The Rate of Air-Sea CO2 Exchange: Chemical Enhancement and Catalysis by Marine Microalgae.

Chapter 5: Operation of steady-state gas exchange tank

This chapter describes the routine operation of the steady-state tank, including the preparation of seawater and algal cultures, measurements of the CO2 fluxes and transfer velocities, and occasional measurement of SF6 and O2 transfer velocities, TCO2, total alkalinity, pH, and chlorophyll. The method of analysing this data will be summarised in the following chapter.

5.1 Note on evolution of the system and choice of factors to vary

The design of the system described below gradually evolved over two years. The major problems encountered and subsequent changes are included in the section elow. However, the optimum set-up and procedures may not apply to some of the earlier gas exchange measurements. Where such differences are significant they are noted in the results chapters.

In particular, the variation in temperature control and stirring motor speed caused many problems. Also, TCO2, Alkalinity and pH measurements were not made routinely until after the Phaeodactylum bloom during which spontaneous calcification must have occurred (see Section 8.3.3 ). The high-CO2 air-supply system evolved gradually, as did the standard measurement routine aimed at minimising possible errors. Trace metals and vitamins were only supplied to the Emiliana Huxleyi, Skeletonema and fourth Dunaliella cultures .

There are many factors influencing the gas transfer velocities which could be varied in the tank: physical factors - water temperature, air temperature, water stirring rate and paddle size, air stirring rate, - chemical factors - salinity, alkalinity, water pCO2 (and hence pH and TCO2 at constant alkalinity), added bovine carbonic anhydrase, - and biological factors - added enzyme inhibitors, nutrients, added zinc species of algae, presence of bacteria, surface algal films, light intensity.

Clearly there are very many possible combinations which could be investigated. As the physical factors influencing air-sea gas exchange have already been investigated in many other studies, most of these using large and more realistic wind-wave tanks or even lakes, initially it seemed sensible to hold the physical factors constant in order to focus on the effect of the enzyme and the algae. Therefore, during the algal blooms the temperature was maintained at 15C and the water stirring speed at 13 rpm (with a small paddle). This stirring speed was chosen because it gave a low transfer velocity (about 3 cm hr-1 for CO2), and the chemical enhancement which is the focus of this study is only expected to be significant at low windspeeds. A series of "blank" calibration measurements were also made for a range of stirring speeds a range of temperatures, a range of water pCO2, and a range of added enzyme concentrations, varying one factor at a time. Acidified seawater was used as a "control" in which there would be no significant chemical enhancement.

5.2 Seawater composition

5.2.1 Sources

Real seawater was used, with added nutrients as appropriate. We had several sources of seawater: the Fisheries Laboratory (MAFF) at Lowestoft, which draws water from the nearby beach through a pipe, surface water collected by bucket from the fishing boat "Florence L" at a station situated 5 miles SE from the mouth of the river Yare at Great Yarmouth , and on one occasion water collected on a MAFF ship returning from the Irish Sea. The water was stored in 30 litre carboys before use, about 50 litres were needed to fill the tank. For the 4th Dunaliella bloom real seawater had to be supplemented with some artificial seawater. This was made up with all the major ions, with bicarbonate and boric acid carefully measured to ensure the correct alkalinity.

Seawater just off the east Norfolk coast tends to have a slightly higher alkalinity than average seawater, due to mixing with fresh water from the river Yare, whose alkalinity is sometimes greater than 3meq/l (my measurements). It is also well mixed with bottom sediment, and so should contain abundant micronutrients. The pCO2 rises substantially as the water is warmed to 15C from the cold winter North Sea. I could not measure pCO2 in situ, although the temperature and approximate pH were recorded.

5.2.2 Filtering and Cleaning

For most purposes, the water was filtered from the barrel into the tank through a Millipore 0.22 micron membrane filter, which took about a day (by gravity). This should remove almost all plankton and bacteria from the water. However, the tank itself was not sterilised, because there was no straightforward way to achieve this. The internal shape is complicated, including the paddle, cooling pipe, electrode, and fan blades. Strong acid would have intefered with vapour pressure measurements made by a colleague in the same laboratory, and the tank was too large for any fume cupboard. So initially the inside of the tank was wiped instead with 60% ethanol, with the intention to kill most bacteria. However, despite allowing the ethanol to evaporate, and then rinsing the tank several times, it seems that a trace of ethanol must have remained, perhaps adsorbed into the PVC walls. When algal cultures were subsequently grown in the tank, the algae grew well to make an intense green bloom, but the water pCO2 rose rather than fell as should be expected, rapidly surpassing 5000ppm which was the limit of detection of the analyser. This occurred on three occasions, significantly delaying the schedule of experiments. Eventually we deduced that the ethanol used to clean the tank might actually be the cause of the problem. A ten micrometre layer of ethanol adsorbed to the tank would have provided sufficient organic carbon to explain the observed pCO2 rise, assuming that either the algae or bacteria which had entered the water at a later stage, were metabolising the ethanol as a food source. To test this, two carboys of unfiltered seawater with added nutrients (as below), one of them previously rinsed with ethanol, were left for a few weeks. The pCO2 in the carboy with the ethanol rose very rapidly, and much DMS (later H2S) was also produced.

Subsequently the tank was only rinsed several times with water between different cultures or seawater samples, scrubbing off adhering algae and also wiping dry before the final rinse. Inevitably a few bacteria were present, but the stock cultures were sufficiently dense that the desired species of algae had a good head-start. The presence of some bacteria may also be beneficial, recycling essential amino acids and other metabolites required by the algae (see Section 8.3.7 ).

5.2.3 Added Nutrients

The major nutrients, nitrate and phosphate, were added to the seawater for growing algal cultures. Approximately 4g NaNO3 and 0.4g NaH2PO4.H2O were added (dissolved in a little milliQ water) per 50 litre tankful, the concentrations then being similar to the f / 2 medium (see table 5-1 ) in which the starter cultures were grown. This is much higher than is typical of seawater. For growing a diatom culture, silicate was also added as in f / 2 medium. The phosphate increased the total alkalinity and had to be taken into account when checking the consistency of the carbonate system speciation (see Section 6.3 ).

