In addition, the CO2 transfer velocity was measured in acidified seawater over the same range of temperatures and stirring speeds (for the latter, distilled water was also used), to provide a control without any significant enhancement, and to demonstrate the effect of varying the physical factors on the diffusion-only transfer velocity. These measurements lie on a straight line in the case of stirring speed ( figure 7-1 ), and on a slight curve in the case of temperature ( figure 7-2 ), and are consistent with a "surface renewal" gas exchange regime (k µ Sc-0.5). The scatter is only about 0.2 cm hr-1.
The curve of enhanced CO2 transfer velocity as a function of stirring-paddle speed ( figure 7-1 ) begins with a reaction-only transfer velocity of about 2 cm hr-1 (at 15C), but this falls to about 0.5 cm hr-1 as the transfer velocity due to diffusion rises up to 5 cm hr-1. The general shape of this curve is as expected (see, for example, the predictions in chapter 3 or in Wanninkhof 1992), however the measured enhancement factors are higher than predicted by the formula of Hoover and Berkshire (1969). The underprediction is slightly greater at higher stirring rates, possibly because this is based on a "stagnant film" rather than a "surface renewal" regime (Keller 1994).
Figure 7-2 shows how the measured enhancement increased as a function of temperature, from 5C to 34C. At 30C, enhancement by reaction raised the CO2 transfer velocity from 3.6 cm hr-1 (acidified seawater control) to 6.7 cm hr-1. The shape of this curve is as expected, since the reaction is known to be much faster at higher temperatures, and for this reason enhancement was predicted to be most important in tropical waters (Boutin and Etcheto 1995, Wanninkhof 1992). However, this is the first systematic experimental demonstration of this effect. Figure 7-3 also demonstrates that the assumption commonly used in many CO2 flux calculations, that the opposite effects of temperature on the transfer velocity and solubility of CO2 cancel, is not valid when there is chemical enhancement. The measured enhancement factors are again higher than predicted by the formula of Hoover and Berkshire (1969), the difference being much greater at higher temperatures. This may be due to the higher concentration of OH-, as the dissociation of water increases with temperature (Kw increases by a factor of 18 from OC to 30C in seawater).
The measured chemical enhancement also increased greatly at lower pCO2, as shown by figure 7-5 . For example, the transfer velocity due to reaction increased tenfold from about 0.4 cm hr-1 at pCO2=600ppm ([OH-]=2μm) to about 4 cm hr-1 at pCO2=15ppm ([OH-]=26μm), while the transfer velocity due to diffusion was about 2.75 cm hr-1 (all at 15C). Plotted as a function of [OH-] ( figure 7-6 ), the "influx" transfer velocities, at least, lie almost along a straight line. This confirms the importance of the OH- reaction pathway and the asymmetry with respect to pCO2, as predicted by Keller (1994), although we must beware that the effect of temperature on Kw dominates in controlling [OH-] in the ocean (see later).
At low pCO2, the measured enhancement factors were much greater than predicted by the Hoover and Berkshire formula. These predictions can be made to match the measurements, by increasing the rate of reaction between CO2 and OH- (the constant kOHKw from Johnson 1982) by a factor of six. Applying this higher OH- reaction rate also helps to reconcile the predictions with the measurements made at varying stirring speed and temperature, and brings all three sets of data together in the plot of measured versus predicted transfer velocities ( figure 7-13 ). This is because the discrepancy was greatest, both at low pCO2, and at high temperatures, where [OH-] is also high. Clearly this topic deserves further investigation, as there is only one other report of measurements of this reaction rate in seawater (Miller and Berkshire 1971), whose rates were slightly higher, and there is also some confusion regarding interpretation of the reported data (see Section 1.5.2 ). However, these experimental results suggest that the much lower values for kOHKw as proposed by Emerson (1995) are unlikely to be correct.
Both "influx" and "efflux" transfer velocities were measured simultaneously, using separate headspaces exchanging with the same water. Although the "efflux" transfer velocities (corresponding to lower air pCO2) were generally less accurate, due to a smaller measured D pCO2 difference (see Section 6.5 ), it is clear that they usually lie slightly above the corresponding "influx" transfer velocities, particularly at lower pCO2 ( figure 7-5 , figure 7-6 , figure 7-7 ). This may be explained, by considering that the air pCO2, as well as the water pCO2, will influence [OH-] within the microlayer, and thereby affect the chemical enhancement. This effect is unlikely to be significant at sea where the air pCO2 variation is much smaller than in these experiments. On the other hand, understanding this phenomenon may aid interpretation of other laboratory measurements (as shown in Section 1.5.4 ). Note that there was no significant air-water heat flux in the insulated tank, ruling out an alternative explanation in terms of heat and matter coupling (see Section 1.4.4 ).
In summary, the systematic discrepancy between measured and predicted transfer velocities might be resolved by using a greater OH- reaction rate, combined with a surface renewal model, and considering the effect of air pCO2. These factors can also explain similar discrepancies among the previously reported measurements in the literature.
The catalysis was also greater at higher temperatures, but the relative increase was not as great as for the uncatalysed reaction, which is to be expected since the enzyme lowers the activation energy and therefore the temperature dependence. Moreover, the enzyme activity declined much more rapidly at higher temperatures, with a lifetime of a few hours at 30C, compared to a few days at 5C.
These experiments confirmed the results of earlier experiments where the concentration of added bovine carbonic anhydrase was measured using the fluorescent marker Dansyl Amide, which also suggested that the enzyme would have a short lifetime in seawater (see Section 3.5 ). It is not known whether the decline in enzyme activity was due to physical or microbiological processes. However, if the enzyme has such a short lifetime, the lack of catalytic activity in stored samples of natural seawater (as reported by Goldman and Dennet 1983 and Williams 1983), does not necessarily imply lack of activity in situ, where the enzyme is produced.
For Dunaliella, catalysis only became significant as the water pCO2 dropped below an apparent threshold of about 220ppm, and increased as pCO2 fell further. For example, the transfer velocity rose from about 5 cm hr-1 when pCO2 was 180ppm, to about 11.5 cm hr-1 when pCO2 was 12ppm at the peak of the bloom, while the equivalent transfer velocities without algae were 3.6 cm hr-1 and 6.5 cm hr-1 respectively. This response to falling pCO2 was expected, since the physiological demand for carbonic anhydrase to aid photosynthetic CO2 uptake is much greater at lower pCO2 (see introduction Section 2.3.4 , and physiological model Section 3.2 ). However, we must bear in mind that the density of algae was also much greater at lower pCO2, as indicated by chlorophyll measurements and the "biological carbon" calculated by mass-balance (see figure 8-16 and the "carbon budget" plots for each bloom).
