Comparing the model projections with empirical evidence
This work seeks to provide a more quantitative assessment of the scope for solar-powered algal biomass production than has hitherto been published. Absolute peak of year-averaged C-biomass productivity in these simulations was 2.4 gC m−2 day−1. More typically, mean AP (Fig. 3) fell between 0.8 and 2.3 gC m−2 day−1, depending on strain configuration and geographic location. Taking the C/dry weight biomass ratio to be 31 % (Heymans 2001) and assuming uninterrupted production could be maintained, this allows a rough estimate of a limit on annual areal production at slightly under 30 t dw ha−1 year−1 for a strain which undergoes one doubling during a 12:12 h light/dark cycle. For strains with faster maximum growth rates (along with a suitable choice of dilution rate), the absolute value of production should rise further.
This projected peak value of 30 t dw ha−1 year−1 from our simulations falls within the mid-range of those reported for different real systems. It is in good agreement with the results of Jimanez et al. (2003) who report annual production of 30 t dw ha−1 year−1 in raceways in Southern Spain. Olguin et al. (2003) report an average production of 11.8 g m−2 day−1 in Mexico over the course of a year cultivating Spirulina using animal waste, which would equate to around 40 t dw ha−1 year−1 of biomass if production could remain uninterrupted over the whole year. Productivity of 60 t dw ha−1 year−1 of Pleurochrysis carterae in flat ponds has been reported (Moheimani and Borowitzka 2006), and cited as an example of maximal productivity (Williams and Laurens 2010), but this weight includes 10 % of calcium carbonate in the form of coccoliths and remains very much the exception rather than the rule. At the other end of the scale, García-González et al. (2003) achieved production of Dunaliella equating to around 6 t dw ha−1 year−1. This compares favourably with the projected AP for the slowest growth rate optimised for in Fig. 1 (0.65 gC m−2 day−1 for U
= 0.346, comparable to that of Dunaliella) which is approximately 7.5 t dw ha−1 year−1. For other system types, there are reports of AP up to 60 t dw ha−1 year−1 for tubular PBRs (Fernández et al. 1998; de Schamphelaire and Verstraete 2009; Rodolfi et al. 2009) and nearly 40 t dw ha−1 year−1 for hybrid systems (Huntley and Redalje 2007), dependent upon the strain cultivated. The more typical model predictions for AP above of 3–8.4 tC ha−1 year−1 (around 10–27 t dw ha−1 year−1 using the above estimation) are of the same order as estimated by Ritchie and Larkum (2012) who measure net photosynthesis for three algae species from measurements of light attenuation in cultures of varying optical depths. In conclusion, this all instills confidence that the model used in this study is producing plausible projections.
Potential for biofuels production
Our simulations indicate a maximum potential biofuels production rate of 0.9 g CexC m−2 day−1, attainable at latitude 15° with a system of optical depth of 0.1 m and a dilution rate of D = 0.25 day−1, under an f/4 nutrient regime (see Fig. 6b). This assumes that all CexC is of use for biofuels production. The maximum CexC content of the simulated microalgae was 63 % of algal C-biomass (compare Fig. 6 with Fig. S3) but this coincided neither with peak AP (with 9 % CexC) or peak AXP (48 % CexC). The typical CexC content using f/4 medium ranged between 10 and 60 % for strain S and 30 and 60 % for strain F, depending on latitude and dilution rate. For comparison with strain S, lipid content of up to 60 % has been measured in strains of Nannochloropsis under N deprivation (Rodolfi et al. 2009) whereas for Scenedesmus (cf. strain F) optimised lipid content has been reported at 58 % (Mandal and Mallick 2009). A review of reported lipid content values is provided by Mata et al. (2010).