It was assumed that the original seawater would contain enough trace metals and vitamins, as it was taken from well mixed coastal waters. The diatom and dinoflagellate cultures grew rapidly with no sign of any nutrient limitation. However the 2nd and 3rd Dunaliella cultures did not grow nearly so well as the first Dunaliella culture, for which the water had not been filtered. This implied some limitation, perhaps lack of vitamins which might have been recycled by bacteria in the unfiltered water. Therefore trace metals and vitamins (as in table 5-1 , measured by volumetric dilution of stock solutions) were also added to the fourth Dunaliella culture, which grew much better, as discussed in Section 8.3.7 . Lack of trace metals is unlikely as the tank contains a steel pipe, screws, and alloy air-stirrer blades.

Zinc by itself was added on a few occasions (noted in results), to ensure that this was not limiting the possible production of carbonic anhydrase. Again, the concentration added was similar to that of f / 2 medium.

5.2.4 Added bovine enzyme and/or inhibitor

Crystals of bovine carbonic anhydrase were obtained from Sigma Chemical Company in 100mg quantities, and stored in a refrigerator at <5 C. When crystalline the enzyme seemed to retain its activity over a few weeks. It dissolved easily in a small flask of seawater, although the protein was much less soluble in MilliQ water, presumably due to the lower pH. Quantities added are given with the results.

The enzyme inhibitor Acetazolamide (typically 50-100mg) was dissolved initially in 50ml 0.1M KOH which was then added to the tank, followed by 50ml of 0.1M HCl which had been titrated against the KOH to ensure equal molarity. Note the Acetazolamide is toxic and thus had to be weighed in the fume cupboard, to only +/-10mg accuracy. A similar procedure was applied on one occasion for another inhibitor, Ethoxyzolamide.

5.2.5 Lowering the water pCO2

The algal cultures all eventually brought the seawater pCO2 down to extremely low levels (sometimes as low as 5ppm), and at this time the highest gas transfer velocities were recorded (see chapter 8). This seemed to fit the prediction of the physiological model developed in Section 3.2 , which suggested that algae would make more carbonic anhydrase at low pCO2. However, as the pCO2 falls, the OH- concentration also rises, which could explain part of the increased transfer velocity. To measure this effect by itself, as a "control" experiment, I had to measure the transfer velocity in seawater with a similarly low pCO2 but without algae. Creating such a low pCO2 in seawater is not as straightforward as it may seem! It is easy to add a lot of CO2 to seawater, by gas exchange or bubbling with air containing several 1000ppm pCO2. However, there is no equivalent method for removing CO2 since the concentration in the bubbled air can go no lower than zero. This, combined with the increasing buffering capacity of the water as pCO2 falls, means that it would take several weeks (as I eventually calculated after waiting several days!), to achieve such a low pCO2 by gas exchange or bubbling. The algae can only achieve this because they are pumping in HCO3- ions to store as organic carbon, or perhaps dehydrating HCO3- with extracellular carbonic anhydrase. Note that despite the pCO2 falling from 350ppm to 5ppm, the total inorganic carbon does not even halve (see carbon budget plots in chapter 8).

So this problem was circumvented by first acidifying the water, sufficiently to raise the pCO2 to several thousand ppm, removing most of this by rapid gas exchange with air in the tank, and then adding the equivalent amount of alkali to restore the original alkalinity, at which point the pCO2 would fall to a much lower level. The acid and alkali were of course titrated one against the other to ensure that the additions cancelled. As the final pCO2 is quite sensitive to the initial alkalinity of the water, to achieve a particular desired pCO2 this procedure had to be repeated several times, with diminishing quantities of acid and alkali.

One possible drawback of this method is that the pH changes might irreversibly affect any dissolved organic carbon in the water, perhaps changing the gas exchange properties. However, it was not necessary to lower the pH below the bicarbonate-buffered range (i.e. below pH 6), so the change is not too dramatic.

An alternative method used for the first set of low-pCO2 blank measurements, involved heating the water to drive off the CO2 (which is less soluble in hot water) and also to raise the gas exchange rate, and then cooling back to normal temperature after equilibration with CO2-scrubbed air. However, the acid method is more straightforward.

5.3 Source of Algal cultures

5.3.1 North Sea spring bloom samples

The spring bloom in the North Sea was sampled four times (30th April, 5th June, 19th June, 3rd July 1996) from the fishing boat "Florence L", with the intention of measuring the effect of any enzyme produced by the algae in the sea itself. Therefore the time interval between collecting the water and transferring it to the tank was kept to a minimum (about 3 hours), as were temperature changes in transit. The 30 litre carboys were lowered over the side of the boat and opened at top and bottom, to minimise CO2 exchange with the air. Each sample was collected at the same location 5 miles SE of the mouth of the river Yare in 28-30m depth of water, although the tidal and weather conditions varied. More details are given for each sample in the results Section 8.3.5 .

5.3.2 Preparation of cultures

The algal cultures in the tank were seeded from starter cultures, which were kindly prepared for me by Gill Malin, or sometimes Gareth Lee. Usually two flasks were prepared, each containing 400ml of "f / 2" culture medium (see table 5-1 ) based on filtered and autoclaved seawater from Lowestoft. These flasks were each inoculated with 40ml of preculture, and then incubated in the light at 10-15C for 10-15 days, until they were visibly well coloured. On one occasion the pCO2 in these flasks was checked crudely by means of a syringe sample passed through the LiCOR analyser, and was in the range 70-100ppm, just before adding to the tank. Sometimes the tank pCO2 was deliberately lowered to a similar level by gas exchange before adding the algae, to reduce the physiological disturbance, but time did not always permit this. The temperature change was usually small on transferring to the tank, and the nutrient concentrations similar.

The culture was thus diluted 110 fold (for 1 flask). From this point it took about 2 weeks to reach a similar density as the original culture, i.e. an average doubling time of about 2 days, although the carbon budget plots in chapter 8 suggested linear rather than exponential growth.