During the Emiliana bloom, transfer velocities of 5-6 cm hr-1 were measured when pCO2 was about 250ppm, compared to an equivalent transfer velocity without algae of about 3.4 cm hr-1 (see Section 8.3.9 and figure 8-13 ). This result is particularly significant, since it is within the pCO2 range typically encountered during algal blooms at sea. Moreover, blooms of Emiliana Huxleyi sometimes raise rather than lower the water pCO2, by forming coccoliths of CaCO3, and recent reports of carbonic anhydrase activity in Emiliana (see references in Section 2.3.2 ) suggest that carbonic anhydrase may play a role in this calcification process. In a calcifying bloom, catalysis of the transfer velocity might therefore increase the flux of CO2 from the sea to the air. However, no significant calcification was observed during the Emiliana bloom in the tank (as confirmed by alkalinity measurements).
During the bloom of the diatom Phaeodactylum, the measured transfer velocity increased dramatically at very low pCO2 but then fell even more rapidly. This fall probably coincided with a spontaneous calcification in the water (CaCO3 is highly supersaturated at such low pCO2), which confused interpretation of the results. However, they are consistent with published studies of Phaeodactylum cultures, which suggest that Phaeodactylum only produces extracellular carbonic anhydrase when TCO2 is lower than 1mM (see Section 8.4.3 )
The Skeletonema bloom was cultured from a strain provided by Prof Merret of Swansea University, which was known to produce carbonic anhydrase, but only when pCO2 in the water falls below a threshold of 130ppm (Nimer et al 1998). Although the pCO2 eventually fell to 65ppm, by this time most of the Skeletonema had flocculated and settled on the base of the tank, so perhaps it is not surprising that no catalysis effect was detected.
The concentration of algae (indicated by measured chlorophyll) was much lower in the untreated seawater samples taken from the North Sea spring bloom, which was dominated by Phaeocystis. The transfer velocity was on average just slightly greater than the equivalent without algae, but this difference was smaller than the scatter of the data.
Thus it seems that some, but not all, species of marine algae can catalyse air-sea CO2 exchange, and there appears to be greater catalysis at low pCO2, as expected. However, it must be borne in mind, that extreme conditions were set up in the tank to maximise the chance of observing any catalysis effect. The nutrient and chlorophyll concentrations during the blooms of algal cultures in the tank were much higher than typical of real seawater, the turbulence was low (corresponding to a diffusion-only transfer velocity of only 2.75 cm hr-1), and the water pCO2 sometimes fell as low as 5ppm, compared to the normal range of 200-500ppm for ocean waters. It is perhaps surprising that these marine algae possess the ability to take in CO2 from seawater, at concentrations far lower than ever found in the ocean today, and we might speculate, that perhaps this ability was more important at some earlier period during the history of their evolution.
In summary, therefore, these experiments showed that it is possible to detect catalysis of air-sea CO2 exchange by marine algae, but these results do not, in themselves, indicate whether this effect would be significant in the real ocean.
The measured CO2 fluxes in and out of headspaces with films were not related to the ΔpCO2 gradient in the normal way, and the transfer velocities calculated from them were very strange, ranging from +86 to -66 cm hr-1 (in the latter case, the flux and ΔpCO2 were in opposite directions). Careful analysis of the data suggested that the microalgae in the films must be taking CO2 directly from the air. The clearest indication of this was given by comparison of pCO2 measurements made simultaneously in the two "equilibrium" headspaces (i.e. without any gas flow). When a film was present in one of these headspaces, it lowered the pCO2 by as much as 100ppm compared to the true equilibrium pCO2 measured in the other headspace without a film (see figure 8-19 , days 39-40). The film lowers the air pCO2 in the headspace, until this active uptake is balanced by a normal diffusion-reaction flux back out of the water. A catalysis process could not explain such a steady-state disequilibrium.
It is possible to estimate the CO2 flux due to the film alone, by subtracting the estimated gas-exchange flux from the total measured flux. The film flux in headspace "B" at the end of the 4th Dunaliella bloom was in the range 5-7 mol m-2 yr-1 (see figure 8-18 ). At the end of the Emiliana bloom the film fluxes ranged from 1-3 mol m-2 yr-1, even when the water pCO2 was still in the "normal" range for ocean water. For comparison, the global average air-sea CO2 flux is only about 3.5 mol m-2 yr-1, if the average transfer velocity is 12 cm hr-1 (parameterisation of Liss and Merlivat 1986) and the average ΔpCO2 is 8ppm. These crude calculations suggest that the CO2 flux due to such films, if they were ubiquitous on the ocean surface, could be of a comparable scale to that due to gas exchange. Nutrient concentrations in these experiments were of course much higher than in normal seawater, but the sea-surface microlayer is also enriched in nutrients (see Section 2.4 ), so this may not be a limitation.
On the other hand, many researchers have suggested that photosynthesis by microalgae will be inhibited by toxic levels of ultraviolet light in the sea-surface microlayer, and that respiration by heterotrophic bacteria and zooplankton will therefore dominate, raising rather than lowering the microlayer pCO2 (see references in Section 2.4.3 ). This was clearly not the case in these laboratory tank experiments, but this film was not exposed to short-wavelength ultraviolet, being lit by a fluorescent tube shining through a perspex lid (see Section 5.3.3 ). If respiration does dominate in the films at sea, then the overall effect would be to decrease rather than increase the net global air-sea CO2 flux (as suggested by Hardy et al 1997).
Clearly, further work is needed on this topic, but we should not ignore the possibility, that surface films may provide a direct pathway for CO2 transfer between the atmosphere and dissolved organic and particulate carbon. This additional carbon flux would not be accounted for by measuring ΔpCO2 and calculating the gas exchange flux in the normal way, and might play a significant role in the global carbon cycle.
