Performance of upflow anaerobic sludge blanket (UASB) reactor and other anaerobic reactor configurations for wastewater treatment: a comparative review and critical updates

Aerobic treatment of wastewater has been traditionally used for decades for high concentration waste. However, sludge disposal is becoming increasingly difficult due to stringent laws and aerobic processes producing a large amount of sludge (Christensen et al. 1984; Khan et al. 2011). This is with the exception of conditions where anoxic-aerobic digesters are adopted for waste activated sludge treatment. In this case, the energy savings of anaerobic digesters are sacrificed for lower technical competence requirements. This, however, encourages further nutrient removal through nitrification and phosphorus precipitation. However, for the treatment of influent primary sludge, anaerobic digestion has been accepted as a desirable alternative, the likes of which are used worldwide for the treatment of a variety of different wastewaters (Cronin & Lo 1998; Mahmoud et al. 2004; Ince et al. 2005; Miranda et al. 2005; Gohil & Nakhla 2006; Akbarpour Toloti & Mehrdadi 2010; Kerroum et al. 2010). Moreover, aerobic processes require large amounts of energy to ensure proper aeration and hence are not only expensive to install, but also costly to run. Anaerobic processes are smaller than aerobic processes and therefore require a smaller area of land in which to be installed (Christensen et al. 1984; Rajakumar et al. 2011; Chong et al. 2012). Further, they require a smaller power supply (Lettinga et al. 1980; Christensen et al. 1984; Akbarpour Toloti & Mehrdadi 2010; Chen et al. 2011; Khan et al. 2011), cost less to run and produce less sludge which can be preserved for a number of months, reducing the need for sludge disposal (Lettinga et al. 1980; Sawayama et al. 1999; Rajakumar et al. 2011; Chong et al. 2012). In addition, anaerobic processes produce biogas which is mostly a mixture of CO₂ and methane that can be burned as a replacement of fossil fuels (Batstone et al. 2002a, 2002b; Chen et al. 2011; Thamsiriroj & Murphy 2011).

Upflow anaerobic sludge blanket (UASB) reactor is a widely used anaerobic treatment of wastewater that has numerous benefits over its alternatives, including aerobic wastewater treatments (Lettinga et al. 1980; Christensen et al. 1984; Khan et al. 2011) and other anaerobic treatments (Hamdi 1996; Rajeshwari et al. 2000; Rajakumar et al. 2011; Ayaz et al. 2012). UASB was first developed by Lettinga et al. (1980) and is now one of the most widely used high rate anaerobic wastewater treatment systems due to its simple design, construction (Chen et al. 2011; Rajakumar et al. 2011) and low operating costs (Christensen et al. 1984; Akbarpour Toloti & Mehrdadi 2010; Chen et al. 2011). For these reasons, and because it is able to withstand the fluctuations in pH, temperature and influent composition that are so common in industrial wastewater, full-scale UASB reactors have come into operation to treat a variety of wastewaters since its introduction. These include breweries (Ahn et al. 2001; Parawira et al. 2005), slaughterhouses (Miranda et al. 2005), canning (Puñal & Lema 1999), alcohol distilleries (Ince et al. 2005) and pharmaceutical wastewaters (Chen et al. 2011) as well as sewage (Schellinkhout & Collazos 1992; Schellinkhout 1993). Because of its promise as a high throughput reactor, the UASB reactor has been studied under many different conditions, including varying types of wastewater, temperature, organic loading rate (OLR) and pH. The effects and importance of mixing in the reactor have also been studied. For example, Habeeb et al. (2010) studied the properties attributed to the digestion process in UASB reactors. Seghezzo et al. (1998) compared sewage treatment in UASB and expanded granular sludge bed reactors. The impact of granule formation on wastewater treatment in UASB reactors was examined by Abbasi & Abbasi (2012), while Singh et al. (1999) had earlier studied the requirements of nutrients in a UASB system. The success of solids removal in UASB reactors has been described by Mahmoud et al. (2003) while Lin & Yang (1991) carried out a general technical review on the UASB process.

Mathematical modeling of wastewater treatment processes is a key component in the support of decision-making (Daigger 2011; Rodriguez 2011) when designing treatment plants (Henze et al. 2000; Morgenroth et al. 2002). Models can help with technology-based design decisions, optimal sizing of the wastewater treatment plants, prediction of the nature of the effluent (Daigger 2011) and the effect of the plant on the environment (Rodriguez 2011). The effect of changing different parameters on the performance of reactors can be modeled and this can be used during operation to spot and troubleshoot errors (Henze et al. 2000). These models have also been used extensively in teaching and research purposes as they can help researchers with experimental design to indicate where further research is needed. Anaerobic Digestion Model No. 1 (ADM1) is a general model for anaerobic digestion developed by the IWA Anaerobic Digestion Modeling Task Group (Batstone et al. 2002a, 2002b) and has been widely used to model a wide variety of anaerobic digestion processes (Batstone & Keller 2003; Batstone et al. 2006), including the UASB (Batstone et al. 2005; Coelho et al. 2006) since its first implementation in 2002. It must be noted that ADM1 is a tool that allows predictions of sufficient accuracy to be useful. Because of the varying demands in process development, operation, and optimization, a different degree of model calibration and validation is required in each case.

Within the scope of this review, it is intended to collect the experimental studies carried out on the UASB reactor, as well as to compare it with other anaerobic treatment options. Furthermore, common modeling approaches for a UASB process have been reviewed. The most common reactor type used for anaerobic digestion of wastewaters is the continuous stirred tank reactor (CSTR). Therefore, an example case to study the implementation of IWA anaerobic digestion model ADM1 has been performed on a pilot-scale anaerobic CSTR with the implications that it can be extended to advanced anaerobic reactors such as UASB.

A CSTR is a simple and inexpensive design for an anaerobic treatment process (Chamy et al. 2011). As the name suggests, it consists of a tank with a motorized stirring arm inserted usually through the top of the reactor and operated constantly to ensure continuous stirring. It can be continuously fed through the bottom of the reactor (Méndez-Acosta et al. 2010) or intermittently fed in a batch process (Wendland et al. 2007) and is often seeded with sludge before operation (Oz et al. 2003; Boušková et al. 2005; Wendland et al. 2007). Figure 1 shows a typical CSTR reactor for anaerobic treatment of wastewater. The anaerobic CSTR requires high hydraulic retention time (HRT) of between 15 and 20 days, compared with often less than 1 day for processes such as UASB. These reactors can also only cope with much lower OLRs (Rajeshwari et al. 2000). Good et al. (1982) found that a fixed film reactor could cope with an organic loading rate five times greater than that of a CSTR. Furthermore, the complexity of the influent to be treated affects the performance of the reactor more than in the other types of treatment described. It has been found that attached biomass reactors, such as the fluidized bed reactor, is superior in coping with shock loads. In spite of these disadvantages, Boušková et al. (2005) has shown that the anaerobic CSTR can cope well with step changes in temperature, and it has been used to treat a number of wastewaters including blackwater from vacuum toilets and kitchen refuse (Wendland et al. 2007), glucose-containing wastewater (Oz et al. 2003) and tequila vinasses (Méndez-Acosta et al. 2010), although tequila vinasses has only been treated on a pilot scale and has not yet been successfully implemented full scale. Oz et al. (2003) found that the anaerobic CSTR was not successful in treating pharmaceutical wastewater due to its complexity, and that an anaerobic filter reactor was much more successful.

The anaerobic contact process is an extension of the anaerobic CSTR in that it requires a continuously stirred digester followed by a settling tank (Nähle 1991; Cakir & Stenstrom 2007), as shown in Figure 2. In this settling tank, the bacteria is allowed to settle from the effluent and is able to be scraped from the bottom and recycled back to the digester (McCarty 1964; Endo & Tohya 1988; Hamdi 1996). This allows for an increase in HRT and a decrease in solid retention times (SRT) (Nähle 1991; Cakir & Stenstrom 2007). A high efficiency is needed in this recycling (McCarty 1964) which can be problematic: a degasifier is often required between the digester and the settler to overcome this problem (Cakir & Stenstrom 2007). Biogas is removed from the sludge in the degasifier to achieve better settleability of the biomass in the settling tank, but this can be expensive (Nähle 1991). The mixing in the digester can be achieved through mechanical means or by the recirculation of the biogas produced in the process (Nähle 1991; Cakir & Stenstrom 2007), but care must be taken as excessive mixing can favor the growth of acidogenic bacteria over methanogenic bacteria which can lead to lower feasible OLR (Hamdi & Garcia 1991; Hamdi 1996). The loading rates that can be applied to the contact process are higher than those that are achievable in the aerobic activated sludge process (Endo & Tohya 1988). The contact process can be quick to reach steady state, however this state can be more unstable compared to other processes (Hamdi & Garcia 1991). This process has been used to treat a variety of wastewaters including distilleries, breweries, meat packing, sugar, dairy and vegetable canning (McCarty 1964; Nähle 1991; Cakir & Stenstrom 2007) wastewaters.