5.3.3 Light

At first a tungsten desk lamp was used to illuminate the tank, but this was not sufficient to grow an algal bloom. Instead a bank of five 60cm-long fluorescent strip lights was used at a height of about 20cm above the tank, surrounded by angled aluminium foil reflectors to direct the light into the tank. The Photosynthetically Available Radiation (PAR) was measured using a hand-held meter. This gave values much lower than full sunlight, but within the range used for algal cultures. In any case, this light proved quite sufficient for the growth of intense algal blooms. I decided to leave the lights on continuously, because a diurnal cycle would have added yet another variable to the interpretation of the gas exchange measurements, both through algal physiology and through radiation affecting the temperature and dynamics of the surface microlayer.

5.4 Chlorophyll measurement

Occasionally I took samples for chlorophyll analysis to give an order-of magnitude indication of the cell density during the blooms, to compare with calculated organic carbon uptake by the algae. During the final stages of the blooms some of the algae had settled out or grew on the surface, so chlorophyll samples were taken before this aggregation was noticeable.

Water was extracted from the tank with a 100ml syringe, and a known volume (depending on algal density) was then slowly forced through a Whatman glass fibre filter, the water being returned to the tank. The filter papers were folded, wrapped in aluminium foil, and frozen until analysis. Four samples were taken on each occasion. One day prior to analysis each filter paper was placed in a vial containing 10ml 90% acetone, into which the chlorophyll was extracted

The samples from algal cultures were analysed with a spectrophotometer. The absorbence was measured at 630, 645, 665 and 750 nm wavelengths, and the 665 and 750 nm measurements were repeated after acidification by 2 drops 8% HCl (which knocks the Mg atom from the centre of chlorophyll-a leaving phaeophytin). The instrument calibration was checked with a chlorophyll standard.

The chlorophyll concentration in the extract was calculated according to the formulae of

SCOR: [Chl a] = 11.6 (A665 - A750) -0.14 (A630 - A750) -1.31 (A645 - A750)

and Lorenzen: [Chl a] = 26.7 [(A665 - A750) - (acidified A665 - acidified A750)]

The mean and standard deviation were calculated for each set of four samples. The results are shown on the carbon budget plots in chapter 8, which show both the SCOR and Lorenzen figures (typically these are very close). The standard deviations ranged from 1% to 16%. The relationship between the measured chlorophyll, and biological carbon calculated by mass balance (see Section 6.7 ) is shown in figure 8.1.

The North Sea spring bloom samples were more dilute and had to be analysed on a fluorimeter (Turner model 112, excitation filter BG12, emission filter RG630). Fluorescence was measured before and after acidification as above, the acidified reading being subtracted from the unacidified reading. The ratio of these readings gives an indication of the phaeophytin concentration. The fluorimeter was calibrated using a set of 5 dilutions of a chlorophyll standard (checked with the spectrophotometer as above), spanning the sample range. The mean and standard deviation were calculated for each set of four samples, and the standard deviation ranged from 3% to 47%, the latter from the 30th April sample where chlorophyll was still low.

5.5 pCO2 measurements

5.5.1 Overview of LiCOR Non-dispersive infra-red gas analyser

The LiCOR analyser compares the amount of infra-red radiation absorbed by an air-sample, with that of a CO2 reference gas, as illustrated by figure 5-2 . The absorption corresponds to transitions between vibrational and rotational energy levels of the CO2 molecule, the same process as causes the "greenhouse effect" in the atmosphere. A filter selects the wavelength range 4.26 ± 0.15 mm before detection. The same light source and detector are used for both sample and reference, a rotating chopper disc directing the light to pass through each in turn. The cell volume is about 12cm3 and the inside of the cell is gold-coated to maximise internal reflection. The trapped gas surrounding the chopper motor and detector is purged of CO2 and water vapour by a bottle containing soda lime and magnesium perchlorate, which was changed regularly according to manufacturers directions. Gas flows continuously through the cells, and the response time is very quick, about 0.1 seconds. The output range for the difference between sample pCO2 and reference pCO2 is actually just over +/-5000ppm, although the factory calibration is specified as only for +/- 3000ppm. The specified accuracy is 1ppm for concentrations similar to air, and the repeatability about 0.2ppm.

The analogue electrical output of the analyser is proportional to the difference between the detector response for the sample cell and that for the reference cell. This response is not linear, but has been calibrated as a third order polynomial. The analyser contains sophisticated electronic circuitry which converts the analogue output into a digital reading of pCO2on a display, which can be passed on to a computer. A correction for sample cell temperature is made automatically, but not for atmospheric pressure, which will be considered later.

5.5.2 Water vapour

Our model "6262" LiCOR analyser also measures water vapour by the same method, detecting absorption instead at 2.59+-0.15 mM. Water vapour interferes with the CO2 measurement by changing the "pressure broadening" of the CO2 infra-red absorption band, so the water vapour measurement is used to automatically correct the CO2 measurement for this effect.

The LiCOR could also make an automatic correction for the dilution of the dry air by water vapour, to provide the pCO2 as a mole fraction of dry air rather than the actual pCO2 of the sample. However, in these gas exchange experiments the air-water gas flux is calculated from the pCO2 flowing in and out of the headspace, multiplied by the flow rate, so clearly the actual pCO2 rather than the dry-air pCO2 is required and the correction was not applied. It was necessary to wait for the H2O reading to stabilise before noting the pCO2 reading, thus ensuring that the inlet filter had equilibrated with the H2O in the sample air stream and hence was not adding or removing water vapour.

5.5.3 Reference gases

We had three reference gas cylinders of CO2 in nitrogen kindly made up for us by David Cooper, then at Plymouth Marine Laboratory. Our standards were determined by another LiCOR analyser in PML, which was itself calibrated with CO2 standards traceable to NIST standards. The three cylinders contained 272.2 ppm, 438.3 ppm, and 2010 ppm CO2. The pressure was initially about 140 bar. One of these, usually the 438.3 standard, was flowing slowly through the reference cell whenever measurements were being made. Others were used from time to time to set the zero and span controls of the analyser (manually adjusted). Typically, the zero was set by flowing the 438.3 standard through both cells (in parallel to ensure equal pressures), and the span set using either pure nitrogen (i.e. zero CO2) or sometimes the 2010 standard when high pCO2 concentrations were being measured. However, the 2010 standard was beyond the range of the PML standards and so was less well defined. Our analyser, with the zero and span set by the 438.3 and 272.2 standards, then read about 2000ppm for the high standard.