The shallow nature of the optimal depth required to assure a production of biofuels becomes a challenge if considering flat raceways for cultivation; requirements for adequate mixing and high susceptibility to evaporation and temperature fluctuations in such shallow ponds mean that it is often impractical to operate raceway systems with depths less than 0.15 m (Tredici 2007; Ritchie and Larkum 2012). Whilst not detrimental to AP per se, the subsequent lower VP resulting from this pragmatic limitation decreases the potential profitability by increasing demand for water and nutrients and increasing harvesting costs. Furthermore, the results in Fig. 7a imply there is also a direct adverse effect on AXP. Increasing the optical depth to 0.15 m (while keeping D fixed) leads to a decrease in AXP of between 10 and 25 % (depending on latitude) compared to the potential peak value. Increasing depth further to 0.2 m results in a halving (or worse) of AXP compared to production under optimal conditions. To some extent, this can be mitigated by adjusting the dilution rate appropriately, as Fig. 8b shows; slowing the dilution rate from D = 0.25 to 0.1 day−1 limits the decrease in AXP from peak values to about 20 %. Even so, if the system which produced the peak value in AXP quoted above was limited in practice to a depth of 0.2 m with dilution slowed to D = 0.15 day−1, peak AXP would not exceed 0.8 g CexC m−2 day−1.
These factors, and given that the model is producing results consistent with data from real systems (“Comparing the model projections with empirical evidence” section), appears to provide a robust estimation of the upper potential for solar-powered microalgal biofuels production of 3 t biofuels ha−1 year−1, which equates to ~4,000 L ha−1 year−1 assuming a carbon fraction of 720 gC L−1 (which is typical for diesel fuels (Miguel et al. 1998)) and that all of the excess C can be recovered and is in the form of lipids. While outperforming many land-based crops, these results imply algae are not appreciably more productive for biofuels and can be even less so in comparison with, as an example, palm oil (Chisti 2007; Schenk et al. 2008; Mata et al. 2010; Scott et al. 2010). This upper limit is in agreement with the calculation performed by Walker (2009) and far below many estimates of theoretical limits (Weyer et al. 2010). In reality, it is unlikely (if not impossible) that optimal culturing conditions can be maintained long enough to achieve the kind of results for biomass and biofuels production obtained in these simulations. To be able to quantify this further requires a detailed sensitivity analysis of risk factors. Even so, should it be possible to overcome the technical difficulties, the physiological limits on cell growth constrain the potential for algae as a feedstock for biofuels.
As a result, the potential for biofuels production from microalgae appears of questionable commercial viability, unless a step change can be attained in algal physiology through GM, with all of its attendant risks. For instance, Flynn et al. (2013) demonstrated through simulation how engineering strain characteristics to allow greater capacity for photosynthetic efficiency coupled with a decrease in the maximum Chl:C ratio could boost productivity by up to five times that of natural strains. They projected a maximum CexC production rate of AXP = 7.5 g CexC m−2 day−1 = 20,000 L ha−1 year−1 of biodiesel. At the same time, they also demonstrated how the creation of such unpalatable, highly productive strains (desirable traits for biofuels production) could easily lead to harmful, even catastrophic, blooms if they escaped into nature.
To date, even with more optimistic production estimates, there remains much uncertainty for the economic potential for microalgal biofuels production (Liu et al. 2012; Sills et al. 2013). In life cycle analyses, this uncertainty is dominated by sensitivity to the algae’s lipid content and growth rate (Stephenson et al. 2010; Davis et al. 2011). Unfortunately, the biological modelling components within otherwise complicated LCA scenarios are invariably based on assumptions and generalisations derived from literature which lead to projections of several hundreds of tons of biomass produced per hectare per year (Williams and Laurens 2010) whereas in practice (as seen from the references above) 60 t dw ha−1 year−1 is the highest claimed to date. Even if 60 t could become the rule rather than the exception, only a fraction of that (ca. at most 50 %) can be expected to constitute stock for biofuels. Our results indicate production of biomass AP below 30 t dw ha−1 year−1 and biofuels feedstock AXP up to 8 t dw ha−1 year−1.
In consequence of all of these interacting events, conducting a full LCA on the commercial viability of the whole process (whether for biomass, biofuels or other products) requires an integrated approach taking into account the physiology of the microalgae that lay at the heart of the whole endeavour. At present, LCAs take scant regard of this issue and in consequence may be at significant variance from reality. It is likely that the inherent uncertainty will remain unresolved until LCAs become coupled to mechanistic models (such as the one used here) that can more adequately capture the dynamic physiological subtleties of microalgal growth. Combining these informative but differing computational approaches can provide a powerful tool that will allow operators to explore realistic options leading towards improved production.