The anaerobic filter (AF) consists of a column which is filled with a packing material (Chian & De Walle 1977; Hickey et al. 1991), as can been seen in Figure 3. This packing media promotes biomass retention in the reactor, as the microorganisms can attach to the media or become entrapped. Along with those attached to the packing media, microorganisms also exist as granules or flocs in suspension, and this is where much chemical oxygen demand (COD) removal occurs (Hickey et al. 1991; Rajakumar et al. 2011). The choice of packing media can greatly affect the performance of the reactor (Manariotis & Grigoropoulos 2006a, 2006b). Characteristics of the media that are particularly important include the texture, porosity and specific surface area. It has been found that a media with high porosity is more effective than one with a lower porosity and higher SSA (Manariotis & Grigoropoulos 2006a, 2006b). The most important factor when choosing a packing media is that it is one to which bacteria can easily attach or become entrapped. Hickey et al. (1991) found that the AF reactor would not recover well from large pH and temperature upsets but Manariotis & Grigoropoulos (2006a, 2008) found that the AF responded well to small changes in operating conditions and restarted reasonably well after long idle periods of up to two years, although this restart time did vary with packing material and previous history of the reactor. A common issue found in literature for the AF reactor is that sludge accumulation can cause clogging (Hickey et al. 1991; Manariotis & Grigoropoulos 2006a) and hence periodic flushing of the reactor may be necessary. The AF reactor can be operated with up- (Hickey et al. 1991; Manariotis & Grigoropoulos 2006a; Rajakumar et al. 2011), down- or horizontal-flow (Manariotis & Grigoropoulos 2006b) and has been seen to treat low strength domestic (Manariotis & Grigoropoulos 2006b), high and low strength poultry (Rajakumar et al. 2011) and high acidic (Chian & DeWalle 1977) wastewaters, as well as other low and high strength (Hickey et al. 1991; Manariotis & Grigoropoulos 2008) wastewaters.

The anaerobic fluidized bed reactor (AFBR) consists of a column that contains a bed of fluidizing media, often sand (Heijnen et al. 1989; Converti et al. 1990; Maragno & Campos 1992; Marín et al. 1999). The configuration of an AFBR is shown in Figure 4. Start-up conditions are set to promote colonization of microorganisms on the bed particles (Hickey et al. 1991). The flow of wastewater is introduced at the bottom of the reactor and is high enough so as to cause fluidization of the bed (Heijnen et al. 1989; Hickey et al. 1991; Marín et al. 1999): the media particles become covered in a biofilm and hence become heavy and require significantly higher upflow velocities than in other reactors to become fluidized. The growth of the film on the media gives rise to a high biomass concentration (Tseng & Lin 1994; Mendonça et al. 2004) and high specific surface area (Carbajo et al. 2010), meaning high mass transfer efficiency (Marín et al. 1999) and reaction time (Converti et al. 1990) with no clogging of the reactor. The control of this biolayer thickness must be included in operation. A thick layer has been shown to cause washout, therefore cleaning of the media may be needed, which could incur extra expense (Heijnen et al. 1989). However, the AFBR generally finds it easy to retain the biomass in the reactor (Heijnen et al. 1989; Converti et al. 1990; Marín et al. 1999), requires smaller volume than other anaerobic wastewater treatment processes (Mendonça et al. 2004; Carbajo et al. 2010) and can resist changes to environmental conditions and toxic substances (Tseng & Lin 1994). AFBR has been used to successfully treat sewage (Mendonça et al. 2004), wine wastewater (Converti et al. 1990) and phenol-containing wastewater (Carbajo et al. 2010) as well as other low and high strength wastewaters (Tseng & Lin 1994). Wastewater from textile production has also been treated, although an additional carbon source needed to be added (Haroun & Idris 2009).

The UASB reactor is generally a tubular shaped reactor, consisting of four zones (Lin & Yang 1991). The sludge bed is a layer of biomass that forms at the bottom of the reactor, and contains granules of a high settling velocity. It is where the main biochemical reactions occur (Alibhai & Forster 1986; Seghezzo et al. 1998; Schmidt & Ahring 2000). Above this forms a blanket of finely suspended flocs with lower settling velocity (Singhal et al. 1998; Kalyuzhnyi et al. 2005; Chong et al. 2012). This suspension forms due to the production of biogas in the reactor – formed from the degradation of soluble organic compounds – and is a mixture containing these gases (Lin & Yang 1991; Schmidt & Ahring 2000). Above this is the settling zone, in the upper part of the reactor. This is a quiescent zone where the granules that have been dislodged from the bed by the upflow of wastewater or gas production with a high enough settling velocity will settle back to the bed (Lettinga et al. 1980; Singhal et al. 1998). Dispersed bacteria can be washed out (Schmidt & Ahring 2000; Chong et al. 2012), but this can be minimized and even completely prevented through correct operation and conditions (Seghezzo et al. 1998). The fourth zone is the integrated three phase separator (Sawayama et al. 1999; Schmidt & Ahring 2000; Chong et al. 2012). The liquid will flow through the overflow and gas escapes through a tube (Christensen et al. 1984). This inclusive separator means that the biogas produced can be collected and used as a replacement for fossil fuels in energy production. The reactor is fed from the bottom, and the wastewater flows upwards through the reactor. This upflow and the production of biogas causes mixing and agitation, and hence provides a good contact between the wastewater and the biomass (Seghezzo et al. 1998; Chong et al. 2012). A schematic diagram of the UASB reactor is shown in Figure 5.

The COD removal occurs over the bed and blanket zones of the reactor. The main biochemical reactions occurs in the bed, and further biodegradation occurs in the blanket (Alibhai & Forster 1986; Chong et al. 2012). UASB reactors have high biomass concentrations, and hence the removal of COD is good even at highly concentrated influents, or a high volume of waste (Lin & Yang 1991; Liu et al. 2003). Rajakumar et al. (2011) found that a UASB reactor could achieve a higher removal efficiency than a similar anaerobic filter reactor and Hamdi (1996) found that the UASB could cope with much higher organic loading rates than that of the anaerobic contact process and in general it was much more efficient at treating olive mill wastewaters. Short HRT, and the fact that HRT is independent of SRT in the UASB reactor, means that treatment time is reduced significantly (i.e. from days to hours). To reduce start up times, the reactor may be initially seeded: problems may occur if the seed used is not cultivated on a similar composition to that of the influent to be used (Christensen et al. 1984; Chong et al. 2012).

Successful operation of a UASB reactor depends greatly on the formation of the sludge granules that make up the sludge blanket. These granules must be well settling and significantly aggregated in order to remain in the bed in spite of the upflow velocity (Seghezzo et al. 1998; Schmidt & Ahring 2000; Liu et al. 2003). The mechanisms for the development of the granules are complicated and are not fully known, however it is suggested that the bacteria attach themselves to heavy sludge ingredients, and the growth of the bacteria concentrates around these and forms granules (Lin & Yang 1991; Liu et al. 2003).