It should be stressed that when the gas exchange rate is calculated from the steady state flows through a tank headspace, errors in the calibration of both the zero and the span cancel entirely, so long as the response to pCO2 remains linear. For the same reason, an atmospheric pressure variation also cancels (see also Section 4.3 ).

5.5.4 Computer readout.

The output from the LiCOR was fed continuously into a computer, using a basic program which I wrote specifically for this purpose. In normal operation, the concentrations of CO2 and H2O, a digitised external analogue input (usually from the oxygen electrode but sometimes from the SF6 GC), and value of the LiCOR's own timer were passed to the computer 5 times a second, and displayed as a graph on the screen. The CO2/ H2O/external input (vertical) and time (horizontal) scales of this graph for each gas could be changed at any time, and the graph wrapped around on reaching the end of the screen. The data was not saved in memory, but could be saved continuously onto the hard disc when required. The program could also integrate under a peak of any of the inputs, the user hitting a key to start and end the integration and a straight baseline between these points being subtracted if desired. A typical "snapshot" of this computer program in operation is given in figure 5-4 , corresponding to the "normal procedure" for measuring CO2 transfer velocities described in

5.5.5 Contamination problem

The inflowing gases passed through air-filters to prevent dust and aerosols entering the analyser. The sample and reference cells in the LiCOR are heated to about 30C in order to prevent condensation of water vapour inside the cell, which would drastically affect the infra-red absorption. Despite these precautions, several times it seems that the sample cell got contaminated such that it was impossible to set the zero and span controls. This was prone to occur whenever the tank water temperature exceeded the lab air temperature, even by just one degree, creating condensation in the tubes leading to the air filters.

Rather than losing several weeks by sending the analyser back to the USA for servicing, it was decided to wash out the sample cell by rinsing and drying it several times. To rinse the cell about a litre of warm MilliQ water was slowly siphoned through. Initially this also contained a mild detergent. Between rinses the cell was dried with a fast flow of nitrogen gas. Eventually the contamination was reduced sufficiently to bring it back within the range of the zero and span controls for CO2. For water vapour the zero was sometimes still slightly out of range - in this case the analyser had to be told that the water vapour concentration in the reference gas was slightly negative, thus it would calculate the correct value for the sample gas. This should not have significantly affected the water vapour correction to the pCO2.

5.5.6 Memory problem

The analyser also developed a puzzling "memory" for the gas which had previously been flowing through it. This is best illustrated by a typical sequence of events. For example, if I had been measuring 600 ppm for half an hour or so, and then introduced a different flow of 300ppm, the reading might fall rapidly to 310ppm but then only slowly settle down to 300 ppm. On the other hand, if the first gas had been 100ppm, the 300ppm gas would initially read about 295 ppm. If the flow of the second gas was stopped soon after being introduced, the pCO2 would creep back up almost to the level of the first gas.

It was clear that this effect was not caused by laboratory air leaking into the flow, because the direction of the memory effect depended on the previous sample. Also, this effect could not be due to slow equilibration in the filter, because if the flow was stopped altogether, the gradual pCO2 change continued despite the filter being some 30cm upstream, whereas on restarting the flow the pCO2 in the intermediate tubing was found to be unchanged.

This behaviour might be explained if a small reservoir of trapped air somewhere within the LiCOR slowly equilibrated with the air in the sample cell. A substance which could absorb or adsorb CO2 would have a similar effect. Exchange with the reservoir would then have greater effect on the sample cell when the air flow was slow, as observed.

This memory effect was frustrating because it forced me to wait many minutes before obtaining a stable reading from a slow gas stream such as that out of a tank headspace. Despite this delay, the memory inevitably introduced a slight error. This problem has been seen by others using these LiCOR analysers (David Cooper, personal communication), but would probably not be a problem during routine pCO2 measurements at sea, for example. This is because there is not a large rapid change of pCO2 in the sample cell (the range of pCO2 measured is much smaller and changes gradually), and also because the gas streams would typically flow much faster during continuous field measurement.

5.5.7 Pressure variation

The LiCOR analyser measures the CO2 concentration in the sample cell, not the CO2 mole fraction, although it's output is expressed as a mole fraction (parts per million). This output is calculated from the concentration using a value for atmospheric pressure stored in memory. For some applications this value can be set by an external pressure sensor. In this case, however, a fixed value was used, i.e. p = 105 Nm-2.

This is justified because the atmospheric pressure cancels in the calculation of the transfer velocity, the key quantity of interest. This was demonstrated by the equations in

5.6 Gas flow rates

Gas flow rates were typically between 50 and 100 ml min-1 and were measured using improvised bubble flow meters based on glass burettes. Although these are very simple, they provided a sufficiently accurate flow measurement which did not require any calibration (assuming the burette is reliable). Soap solution from burst bubbles collected in a trap to avoid contamination of the tank. The flow was measured between the CO2 supply system and the tank (see Section 5.7 below, also figure 5-3 ), thus ensuring a constant pressure drop. It was assumed that the flow out of the headspace must be the same as the flow into the headspace, although the outflow was more difficult to measure. This is due to the large volume of the headspace, and the correspondingly large volume of air which must be introduced to increase the pressure in the headspace sufficiently above the atmospheric level to push the bubble, once introduced, up the tube. Such pressure differences are tiny, but even so they take a while to achieve at low flow rates.

A stopwatch was used to measure the time taken (typically 10 - 30 s) for the soap film to move either 25 or 14 cm3. Readings were repeated until at least three agreed to within 0.1s.

The flow of the reference gas was also monitored to ensure slow flow, such that the LiCOR cell was flushed but the cylinder was not depleted too fast.

