low values of Q (large
patch size) vertical diffusion out of the patch becomes comparable to the horizontal diffusion
and cannot be neglected.
The Waite and Johnson model highlights the spatial and temporal complexity of export and
phytoplankton concentration in fertilised patches, and therefore the inadequacies in field
sampling, especially in the in-situ enrichment experiments conducted so far. Furthermore, it
demonstrates that such experiments are not likely to be testing the second tenet of Martin’s
iron hypothesis, that is, export to intermediate and deep waters, because patch size is too
small to initiate export.
The temporal and spatial complexities in phytoplankton dynamics implied by the Waite and
Johnson model are inherently interesting, but especially so in the context of how these
complexities may affect the way that future in-situ experiments and non-scientific projects are
conducted. To date, the use of numerical models to simulate carbon export behaviour of a
phytoplankton system has been progressive but needs to be continued. Diffusion-growth-
export models address the key processes identified by in-situ experiments as determining
phytoplankton dynamics. The next step is too fully interrogate such models as to how these
processes create the phytoplankton and export behaviour observed.
Temporal variations in the way nutrients were applied were investigated, in particular, the
total export and export time series, by applying nutrients to the domain more than once.
Multiple applications of iron were a feature of both IronEx II and SOIREE.
A further function which encompassed the diffgrow model in a user-specified repeat
procedure was created (coding in Appendix 2). The domain was repeatedly fertilised with the
same fertilising patch variant over an 18 day period, followed by a period of 15 days ‘cooling
down’ without fertilisation. The period between nutrient applications was varied from 1 day
Figure 3.1 – Example initial phytoplankton distributions at the different random length scales input - a) L/10, b) L/5, c) L/2 and deterministic distribution d) uniform. The basic passes of the model use the distribution shown in a).
a
dc
b
Figure 3.2 – Nutrient distribution of fertilising patch variations – length scales a) uniform, b) L/10, c) L/5. Green indicates the presence of nutrients and blue the absence. The basic passes of the model use the distribution shown in a).
4.1 Variations of growth and fertilising concentration parameters Export per unit area increases both with increasing fertilising concentration and increasing
maximum growth rate (Figure 4.1). Export increases linearly with applied nutrient
concentration, but is non-linearly affected by maximum growth rate. This means that the
effect of increasing growth rate is greatest when growth rate is small and tapers off when
growth rate is larger.
Figure 4.1 – The effect of variations of maximum growth rate, µmax, (on axis from 0 to 1) and fertilising concentration (on axis from 0 to 10) for the three different patch lengths shown. The plots represent, on the z axis, export per unit area (mmol C m-2) in the left column, time to nutrient depletion, tN (days) in the central column and tN x µmax in the right column. Derived from 10-run ensemble.
Increasing the patch size is also important for the amount of export generated. The frontier for
export to occur moves towards lower concentrations and lower growth rates as patch size
increases. That is, export occurs for a greater range of growth and fertilising concentration
parameters at larger length scales than it does at smaller lengths scales.
By normalising the export by the fertilising concentration, export still increases with
fertilising concentration, however the curvature of the export slope with fertilising
concentration is no longer linear. Instead it shows concentration increase to be more important
when concentration is low than when it is high.
4.1.1 Time scales of export
Figures 4.5 and 4.1 demonstrate that time to first dump and time to nutrient depletion are
unaffected by changes in fertilised concentration. These parameters are significantly affected
by growth rate and appear to be bound by an inverse relationship (Figures 4.1 and 4.4). This
means that the inverse of growth rate can be considered an important timescale in export. This
proposal is partly supported by Figure 4.1 which demonstrates a somewhat constant
relationship between the maximum growth rate, µmax multiplied by time to nutrient depletion,
tN, and maximum growth rate. An inverse relationship between µmax and tN would lead to a
constant relationship between µmaxtN and µmax, however, Figure 4.1 does not show an entirely
constant relationship. Deviations from this pattern occur when growth rate is large and where
fertilising concentration is large and these are propagated to lower maximum growth rates and
lower fertilising concentrations with increasing patch length scale.
Figure 4.2 – Export per unit area per unit concentration of nutrient applied, generated by variations of maximum growth rate, µmax (on axis from 0 to 1) and fertilising concentration (on the axis from 0 to 10). Patch length is a) 10 km, b) 25 km and c) 50km.