Guiot & van den Berg (1985) first introduced the idea of a hybrid UASB reactor by keeping the bottom two-thirds of the reactor as a UASB similar to that described, and incorporating an anaerobic filter (AF) in the top third of the reactor to treat sugar wastewater. Figure 6 shows the configuration of the hybrid reactor. A COD removal efficiency of 93% was achieved for OLRs of up to 26 kg COD m^–3 d^–1. The authors also commented that the filter section achieved effective biomass retention. Hutňan et al. (1999) compared laboratory-scale UASB and hybrid UASB-AF reactors treating synthetic wastewaters. They found that comparable COD removal efficiency was achieved in both reactors. Biomass washout occurred in the UASB reactor at an OLR of 6 kg COD m^–3 d^–1 but did not occur in the hybrid reactor until an OLR of 12 kg COD m^–3 d^–1 was applied, which is in agreement with the observation made by Guiot & van den Berg (1985). Similar reactors were also compared by Kalyuzhnyi et al. (1997) for the treatment of soft-drinks wastewater. Again, both reactors performed similarly for applied OLRs up to 8 kg COD m^–3 d^–1 achieving COD removal efficiencies of over 80%. However, when an OLR of 12 kg COD m^–3 d^–1 was applied to both reactors, the efficiency of the UASB dropped to 60–70%, whereas it remained above 80% in the hybrid reactor. This was attributed to the superiority of the hybrid reactor for biomass retention. Using coal wastewater, Ramakrishnan & Surampalli (2012) observed COD removal efficiency for HRT ranging from 0.75 to 0.33 days to both reactors. The UASB reactor exhibited slightly less removal efficiency than the hybrid reactor. For example, at HRT of 0.39 days the removal efficiencies achieved were 88 and 92% in the UASB and hybrid reactors respectively. The authors attributed this to better granulation and biofilm establishment in the hybrid reactor due to the AF section. These studies highlight that UASB reactors can successfully be combined with an anaerobic filter to reduce biomass washout, with no need for an additional unit.

Low strength influent wastewater can cause a substrate shortage inside anaerobic treatment reactors, and this can cause deterioration of the sludge in the reactor (Yoochatchaval et al. 2008a, 2008b). To counter this, a modification of the UASB reactor can be considered, called an expanded granular sludge bed reactor (EGSB). It is operated in a similar way to the UASB, and the configuration of the reactor is the same as in Figure 5. However, it is operated with a higher upflow velocity, which causes the bed to expand (Seghezzo et al. 1998; Kato et al. 1999). This can increase the sludge-wastewater contact and hence reduces the problem caused by substrate shortage (Yoochatchaval et al. 2008a). The hydraulic mixing in the reactor is intensified and hence the reactor can be considered completely mixed (Seghezzo et al. 1998). The increase in upflow velocity can be achieved by effluent recirculation. Some advantages of the EGSB include the high solid retention times and the low accumulation of fats (Núñez & Martínez 1999). This type of reactor has been successfully used to treat wastewaters including: slaughterhouse (Núñez & Martínez 1999), domestic (Chunjuan et al. 2009), sucrose-based (Yoochatchaval et al. 2008a) and brewery (Xing et al. 2009) wastewaters, as well as other low-strength wastewaters (Yoochatchaval et al. 2008b). However, the increased upflow velocity required could mean higher operating costs than that of a UASB and taller reactors are required to allow for bed expansion (Kato et al. 1999), meaning higher initial costs as well. The COD removal in EGSB has been seen to drop at low temperatures such as 15 °C (Xing et al. 2009).

The anaerobic baffled reactor (ABR) is a high rate anaerobic treatment that can be described as a series of UASBs (Bachmann et al. 1985; Hutňan et al. 1999; Dama et al. 2002; Manariotis & Grigoropoulos 2002; Krishna et al. 2007; Gopala Krishna et al. 2008; Liu et al. 2012) separated by baffles that force the wastewater to flow over and under them as can be seen in Figure 7. A mixed culture of microorganisms (Hutňan et al. 1999) form flocs in the upward and downward chambers (Polprasert et al. 1992; Manariotis & Grigoropoulos 2002) and these rise and settle with gas production, but their movement through the reactor is slow (Gopala Krishna et al. 2008). This means that the wastewater can come into contact with a large amount of active biomass while inside the reactor (Liu et al. 2012). The design of an ABR is simple and inexpensive to manufacture (Polprasert et al. 1992; Manariotis & Grigoropoulos 2002). However, if large numbers of compartments are needed, costs can rise significantly. The number of compartments in an ABR can vary from three to 11, but 4–5 compartments is generally recommended (Manariotis & Grigoropoulos 2002). Often most of the COD removal takes place in the first two compartments. An advantage of the ABR is that the methanogenesis and acidogenesis often separate into individual compartments, allowing for specific microorganisms to dominate in each compartment (Barber & Stuckey 1999; Hutňan et al. 1999; Langenhoff & Stuckey 2000; Dama et al. 2002). ABRs are also tolerant to shock loads (Langenhoff & Stuckey 2000) and have a small amount of washout of biological solids (Bachmann et al. 1985; Liu et al. 2012) compared to similar UASB and hybrid UASB reactors (Hutňan et al. 1999). It has been found that ABRs should be avoided if there is a large number of particulates in the wastewater to be treated (Bachmann et al. 1985; Barber & Stuckey 1999). The solids can displace the biomass in the reactor and build up inside, causing reduced contact between the wastewater and biomass (Barber & Stuckey 1999). Hutňan et al. (1999) found the amount of NaHCO₃ that was needed to be added to the ABR is double compared to the UASB and hybrid reactor to maintain the pH. It has also been found that UASB reactors, when compared with ABR, perform better at lower temperatures (Ayaz et al. 2012). ABR reactors have been shown to successfully treat low (Barber & Stuckey 1999; Krishna et al. 2007), medium (Barber & Stuckey 1999) and high strength (Barber & Stuckey 1999; Langenhoff & Stuckey 2000) wastewaters as well as slaughterhouse wastewater (Polprasert et al. 1992). As with other waste treatment unit operations, the selection of appropriate anaerobic systems is specific to the waste characteristics and regional constraints. Various modifications of anaerobic digestion reactors have been utilized in the past to ensure that the system could cope well with variations in temperature, and organic loading rates (e.g. CSTR), increased solid retention times (e.g. with the addition of settling tanks to CSTRs or utilization of UASB type systems), treatment of complex wastewaters (e.g. AFBR), accommodate for substrate shortage in low strength wastewaters (e.g. EGSBR and ABR).

In the past, anaerobic treatments of wastewater were not used in industry since they were not able to handle high organic loading rates and could be unstable in terms of treatment efficiency. The UASB solves these issues as it can handle high organic loading rates and adding a recycle increases the biological population development and hence start-up times are reduced (Christensen et al. 1984). It is also simple in design and construction, which gives rise to low initial costs, and has a low energy consumption, and therefore low running costs (Akbarpour Toloti & Mehrdadi 2010; Chen et al. 2011; Khan et al. 2011; Rajakumar et al. 2011; Chong et al. 2012). It also exhibits the classic advantages attributed to most anaerobic wastewater treatment processes such as biogas production (Sawayama et al. 1999; Kerroum et al. 2010), low sludge production (Khan et al. 2011; Rajakumar et al. 2011) and small land requirements (Christensen et al. 1984; Chong et al. 2012) compared to aerobic processes. The UASB can also withstand fluctuations in pH and hence no extra chemicals need to be added to maintain a desired pH (Akbarpour Toloti & Mehrdadi 2010; Chen et al. 2011). Fluctuations in influent concentration and temperature can also be handled by this reactor. The UASB has a short HRT and a long SRT compared to other processes, and hence can process large amounts of wastewater in a short time (Chong et al. 2012).

There have been a wide range of publications related to the treatment of various types of wastewater by UASB reactors, at full, pilot and laboratory scale and some of these have been summarized in Table 1. Parawira et al. (2005) found that the full-scale UASB reactor used for their studies treated the wastewater from a brewery brewing opaque beer to a satisfactory quality so that it could be discharged into public waters. This treatment included an average COD removal of 57%; total solids reduction of 50% and settleable solids removal of 90%. They suggested brewery wastewater provided ideal conditions for the UASB influent due to its composition and strength and that the UASB reactor presented a more suitable process for the treatment of brewery wastewater than aerobic processes, due to the excessive amount of energy required for aerobic treatment of high strength wastewaters such as those from breweries. They also found that pilot-scale UASB reactors treating similar wastes could provide higher COD removals, up to 90%. Ahn et al. (2001) also studied a full-scale UASB reactor treating brewery wastewater, which was built for pre-treating the wastewater before discharging it to a municipal wastewater treatment plant. The desired treatment level was achieved, giving an average COD removal of 80% over the seven years it was studied. Hack (1985) studied the UASB reactor as a possible pre-treatment for an overloaded aerobic treatment plant on a brewery site. A pilot-scale UASB reactor was built to study its suitability, using the effluent wastewater from the brewery. The concentration of the influent to the reactor was, on average, 1,230 mg COD L^–1 and it was operated at an average temperature of 22 °C. These conditions were similar to those at which the full-scale reactor would be operated and they achieved a COD reduction of 80%, meaning this anaerobic pre-treatment is a promising solution for extending the overloaded aerobic treatment plant. Cronin & Lo (1998) studied the treatment of brewery wastewater by two pilot-scale UASB reactors seeded with different amounts of activated sludge. The reactor seeded with 5.93 g volatile suspended solids (VSS) per liter of activated sludge achieved a satisfactory performance according to the requirements of the authors, reaching maximum COD and VSS removals of 91 and 81% respectively at an HRT of 18 hours. Yan & Tay (1996) studied the treatment of brewery wastewater by a laboratory-scale UASB reactor operating at an ambient temperature of 21.8 °C, an OLR of 12 g COD L^–1d^–1 and an HRT of 4 hours, and achieved a COD removal efficiency of 89.1%.