5.7 Gas flow system for supplying air with controlled pCO2

The design of the steady-state tank required two streams of continuously flowing air, one with pCO2 greater than that of the water, and one with pCO2 less than that of the water. Both the pCO2 and the flow rate of these two air streams would need to be varied, so as to minimise the error in the transfer velocity, and to control the water pCO2.

The gas flow also had to be comparable to the expected flux across the air-water interface. Calculations showed that continuous operation at such flow rates would soon exhaust CO2 reference cylinders. Therefore the high-CO2 supply would have to be made by diluting pure CO2 into a gas stream. Also, the gas could not be made up with pure nitrogen, as oxygen was required for the growth of the algae. Therefore the basis of the supply had to be air pumped in from outside the building. As the diaphragm pump produced a rapidly oscillating pressure, a 2-litre bottle was inserted after the pump to remove this oscillation. Since the pCO2 of such air varies (mainly due to traffic on the road outside), the incoming air was first scrubbed with sodalime to remove the CO2. This stream was split in two, and CO2 added to one stream only, the other having a pCO2 near to zero.

The quantity of CO2 required from the pure-CO2 cylinder was extremely small, and could not be delivered by a high-pressure regulator, particularly as high-pressure CO2 does not behave as a perfect gas. A double-dilution system was therefore devised, as depicted in figure 5-3 .

The pressures were regulated with the valves, to ensure such that the gases flowed were in the required direction, which is generally from left to right in the figure, except where indicated otherwise. A stream of pure CO2 was vented through a valve (A), upstream of which a smaller flow was diverted through a 3-way needle-valve (B) into a stream of air. Most of this stream was also vented through a valve (D), but another 3-way needle -valve (C) allowed a small flow into the high CO2 branch of the scrubbed air stream. Upstream of this, the flow rates of both the scrubbed air streams were controlled by two more valves (E,F). Thus it was possible to control the pCO2 and flow rates independently.

To achieve the correct direction of flow the relative pressures must be maintained in the order: pure CO2 > air venting through valve D > air leaving valve E (towards the tank). To control the pCO2 produced by this double dilution system required much trial- and error, aided by the pCO2 trace preserved on the computer screen, two pressure gauges, and a ball flowmeter for the venting air.

A slight pCO2 oscillation was always produced, which became large if any one of the valves was dominating the pressure drop across the system. Generally the amplitude of this oscillation was about 1% of the average pCO2, and the wavelength a few minutes. The average pCO2 was noted. This was done by eye if the oscillation was regular, or by integrating the pCO2 on the computer if the oscillation was particularly irregular. Thus the error was smaller than the amplitude of the oscillation. When the error seemed large, this was noted and incorporated in the error calculations (see Section 6.5 ).

Typically the pCO2 of this high-CO2 air supply was about twice the atmospheric level, i.e. 700ppm. Sometimes a higher pCO2, up to 3000ppm, was used to raise the pCO2 of the tank water, or to offset vigorous biological uptake.

Because the measurement procedure requires frequent rearrangement of the "plumbing" (see Section 5.8 ), plastic tubing and connectors were used for connections in and out of the tank. Note that these tubes are all very close to atmospheric pressure. Stainless steel tubing and Swagelock fittings were used for the upstream air supply system and reference gases which were at higher pressures.

5.8 Normal procedure for CO2 gas exchange measurement

The normal procedure for measuring CO2 gas exchange was as follows

Assume for this example that Headspaces A and D are the efflux and influx headspaces respectively, and the air leaving A is already flowing through the LiCOR sample cell:

  1. Watch trace on computer (last 2 hours) until Headspace D appears to reach steady state
  2. Flush out reference cell with fast flow from reference cylinder, then reduce to very slow flow once pCO2 stabilised. Calibrate LiCOR if first measurement this day.
  3. Meanwhile measure flow rates of gases into headspaces
  4. Note pCO2 reading from headspace D.
  5. Connect headspace A outflow to LiCOR, note pCO2 reading once stabilised.
  6. Connect headspace B in closed loop to LiCOR and back into headspace. Pump air around loop until reading stabilises. Switch off pump, note pCO2 reading.
  7. Connect headspace C in closed loop to LiCOR and back into headspace, using same tubing as for B (B has thus been used to flush the tubing). Pump air around loop until reading stabilises. Switch off pump, note pCO2 reading. This reading indicates the water pCO2.
  8. If pCO2 in the water is low, check LiCOR calibration using Nitrogen, note pCO2 reading once stabilised and subtract from all other measurements.
  9. Connect headspace A inflow to LiCOR, note pCO2 reading once stabilised. Reconnect inflow to headspace.
  10. Connect headspace D inflow to LiCOR, note pCO2 reading once stabilised. If this comes from the high-CO2 supply, observe a few cycles of the wobble to determine average (see Section 5.7 ). Reconnect inflow to headspace.
  11. Measure tank water temperature and motor speed (and at regular intervals to check they are constant)
  12. Calculate 60 * (flow rate) * (pCO2 in - pCO2 out) / (pCO2 out - pCO2 water) for each headspace. This is approximately 1000 times the transfer velocity, and should be similar for both headspaces. If not why not - check for problems.
  13. Make any changes to pCO2 inputs, flow rates etc. Air flows should match transfer velocity such that pCO2-out is approximately half way between pCO2-in and the water pCO2 (this minimises the error). pCO2-in can be used to control changes in water pCO2.
  14. Reconnect headspace A outflow to LiCOR to monitor adjustment to next steady state.
  15. Take any samples for measuring TCO2, Alkalinity, pH, Chlorophyll etc.

Note that the flowrates are measured at the beginning. This avoids any change due to the small pressure drop across the LiCOR and tubing. The order of the measurements may be changed in order to minimise the change in pCO2 between readings. This helps to reduce the "memory effect" ( Section 5.5.6 ).