The crude calculations in Section 3.3 using simple hypothetical pCO2 and windspeed distributions, and also the thought experiment of Keller (1994) based on an even simpler bimodal pCO2 distribution, both demonstrated that the effect of chemical enhancement could be greatly magnified by its intercorrelation with water pCO2. However, neither of these calculations considered the importance of temperature variation. The experimental measurements in the tank (see figure 7-2 and Section 10.1 above) showed that chemical enhancement is much greater in warmer water, and in the ocean this tends to be associated with higher pCO2. This bias towards the tropics was anticipated by Wanninkhof (1992), who also noted that chemical enhancement would be favoured by the lower average windspeeds in these regions.
The first thorough calculation of the effect of chemical enhancement on the net global air-sea CO2 flux, was made by Boutin and Etcheto (1995). They applied the formula of Hoover and Berkshire (1969) (see Section 1.5.4 ) to calculate the enhanced and unenhanced transfer velocities and fluxes for every 1 degree square of the ocean for every month, using satellite windspeed and temperature data, and a pCO2 map based on an ocean carbon cycle model. The short-timescale variation in windspeed was also taken into account. Using the Liss and Merlivat (1986) parameterisation of the transfer velocity, they found that chemical enhancement increased the global average transfer velocity by about 7%, but decreased the net global air-sea CO2 flux by about 5%, because most of the enhancement was in the tropics.
However, the calculation of Boutin and Etcheto (1995) did not take into account the effect of pH / pCO2 variation on the OH- concentration. Boutin et al (1997) state that varying the pH appeared to make very little difference to the chemical enhancement, but this was based on the formula for kOHKw taken from Emerson (1995), who effectively divided the rate measured by Johnson (1982) by five. I believe this adjustment is incorrect, as explained earlier ( Section 1.5.2 ). Indeed, the experimental results from the laboratory tank, suggest that the OH- reaction rate may actually be much greater than measured by Johnson (1982). Only when this OH- reaction rate was multiplied by a factor of six, did the predicted enhancements match the measured enhancements (see Section 7.7 ).
Therefore a similar net global air-sea CO2 flux calculation to that of Boutin and Etcheto (1995) was made, using this much higher OH- reaction rate to represent the experimental measurements from the laboratory tank, and varying the OH- concentration based on the pH / pCO2 data. The transfer velocity parameterisations of both Liss and Merlivat (1986) and Wanninkhof (1992) were investigated, as was the effect of using the Rayleigh distribution to simulate short-timescale windspeed variability. In addition, the effect of enzyme catalysis was investigated, using the formulae developed in chapter 2, and five global average enzyme concentrations ranging from 0.21 to 53 nM. This enzyme was distributed according to satellite chlorophyll data, which could also be combined with physiological demand. The latter was predicted by calculating the increase in the growth rate of algae due to catalysis at the cell surface, using a reaction-diffusion model adapted from Riebesell (1993) as described in Section 3.2 , and balancing this benefit against an arbitrary cost of producing the enzyme (high cost or low cost). There were many possible combinations of all these different calculation options, and altogether 1200 global maps were produced, incorporating 15 billion chemical enhancement calculations (see Section 9.5.1 ). It should be noted, that no previous studies considered such wide range of scenarios, nor has there been any previous discussion in the literature of the potential effect of the distribution of enzyme catalysis, on the global air-sea CO2 flux.
In the absence of standard deviation data, an attempt was made to split the average weekly windspeeds using the Rayleigh distribution (which is based only on the mean). However, the spread of this distribution was too broad (see figure 9-8 ), and it resulted in a global average transfer velocity greater than that predicted by Boutin and Etcheto (1995). On the other hand, use of the Rayleigh distribution had the greatest effect on chemical enhancement, not in the tropics, but in the windier midlatitudes where the net CO2 flux was from air to sea, and consequently it reduced the net sea-air flux due to chemical reaction.
When the OH- reaction rate was multiplied by six, chemical enhancement of the global average transfer velocity was 70% greater, and the net sea-air flux due to chemical reaction was 49% greater (in total, 54 Mt C yr-1). It is important to note that the extra reaction via the OH- pathway is still greatest in the tropics where water pCO2 is high, rather than increasing the flux more in low pCO2 regions as predicted by Keller (1994). This is because water dissociates (Kw increases) much more readily at higher temperatures, and consequently both pH and pOH are higher in tropical waters (see figure 9-2 ). Most of the sea-air flux due to reaction in the tropics is cancelled, however, by the air-sea flux due to reaction in the midlatitudes, and so the overall effect on the net CO2 flux is less than the effect on the global average transfer velocity, although these effects are in opposite directions. Applying the Rayleigh windspeed distribution, which increases enhancement most in the midlatitudes, further reduced the net effect of the OH- reaction on the CO2 flux.
Chemical enhancement was much lower when using the Wanninkhof (1992) parameterisation, because enhancement due to reaction is lower at higher transfer velocities. However, in this case the net sea-air flux due to chemical reaction was slightly increased when the Rayleigh distribution was applied, and the relative effect of the higher OH- reaction rate was greater, although the actual increase was smaller.
In the case of the Liss and Merlivat parameterisation, with weekly average windspeeds, the lower OH- reaction rate, and the "low-cost" physiological enzyme distribution, the net global air-sea CO2 flux due to reaction increases from -36 Mt C yr-1 with no enzyme, crosses through zero at an average enzyme concentration of about 2nM, and rises up to 650 Mt C yr-1 at an average concentration of 53nM. The levelling off at higher enzyme concentrations, is similar to that observed for the measurements in the tank with added bovine enzyme ( figure 7-10 ). This is due to increased competition between reaction and diffusion processes in the gas exchange model.
The global average transfer velocities are similar for all three enzyme distributions. However, they are just slightly higher for the low-cost physiological distribution, compared to the chlorophyll-only distribution. This is because the physiological model places more of the enzyme, where it is more effective for aiding photosynthetic CO2 uptake. Although this favours low pCO2 regions, it also favours warmer temperatures, where the enzyme is likewise more effective at catalysing air-sea CO2 exchange. The maps show that it removes all the enzyme from the coldest polar seas, and especially favours the warm low-pCO2 waters around South-East Asia (see figure 9-21 ). This is also true for the high-cost distribution, but in this case most of the enzyme is concentrated in just a few most-favoured areas, and therefore the competition effect mentioned above lowers the average transfer velocity compared to the chlorophyll-only distribution, particularly at higher average concentrations.