Puñal & Lema (1999) studied the performance of a full-scale UASB reactor treating fish-canning wastewater. They found that it performed best when the influent contained a mixture of mussel- and tuna-processing wastewater compared to when it only contained tuna-processing wastewater: mussel-processing wastewater contains carbohydrates in the majority, which are easily degradable by the reactor. They found that the UASB reactor performed acceptably, yielding COD removal efficiencies of 70–90% at OLR from 2–8 kg COD m^–3 d^–1 but this could be achieved only through the control of the alkalinity in the reactor. Wongnoi et al. (2007) used pilot-scale UASB reactors to treat the wastewater from a fruit-canning factory and managed to achieve a good COD removal efficiency of between 90 and 95% when running at an HRT of 35 h and an OLR of between 1.2 and 3.9 kg COD m^–3 d^–1. Diamantis et al. (2003) studied a pilot-UASB reactor treating wastewater from fruit canning, specifically peach canning. They found that the UASB was very effective in terms of COD removal for the examined wastewater, as it managed to achieve an average of 86.3% COD removal at an HRT of 10 h. The OLR varied between 4 and 12 kg COD m^–3 d^–1. Despite this high removal, a polishing step was needed to satisfy the treatment requirements. Gohil & Nakhla (2006) studied a laboratory-scale UASB reactor for treating tomato-canning wastewater. They reached a maximum removal efficiency of 96% when operating at an HRT of 2.66 days. Moosbrugger et al. (1993) studied the treatment of grape wine distillery waste in a laboratory-scale UASB system, and achieved a COD removal efficiency of 85% even at loading rates as high as 41 kg COD m^–3 d^–1. From this study, they concluded that grape wine distillery waste is suitable for treatment in UASB systems. Akarsubasi et al. (2006) compared the performance of two full-scale UASB reactors treating alcohol distillery wastewaters of similar composition. They called the two reactors IUASB (refers to the UASB reactor in Istanbul, Turkey (Akarsubasi et al. 2006)) and TUASB (refers to the UASB reactor in Tekirdag, Turkey (Akarsubasi et al. 2006)), and the latter reactor was seeded with sludge from the former reactor. IUASB performed better in terms of COD removal than TUASB by achieving 90% removal at OLRs between 6 and 11 kg COD m^–3 d^–1. TUASB only reached 60–85% removal at 2.5–8.5 kg COD m^–3 d^–1, the maximum removal of 85% being achieved at 8 kg COD m^–3 d^–1. The authors concluded that the reason for the difference in performance was due to the smaller OLR used in TUASB and recommended that higher OLRs should be used to improve the performance of that reactor. Ince et al. (2005) compared the performance of these two reactors with a third reactor CUASB (refers to the UASB reactor in Canakkale, Turkey (Ince et al. 2005)) which was also seeded with sludge from IUASB. The CUSAB reactor achieved removal efficiencies of between 70 and 85% when operated with OLRs of between 1 and 4.5 kg COD m^–3 d^–1. The efficiency was highest at the largest loading rate, but was still lower than that of the IUASB and so the authors again recommended a higher loading rate be applied to increase the efficiency. Wolmarans & De Villiers (2004) studied the seasonal treatment of alcohol distillery wastewater by a full-scale UASB reactor. The OLRs applied were between 4 and 18 kg COD m^–3 d^–1 and COD removal efficiencies greater than 90% were achieved. They noted that the stop–start nature of seasonal treatment posed no problems for the plant, suggesting that the UASB process should be considered as a viable option for the treatment of seasonal wastes. Zheng & Hu (2002) compared two methods of start-up and subsequent operation of laboratory-scale UASB reactors treating pharmaceutical wastewaters. The first process involved supplying the reactor with a synthetic wastewater to achieve the start-up before swapping to the pharmaceutical wastewater to be treated subsequently. Once the reactor had begun to be supplied with the pharmaceutical wastewater, it achieved a COD removal efficiency of 78% when operated with an OLR of 4.79 kg COD m^–3 d^–1. The second process involved supplying the reactor with pharmaceutical wastewater from the start of operation. Start-up in this way managed to achieve a COD removal efficiency of 77.7% while operating at an OLR of 4.76 kg COD m^–3 d^–1. The authors concluded that both these processes are viable for the start-up and subsequent operation of UASB to treat pharmaceutical wastewater. Akbarpour Toloti & Mehrdadi (2010) also studied a laboratory-scale UASB reactor treating chemical synthesis pharmaceutical wastewater. The UASB reactor was chosen for this particular study as the authors had noted that it had shown itself to be effective in treating high strength wastewaters, and pharmaceutical wastewaters are generally high strength. It was shown that generally an increase in OLR and a decrease in HRT increased the removal efficiency and the optimum operating conditions of an OLR of 10.81 kg COD m^–3 d^–1 and an HRT of 33.7 hours yielded a COD removal efficiency of 54%. Satyanarayan et al. (2009) set up a pilot-scale UASB reactor for treating wastewater from a herbal pharmaceutical plant. They selected the UASB for reasons in agreement with Akbarpour Toloti & Mehrdadi (2010): herbal pharmaceutical wastewaters are high strength and therefore should be well treated in a UASB system. The reactor achieved COD removal efficiencies of 86.2–91.6% when operating at OLRs between 6.26 and 10.33 kg COD m^–3 d^–1 and so the authors concluded that the UASB reactor is feasible for herbal pharmaceutical wastewater treatment, adding that the process is very cost effective and economical. Chen et al. (2011) studied the performance of a full-scale UASB reactor, also treating chemical synthesis pharmaceutical wastewater. The OLR for the full-scale reactor was between 12.57 and 33 kg COD m^–3 d^–1 and the average COD reduction was 52.2%, with the maximum reaching 85%.

Borja et al. (1994) studied a laboratory-scale UASB reactor that was treating slaughterhouse wastewater. The reactor managed to achieve a COD removal efficiency of 99% when operated at an HRT of 4 days and an OLR of 1.27 kg COD m^–3 d^–1. The HRT was then reduced from 4 days and this had no significant effect on the removal efficiency until the HRT reached 0.8 days, where a significant drop of COD removal occurred. However, for HTRs above 0.8 days, the removal remained above 95%, and hence the laboratory-scale UASB was successful in treating the slaughterhouse wastewater. It is also important to study the performance of full-scale reactors to demonstrate the effectiveness of the system in industry applications. A pilot-scale UASB reactor was studied by Nacheva et al. (2011) for its performance in treating pig and cattle slaughterhouse wastewater from a slaughterhouse in Mexico. The reactor was operated at ambient temperature and the influent to the reactor was pre-treated for solids separation. They found that the COD removal efficiency increased with increasing OLR, reaching 90% at an OLR of 15 kg COD m^–3 d^–1. In addition, the remaining total suspended solids (TSS) were reduced by more than 40% and retained in the reactor, and this did not affect the performance. Miranda et al. (2005) also found that a pre-treatment of the wastewater was needed to reduce the concentrations of oil and grease in the influent to the reactor, the build-up of which led to a system failure before the implementation of the pre-treatment. Once this pre-treatment had been introduced, COD removal efficiencies of between 70 and 92% were achieved when OLRs between 0.86 and 2.43 kg COD m^–3 d^–1 were applied. As a reference to the performance of the reactor, the authors also added that for the year after the pre-treatment was introduced, the concentration of COD in the reactor effluent was consistently below the emission standards value. However, the oil and grease concentrations remained above these standards, leading to the need of post-treatment, as also agreed by Nacheva et al. (2011).