Figure 5-4 shows a typical trace from the computer program described earlier ( Section 5.5.4 ), following this normal measurement procedure. The numbers in circles have been superimposed to indicate the stages in the procedure as listed above. The vertical scale (pCO2 range) has been changed several times, in order to show in detail the gradual equilibration within the LiCOR cell, due to the memory effect described in Section 5.5.6 . These vertical scales are shown at the bottom of the figure. In this example the steady-state pCO2 in D is about 400ppm, the steady-state pCO2 in A is about 200ppm and the equilibrium pCO2 is about 300ppm. The N2 cylinder gives 3.5ppm, the scrubbed low pCO2 input is 8ppm, and the high pCO2 input is 600ppm with a slight oscillation of about ± 3ppm.

Note the effects of varying the gas-flow rate: when the reference gas is first flushed rapidly through the reference cell, the trace for the sample gas drops instantly with no memory effect (in this case we assume the gas previously in the reference cell had a pCO2 just slightly lower than the reference gas, thus raising this pCO2 causes the sample gas trace to drop accordingly). However when the flow rate is slowed (to preserve calibration gas) the trace jumps back up and gradually falls again. When the equilibrium pCO2 is being sampled (headspaces B and C) the pumped flow is sufficiently fast to cause a slight drop in pressure and hence pCO2, but when the pump is switched off the reading jumps back up.

5.9 Total CO2

Total CO2 in a water sample is measured by acidifying the water such that all the CO32- and HCO3- is converted to CO2 , which is then extracted in a stream of nitrogen and measured using the LiCOR analyser.

Nitrogen gas was set up to bubble through the sample, through the LiCOR, and through a flowmeter. This system is illustrated in figure 5-5 . The flow is kept constant, while the pCO2 reading from the LiCOR is integrated by the computer. The procedure was as follows:

TCO2 = 10-12 *(1/1.035) *10 *integral *flowrate *air-pressure /( 60 * R * air- temperature)

Units and conversion factors in this formula are given in the table below:


mol kg-1


ppm. s


ml min-1

Air Pressure

kg m-1 s-2

Air Temperature


R (gas constant)

kg m2 s-2 K-1 mol-1


ppm => mol fraction

ml => m3

1/ 60

min-1 => s-1


l-1 => kg-1

TCO2 measurements were not made every day, but were used for occasional comparison with the TCO2 calculated from pCO2 and Alkalinity (see Section 6.3 and figure 6-1 ). This provided a check on the internal consistency of the carbonate system.

Repeated measurements showed that the error in measured TCO2 was approximately 2%. This is a combination of variation in the flow rate and error in the pCO2 beyond the calibration range of the reference CO2 standards. The flow rate (typically 300ml / min) was adjusted such that the maximum pCO2 was less than 3000ppm, the linear range of the LiCOR specified by the manufacturer. Note that the integration had to be done by computer, as the LiCOR's internal integrator had a slow response and insufficient memory.

Whilst it is possible to make much more accurate TCO2 measurements (see Section 1.3.2 , Section 3.4.2 , and references therein), this method was convenient and sufficient for the purposes of these experiments in which the observed changes in TCO2 were large.

5.10 Alkalinity and pH

Total Alkalinity was measured by a simple titration, whose endpoint was found using a pH electrode.

The initial pH was measured immediately after taking a 100ml sample from the tank by syringe, thus minimising gas exchange. The pH electrode was calibrated in pH 7 and pH 9 buffers before and after each sample, and was readable to the nearest 0.025 pH units. The sample was then titrated with 0.01M HCl from a burette. The pH was measured at 1ml intervals between 20ml and 30ml. Extrapolating between points, the volume of acid required to bring the sample to pH 4.5 was used to calculate the Total Alkalinity. The 0.01M HCl was made up by volumetric dilution of commerical AnalaR 1M HCl and then titrated against 0.01M KOH (diluted from commericial AnalaR 1M KOH) to confirm its molarity. The two solutions matched to within 1%.

This simple method is based on the definition of Total Alkalinity but is limited by the accuracy of the pH electrode. A more accurate method would be needed if alkalinity were used to calculate pCO2, as a small change in alkalinity corresponds to a large change in pCO2. However, as pCO2 was measured directly, these alkalinity measurements were only required to check that there was no significant change in Total Alkalinity during the course of an algal bloom in the tank. Air-water CO2 exchange and photosynthetic uptake or respiration of CO2 do not affect the total alkalinity, since the contribution of any extra bicarbonate and carbonate ions created by dissociation of carbonic acid (hydrated CO2) is cancelled by the hydrogen ions also created. Only removal of carbonate ions through precipitation of solid CaCO3 would significantly affect the alkalinity. In most cases no significant change was observed and the average of the measurements was used for calculating the carbonate system speciation, as described in Section 6.3 . Where the alkalinity did change (during the Phaeodactylum bloom, and after any addition of phosphate) this was incorporated into the calculations as described later (e.g. Section 8.3.3 ).

For the purpose of determining the speciation of the carbonate system, the pH was the least accurate of the four measured parameters (pCO2, TCO2, Alk, pH) and was therefore plotted on the graphs for comparison, but not used in the calculations.

5.11 SF6 gas exchange

SF6 transfer velocities were measured occasionally by following over time the gradual transfer of SF6 from the water into a closed headspace, after an initial spike had been added to the water. The principle behind the method has already been derived in Section 4.5.2 . Here the practical procedure is described and some typical plots used to calculate the transfer velocity are shown.

5.11.1 Spiking the tank to create a disequilibrium

The quantity of SF6 in the spike had to be such that when diluted in the tank headspaces, the final concentration would be less than 11 parts per billion, the top of the range for which the Gas Chromatograph could calibrated (see below). The dilution of pure SF6 by a factor of 108 or more is made possible by its very low solubility in water ("a" is typically about 0.003 depending on temperature). Diluting 60 microlitres of SF6 saturated seawater into all four headspaces of the tank (volume 45 litres approx), gives a final concentration of about 4 ppb.