Changing the enzyme distribution has a much greater effect on the net global air-sea CO2 flux, because the physiological model favours areas where water pCO2 is low. This bias is much greater for the high cost distribution. For example, in the case of the Liss and Merlivat parameterisation, with weekly-average windspeeds, the lower OH- reaction rate, and an average enzyme concentration of 3.2nM, the extra fluxes due to catalysis are 26 MtCyr-1, 63 MtCyr-1 and 201 MtCyr-1 for the for the chlorophyll-only, low-cost and high-cost distributions, respectively.
For the low-cost and chlorophyll-only distributions, the effect of using the faster OH- reaction rate diminishes slightly at higher enzyme concentrations, due to the competition effect. With the high-cost distribution, however, this competition from enzyme reduces the effect of the OH- only in low pCO2 waters where the flux is from air to sea, which has the overall consequence of increasing the net flux due to the faster OH- reaction, which is from sea to air.
As for uncatalysed enhancement, the effect of enzyme catalysis is greater for the Liss and Merlivat than for the Wanninkhof parameterisation, due to the lower diffusion-only transfer velocities. This difference is lower for the physiological distributions, because these favour the midlatitude regions with mid-range windspeeds, where the two parameterisations are closest. Using the Rayleigh distribution for local windspeeds also increases the potential for chemical enhancement, and thus increases the effect of enzyme catalysis, particularly at low enzyme concentrations.
There are many other factors besides chemical enhancement which affect the transfer velocity, including bubbles, the physical effect of surface organic films, the thermal skin effect, evaporation and condensation, and possibly the air-sea heat flux (more details of all these in Section 1.4 ). However, chemical enhancement is the only process which obviously favours CO2 (and hence 14C), compared to the inert tracer gases. Unfortunately, bubble-mediated transfer is more important for the low-solubility inert gases than for CO2, and would therefore increase the discrepancy.
When considering this discrepancy, it is critical to bear in mind that the global air-sea flux of both natural and bomb 14C was effectively one-way (from the atmosphere to the ocean) and is therefore little affected by the small variations in water pCO2, which are so important in calculating the net global air-sea 12CO2 flux. Consequently, the 14C budget tells us only about the global average transfer velocity. Intercorrelation between the kinetic processes and water pCO2, such as the bias of the physiological enzyme distribution towards low pCO2 regions, or the bias of uncatalysed chemical enhancement towards warmer high pCO2 regions, may have a big impact on the net global air-sea 12CO2 flux, but will not significantly affect the global 14C flux.
Therefore, we need to ask firstly, how much catalysed and uncatalysed enhancement would resolve the discrepancy in the average transfer velocity, and secondly, how would this amount of enhancement affect the net global air-sea 12CO2 flux? The extrapolations presented in table 9-5 attempt to answer these two questions. The average concentration of enzyme required to bring the global average transfer velocity calculated with the Liss and Merlivat (1986) parameterisation, up to the level derived from the 14C budget, is about 78-137nM, depending on the various calculation scenarios. Note that the average windspeed used in this dataset was 7.0 ms-1, lower than the usually-quoted figure of 7.4 ms-1, which slightly increases the amount of enzyme needed.
The effect of this amount of enzyme on the net global air-sea CO2 flux is much more dependent on the distribution used, and ranges from +33% to +202%. The higher figure, derived from the high cost physiological distribution, the lower OH- reaction rate, and average weekly windspeeds, could be considered as a constraint on the maximum possible CO2 flux due to enzyme catalysis, assuming the transfer velocity is constrained by the 14C data. However, this high air-sea CO2 flux (5.65 Gt C yr-1) is far too large to be accommodated within the global carbon budget. On the other hand the lower figure, derived from the chlorophyll-only distribution, is smaller than the relative increase in the average transfer velocity (63%), indicating that in this case much of the air-sea flux due to catalysis in low pCO2 regions is cancelled by a sea-air flux in high pCO2 regions.
We should note that some coastal seas were excluded from the pCO2 map, which was derived from a carbon cycle model with a coarse 2.5o grid. In these regions, which only cover a small area, but include about 17% of the total enzyme due to their high chlorophyll levels, ΔpCO2 was taken to be zero. Effectively, therefore, the calculated net global air-sea CO2 fluxes correspond to about 17% less enzyme than reported.
Even when this is taken into account, the effect of the enzyme on the net CO2 flux is much lower than was predicted by the crude calculations in chapter 3, in which the temperature variation was not considered, and the hypothetical pCO2 distribution was too broad. However, the average concentration required to explain the discrepancy in the transfer velocity is similar to that predicted earlier.
It is very improbable that there is enough enzyme in the microlayer to increase the average transfer velocity by 60%. Even in extreme high-nutrient, low pCO2, and low turbulence conditions of the blooms in the tank, the measured enhancement attributed to algal catalysis was usually less than this. So the main purpose of the global flux calculations, was to illustrate the importance the distribution of enzyme, due to the intercorrelation between enzyme catalysis and pCO2. The greater physiological demand for enzyme at lower pCO2, as observed in the tank, can substantially increase its effect on the net global air-sea 12CO2 flux. It is important to bear in mind, that the high-cost physiological distribution puts most of the enzyme in just a few areas with low pCO2, warm water and high chlorophyll, and consequently the local enzyme concentration in these grid cells is much higher than the average concentrations reported above. On the other hand, in areas with high-chlorophyll we also expect to find higher levels of nutrients, including the trace metal zinc which is a key component of carbonic anhydrase. Moreover, where algal activity is intense, surface films will be more prevalent, and so the microlayer enrichment factor may be greater.
Given that many researchers have been trying for many decades to resolve the discrepancies between the different methods of measuring the transfer velocity and calculating the net global air-sea CO2 flux, it is unlikely that there is any single, simple answer to this problem. Instead, the solution probably lies in the combination of a variety of factors and processes, which certainly include chemical enhancement, and may include catalysis by marine algae.
Indeed, Nightingale et al (1999) suggest that the discrepancy may be much smaller than was previously thought. Their new triple-tracer measurements at sea, combined with a reinterpretation of the earlier dual-tracer data paying particular attention to the uncertainty and variability of the windspeeds during the flux measurements, substantially raises the global average transfer velocity predicted from the inert gases. To this, we can fairly confidently add at least 5% due to uncatalysed chemical enhancement, or about 9% using the faster OH- reaction rate. Meanwhile, it has also been suggested that the 14C transfer velocity should be lowered by up to 25% following a recalculation of the stratospheric 14C inventory (Hesshaimer et al 1994), although there is some dispute about this (see Section 1.2.11 ). In this case, the remaining discrepancy would be small, compared to the errors in both estimates.