Coelho et al. (2003) investigated the possibilities of using a UASB reactor to replace a septic tank treating sewage from a single family. The pilot-scale reactor reached a COD removal efficiency of 85% and a TSS removal efficiency of 95%, meaning that no post-treatment of the effluent was necessary. These results, when compared to the removal efficiency of the septic tank (75% COD and 75% TSS) and coupled with other benefits of the UASB, such as smaller required area and cost (in Brazil) and little maintenance compared with the ST, led the authors to conclude that the UASB can substitute the septic tank as a single-family wastewater treatment system with significant advantages. Mahmoud (2008) studied the treatment of sewage in two similar systems operating in summer. The UASB-digester system again showed significant improvement of COD removal compared to the single-stage reactor, achieving 72% COD removal efficiency as compared to 54%.

The effect of temperature on the efficiency of a UASB process has been studied, with many researchers focusing on the reactor's ability to treat wastewater at low temperatures. Some of these studies have been summarized in Table 2. Uemura et al. (1995), however, studied the effect of increasing the temperature of a thermophilic UASB reactor that was already operating at a temperature of 55 °C and a loading rate of 18.3 kg TOC m^–3d^–1. The wastewater contained sucrose, and as the temperature was raised to 65 °C the efficiency dropped dramatically to around 65% and started to increase slightly with the reduction of loading rate to 10 kg TOC m^–3d^–1. However, it never returned to its previous efficiency of over 90%. An increase in methane production (from 100 to 130 L d^–1) was observed initially after the increase but then decreased drastically (to 40 L d^–1) after a period of time. This behavior, along with the failure to achieve the initial efficiency, was attributed to the disintegration of the sludge at the higher temperature. Farajzadehha et al. (2012) studied a laboratory-scale UASB reactor treating municipal wastewater. The wastewater was enriched with molasses and milk powder to simulate the treatment of municipal wastewater mixed with industrial wastewater (in this case, wastewater from the dairy industry) and the OLR was kept between 7 and 11 kg COD m^–3 d^–1. The temperature of the reactor was decreased from 30 to 25 °C, then further to 20 °C and the COD removal efficiencies were 85, 81 and 75% respectively, indicating that the COD removal decreases with a decrease in temperature. However, 75% is still a high removal efficiency which could be satisfactory for the treatment of the wastewater. The TSS removal was observed to be slightly higher at the higher temperature (30 °C), but this was not a significant difference. The effect of reducing the temperature in a laboratory-scale UASB reactor treating municipal wastewater was also studied by Singh & Viraraghavan (2003). Two reactors were used in this study and were operated with an HRT of 6 hours. The reactors were then studied for COD removal efficiency at 6, 11, 15, 20 and 35 °C. The efficiency decreased with decreasing temperature (79, 81, 84 and 87% for 11, 15, 20 and 35 °C respectively) but these efficiencies were still considered satisfactory. However, when operating at 6 °C the removal efficiency decreased to 40%. The production of biogas accordingly decreased with the temperature. A decrease in COD removal efficiency was also observed when reducing the HRT, and it was noted that this decrease was more severe at lower temperatures. The removal efficiency of TSS was not significantly affected by the decrease in temperature, as agreed with by Farajzadehha et al. (2012). Singh & Viraraghavan (2003) attributed this to the successful formation of granules during start-up, hence causing successful entrapment of TSS. This meant that the effluent from the reactors met the criteria for discharge in terms of suspended solids, and required only a simple aerobic step to meet the criteria in terms of COD, even at temperatures as low as 11 °C. A comparison between two laboratory-scale UASB reactors treating greywater (refers to wastewater produced in households which does not include toilet wastewater, i.e. water from washing and cooking), one operating at ambient temperature (14–25 °C) and one operating at 30 °C, and a septic tank was carried out by Elmitwalli & Otterpohl (2011). They observed that the maximum COD removal efficiency was achieved by the reactor operating at 30 °C (84%), but that the removal achieved in the reactor operating at ambient temperature still greatly exceeded that achieved in the septic tank (79 and 14% respectively). This leads to the understanding that the UASB reactor is preferable to the septic tank for treating greywater, even at low temperatures. The authors also recommended a lower HRT if operating the UASB at lower temperatures. Recent experiments on a full-scale UASB reactor (unpublished results) at Waterleau showed that although the optimum temperature for operation was at 35 °C in terms of biogas production, no significant reduction in COD removal was observed at lower temperatures of around 20 °C. These studies highlight that the UASB can be used for treatment at a variety of temperatures, though generally higher temperatures yield higher COD removal efficiencies.

The organic loading rate (OLR) applied to UASB reactors can affect the performance of the reactor. Some of the studies on the effect of varying the OLR have been summarized in Table 3. Trnovec & Britz (1998) used a laboratory-scale UASB treating canning wastewater to study the effects of changing the OLR. The OLR was increased from 2.28 to 10.95 kg COD m^–3 d^–1. The increase to 3.95 kg COD m^–3 d^–1 was achieved by the increase of effluent COD. The increase from 3.95 kg COD m^–3 d^–1 to the maximum of 10.95 kg COD m^–3 d^–1 was achieved by the decrease of HRT. Up to the OLR of 4.36 kg COD m^–3 d^–1, the removal efficiency showed a definite increase from 88 to 96%. When the OLR was increased further, the efficiency reduced but still remained high (within 88–96%). The rate of removal increased steadily with OLR. Here, the authors showed that a single UASB can cope with a wide variety of OLRs. Kalyuzhnyi et al. (1996) also studied a laboratory-scale UASB reactor. This was fed with synthetic wastewater at a wide variety of OLRs between 3.4 and 44.9 kg COD m^–3 d^–1. At OLRs below 11 kg COD m^–3 d^–1, the COD removal efficiency remained above 98%. At higher OLRs the removal efficiency was reduced, but still remained above 96%. However, at these higher OLRs, the time taken to reach the maximum efficiency was up to three times as long as at the lower OLRs. In addition, temporary disturbances such as large gas bubble formation and sludge flotation were observed immediately after a change in OLR. These results show that the UASB reactor is capable of coping with a wide range of OLRs, given the time to reach maximum removal efficiency. González et al. (1998) studied a laboratory-scale UASB reactor for the treatment of sugar cane molasses. They increased the OLR in a relatively small range of 2.3–7.1 kg COD m^–3 d^–1. The COD removal increased from 59 to 91% while increasing the OLR from 2.3 to 5.6 kg COD m^–3 d^–1. However at 7.1 kg COD m^–3 d^–1, the removal efficiency dropped to 78%. The authors also noted that the biogas production increased with OLR: a useful observation when considering the utilization of biogas produced. Torkian et al. (2003) used a pilot-scale UASB reactor for treating wastewater from a slaughterhouse to study the effect of OLR. The first stage of the study saw the OLR increase from 10 to 18 kg COD m^–3 d^–1 and a corresponding increase of COD removal efficiency from 62 to 92%. The authors, in agreement with Kalyuzhnyi et al. (1996), noted that after a change in OLR there incurred disturbances within the reactor that led to an initial decrease in removal efficiency. However, the system recovered quickly and hence a general increase in removal efficiency was observed. The second stage of the study saw the OLR increase to 27 kg COD m^–3 d^–1. Again a decrease in removal efficiency was observed immediately after this increase, but once the system had recovered a removal efficiency of 93% was achieved and no adverse effect was caused by the large increase in OLR. Although the OLR was increased to above 30 kg COD m^–3 d^–1, the removal efficiency did not increase above 68% and the methane production decreased at OLRs above 30 kg COD m^–3 d^–1. It was therefore concluded that the upper limit for the OLR was 30 kg COD m^–3 d^–1. However, even at OLR of 40 kg COD m^–3 d^–1, no sludge washout was observed. Zhou et al. (2009) studied a full-scale UASB reactor treating distillery wastewater. The primary aim of their study was to observe the effects of low OLRs between 0.2 and 0.8 kg COD m^–3 d^–1. High removal efficiencies (96–99%) were observed even at such low OLRs, and there was a general increase in removal efficiency with an increase in OLR. Also, according to the standards imposed by local authorities for effluent discharge, no extra treatment was needed to achieve the required effluent quality. Recent experiments on a UASB pilot reactor at Waterleau (unpublished results) saw COD removal efficiencies of around 84% at OLR up to 16 kg COD m^–3 d^–1. These results indicate that the UASB reactor technology is able to cope with a variety of loading rates, and can perform with high COD removal efficiencies at low and high OLRs.