However, it was critical not to allow any of that pure SF6 to contaminate the lab, or other labs in the building where parts per trillion of SF6 were measured in background air samples. Therefore the SF6 cylinder was only ever opened outside well away from air ducts. 50 ml of SF6 was let into a gas syringe containing 50ml of seawater (and no air). This was shaken vigorously for a few minutes to ensure equilibration. The SF6 gas was then driven out and a 200 microlitre syringe inserted into the gas syringe to withdraw a small sample of SF6 saturated water, whilst being extremely careful to avoid taking in any bubbles of gas. The syringe water temperature was then noted. The microlitre syringe was brought inside and used to spike the tank water with the required volume, again being careful to avoid forming any bubbles. Note that just 0.2 microlitres (a very tiny bubble) of pure SF6 would be equivalent to the whole quantity injected.

5.11.2 Measurement procedure

A six way valve was used to isolate the sample in a small sample loop before connecting it in line with the GC column. This ensured that the same sample volume was always used. A pump was used to flush air from a tank headspace through this sample loop and back into the headspace again, as shown in figure 5-6 . Before entering the sample loop, the air was dried by passing through a tube packed with magnesium perchlorate. The pump was operated for at least 30 seconds to ensure equilibration within the pipes, and then switched off for 15 seconds to allow the sample to return to atmospheric pressure, before switching the six-way valve. The sample was then taken up in a stream of nitrogen carrier gas flowing through the GC column.

The set up of the GC was very simple. The column was made from a 5m length of 1/8th inch stainless steel tubing packed with molecular sieve 5A. The SF6 was detected with an electron capture detector maintained at 100C, which was connected to a chart recorder and also to the computer. The attenuation was set to 2, the "range" to 1. The oven temperature was kept just above room temperature at 26C, because at higher temperatures the SF6 peak rose up the tail of the large oxygen peak. To resolve much lower concentrations of SF6 such as ambient background levels, oxygen would have to be removed (e.g. by cryogenic trapping). However for this spiked system the SF6 peak from dry air was clearly distinguishable and appeared after about 12 minutes, on a sloping baseline following the oxygen peak. The peak height measured by hand from the chart recorder trace was compared with the peak area integrated by the computer (both taking into account the sloping baseline), and the correlation was extremely good. Thereafter only the chart recorder was used, to free the computer for CO2 and oxygen measurement.

5.11.3 Calibration

Three calibration gases of SF6 in Nitrogen were provided by Phil Nightingale, at 11.1 ppbv (± 1.4%), 3.15 ppbv (± 1.3%), and 0.826 ppbv (± 3.0%).

However, it was generally more convenient to use fractions of the highest calibration gas, diluted with air in a gas syringe, and then injected into the sample loop through the magnesium perchlorate. The air mixed in provided an oxygen peak similar to (although slightly smaller than) that of the oxygen in the sample air. The calibration curve was non-linear as expected for an electron capture detector, but the fit was good as shown in figure 5-7 . Note that the peak height increases rapidly at about 11 ppb (right hand plot). The best fit curves are either quartic or cubic polynomials calculated by the computer. These formulae were then used to estimate the SF6 concentration in tank samples.

The peak height was found to be very sensitive to column temperature variation, even of just one or two degrees. Initially a formula was used to compensate for this, based on a series of calibrations at a range of temperatures from 25-30C. Later, temperature control was improved and reference gas samples were interspersed with samples from the tank to ensure a reliable calibration under identical conditions. To avoid large extrapolation, tank samples and calibration gases were usually alternated, the dilution of the calibration gas being chosen to match approximately the expected concentration of the next tank sample. Typically, three tank sample peaks and 2 calibration peaks might be recorded every hour.

5.11.4 Calculation of transfer velocity

Typically for one determination of the SF6 transfer velocity, about 12 tank samples were measured, interspersed with calibration measurements, over a period of 3 - 8 hours. The calculation method is based on the equation derived at the end of Section 4.5.2 . Peak heights were typed into a spreadsheet from which the logarithm of [SF6 (end) - SF6 ] was plotted against time, and a least-squares linear regression was used to derive the best fit line, from whose gradient the transfer velocity can be determined. Figure 5-8 shows all the plots used to derive the transfer velocities reported later in table 7-2 ( Section 7.4.1 ). It can be seen that the points generally lie well on a straight line.

5.11.5 Uncertainty regarding the equilibrium SF6 concentration

However for each plot in figure 5-8 , two sets of points and lines are shown. The circles show [SF6 (end) - SF6 ] calculated using the equilibrium SF6 concentration (SF6 (end) ) calculated from the known spike that was added to the tank. The diamonds show [SF6 (end) - SF6 ] calculated using the equilibrium SF6 concentration measured in the headspace about 24 hours after the spike had been added, by which time it was assumed that the system would be almost at equilibrium. The figures shown on the plots in figure 5-8 were calculated using the measured endpoints. The values of both endpoints and the transfer velocities calculated from them will be given later, in table 7-2 ( Section 7.4.1 ). Here it is sufficient to observe that the two lines in each plot sometimes have quite different gradients. Clearly there is a problem as the calculated and measured equilibrium concentrations ought to be the same. However there was no consistent pattern: sometimes the measured endpoint is higher than that predicted from the spike, sometimes it is lower.

To check the assumption that measurements the following day should give a reasonable approximation to the equilibrium level, we can recall that the time constant for exponential approach to equilibrium as derived in Section 4.5.2 is k / d*. The transfer velocity k was typically in the range 1 to 3 cm hr-1 and the effective depth d* is 10.94 cm2 (see figure 4-5 ), so the e-folding time should be about 3-10 hours.

Since the calibration curves seem quite good and there is not too much scatter between the points in figure 5-8 , it seems more likely that the problem lies in the estimation of the amount of SF6 added in the spike.

Here there are several possible sources of error - for example under or over saturation of the initial seawater in the gas syringe, or uncertainty regarding the temperature of this seawater at the time of saturation (recall this was done outside the building to avoid contamination, but with water which had been stored inside the building at a different temperature). This has to be known in order to calculate the solubility and hence the concentration of SF6 in the injected spike. Sometimes a tiny air bubble may have been trapped in the syringe and injected into the tank, although care was taken to avoid this.