As maximum growth rate increases the time until the first export event reduces. There is a
significant reduction from approximately 10 days, when export begins to occur, to close to 2
days at maximum growth rates of 0.7 /day. The weighted average of export flux moves
closely with the time to first export since the duration of the export event is short and
reasonably constant.
At higher growth rates there is evidence of two export peaks. The maximum export flux
increases with growth rate but there is also a secondary event with a smaller export flux. This
secondary event usually occurs after the maximum export event, however it may also occur
before (Figure 4.3e)).
a c b
d e f
Figure 4.3 – Ten-run ensemble of time series of export showing variations with maximum growth rate. Maximum growth rates a) – f) are 0.2, 0.3, 0.4, 0.5, 0.6 and 0.7 /day respectively. The fertilising concentration is 4.0 µmol m-3. Patch length, L = 25 km.
Figure 4.4 demonstrates the behaviour of the export time series with changing maximum
growth rate by considering some descriptive parameters. The time of the first export is related
to maximum growth rate in an apparently inverse relationship. The mean time of export,
weighted by export flux closely resembles this pattern, because as shown, export occurs
reasonably instantaneously and only increases slightly with µmax. This increase is due to the
multiple peaking demonstrated in the complete time series (Figure 4.3).
4.1.3 Fertilising concentration
Increasing fertilising concentration has little impact on the time scales involved in the export
time series (Fig 4.5). For the maximum growth rate of 0.4 day--1 used, the timing of the first
export event falls consistently about 4.2 days. The fertilising time scale does, however, appear
to have a more significant impact on the distinction of export ‘events’ that occur. Two events
are clearly distinguished for concentrations of 6.0 and 7.0 µM m-3 that are far more distinct
than the multi-peaking seen in Figure 4.3 for high growth rates. Less distinct multi-peaking of
the kind seen in Figure 4.3 are, however, evident in both the 4.0 and 5.0 µM m-3 and also
within the first and second distinguishable peaks. The multi-peaking at high fertilising
Figure 4.4 – Three parameters of export time series behaviour and how they change with increasing maximum growth rate. Weighted average is the mean time of export weighted by export flux.
0123456789
10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Maximum growth rate (/day)
Tim
e (d
ays)
Weighted average Time of first export Export duration
concentrations is distinguishable due to the long duration of both peaks. This means that
although the initial export timing is unaffected by changes in fertilising concentration the
weighted average of export occurs more than 1 day later for the 7 µM m-3 than it does for the
3 µM –3 case. In contrast to the multi-peaking of short duration (Figure 4.5 c) and d)), the
distinguishable peaks are of relatively even export, indicating two major export events in their
own right. The characteristics of the second export event are of an increasing export flux
leading to the second peak. Animations created by the model show phytoplankton export
occurs at peripheral concentrated nodes, before the central area removed by the first export
event has recovered sufficiently from growth and diffusion from other areas to be at the
critical export concentration again, thus leading to the second export flux peak. The maximum
export flux of the time series is not substantial and if anything reduces as fertilising
concentration increases. Since export is sustained for a greater duration, however, the total
exported increases linearly as demonstrated in Figure 4.1.
a c b
d e f
Figure 4.5– Ten-run ensemble of time series of export showing variations with fertilising concentration. Fertilising concentrations a) – f) are 2.0, 3.0, 4.0, 5.0, 6.0, and 7.0 µmol m-3 respectively. The maximum growth rate is 0.4 /day. Patch length, L = 25 km.
Figure 4.6 demonstrates that after fertilising concentration is significantly large enough to
generate export, at 2 µmol m-3, it has little effect on the timing of the first export. There is,
however, a significant jump in the export duration at 6 µmol m-3 when the distinguishable
peaks in export begin to occur. This causes the weighted average to increase.
4.1.4 Variation of the initial phytoplankton distribution length scale of randomness
The random length scale of initial phytoplankton distribution does not cause any obvious
patterns of difference for export time series at smaller length scales, however as length scales
increase more export events occur (Figures 4.7 - 4.10). At 25km, single peaks occur for all
length scales.
Figure 4.6 - Three parameters of export time series behaviour and how they change with increasing fertilising concentration Weighted average is the mean time of export weighted by export flux.