Pol & Lettinga (1986) explained a need to achieve sufficient contact between the biomass in the reactor and influent wastewater in order to yield satisfactory COD removal, and that mixing caused by the rising of biogas bubbles produced during the process could provide this contact. They suggested that in low-rate UASB reactors, an effluent recycle or mechanical mixer could be added to the reactor configuration if insufficient mixing is achieved by biogas production. Table 4 summarizes the effects of mixing observed during some studies found in the literature. During their experiments, Lettinga et al. (1980) included a mechanical mixer in the form of an axle with attached blades in the reactor. The mechanical agitation was kept to a minimum, between 10 and 30 rpm, for 1 minute/10 minutes: any more would be detrimental to the sludge granules. However, after studying the results of COD removal efficiencies, the authors concluded that there seemed to be no need for any extra mixing in addition to that already achieved by biogas production. Nadais et al. (2006) suggested a different, but still minimal, mixing regime of 40 rpm for 10 minutes/hour, and this helped to prevent channeling in the reactor. Lin & Yang (1991) agreed that mixing achieves better contact between the sludge and wastewater and minimizes channeling in the reactor, and helps to enhance the treatment efficiency. They suggested that, in general, sufficient mixing can be achieved by the rising gases produced in the reactor, but this may not be enough at the beginning of the process, when gas production is minimal. As such, mechanical mixing may be required, the amount of which should not exceed 30 rpm for 1 minute/10 minutes, in agreement with Lettinga et al. (1980). They also suggested that an increase in mixing, should it be required, could be achieved by a high hydraulic load, the addition of effluent recycle or the addition of gas recycle. Hickey et al. (1991) also suggested the use of an effluent recycle to enhance the mixing if needed, but also stated that the biogas played a greater role in the mixing than the upflow velocity. Quaff & Guha (2011) studied the performance of laboratory-scale UASB reactors treating domestic wastewater at different upflow velocities. They found that, generally, the higher flow (and hence increased mixing) gave an increase in removal efficiency: 86% COD removal was achieved at a flowrate of 700 mL h^–1 compared to 78% at 100 mL h^–1. Statistical tests confirmed that the removal efficiencies were sufficiently different to suggest that the mixing had an effect on the performance. The authors also noted that a higher flowrate encouraged methane production. They concluded that at low flowrates sulfidogenic bacteria is not washed out and hence is able to colonise in the reactor and compete with the methanogenic bacteria. Peña et al. (2006) studied a full-scale UASB reactor treating municipal wastewater in which different upflow velocities were applied to increase the mixing in the reactor, and an increase in COD removal was observed with an increase in mixing. Statistical tests confirmed that the hydraulic loading rate, and hence mixing, affected the performance of the process.

Souza (1986) found the optimum pH for anaerobic digestion to be around 7. Any pH below 6.5 or above 7.5 may be harmful to the bacteria. However, they also mention that wastewaters with low and high pH could be treated in UASB reactors, due to the buffer capacity of the process although a pH control system may be needed. There have been some studies on the effect of varying pH in UASB reactors, and some of these are summarized in Table 5. Liu et al. (2010) studied three UASB reactors, each operating at an HRT of 6.9 hours and an OLR of 5 kg COD m^–3 d^–1. One of the reactors was kept at a pH of 7, one at 6 and one at 5, and COD removal efficiencies of 89, 49 and 10% respectively were achieved. As well as the drop in removal efficiency, a significant drop in methane production was observed, as well as high accumulation of volatile fatty acids (VFAs). Their results are in agreement with Souza (1986) that the optimum pH is around 7. A laboratory-scale UASB reactor treating winery wastewater was used by Ronquest & Britz (1999) to investigate the effect of low pH on the performance. The effect of pH is particularly important for the treatment of winery wastewaters, as they often have to undergo neutralization which is an extra cost. The HRT in the reactor was maintained at 14 hours and the pH of the influent was reduced from 6.7 to 5.0 in gradual steps. The effluent pH was measured, and was consistently around 7 for all influent pH values, which suggests a good buffer capacity in agreement with Souza (1986). However, the COD removal started decreasing significantly at the influent pH of 5 and a slow recovery and drop in biogas production suggested that it was the lowest operational pH value. This meant that the authors could conclude that it is feasible to treat winery wastewater in a UASB reactor with little or no neutralization (depending on the source) and hence reduce processing costs. Visser et al. (1993) fed three thermophilic laboratory-scale UASB reactors at 55 °C with synthetic wastewater. They were all operated at HRT of 6 hours and OLR of 6.6 kg COD m^–3 d^–1. The pH in each reactor was varied, so that a maximum pH of 8.3 and a minimum pH of 6.75 were used. At a pH of 8.3, the COD removal was 77%, compared to at pH of 7.2 where the maximum COD removal of 98% was achieved. Sandberg & Ahring (1992) also studied a laboratory-scale UASB reactor. This was used to treat fish-treatment wastewater, and was operated at a loading rate of 16 kg COD m^–3 d^–1. The pH was varied between 7.3 and 8.3. Between pH of 7.3 and 7.9 the COD removal remained around 97%. At pH of 8.1 the removal reduced to 79% and at pH 8.3 disintegration of the sludge granules was reported resulting in washout of the biomass. Methane production also steadily decreased with an increase in pH. These studies indicate that although UASB reactors can cope with low pH relatively well, a high pH (>8) seems to cause serious problems resulting in reduced performance.

UASB systems can be useful in the treatment of high strength wastewaters such as brewery wastewater, carbohydrate-rich food wastes, grape wine distillery wastes, slaughterhouse wastewater and domestic wastes. However, adequate design and careful control and monitoring of the system during operation are required to ensure a successful start-up and subsequent stable operation. Optimum operation of UASB systems is noted to occur with some mixing (sufficient mixing can be achieved by the rising gases produced in the reactor) at temperatures of 35 °C (COD removal decreases with decreased temperatures) and pH of 7 (although UASB reactors can cope well with lower influent pH due to its buffering capacity, but experiences inhibitions at pH values of above 8). The UASB systems generally obtain good COD removal efficiencies, even with varying organic loading rates. However, as with other biological systems, the capacity to efficiently carry out COD removal is greatly dependent on the retention times (SRT, to ensure that sufficient biomass is retained for duration of treatment process; and HRT, to ensure that the liquid waste is in the system long enough to be treated) as dictated by the general sizing and configuration of the reactor. Moreover, it is important to note that in situations where the effluent is required to be discharged into receiving water bodies, organic loading rates require careful selection and, in most cases (depending on the regional environmental standards) a post-treatment step is required to ensure that the final effluent is compliant.

An axial dispersion model was proposed by Singhal et al. (1998) to describe the fluid flow through the UASB reactor. The aim of this model is to simplify the previous models describing the fluid flow, as they used multiple compartments inside the reactor. The axial dispersion model uses only two compartments that are each assumed to represent axially dispersed zones. The principles of the model rely on the Peclet number (Pe) of each compartment (Pe → ∞ causes minimum axial dispersion and hence it can be assumed to be plug flow; Pe → 0 causes maximum axial dispersion and hence exhibits ideal well-mixed behavior). It can also be assumed that some liquid will bypass the first zone, and this is included in the model. Peña et al. (2006) implemented a combined dispersion-compartmental model to describe the mixing properties inside their full-scale UASB reactor. Residence time distribution (RTD) curves were derived from tracer tests on the UASB for HRTs of 5, 6, 8, and 10 hours. At HRTs of 8 and 10 hours, the curves show that the UASB displayed characteristics of a poorly mixed reactor with dead volumes. However, at the lower HRTs of 5 and 6 hours, the curves showed better mixing in the UASB, and the curves were much closer to that of the theoretical CSTR. This shows that under-loading (longer HRTs) causes less mixing in the UASB. Kalyuzhnyi et al. (2005) derived a model that was intended to combine granular sludge dynamics, solid–liquid–gas interactions, biological conversions and liquid-phase equilibrium chemistry. Their plug flow dispersion model is based on an assumption that the processes involved inside the reactor depend on the distance from the bottom of the reactor, and hence in any given cross-section of the reactor, the process characteristics are constant. The validity of the model was tested with results obtained from the pilot-scale UASB reactor treating cheese whey. It was seen that the model outputs correlated well with experimental data obtained, with the exception of effluent COD concentration at a particular time during the operation of the UASB. The model underestimated the effluent COD concentration, and therefore overestimated the methane production at the corresponding time. The authors conclude this on the simplification of the description of granule transport by the model.