Nevertheless, we still need to reduce these errors, as climate models become more sensitive, and the cost of CO2 emissions rises. This may require that algal catalysis be included among the various minor process affecting the air-sea CO2 flux, and will also require very careful consideration of the subtle effects of intercorrelation of the kinetic and thermodynamic processes.
Clearly, more thorough flux calculations are necessary, incorporating all the "minor" kinetic processes, and based on concurrent measurements of all the parameters. In a rapidly changing world where the ocean is warming and pCO2 levels are rising, we also need to keep updating such calculations. This is a large task occupying many researchers around the world (see Section 1.3 for a discussion of the difficulties of obtaining sufficient data).
However, given the apparent complexity of the calculations, there is a danger that future investigators may be tempted to exclude chemical enhancement and other "minor processes" from their main calculation, and account for them by tacking on corrections using "off-the-shelf" figures taken from separate studies of each process. It should be evident, even from examples discussed earlier such as the competition between the OH- and enzyme-catalysed reaction pathways, that the extra fluxes due to each "minor process" are not additive. The only valid way to account for all these interacting processes, is to include them all together in the main calculation -which is not beyond the capability of modern computers, and requires little additional data. Then, if we wish to investigate the specific contribution made by one individual process, this can be determined by excluding that process from the calculation. This more holistic, subtractive process is common practice in ecological modelling of interactive systems.
While awaiting the results of such a calculation, we can attempt to anticipate a few potential intercorrelations between these processes.
For example, the thermal-skin effect might be expected to reduce chemical enhancement slightly, because the reaction is slower in colder water. On the other hand, the skin effect is greatest in the Gulf Stream and Kuroshio currents in winter (see VanScoy et al 1995), when the water is still relatively warm but the pCO2 is fairly low. In these conditions the physiological demand for carbonic anhydrase is particularly high, as shown by a close examination of the maps in chapter 9, so potentially the interaction between these processes might increase the net CO2 flux.
We might expect a strong intercorrelation between chemical enhancement and surface organic films, since these are both more important when the windspeed is low, and when algal productivity is high (if catalysis of enhancement is significant). Locally, the reduction in the diffusion-flux due to a film might be partially offset by an increase in the reaction-flux. It is hard to predict the overall effect on the global average transfer velocity, but this interaction is likely to decrease the net global CO2 air-sea flux, because the enhancement would be greatest in warm, high pCO2 regions.
We cannot account for the effect of such local short-timescale intercorrelations on the net CO2 flux, unless we use local concurrently measured windspeed and pCO2 data. Such data is usually available for specific ship or even buoy measurements, but this information is lost, whenever the data is interpolated over time or space to form global maps. There are many factors which could introduce significant local intercorrelations. For example, during calm conditions when the mixed layer is shallow, the depletion of pCO2 by an algal bloom may be much greater than during windy conditions. This local intercorrelation between windspeed and pCO2 would increase the net air-sea CO2 flux due to chemical enhancement, especially if this is augmented by increased enzyme-catalysis, but it would also reduce the unenhanced flux. During stormy weather, on the other hand, the surface water pCO2 is likely to change locally, both as a result of faster equilibration with the atmosphere, and also as a result of deepening of the mixed layer by increased wave mixing.
We should also beware of local intercorrelation between windspeed and temperature, which might strongly affect chemical enhancement, as well as the thermal skin effect. The diurnal cycle may also introduce systematic errors, if pCO2 and windspeed maps are based predominately on measurements made during daylight hours.
It is unlikely that there is any significant intercorrelation between chemical enhancement and bubble-mediated transfer, because these occur at opposite ends of the windspeed spectrum. Gas transfer across the surface of the bubbles themselves would not be significantly enhanced by reaction, because the turbulence is too great.
We should beware, however, that formulae for predicting chemical enhancement (see Section 1.5.4 ) generally require an equivalent "stagnant-film thickness" or "surface-renewal time". This is usually calculated from the unenhanced transfer velocity using a parameterisation as a function of windspeed. If the latter includes bubble transport, the "stagnant-film thickness" and consequently the enhancement factor will be underestimated, but on the other hand, this enhancement factor will be incorrectly multiplied by the whole transfer velocity including bubble transport. Similarly, we should beware of multiplying a ΔpCO2 which has been corrected for the thermal skin effect, by a transfer velocity which includes a bubble flux, since there is no thermal skin on the surface of bubbles!
All the well-known parameterisations of the transfer velocity have included the surface-diffusion flux and the bubble flux in one formula. However, it has long been realised, that this leads to errors when interpolating between gases and temperatures, because the dependence on the solubility and Schmidt number is different for each process. The "variable Schmidt number" approach adopted as a compromise by Wanninkhof (1993) caused much confusion in interpretation of the dual-tracer measurements (see Section 1.4.1 and references therein).
Incorporating chemical enhancement into a general transfer velocity formula could be even more misleading. The temperature dependence of transfer by reaction is quite different to that of transfer by diffusion (recall figure 7-2 ), and the intercorrelation with water pCO2 would be ignored entirely if enhancement were incorporated into a formula solely as a function of windspeed and Schmidt number. The extra CO2 flux due to enhancement calculated using such a formula would not only be of the wrong magnitude, but might also have the wrong sign (recalling that the uncatalysed reaction decreases the net global air-sea CO2 flux)! Wanninkhof (1992) instead proposed a compromise formula for chemical enhancement based solely on temperature, which was simply added to the main transfer velocity due to diffusion. However, this greatly overestimated enhancement at higher windspeeds, as shown by the global calculations of Boutin and Etcheto (1995).
So we should abandon attempts to parameterise the transfer velocity with one simple curve, and instead calculate each process with a separate formula. It would also be preferable to use wave height (or slope), rather than windspeed, as the basis for such formulae, as this is more closely related to turbulence in the microlayer, can be measured directly by satellite, and incorporates the effects of wave-damping by surface organic films (see Section 1.4.5 ).
Finally, we should be wary of using terms such as "effective average transfer velocity" to incorporate the effects of intercorrelation between kinetic processes and ΔpCO2. This is particularly confusing when interpreting the global flux of 14C, because the relative variability of Δ14CO2 is much lower than the variability of Δ12CO2. For example, Keller (1994) correctly pointed out that the asymmetry of chemical enhancement with respect to [OH-] might have a large effect on the 12CO2 flux, but mistakenly suggested that this would also apply to 14C, and might therefore explain the transfer velocity discrepancy.