0
1
2
3
4
5
6
0 2 4 6 8
Fertilising concentration (umol m^-3)
Tim
e (d
ays)
Weighted average Time to first export Export duration
Figure 4.7 – Patch length 25 km, 10-run ensemble of export time series for four length scales of randomness in initial phytoplankton distribution. Random length scale is (a) L/10, (b) L/5, (c) L/2 and (d) uniform.
a
c
b
d
Time since nutrient fertilisation (days)
Exp
ort (
mm
ol C
m-2
day
-1)
Figure 4.8 – Patch length 40 km, 10-run ensemble of export time series for four length scales of randomness in initial phytoplankton distribution. Random length scale is (a) L/10, (b) L/5, (c) L/2 and (d) uniform.
a
c
b
d
Time since nutrient fertilisation (days)
Expo
rt (m
mol
C m
-2 d
ay-1
)
Figure 4.9 – Patch length 70 km, 10-run ensemble of export time series for four length scales of randomness in initial phytoplankton distribution. Randomlength scale is (a) L/10, (b) L/5, (c) L/2 and (d) uniform.
a
c
b
d
Time since nutrient fertilisation (days)
Exp
ort (
mm
ol C
m-2
day
-1)
Figure 4.10 – Patch length 85 km, 10-run ensemble of export time series for four length scales of randomness ininitial phytoplankton distribution. Random length scale is (a) L/10, (b) L/5, (c) L/2 and (d) uniform.
The number of export peaks occurring in each of the fertilising length scales is summarised in
Table 4.1. It shows that the number of peaks increases as the patch length scale increases but
this then decreases at the largest length, 85km.
Table 4.1 – Number of export peaks in 10-run time series ensemble.
Growth rate is 0.4 /day.
L/10 L/5 L/2 Uniform
25 km 1 1 1 1
40 km 2 2 3 1
70 km 3 3 4 2
85 km 1 2 3 1
The timing of the end of export is constant for each of the background phytoplankton
distributions, however, the duration of export is greater when the randomness length scale is
L/2 than in other cases. Consequently the time of the first export is earlier in the case of the
L/2 randomness length scales distribution.
Figure 4.9 demonstrates that there is very little difference between the amounts of export per
unit area generated by any of the different randomness length scales. The export per unit area
increases with patch length at the same rate for each of the random phytoplankton distribution
length scales, however there is a great deal of variation about this trend.
Figure 4.11 – Export per unit area generated over 10 days from a single fertilisation for different length scales of randomness of the initial phytoplankton distribution.
4.1.5 Variation of the fertilising concentration length scale
Figures 4.12-4.15 show that the uniform application of nutrients mainly generates single
peaks. By contrast, the smaller subpatch length scales are associated with multiple peaks at all
patch length scales. The maximum export flux of the different distributions is greatest at the
smallest (L/10) length scale. The L/5 length scale generally has a smaller maximum export
flux than either of the other distributions but has a greater duration of export than the uniform
nutrient application and equal or greater duration than the L/10 length. Note that the scales of
the ‘export’ axis vary between Figures 4.12 to 4.15 (but not within a figure).
The length scale of the fertilising subpatches has a significant effect on the export per unit
area (Figure 4.16). There is greatest divergence between the exports generated for different
fertilising distributions when the fertilising length scale is large.
Figure 4.12 – Patch length 25 km, 10-run ensemble of export time series for three length scales of fertilising subpatch application. Length scale is (a)uniform, (b) L/5 and (c) L/10.
a
c
b
Time since nutrient fertilisation (days)
Expo
rt (m
mol
C m
-2 d
ay-1
)
Figure 4.13 – Patch length 40 km, 10-run ensemble of export time series for three length scales of fertilising subpatch application. Length scale is (a) uniform, (b) L/5 and (c) L/10.
Figure 4.14 – Patch length 70 km, 10-run ensemble of export time series for three length scales of fertilising subpatch application. Length scale is (a) uniform, (b) L/5 and (c) L/10.
a
c
b
Time since nutrient fertilisation (days)
Exp
ort (
mm
ol C
m-2
day
-1)
Figure 4.15 – Patch length 85 km, 10-run ensemble of export time series for three length scales of fertilising subpatch application. Length scale is (a) uniform, (b) L/5 and (c) L/10.
a
c
b
Time since nutrient fertilisation (days)
Exp
ort (
mm
ol C
m-2
day
-1)
0
0.05
0.1
0.15
0.2
0.25
0 20 40 60 80 100
Patch length (km)
Expo
rt (m
mol
C m
^-2)
uniform
length L/5
length L/10
Figure 4.16 – Export per unit area generated over 10 days from a single fertilisation for different length scales of fertilising nutrient subpatch. Note significant divergence between length scales.