Areal vs volumetric production
The emphasis above has been upon areal production of biomass (AP) and of excess C as biofuels (AXP). In the Appendix are the corresponding volumetric production values (VP, VXP, respectively). For commercial operations, it is important to maintain an optimum balance between AP and VP. However, this ideal is conflicting as the highest VP requires very low optical depths, which do not then permit high AP (see Flynn et al. 2010a). In oceans, with optical depths of many tens of meters, VP is extremely low, but AP by fast-growing phytoplankton at upwelling zones (ca. 3–4 gC m−2 day−1 (Field et al. 1998)) can match rates in shallow ponds (Flynn et al. 2013) in short bursts during spring blooms. Figure S1 illustrates how a high VP requires a shallow optical depth to prevent light limitation and this has the added benefit of diminishing the demands for nutrients and water, and hence harvesting costs. The trade-off comes as the resulting low volume minimises AP. Increasing depth to boost AP suppresses VP but the rate of depth increase initially outpaces the rate of VP decrease and so the AP continues to rise. After a certain point, VP decreases in proportion to the increase in depth which leads to the saturation of AP seen in Fig. 4.
For CexC, the corresponding rise in AXP (Fig. 7) as VXP falls (Fig. S4a) initially follows the trend for AP as optical depth increases. However, beyond a critical optical depth (~0.1 m) light rather than nutrient becomes the limiting factor; CexC production is suppressed and VXP decreases faster than the depth increases leading to the fall in AXP. Production of C-rich products (e.g. biofuels) is, therefore, more sensitive to the conflict between areal and volumetric production than is biomass production. This conflict will extend directly to costs for space (and/or PBR infrastructure) and in preparing and handling different volumes of water/algal suspension and nutrient loadings. Flynn et al. (2010a) described this interaction using a function relating AP and VP to what they termed a commercial production index. While this provided a single index, in the commercial world the costs of land, energy, nutrients, and other resources would apply differential weights to AP vs VP.
The results from the simulations presented here enable a number of useful conclusions to be drawn in this regard. Most notably, the routes to maximising production of biomass are not the same as those for maximising fatty acid/biofuels production. That said, while individual needs may vary, the optimal depth for commercial cultivation of wild-strain (non-GM) phototrophic microalgae in a facility intended for multiple applications should be approximately 0.1 m (a value consistent with that suggested by García-González et al. (2003) and Ritchie and Larkum (2012) who also place an upper limit on useable pond depth around 25 cm) coupled with the use of nutrient loads of around f/2 containing 12.35 mg N L−1 and 1.11 mg P L−1 (Guillard and Ryther 1962) for biomass production, and f/4 for biofuels production. Further, in general, stimulating biofuels production requires the combination of a fast-growing strain, and nutrient deficiency, which is promoted by shallow optical depth and relatively slow dilution rates.
Not surprisingly, production at high latitudes is projected to be far more seasonally dependent than at lower latitudes, as seen for both biomass in Fig. 2 and CexC in Fig. 5. However, while biomass production in mid-winter may be so poor as to likely not be commercially viable, the longer summer days have potential to provide a window for increased production over the summer months sufficient to ensure viability. That may be especially so if the intended use of the biomass in support of seasonal aquaculture activities. This paints a qualitatively similar picture to the gross photosynthesis calculations of Ritchie (2010) and to Williams and Laurens (2010) who suggest a lack of sufficient irradiance over winter months restricts areal production at high latitudes to little more than half of that possible at equatorial latitudes, compared to our prediction of around 60 % of maximum. However, Williams and Laurens’ estimate of absolute values for production (obtained by assuming either a 3 or 10 % bioenergetic yield with a biomass calorific value of 24.7 kJ g−1 dw) are larger than the mechanistic model calculations by a factor of 10. Our values are more keeping with empirical values from the literature (‘Comparing the model projections with empirical evidence’ section).