However, such films which prosper in the sheltered environment of the laboratory tank, may not occur at sea where nutrients are scarcer, turbulence is greater, there are more predators, and algal photosynthesis may be inhibited by toxic levels of ultraviolet light. Unfortunately, it is very difficult to make direct measurements of the air-sea CO2 flux at sea (as explained earlier -see Section 1.2.9 ), although the eddy-correlation technique on the air-side of the interface (see Section 1.2.10 ), and carbon budgets of the mixed layer (see Section 1.2.9 ) may eventually become sufficiently accurate to detect enhanced fluxes due to films.
Meanwhile, it may be possible to detect CO2 uptake or release by a film, by measuring the CO2 fluxes in enclosed headspaces floating on the sea surface. This technique has already been used for measuring air-sea CO2 fluxes by Frankignoulle (see references in Section 1.2.10 ), and has proven to be particularly useful for measuring very localised fluxes, such as those due to the metabolism of coral reef communities. Wanninkhof and Knox (1996) also used a floating headspace to measure chemical enhancement across the surface of a lake. However, to investigate film fluxes, it would be necessary to have at least two such headspaces, one of which was periodically "cleaned" to remove or diminish the effect of any film, and thus serve as a simultaneous control experiment. Since cleaning the surface might also affect the flux due to diffusion, the flux of an inert gas (preferably N2O, whose diffusivity is very close to CO2) would have to be measured simultaneously in both headspaces. Moreover, to investigate photosynthetic uptake by the film, the headspaces would have to be translucent. This raises the problem of temperature control, since we would effectively have created a floating greenhouse, and in such circumstances condensation is also likely, which can significantly affect gas exchange (see Section 1.4.3 ). However, such problems are not insurmountable, as a feedback system could be designed to control the air temperature and humidity automatically, and also to compensate for pressure changes due to wave action. A location with low winds, low pCO2, and high algal productivity would be preferable.
Of course, the transfer velocities measured under such a headspace would bear little relation to those outside, where the surface is ruffled by the wind. Nevertheless, such experiments might indicate whether a significant film flux is possible at sea. The headspace would inevitably affect the ecological balance of the film, although this might be minimised by keeping the experiments as short as possible.
One very important factor, which may be hard to preserve inside the headspace, is the spectral composition of the sunlight. Although glass and perspex transmit sufficient light for photosynthesis, they absorb the shortest wavelength ultraviolet light. Many investigators (see references in Section 2.4.3 ) have suggested that this ultraviolet light may be toxic to many species of microalgae, and therefore predicted that respiration would dominate over photosynthesis in the surface microlayer. This would raise the microlayer pCO2 and cause an extra flux into the air, the opposite effect to that observed in the tank. However, there is little direct evidence for this, other than cell counts of selected species. Clearly, preliminary laboratory experiments using different light spectra would be essential.
An alternative experimental approach was pioneered by Garabetian (see references in Section 2.4.3 ), who used glucose labelled with 14C to investigate uptake or release of CO2 by microorganisms in the microlayer. It would be difficult to extend such experiments to the open sea, but as our understanding of natural isotope ratios improves, it might be worth looking for isotopic evidence of direct CO2 uptake.
So the ultimate aim of such experiments would be to derive a parameterisation of the direct film flux, in terms of temperature, waveslope, sunlight, chlorophyll. Some of the other pigments which are now measured by the new SEAWIFS satellite might also be useful, as the spectral signal is dominated by the water at the surface. These data could then be combined in global flux calculations, similar to those for chemical enhancement in chapter 9.
The microlayer samples collected by Hardy and Apts (1984) suggested evidence of photoinhibition of algae in summer, but not in winter when ultraviolet light levels were lower. If this pattern were repeated across the ocean, the net global air-sea CO2 flux due to surface films may be the balance of larger seasonal and regional film fluxes, with a similar distribution to the normal diffusive CO2 exchange.
We should recall that the film fluxes measured in the tank were of comparable magnitude to the average air-sea CO2 flux (see Section 8.5 ). Moreover, the film flux is independent of ΔpCO2, since it forms a direct pathway between the atmosphere and dissolved and particulate organic carbon. Although the latter may be measured by sediment traps, such measurements are not yet sufficiently accurate to constrain the potential impact on the air-sea flux.
Using a higher OH- reaction rate also helps to explain the influence of air pCO2 on the enhancement measured in the tank. The air pCO2 affects the profile of pH in the microlayer, which has been the subject of much theoretical debate, since it is the key difference between the algebraic and more the complex iterative reaction-diffusion models (see Section 1.5.4 ). It might be interesting to attempt to measure this vertical pH profile in situ during steady-state CO2 exchange, perhaps using an indicator dye or microsensor technique. However, although further investigation of the affect of air pCO2 might be interesting from a theoretical point of view, it would probably not significantly improve estimates of chemical enhancement at sea, where air pCO2 is not so variable.
A closer examination of the temperature profile across the microlayer is also needed, to resolve the theoretical argument between Phillips and Doney (see Section 1.4.4 ) regarding the interaction between heat and matter transfer during gas exchange. Phillip's claims of "experimental evidence" for his theory depended on the temperature of an arbitrary boundary layer on the air-side of the interface, and it is hard to distinguish the influence of the "heat of solution", from the effect of the thermal skin on the solubility. The temperature profile might be examined, using microscale temperature sensors, or remote thermal imaging techniques. If the theory developed by Phillips (1991a) is correct, then the heat of solution and the heat released by the chemical reaction of CO2 with water might both affect the effective thermodynamic driving force for gas exchange. In this case, we might expect this effect to be closely intercorrelated with chemical enhancement. It should be possible to measure such an effect in a tank, specifically designed to maintain a steady-state heat flux.
The experiments with algal blooms in the tank, reported in chapter 8, were set up to maximise the chance of detecting any algal catalysis of air-water CO2 exchange. High nutrient concentrations encouraged intense algal growth, which brought pCO2 down to levels much lower than typically found at sea, thus increasing the demand for carbonic anhydrase for photosynthetic uptake. However, Raven (1995) pointed out that internal carbonic anhydrase helps to reduce the requirement for nitrogen, by increasing the efficiency of the large enzyme RuBISCO which has a particularly high N:C ratio. In this case, low levels of nutrients might also stimulate carbonic anhydrase production, possibly within the normal pCO2 range (as observed during the Emiliana Huxleyi bloom).