Figure 4.17 – Maximum export flux occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L.
Figure 4.18 – Mean time of export weighted by export flux occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L.
The greatest export flux at most L occurs at the smallest, L/10 subpatch length scale (Figure
4.17). This is in addition to generating the greatest export per unit area (Figure 4.16) The
deterministic, uniform initial phytoplankton distribution also demonstrates large export flux
associated with simultaneous sinking of the entire patch which occurs, as Figure 4.20 shows,
over small durations relative to the rest of the spatial distributions. The duration of export
does not increase greatly even as the length scale increases for the deterministic case,
however, in most other spatial variants there is an increases with the length scale, and in the
case of the L/5 scale subpatch, a significant jump in the export duration that coincides with
Figure 4.19 – Timing of the first export occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L.
Figure 4.20 – Timing of the first export occurring in different spatial scales of initial phytoplankton distribution and fertilising subpatch scale and how these vary with the length scale, L.
The export time series produced by multiple fertilisation events are complex and appear
highly random (Figure 4.23 and 4.24). There is not clear evidence of regular periods of
export, for any of the fertilising frequencies, in the 18 day pass, however, regularity of export
is most obvious when the patch length is small (e.g. – 10km). At small length scales, export
Figure 4.22 – Export per unit area generated by the uniform fertilising distribution. Q varied by µmax. Note that the range of Q, especially where the downturn occurs, is not shown in Figure 4.21.
Figure 4.21 – Export per unit area generated by three fertilising distributions – at length scales of L/10 and L/5 – and uniform. Q varied by patch length, L. µmax = 0.4 /d.
flux variation is large in comparison to larger patch lengths (80km) where the export flux is
smaller but more consistent. At inter-fertilisation periods of 9 days it is possible to see the
distinctive patterns of individual fertilisation events as export ceases and then resumes
following fertilisation. Export flux is often greatest when nutrients are reapplied following a
period of no export.
Figure 4.23 – Ten-run ensemble of export time series for multiple fertilisations at different patch length scales corresponding to Q = 0.0431, 0.0271, 0.0207, 0.0171, 0.0147, 0.0130, 0.0118, 0.0108 respectively. Growth rate is 0.4 /day. The period between fertilisations is 1 day (upper left), 3 days (above) and 9 days (left) and is indicated by the dotted line.
Variations of µmax clearly alter the time scales in the export time series, most obviously the
time to the first export event (Figure 4.24). Large export flux spikes are observed in low Q
values when µmax is varied, in contrast to the same being observed in high Q when length
scale is varied. At high Q (low µmax) export is still variable but the amplitude of export flux is
smaller. Export peaks are evident following re-fertilisation after a period of no export.
Figure 4.24 – Ten-run ensemble of export time series for multiple fertilisations at different patch maximum growth rates corresponding to Q = 0.0936, 0.0624, 0.0468, 0.0374, 0.0312, 0.0267, 0.0234, 0.0208, 0.0187, 0.0170 and 0.0156 respectively. Patch length is 25km. The period between fertilisations is 1 day (upper left), 3 days (above) and 9 days (left) and is indicated by the dotted line.
Power spectral analysis of a 180 day re-fertilisation length reveals dominant periods of export
at a variety of values. These can be seen in Figure 4.25, and are summarised from closer
investigation, in Table 4.2. The dominant period coincides with the fertilising period,
however, other periods are also important. Multiples of the forcing period are evident. Export
periodicity of approximately 2 days is evident in Figure 4.25 at all fertilising periods.
Powerful periods are densely distributed at lower values. The dominant export periods that are
evident for each of the different fertilising periods are summarised in Table 4.2.
When the fertilising period is 1 day the dominant export periods are approximately 1.5 days
and 0.5 days. Figure 4.25 demonstrates how the fertilising periodicity represents a period of
low power, not high, at fertilizing periods of 1.