The optimal dilution rate depends primarily on the maximum growth achievable by the strain cultivated. Even though growth may be conducted at low dilution rates, a high U
is a desirable trait to select or engineer into microalgal strains (Flynn et al. 2013). However, this trait is likely to be selected against during long-term enforced slow growth in continuous culture systems (Flynn 2009). Away from tropical latitudes, location becomes an additional factor. The extent to which it does so also depends upon the maximum growth rate; production using slower growing strains is more sensitive to the choice of dilution rate with increasing latitude than it does for a faster growing strain; the optimal dilution rate progressively decreases the further from the equator the facility is situated (Figs. 3 and 6). Furthermore, as faster-growing strains are less sensitive to the choice of dilution rate, the most appropriate dilution rate may not necessarily be the one that supports maximum biomass yield. Figure 1a shows that decreasing the dilution rate for the strain with U
= 1.386 day−1 from D = 0.35 to 0.2 day−1 uses <60 % of the nutrients and water but still returns 90 % of peak AP (1.8 cf. 2 gC m−2 day−1). Such a saving of resources is likely to impact significantly on commercial viability, especially as nutrient prices increase.
The results from Fig. 1, and indeed the seasonal variability in production at high latitudes, imply a single objective optimisation method may be insufficient. A more sophisticated and effective method could be to use multi-objective optimisation to balance between maximising biomass production and minimising resource consumption, and also consider financial inputs and outputs. At extremes, one could consider changing algal strains or system optical depths (Olguin et al. 2003; Moheimani and Borowitzka 2007), but more readily changed are facility operational procedures such as dilution rate.
While higher-plant crops are grown in what amounts to discontinuous culture, industrial-scale microbial growth is most commonly within continuous culture systems. All of the simulations performed here have been run in such a chemostat mode, with continuous dilution and harvesting at a constant rate. While there are problems with using such an approach (notably strain selection to match maximum growth rate to dilution rate, and the risk of establishing pest-predators), there are also distinct logistic problems in discontinuous approaches. These include the necessity to rapidly drain and harvest large volumes of medium containing the biomass, and replace the same volume with fresh medium and nutrients. Exposure of the newly diluted remaining culture to high light then risks photodamage, especially if the organisms are nutrient stressed (Geider et al. 1993).
Improvements to the use of continuous dilution methods as simulated here would be to consider seasonal changes to the dilution rate and/or through the introduction of discontinuous harvesting methods into the simulations. From our results, it appears that an optimisation of dilution rates separately for winter and summer months could be sufficient. For instance, from the results in Fig. 2b
i and iv, it seems more appropriate to run with a substantially slower dilution rate in winter at high latitudes and increase dilution as production ramps up later in the year.
To fully automate the optimisation process, a real-time regulation of production rates is needed. Computationally, this can be achieved using a simple predictor-corrector. At regular intervals, a prediction is made of the output at the end of a time period t + m∆t based on the current dilution rate and compared to the output for a set of dummy rates. Whichever gives the best result becomes the rate for that time period. Methods of intelligent harvesting (whether manual or automated) would have to rely on real-time monitoring of culture conditions, with several measurements taken simultaneously. Monitoring and regulation of external nutrient levels would be quite straightforward, deploying nutrient probes directly exposed to the culture medium. Regulating dilution based on algal physiology is more problematic, though monitoring of biomass and photophysiology using F
to monitor the efficiency of the PSII photosystem are obvious starting points. Measurement of the fluorescence emission spectrum can reveal levels of nutrient stress in the culture (Masojídek et al. 2000); a threshold value could be used as a trigger for further nutrient injection. Light absorption is an indicator of total biomass (Griffiths et al. 2011) and measurement of changes in turbidity can provide an estimation of growth rates, while analysis of the absorption spectrum can quantify the amount of chlorophyll per unit of biomass. Thus, if a measurement of turbidity and/or F
indicates an aberrant change in production then the dilution rate can be automatically adjusted to compensate.
Intelligent control of dilution rates and nutrient addition is one way to optimise yields, but a more radical control scenario would enable switching between periods of continuous and discontinuous operation. This leads to further questions as to what the optimal dilution and harvesting strategies may involve in different modes, and how these may relate to other factors such as the control of pests. These topics will be considered in future papers.