In any case, further experiments in more "realistic" conditions would be valuable, although the accuracy of the transfer velocity measurements would probably have to be improved. For example, with hindsight it would have been better to reduce the volume (but not the surface area) of the headspaces, in order to reduce the time taken to reach steady-state. Improvements could also be made to the flow system: for example, the difference between pCO2 flowing in and out of the headspaces could be measured directly using two cells of the analyser, if a more reliable supply of "high pCO2" gas was available.
Monocultures of one species of algae are not only unnatural, but they are also unsustainable because metabolic products cannot be recycled -hence the need to supply vitamins when filtered seawater was used (see Section 5.2.3 ). It might be preferable to grow cultures from samples taken from blooms at sea, which would include not only the dominant species of algae, but also bacteria and zooplankton grazers. A general problem with interpreting the data during the algal blooms, was that the concentration of algae was generally greater at lower pCO2, so it was difficult to distinguish the influence of these two factors (see Section 8.4.2 ). It would be relatively straightforward to keep the water pCO2 more constant, if desired. With more data, it would also be interesting to separate the different stages of the bloom.
The physiological model used for distributing the enzyme in the global flux calculations used arbitrary parameters to calculate the concentration of carbonic anhydrase in the microlayer, combined with kinetic data from a freshwater alga. To make such calculations more realistic, we would need direct measurements of catalysis of CO2 hydration in the microlayer, during an algal bloom. However making such measurements may be a daunting task. The enrichment of CA just below the air-sea interface may be greater than in a thicker microlayer sample, due to its surface activity, and its catalytic activity may be affected by the complex web of macromolecules in the microlayer in situ (these problems were discussed in Section 2.4 ). Moreover, although the techniques for measuring CA are continuously being refined and the detection levels are falling, these are generally designed for extracts from algal cultures, not for measuring CA dissolved in seawater itself. It would be useful, at least, to use such techniques to derive genuine enzyme kinetic constants for extracellular CA from common marine microalgae. Meanwhile, measurements of pH changes in the microlayer itself (as already mentioned above), might be an alternative approach for measuring catalysis in situ.
The fluorescence technique using dansyl amide ( Section 3.5 and appendix) gave promising results in artificial or filtered seawater, but unfortunately the signal:noise ratio was much poorer in microlayer samples, and in unfiltered natural seawater, due to light scattering. The scattering problem might possibly be avoided by using a more sophisticated optical system. Ideally, optical fibres might be used to pick up the fluorescent light directly from the microlayer in situ (this would have to be done with a tank in the dark).
Even without new data, some improvements could be made to the physiological model for distributing enzyme catalysis in the global flux calculation. More sophisticated theoretical reaction-diffusion models have recently been developed (see references in Section 3.2 ), and might be adapted for this purpose. A more general iterative reaction-diffusion model (such as that of Emerson 1975b for the air-sea interface) might also be useful to test some of the assumptions necessary in algebraic models.
In the model developed in Section 3.2 , carbonate kinetics and thermodynamics were temperature dependent, but the growth constant Vmax was constant, which is clearly not realistic in the ocean, or even at different stages of one algal bloom. The maximum photosynthetic uptake should be expressed in the model as a function of temperature and light. Further tank experiments measuring the effect of algal blooms on the CO2 transfer velocity should also be made at a range of different temperatures, although careful analysis would be necessary to distinguish all the different ways in which temperature may affect catalysed chemical enhancement.
For carbonic anhydrase to effectively catalyse air-water CO2 exchange, it must be released by algal cells and dissolved in the microlayer. Therefore, although the bloom as a whole might benefit from catalysis of air-water CO2 transfer and thereby increasing the pCO2, the marginal benefit for the individual cell that released the enzyme would be negligible. On the other hand, the cost of producing carbonic anhydrase is considerable, because each molecule has a mass of about 35,000Da (see Section 2.2.2 ). Consequently, if the bloom is deliberately catalysing air-sea CO2 exchange, this implies the evolution of cooperative behaviour between the individual cells.
This is not as implausible as it may seem, because algal blooms at sea arise from just a few seed cells, which happen to be in the right place at the right time, and multiply very rapidly. Therefore, a genetic characteristic which benefits the whole bloom, might result in the creation of more seed cells when the bloom decays and disperses. There are other examples, to which a similar argument might apply. For instance, the white coccoliths produced by blooms of Emiliana Huxleyi reflect a lot of sunlight, thereby altering the temperature profile within the water, and denying light to other species. The dimethyl sulphide released by many algal blooms may affect the weather by seeding clouds (Charlson et al 1987), although this effect would be regional rather than local due to the time taken to oxidise DMS to sulphate and form cloud condensation nuclei. Recent evidence also suggests that the release of large quantities of chelating agents (probably siderophores) by algal cells, helps to reduce the loss of iron from surface waters (iron being a limiting nutrient in certain areas of the ocean, as demonstrated by Coale et al 1996).
On a smaller scale, but nevertheless much larger than the individual cell, many bacteria release digestive enzymes into their environment, and various micro-organisms release toxins to discourage grazing by predators. We should recall that the sea-surface microlayer is a microhabitat whose small volume, relative stability, and high density of organisms, would be particularly conducive to the evolution of such cooperative behaviour. If the cells living in the microlayer release carbonic anhydrase, most of it, being surface-active, is likely to remain nearby, and so it might significantly change the pCO2 within the layer.
This is only a speculative hypothesis, and there are alternative explanations for release of carbonic anhydrase by algal cells. During the period that the enzyme remains near to the cell, it helps to increase the CO2 flux towards the cell wall (as shown by the calculations in Section 3.2 ). The complex spiky structure of many marine algae, particularly diatoms, may help them to trap pockets of water near to the cell, in which released enzymes are retained for sufficient time to be useful. Many metabolic products are also released when cells die. However, if this were the case, we should expect little catalysis of air-water exchange during the "exponential growth" phase of a bloom when few cells die, which was not the case in tank experiments.