Mathematical models for a unit operation such as anaerobic digestion aim to virtually replicate the process such that they can provide expert guidance for aiding in system design and process optimization. Most developed anaerobic digestion models incorporate a number of kinetic and stoichiometric expressions, which represent the biological interactions that are hypothetically proposed to occur during the anaerobic digestion process. The hypotheses on which these expressions are based are tested using experimental data that is tailored towards determination of the system response (i.e. simulated predictions get compared to the experimental data). Modeling approaches can either be empirical or mechanistic. Empirical models are based upon observed correlations between the performance of the plant and its main design and operating variables (i.e. established by observation when the mechanisms and/or processes operating in the system are not known or are ignored). In contrast, mechanistic models are based on some conceptualization of the biological/physical mechanisms operating in the system (Sam-Soon et al. 1991). In developing the mechanistic model, these determined conceptual mechanisms are used to mathematically formulate the process rates and their stoichiometric interactions with the compounds.

Dynamic modeling of anaerobic digestion has been an active area of research in the past few decades. Based on the Haldane model, a model with a single bacterial population (aceticlastic methanogens) was proposed by Graef & Andrews (1974) in which the conversion of fatty acids into biogas is considered limiting and an increase in fatty acid concentration caused a drop in pH and a rise in undissociated acetic acid concentration. However, the model was also able to predict the digester response to the entry of an external inhibitor. Hill (1982) introduced a model that was developed to describe digestion of manure and animal wastes. The model assumed that methanogenesis depends on the total fatty acids and inhibition by the total fatty acid concentration. According to this model, anaerobic digestion is inhibited with an accumulation of VFAs, causing a decrease in the rate of VFA consumption and leading to acid accumulation. Mosey (1983) introduced a four-population model with one acidogenic reaction, one acetogenic reaction, and two methanation reactions. According to their model, a sudden increase in the OLR is expected to cause an accumulation of VFAs, given that acetogens grow at a slower rate than the acidogens. Angelidaki et al. (1993) considered hydrolysis, acidogenesis, acetogenesis, and methanogenesis to describe the behavior of manure-fed digesters. In this model, free ammonia is assumed to inhibit methanogenesis; acetic acid is assumed to inhibit acetogenesis, and total VFA is assumed to inhibit acidogenesis. It must be noted that all biokinetic models described so far consider organic matter as a whole and do not account for the nature of the organic macromolecules in the feed composition.

The Anaerobic Digester Model 1 (ADM1) was developed by the IWA Anaerobic Digestion Task Group (Batstone et al. 2002a, 2002b, 2006) and is a generalized model for anaerobic wastewater treatment. The ADM1 includes multiple steps and these describe both the biochemical and physico-chemical processes. The first biochemical process is an extracellular disintegration step, where the complex organic waste disintegrates to carbohydrates, proteins, lipids and inert material. This can include many different processes. The second step is extracellular hydrolysis, where these carbohydrates form sugars, proteins form amino acids and lipids form long-chain fatty acids. This hydrolysis is the rate determining step and both these extracellular steps are assumed to be first order. The sugars and amino acids undergo acidogenesis to form organic acids, hydrogen and carbon dioxide. The organic acids, including the long-chain fatty acids, are converted to acetate, hydrogen and carbon dioxide by acetogenesis. Hydrogen and acetate then undergo methanogenesis. The methane produced, along with the CO₂, forms the biogas which can be used to provide process energy. Figure 8 summarizes these processes. The physico-chemical processes included in the anaerobic digester model are the ion association and dissociation and gas transfer. The ion association and dissociation are very rapid and can be considered as equilibrium processes and hence can be described as algebraic equations. All organic species are described in the model in terms of COD (Batstone et al. 2002a, 2002b; Coelho et al. 2006), and the nitrogenous (NH₃ and NH₄⁺) and inorganic carbon (carbon dioxide and bicarbonate) compounds are described in terms of their molar concentration. In Figure 8, S denotes soluble species and X denotes particulate species. There are many advantages to using a generalized model. It can be applied to full-scale plant design and optimization, provide a common base for any further development of the model to specific applications, be used to compare outcomes to other investigations and includes the physico-chemical system. This last point is important as the physico-chemical transformations can be used to optimize the major performance variables (such as gas flow). It also allows for inhibition factors to be expressed such as pH and dissolved gas concentrations and for a pH control set point to be calculated. Based on these processes, the Task Group (Batstone et al. 2002a, 2002b) produced a set of equations used in the modeling of anaerobic processes. A mass balance for each component in the liquid phase is produced, including terms for the kinetic rates of each process and the rate coefficients for the specified component on each process. Mass balances for the gas phase components and rate terms for the transfer of gas components in the reactor to the headspace are also included. The physico-chemical equations are also included in the model, and can be implemented as algebraic or kinetic rate equations.

The ADM1 model has been used to model some of the various anaerobic wastewater treatment processes, including CSTR (Kerroum et al. 2010; Thamsiriroj & Murphy 2011) and UASB (Batstone & Keller 2003; Batstone et al. 2005; Coelho et al. 2006; Tejasen & Taruyanon 2010) reactors. Batstone et al. (2005) modeled a laboratory-scale UASB reactor using ADM1. The model was implemented in AQUASIM – a computer program designed for the simulation of aquatic systems (Reichert 1994). The response of the system to a pulse in influent acetate was studied. The model could not simulate the response of the system for a pulse of 0.8 g COD acetate at HRT of 2 days as the response was too small. However, when the pulse was increased to 1.6 g COD acetate and an HRT of 1 day, the response was larger and could be modeled. Using AQUASIM, Coelho et al. (2006) studied the accuracy of ADM1 model for a laboratory-scale UASB reactor treating dairy waste. The methane production was modeled instead of biogas for ease of implementation. The values produced by the model for methane production and effluent COD concentration correlated well with the experimental values, except for on day 80, where unexpected washout from the reactor caused experimental effluent COD concentrations to be larger than those predicted by the simulation. Batstone & Keller (2003) studied the industrial applications of ADM1, including the application to a full-scale UASB reactor treating paper mill wastewater in AQUASIM. The aim was to help towards making a decision whether dosing the reactor with HCl was economically effective: it would decrease the pH and hence decrease the precipitation of calcium carbonate to which this particular system was liable. Real measurements and model predictions of the gas flow from the reactor were plotted to validate the model which showed a good correlation (±10%). Tejasen & Taruyanon (2010) studied a laboratory-scale set up of a CSTR followed by a UASB reactor treating wastewater from an ethanol distillery using ADM1 (implemented in AQUASIM) and included an extension to allow for the reduction of sulfate. The CSTR and UASB were modeled separately, and it was found that the simulated and the experimental results for soluble COD and volatile fatty acids in the effluent correlated well. The sulfate removal was also accurately simulated by the model. However, pH measurements and gas flow from the reactor were simulated at lower values than those obtained experimentally. Despite this, the authors agreed that ADM1 was valid, and also that it is sufficient for application in the design and operation of full-scale systems. It can be concluded that ADM1 has been quite successful in the simulation of UASB processes at laboratory scale, though research into the application of the model to full-scale processes is limited.