Figure 4.26 demonstrates the effect of the patch length on the dominance of certain periods,
for the 6 day inter-fertilising period. The 6 day fertilising period is important but generally of
equalled in dominance by the 2 day period at all patch sizes. Important periods between 0.2
days and 2 days that are evident at the 10km patch length are not important as the length scale
1 day
3 days
6 days
9 days
18 days
2 days
Figure 4.25 – Power spectrum density of export time series of 180 day pass. Analysis of final 160 days. The 6 fertilising periods are shown. Power is on a logarithmic scale.
increases. For small length scales, a greater variety of small periods are dominant in the time
series, in comparison to the larger patch lengths. The lower extent of these periods exists at
approximately 0.2 days, corresponding to the timestep, dt.
Table 4.2 – Distinguishable periods of export from spectral analysis of export time series.
Fertilising period (days) Important export periods (days) 1 1.5 2 2 3 1.5, 3 6 1, 1.5, 2, 3, 6 9 1, 1.125, 1.25, 1.5, 1.7, 2.25, 3, 4.5, 9 18 Several small periods, 1.5, 1.6, 1.75, 2, 2.25, 2.6, 3, 3.6, 4.5, 6, 9, ~18
The spectral density plots of each of the spatial distributions of fertiliser application (Figure
4.27) do not show consistent periods of export between them. Phytoplankton initially applied
uniformly at a specified concentration has the most numerous and powerful period peaks.
Periods of 0.5 and 2 days appear dominant in many cases.
10 km
40 km
50 km
60 km
70 km
80 km
30 km
20 km
Figure 4.26 - Power spectrum density of export time series of 180 day pass. Analysis of final 160 days. 8 patch lengths for a fertilising period of 6 days. Power is on a logarithmic scale.
Figure 4.27 - Power spectrum density of export time series of 18 day pass of different spatial distributions. Power is on a logarithmic scale with lower cut-off at 10,000. ‘Basic’ represents phytoplankton randomness at a length scale of L/10 and uniform fertilising distribution. ‘Phyt. L/5’ represents initial phytoplankton randomness at a length scale of L/5 and uniform fertilising distribution. ‘Phyto. L/2’ represents initial phytoplankton randomness at a length scale of L/2 and uniform fertilising distribution. ‘Phyt. deterministic’ represents uniform initial phytoplankton distribution and uniform fertilising distribution. ‘Fert. L/5’ represents initial phytoplankton randomness at a length scale of L/10 and fertilising distribution at sub-patch length scale of L/5. ‘Fert. L/10’ represents initial phytoplankton randomness at a length scale of L/10 and fertilising distribution at sub-patch length scale of L/10. a) – f) are for fertilising periods of 1, 2, 3, 6, 9 and 18 respectively. Patch length, L = 40km, µmax=0.4 /d.
Appendix 1 “diffgrow2.m” script for growth-diffusion-export model
function [Cout,Eout,Nout,Et,tNdeplete,Cc,Cpua]=diffgrow2(C,Nf,Nb,K,G,T,dx,dt,steps,bound,movname,p) %diffgrow2 - diffgrow with nutrient limitation %usage: = %[Cout,Eout,Nout,Et]=diffgrow2(C,Nf,Nb,K,G[3],T,dx,dt,steps,bound,) % %solves the diffusion-growth-threshold equation: %dC/dt = K*del^2(C) + m*C - F(T,C) %over an arbitrary rectangular domain of the same size as input concentration matrix [C] with %grid size of [dx] (same for x and y) for [steps] timesteps of size [dt] % %[K] is the diffusion coefficient, and m is the growth rate - this is determined by the Monod equation: %m=G(1)*(N/G(2)+N); %where N is the total nutrient N=Nf+Nb , [G(1)] is the maximum growth rate %and [G(2)] is the half-saturation constant. %[G] contains the growth parameters and must be a three component vector % %Nutrients are put in as a background quantity [Nb] which remains constant %and a fertilized amount[Nf] which is used up as plankton grow and obeys: %d(Nf)/dt = K*del^2(Nf+Nb) - G(3)*m*C %where [G(3)] percentage nutrient usage per unit growth. % %[T] is the threshold criteria %so that: C=0 if C>=T % %[bound] defines the boundary condition (0=open 1=closed 