This discrepancy was not resolved, despite attempts to refine the measurement and spiking techniques. Transfer velocities calculated using both endpoints are shown later in table 7-2 and figure 7-4 , and are discussed in Section 7.4.1 . It can be seen that neither set fits well enough to the corresponding oxygen transfer velocity measurements to be useful in determining the "Schmidt number dependence" of the gas exchange regime in the tank, as initially hoped. However, the response to varying temperature provided an alternative means of gauging the "Schmidt number dependence" ( Section 7.4.2 )

5.12 Oxygen gas exchange

The oxygen transfer velocity was measured occasionally, in a similar way to that for SF6, by creating a disequilibrium and then following the gradual reequilibration over time. In this case, however, measurements were made in the water phase using an oxygen electrode. The principle behind this approach has already been discussed in Section 4.5.3 . Here, the practical procedure will be given. The bubbling procedure described below would clearly be disruptive for any algae growing in the tank, and would also affect any surface films. However since the oxygen concentration in the water phase would be affected by any biological activity, and is not buffered chemically like CO2, the transfer velocity of oxygen measured using a probe in the water would be inaccurate anyway in such circumstances. Therefore the main reason for measuring the oxygen transfer velocities was for comparison with SF6 and CO2 in control experiments to attempt to determine the "Schmidt number dependence" of the transfer velocity in this particular tank.

5.12.1 Creating a disequilibrium

The disequilibrium was created by bubbling either nitrogen (for an invasion experiment) or pure oxygen (for an evasion experiment) through the water for about 1/2 hour, while the headspaces were open to the atmosphere. All four headspaces were then rapidly flushed with normal air pumped from outside the building. This flushing could be followed by measuring the pCO2 in the air leaving the headsapces using the LiCOR analyser. Since the bubbling gases contained no CO2, when the pCO2 was back to "normal" (about 350ppm) it could be assumed that the oxygen in the headspace air was also at the "normal" level of 21%. After this the headspaces could either be closed or set up for measuring steady-state CO2 transfer velocity as usual. Since the dimensionless solubility of oxygen in water is only about 0.03, the small quantity of oxygen entering or leaving the water during reequilibration would not significantly affect the concentration in the headspaces which could be assumed to be constant.

5.12.2 Measurement

The Hach oxygen electrode was suspended in the water a few centimetres below the surface, just above the stirring paddle, in order to maximise the agitation of the water at the electrode tip. The millivolt output from the oxygen meter supplied with the electrode was connected to the external input of the LiCOR CO2 analyser, which converted it into a digital reading. These digital readings were sent to the computer and saved as a file on the hard disk, with one reading being recorded every few seconds. The frequent measurement is necessary because there was a significant rapid variation in the measured oxygen concentration, (as can be seen from the scatter in the plots of figure 5-9 ). This variation is due to variation in the turbulence in the water as the paddle swept around the tank. Since the electrode actually consumes a little oxygen during operation, this must be constantly replenished by turbulence in the water, hence the effect of variable agitation.

It was checked that the response of the electrode was a linear function of oxygen concentration - thereafter no calibration was necessary since neither the zero nor the span coefficients would affect the gradient of the logarithmic plots which is used to calculate the transfer velocity.

5.12.3 Calculation of the transfer velocity

The computer could be left recording the oxygen measurements overnight, until sufficient time had elapsed that the system had effectively returned to equilibrium (the timescale is similar to that for SF6 as calculated above). The file containing thousands of measurements was then imported into a spreadsheet which was used to calculate the quantity
ln [O2(time t) - O2 (end)] and plot this as a function of time. Eight such plots are shown in figure 5-9 . Note that the scatter in the plots seems to increase with time -this is just a function of the logarithmic vertical scale since as the log values decrease, they correspond to smaller and smaller measured differences between O2(time t) and O2 (end).

A linear least-squares regression was fitted to the data by the computer (excluding the end portion), and from the gradient of this best fit line the transfer velocity can be calculated using the formula already developed in Section 4.5.3 . As for SF6, the equilibrium O2 concentration is needed for this calculation, but in this case it is easy to tell when the measurements have stabilised by looking at the recorded data (before taking the logarithm), and in any case the equilibrium is always close to the normal level in the atmosphere.

The actual transfer velocities calculated from these plots will be reported in table 7-2 and discussed in Section 7.4.1 , as for SF6.

5.12.4 Some problems with the oxygen measurements

Although the plots in figure 5-9 sometimes show considerable scatter, this should not be a problem due to the large number of datapoints used to calculate the best fit line. Of greater concern are the kinks in the whole dataset -particularly apparent in the plots for 21st September and 13th January. In the latter case this was caused by a sudden change in the speed of the water-stirring motor, but no such change was observed in the former case. For these plots two separate best-fit lines were calculated for the different section f data. It was also noted that sometimes the reading from the oxygen electrode seemed to jump suddenly upwards or downwards beyond the normal range of variability, and then return a little later, as can be seen in some of the plots. This may have been due to small bubbles of air trapped in the electrode, or possibly microbiological activity at the electrode tip.

During the influx and efflux measurements made on 4th and 5th March there was considerable biological activity in the water (see Section 8.3.3 ). This probably affected the equilibrium concentration of O2 in the water sufficiently to explain the slight curve apparent in the plots and the different gradients.

Generally, the oxygen transfer velocity measurements were also not sufficiently accurate to act as a "control" from which to calculate the unenhanced transfer velocity for CO2, which was calculated instead using acidified seawater. However, both the O2 and SF6 transfer velocities did remain within the "normal" range even when the CO2 transfer velocity measured concurrently (see table 7-2 ) was much higher, thus indicating at least that these high CO2 transfer velocities could not be due solely to physical factors.

Continue to Chapter 6:
Analysis of measured CO2 data from steady-state tank.

If you jumped into this page from elsewhere, you may find it more convenient to go to the "frames" version of this thesis, alternatively you can use the links below to jump straight to a particular chapter:

Title Page Abstract Contents Figures Overview Chapter1 Chapter2 Chapter3 Chapter4 Chapter5 Chapter6 Chapter7 Chapter8 Chapter9 Chapter10 Appendix References Acknowledgements Links Download

Hits since 21st April 2000