Whatever the reason for the release of carbonic anhydrase by algal blooms, potentially this might be considered a local physiological feedback process between organisms and their environment. The calculations in chapter 9 also illustrated how the intercorrelation caused by greater physiological demand for carbonic anhydrase in low pCO2 regions, could significantly increase the net global air-sea CO2 flux. However, this is unlikely to be significant as a geophysiological process, potentially stabilising the climate or CO2 concentrations on a global scale. This is because air-sea CO2 exchange is not the rate-limiting process for CO2 transfer between the atmosphere and the deep ocean, which is determined more by the transfer of carbon from surface to deep water by the biological and salinity pumps (as explained in Section 1.1.3 ). If carbonic anhydrase significantly catalysed the CO2 transfer velocity, this would cause the mixed layer to equilibrate faster with the atmosphere, and so the increased flux in one region, would be partially offset by a reduced flux elsewhere. This type of compensation is apparent in dynamic carbon-cycle models, which are not particularly sensitive to the parameterisation of the transfer velocity (see Section 1.1.5 and Section 1.3.8 ).
Nevertheless, to test and calibrate such models, we still need to be able to balance the global carbon budget using direct, static CO2 flux calculations such as those in chapter 9, which are strongly influenced by small biases in the transfer velocity. Resolving discrepancies between methods of measuring the transfer velocity, is also necessary to improve calculations of the sea-air fluxes of other climatically important gases.
We also need to understand the kinetic processes in detail, in order to predict how they might change the CO2 flux in the future, as the surface ocean warms and the pCO2 rises. As the pCO2 and climate are changing rapidly, compared to the slow circulation of ocean waters, the compensating adjustments observed when dynamic carbon cycle models are run to equilibrium, may not apply on this timescale.
Uncatalysed chemical enhancement might be expected to increase in a warmer ocean, but the key question regarding the effect on the net CO2 flux, is whether this increase would be greater in high or low pCO2 regions. If the warming is greatest in polar regions (as suggested by most climate models), then increased enhancement might increase the net air-sea CO2 flux a little. However, such changes are likely to be far smaller than those brought about by changes in the thermohaline circulation, particularly the cessation of the main sink for CO2 in the North Atlantic (see Section 1.1.4 ). Such changes would also reduce upwelling of nutrients, and hence the growth of algae. Although the overall effect on the biological pump is still very uncertain, the reduction in nutrient supply would presumably reduce any enzyme catalysis of air-sea CO2 exchange. Moreover, as the average pCO2 rises, there would be less demand for the enzyme. The change in concentration of OH- is harder to predict, since the rise in pCO2 and the rise in temperature would have opposite effects. Again, however, this kinetic effect is likely to be dwarfed by the reduction in the solubility of CO2 as the temperature rises, and the reduced buffering capacity of seawater as the acidity rises. In summary, therefore, chemical enhancement will remain a "minor process", so far as future flux changes are concerned.
Nevertheless, as shown in chapter 9, uncatalysed enhancement may account for a net annual source of 20-50 Mt C to the atmosphere, while catalysis might possibly cause a sink of even greater magnitude. Transnational companies are now eagerly purchasing tropical forests, and investigating industrial processes for capturing and storing CO2, in the expectation that there will soon be a market for tradable carbon credits, with typical costs in the range 10-100$/ton C. On this basis, we could claim that the net flux due to uncatalysed chemical enhancement costs about 0.2-5.0 billion dollars per year. Perhaps this is a convenient argument for requesting more money for research into this topic!
Unfortunately, this type of argument is increasingly driving the research agenda. Large institutions around the world are spending vast sums to investigate ways by which they might claim credit for natural changes in sources and sinks of greenhouse gases, and thereby provide an excuse to postpone any real reductions in emissions from burning fossil fuel. This game may keep some researchers employed for a while, although the greatest beneficiaries are probably diplomats, lawyers and accountants. Thousands jet around the world to conferences on this topic, rarely considering that their flights probably put more CO2 into the atmosphere, than they will ever remove by trading emissions credits created by adjusting hypothetical baseline scenarios.
Others travel to such conferences to promote research into technical fixes, such as promoting the growth of ocean algae to soak up more CO2, or reflecting more sunlight with stratospheric dust. Given the bizarre nature of some of these proposals, I would not be surprised to find, at some later date, that selected figures from chapter 9 of this thesis have been quoted out of context, to justify a research proposal into deliberately seeding the ocean with genetically modified algae designed to release carbonic anhydrase in the microlayer and thereby catalyse air-sea CO2 exchange. The ecological risks of such an intervention would be immense. Even if it did increase the CO2 transfer locally, it would have little long-term effect on the net global air-sea CO2 flux, because the transfer velocity is not actually rate-limiting, as explained earlier. A slightly more mundane proposal might be to use such algae to soak up CO2 directly from flue gases from power stations. However, since photosynthesis requires sunlight, this is effectively a rather inefficient form of solar power. Essentially, many of these proposals are distractions to employ researchers, fool politicians, and put off hard decisions.
Even the best-intentioned researchers are driven by the structure of academic institutions and scientific careers, to delve more and more into obscure details of specific processes, with little consideration of their relevance in the broader picture. I would be the first to admit, that this study, too, could be criticised for focussing on just one process. This constraint is unfortunately imposed upon us by the PhD system. I hope, at least, that the range of topics considered here, from the microscale of enzyme kinetics to the global carbon cycle, is broader than in most theses.
Meanwhile, the real world does not wait for such studies - it is getting warmer, and the pCO2 is rising. There is a danger that the oceans, the atmosphere, the algae and the forests, are changing faster than the data in our computers, which we use to represent them. The climate, driven by a complex interaction of feedback processes, may be evolving faster than the crude models, which we use to predict it. Yet the real challenge for scientists, is not to put more and more effort into collecting data and improving our models to keep up with this rapidly changing world, nor to create high-risk technical fixes to treat the symptoms of the problem. The challenge is to persuade everybody to reduce the emissions which are the cause of the problem, and thus to slow the pace of climate change. The solutions for this are well known, and technically simple, but require all of us, including scientists, to reconsider our lifestyles, values, and priorities, and to act on the basis of the knowledge we already have, as soon as possible.
When we have done this, we might rest awhile, and then continue to investigate the microscale processes in sea-surface microlayer. We could then do this, not necessarily because it is important for the future of humankind, but because the undulating skin through which the sea breathes, and the microscopic creatures which live within it, are fascinating, beautiful, and always full of surprises.
Note: 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