2=periodic) %[movname] required for labelling movie %[p] for plotting and movie p=1. no plotting or movie p=0 % %[Cout] returns final concentration %[Eout] returns total removal due to threshold in a matrix same size as [Cout]; %[Nout] returns total remaining fertilizer %[Et] returns the timeseries of total export %[tNdeplete] returns time at which nutrient becomes depleted %[Cc] returns the timeseries of concentration over the entire test area %[Cpua] returns the time series of the mean concentration over the domain % %By default the plankton color range is set to [0 max(max(C),T)] and the nutrient color to max(max(Nf)); % %To view demo, type diffgrow('demo'); tNdeplete=0; %plotting switch: 1 for plotting, 0 for no plotting plotting=p; if nargin==1 if strcmp(C,'demo'); demo=1; else
demo=strcmp('Yes',questdlg('Run demo?','Demo')); end if (demo) C=rand(61);Nf=zeros(61);Nf(20:40,20:40)=0.9;Nb=0.01;K=0.1;G=[0.2 0.2 0.2];T=3.0;dx=1;dt=0.5;steps=100;bound=2; UIWAIT(msgbox({'Demo:';... 'Fertilization of a random plankton concentration with a square blob of fertilizer of concentration 0.9';' ';... 'Running demo simulation on a 61x61 grid with periodic boundaries';'using normalised parameters of:';... 'Diffusion coefficient K=0.1';'Growth parameters G = [0.2 0.2 0.2]';'Threshold parameter T=3.0;';... 'Input plankton concentration random values between 0 and 1'},'Diffgrow2 Demo','modal')); else return end elseif(nargin~=12) %fix up for different parameter numbers error('Diffgrow: Incorrect number of input parameters'); end [nx,ny]=size(C); nn=nx*ny; Cmax=max(max(max(C)),T); Nmax=max(max(Nf))+Nb; if plotting==1 Ff1=figure;set(Ff1,'NumberTitle','off','Name','Diffgrow2','Position',[50 120 780 500]); end %generate general coefficient matrix: K=K*dt/(dx*dx); m=G(1)*dt; AC=coeffmat(K,0,bound,nx,ny); AN=coeffmat(K,0,bound,nx,ny); %make plotting grid if plotting==1 [yy,xx]=meshgrid(1:dx:dx*ny,1:dx:dx*nx); end Nb=max(Nb,0.000001); Ct=reshape(C,nn,1); Nbmat=Nb*ones(nn,1); Nt=reshape(Nf,nn,1)+Nbmat; E=zeros(size(Ct)); if plotting==1 set(gcf,'DoubleBuffer','on') mov=avifile([ movname '.avi'], 'compression','indeo5', 'fps', 7); end for t=1:steps %calculate growth and m=G(1)*(Nt./(G(2)+Nt)); P=G(3).*Ct.*m; AC=spdiags(1+4*K-0.5*m*dt,0,AC);
Appendix 2 “difffert.m” script for multiple fertilisation events
function [Earea,Etlong,Cclong,Cpualong,Nzero,dt,sumsteps]=difffert(fert,L,alph,g,Nn,C,p) % difffert - runs diffgrow specified number of times % usage: = % [Earea,Etlong,Cclong,Cpualong,Nzero,dt,sumsteps]=difffert(fert,L,alph,g,Nn,C,p) % % Applies nutrients in the concentration pattern given by [Nn], a 10 x 10 matrix over an initial % phytoplankton background given by [C] (50 x 50 matrix). Nutrient fertilisation is done initially % and then at intervals after this denoted by the elements of [fert]. The maximum growth rate is % given by [g] and the [alph] denotes the coefficient in the definition of length scale dependent % horizontal diffusion (see diffgrow2.m). p switches the plotting and movie creation to on (p=1) % or off (p=0). % % [Earea] gives the spatial distribution of total export % [Etlong] returns the time series of total export % [Cclong] returns the time series of total concentration % [Cpualong] returns the time series of concentration per unit area % [Nzero] returns a vector of the times at which nutrients became depleted to zero % [dt] returns the timestep used % [sumsteps] returns the total number of timesteps evaluated movname='pres_movie'; sumsteps=0; %defining length scale dependence K=alph*L^(4/3); dx=L/10; dt=min(0.5*dx*dx/K,0.2); %growth terms G(1)=g; %maximum growth rate G(2)=0.2; %1/2 saturation 3* Nb G(3)=0.2; %uptake ratio iron % fertilizer iron in uM/m3 Nnb=0.0; %concentration of fertilisation in non-targeted cells Nb=0; %background concentration in environment Nf=Nnb*ones(50); Nf(21:30,21:30)=Nn; %fertilised cells %plankton conc in mM C m-3 T=14; %threshold Etlong=[]; Cclong=[]; Cpualong=[]; Earea=zeros(50);