In some of the cases discussed previously (Diamantis et al. 2003; Gohil & Nakhla 2006; Nacheva et al. 2011), the required treatment had not been achieved by the UASB reactor alone and aerobic post-treatment steps had to be incorporated into the process to ensure satisfactory effluent quality. If the UASB process is to be modeled, it would, in these cases, also be useful to be able to model the aerobic post treatment step. The Activated Sludge Models (ASM1, ASM2, ASM2d and ASM3) developed by Henze et al. (2000) are widely used for the modeling of activated sludge treatment of wastewater. Although not a complete general model, the ASM models have grown to incorporate many of the developments in understanding the activated sludge process. The ASM models have been integrated with ADM1 to produce a single model that describes plants involving anaerobic digestion and activated sludge processes, rather than simulating each part individually. Recent approaches to overcome incompatibility problems when integrating different unit processes (in this case activated sludge linked to AD) to develop plant-wide WWTP simulation models include (i) the continuity based interfacing method (CBIM) of Vanrolleghem et al. (2005) and Tejasen & Taruyanon (2010), (ii) the ‘supermodel approach’ of Jones & Takács (2004, cited in Gujer et al. 1999) and Seco et al. (2004, cited in Gujer et al. 1999), (iii) the transformation based approach of Gujer et al. (1999) and (iv) the mass balances based plant-wide WWTP model approach of Ekama et al. (2006). All of these approaches have been aimed at overcoming the model interfacing difficulties caused by state variable incompatibilities. The ADM1 Scientific and Technical Report (STR) (Batstone et al. 2002b) outlines the two differences between the implementations of the two models as being the units and kinetics. It suggests that leaving the ASM unmodified and changing the ADM1 to use ASM kinetics and units could be a way around these differences and these modifications are included in the ADM1 STR. It also suggests the use of a conversion interface as an alternative way of overcoming these differences. An interface with this purpose has been developed by Yasui et al. (2006) for modeling anaerobic sludge digestion of sludge used in an activated sludge process. The results showed that they had successfully integrated ASM and ADM1 to model the water treatment (activated sludge) and sludge treatment (anaerobic) process. Nopens et al. (2009) also developed an interface for the combination of ASM1 and ADM1 for the simulation of an activated sludge plant followed by an anaerobic digester used for the treatment of sludge. They produced an interface that did not involve a general term for the input, as this had been used before and had not been successful in accurate modeling. Validation of the model produced, including the developed interface, was carried out and the model predicted realistic behavior of the plant. Kauder et al. (2007) combined ADM1 and ASM2d to simulate the aerobic and anaerobic phases in a sequencing batch reactor. Although the simulation results are yet to be compared with the experimental data to validate the accuracy, they have concluded that the model could save considerable time and money when it comes to laboratory work for optimization of the process. As water resource recovery modeling complexity increases, Vanrolleghem et al. (2014) have debated on acceptance of mass continuity interface approach as the best approach to connect unit processes in a plant-wide set up or whether adoption of a super model approach (where all components and transformations from the sub-model of each unit process are combined to form the plant-wide model) such as that of Grau et al. (2009) would improve the modeling process. The movement towards the supermodel approach would require description of all components according to their elemental composition, which would allow for the virtual tracking of all material elements in the models, using the principle of mass balances for the stoichiometric processes. Because many species are electroactive, Lizarralde et al. (2015) favored this supermodel approach that links slower reacting components which are simulated using differential equations to the faster physicochemical reactions which are calculated algebraically at each iteration. Although there appears to be no literature on integrating ADM1 simulating a UASB reactor with ASM, the above studies suggest that the interfacing or supermodel approaches are achievable.

Several anaerobic digestion models have accounted for the interaction between biological reactions of the different organism groups and the weak acid-base chemistry of the aqueous phase in which they function using separated bioprocess stoichiometry and thermodynamic equilibria based on mass and charge balance approaches towards pH prediction (e.g. the IWA ADM1 developed by Batstone et al. (2002b) and UCTSDM developed by Sötemann et al. (2005)). However, although it is possible to predict or enter the factors that cause organism inhibition (e.g. pH variations), prediction of model instability (hence early warning for failure prevention) is still a challenge in biokinetic modeling of UASB reactors and other anaerobic systems. This is mainly because most earlier anaerobic digestion models are based on one rate-limiting bioprocess (e.g. the growth of methanogens (Hill & Barth 1977) or hydrolysis of complex organics (Sötemann et al. 2005; Ikumi et al. 2015)). However, to accurately predict anaerobic model dynamics requires models to account for potential multiple rate limiting bioprocesses, which involves carrying out complete calibration for all the anaerobic digestion organism groups. This aspect is being addressed in a separate forthcoming publication by the Water Research Group, University of Cape Town, South Africa. However, recent advances in research and development of mathematical models has given motivation to and greater confidence in the application of anaerobic digestion. Improved models hold the promise of being capable of optimized operation of UASB and other anaerobic systems, including the prediction of model instability before it occurs and so prevent failure. Studies on anaerobic digestion systems have resulted in the development of explicit equations that could be integrated for the construction of mathematical models, by using fundamental kinetic relationships and application of material mass balances (McCarty & Mosey 1991; Ikumi et al. 2015).

A semi pilot-scale upflow anaerobic sludge bed (UASB) reactor treating primary municipal sludge was studied at Waterleau R&D centre at Wespelaar (Belgium). The aim of the test case study was to implement ADM1 using AQUASIM (Reichert 1994). By adding variables and processes and defining the compartments in which these processes occur, AQUASIM is able to simulate values for parameters such as COD, VFA and pH over a given period of time. AQUASIM has previously been used in literature for the implementation of ADM1. Ideal CSTR compartments were used in this implementation and the kinetic and stoichiometric parameters used were the ones recommended in the main body of the ADM1 STR (Batstone et al. 2002b). Those that were suggested to be most likely to be sensitive to the process in the STR were then subject to a sensitivity analysis, to identify the most important parameter values to be adjusted to give the best agreement between simulated and experimental results. This was done in AQUASIM using the absolute-relative sensitivity function. Those with the highest sensitivity were then subjected to parameter estimation which uses the weighted least-squares method. AQUASIM uses the secant method to convert the nonlinear problem to a linear one to allow for parameter estimation (Ding et al. 2010). The layout used in AQUASIM to model the UASB is shown in Figure 9(a). It is implemented by assuming a perfectly mixed reactor followed by a settler, to represent the settling zone. The recycle has also been included. The addition of the recycle stream is needed for a more realistic prediction of the results, as is the addition of the settling zone compartment. Based on a sensitivity analysis, the parameters chosen for estimation were ⁠, ⁠, ⁠, ⁠, and and these are summarized in Table 6. Values of influent flow, soluble COD, particulate COD, reactor volume and temperature were used to form the inputs to the model, as well as the concentrations of VFAs and ammonium in the influent. Figures 9(b)–9(e) show the experimental and simulated results of the laboratory-scale UASB reactor, plotted together for model validation. Figures 9(b)–9(e) indicate that the model can generally predict the trend of each of COD (soluble and total), VFA and biogas production respectively. However, there is generally a small over-prediction in the COD and biogas simulations and an under-prediction in the VFA simulation. This could be due to an inaccurate experimental measurement or the composition of influent. The influent fractions of carbohydrates, proteins and lipids were assumed to be similar to that found in literature for brewery wastewater (Ahn et al. 2001) and therefore more accurate results may be obtained if the influent had been examined for these fractions. The anomalous experimental result at day 61 for VFA and total COD was due to the reactor tending towards failure over a weekend where there was no measurements taken, due to a sharp reduction in temperature to 13 °C. This caused the methanogens to become less active, meaning an increase in VFA production. At day 21, there was a problem with the influent pump and there was no new influent being pumped through the reactor, except that from the recycle. It is suggested that this caused the anomalous results at days 21 and 22. It must be noted that due to problems with the measurement of biogas in the experiment the result of the experiment biogas production should be treated with caution.

The principles and effectiveness of various anaerobic treatment processes have been outlined, including the upflow anaerobic sludge blanket reactor. The importance of the UASB reactor has been discussed as well as its presence in industry. Various studies on the factors affecting the performance of the UASB process have led to some conclusions to be made surrounding these factors. The UASB has been successful in treating brewery, canning, distillery, pharmaceutical and slaughterhouse wastewaters, as well as sewage. It has been used successfully at temperatures in the range of 11–55 °C, but reactors operating with temperatures at the higher end of the range generally performed better. Organic loading rates of between 0.2 and 44.9 kg COD m^–3 d^–1 have been applied to a variety of UASB reactors, providing generally high removal efficiencies. For OLRs at the higher end of the range, the highest removal efficiencies were only achieved after a long period of time. Mechanical mixing is generally avoided in UASB reactors, as rising biogas and upflow velocities provide sufficient contact between wastewater and biomass. However, at times when this has not been enough, minimal mechanical mixing has been applied. UASB reactors have been shown to operate best at around 7 pH, but can cope with a slightly lower value such as 6.5 pH as well. pH values higher than 8 have been shown to cause problems in UASB reactors. Models specific to UASB reactors have been developed to describe flow patterns, mixing patterns and effluent characteristics successfully. ADM1 is a general model to describe anaerobic digestion and has been applied to pilot-scale UASB reactors with fairly good correlation to experimental values. The implementation of ADM1 in AQUASIM successfully provides simulated results for effluent soluble and total COD and effluent VFA, except for at times where there was a sharp reduction in temperature as the model assumed constant temperature at 32 °C.

The authors are grateful to Ron Gerards, Vice President (Technology) of Waterleau NV Wespelaar (Belgium) for providing the pilot-scale data of the UASB. CD is grateful to Gilbert Shama from the Department of Chemical Engineering, Loughborough University, UK for arranging a research stay at KU Leuven Campus Groep T through the ERASMUS exchange program.