7/30/2019 mbr1

1/9

Development of an integrated membrane process for

water reclamation

C.H. Lew*, J.Y. Hu*, L.F. Song*, L.Y. Lee*, S.L. Ong*, W.J. Ng* and H. Seah**

*Environmental Science and Engineering Programme, National University of Singapore, 10 Kent RidgeCrescent, Singapore 119260 (E-mail: cvehujy@nus.edu.sg)

**Public Utilities Board, 40 Scotts Road, Singapore 228231

Abstract An integrated membrane process (IMP) comprising a membrane bioreactor (MBR) and a reverse

osmosis (RO) process was developed for water reclamation. Wastewater was treated by an MBR operated

at a sludge retention time (SRT) of 20 days and a hydraulic retention time (HRT) of 5.5 h. The IMP had an

overall recovery efficiency of 80%. A unique feature of the IMP was the recycling of a fraction of RO

concentrate back to the MBR. Experimental results revealed that a portion of the slow- and hard-to-degrade

organic constituents in the recycle stream could be degraded by an acclimated biomass leading to an

improved MBR treatment efficiency. Although recycling concentrated constituents could impose an inhibitory

effect on the biomass and suppress their respiratory activities, results obtained suggested that operating

MBR (in the novel IMP) at an F/M ratio below 0.03g TOC/g VSS.day could yield an effluent quality

comparable to that achievable without concentrate recycling. It is noted in this study that the novel IMP

could achieve an average overall TOC removal efficiency of 88.94% and it consistently produced productwater usable for high value reuse applications.

Keywords Biomass acclimation; integrated membrane process; membrane bioreactor; reverse osmosis;

slow- and hard-to-degrade compounds; water reclamation

Introduction

Economic development and global industrialization over the past decades have inevitably

heightened the conflict between water demand and water supply. Consequently,

water reclamation has increasingly been receiving attention and seriously explored and/or

exploited in many parts of the world for augmenting freshwater supplies. In most cases,

membrane technology has been employed and that more than one membrane processes

are typically used in order to obtain the benefits associated with multiple barriers

provided by such arrangement and to assure product water with high quality and purity.

One such advanced reclamation technology is the coupling of an MBR with a downstream

RO process whereby the MBR is used to serve as a pretreatment to the RO process. In this

configuration, soluble microbial products (SMP) could potentially be an issue that may

impede the performance of the integrated MBR-RO system (i.e. IMP).

SMP are defined as the pool of organic compounds that are released into solution

from substrate metabolism accompanied by biomass growth, and biomass decay ( Duncan

and David, 1999). SMP has been said to be inert in nature. It has been shown that the

majority of soluble organic matters in effluents from biological treatment processes are

actually SMP, and hence its presence could significantly affect the effluent quality. Simi-

larly, in the treatment of industrial wastewater, the presence of inherent refractory organ-

ics can also significantly affect the quality of the final treated effluent. In view of this, it

could be expected that the major constituents in the effluent of a biotreatment system

treating mixed municipal and industrial wastewater would likely contain SMP and waste-

water inherent refractory organic compounds.

WaterScience

&Technolo

gyVol51

No

67

pp

455463

Q

IWAPublishing2005

455

7/30/2019 mbr1

2/9

One of the potential advantages offered by the MBR is that it could allow the biode-

gradation of slow- and hard-to-degrade organic compounds. This is because MBR could

allow the establishment of specialized microorganisms capable of removing slowly

degradable components (Rosenberger et al., 2002). In addition, MBR could facilitate

selective growth of specific microorganisms for the degradation of hardly degradable

hazardous substances (Yamamoto, 2001). For example, it has been reported that operating

MBR at a SRT greater than 20 days could allow the development of slow-growing micro-

organisms which in turn results in better removal of refractory organic matters ( Cote et al.,

1997). SMP is noted to be biodegradable over time, although the kinetics of degradation

may be much slower than simple substrates. It has been reported that SMP biodegrada-

tion occurred in MBRs after an acclimation period (Huang et al., 2000; Shin and Kang,2003). This observation suggested that one could recycle a fraction of the RO concentrate

back to the MBR operated at a long SRT and exploit the MBR for treating slow- and

hard-to-degrade organic compounds present in the RO concentrate stream produced from

the IMP. That is, carrying out concentrate recycling may facilitate: (i) degradation of any

residual easily degradable organic matters present in the recycle stream, and (ii) degra-

dation of a portion of the slow- and hard-to-degrade organic compounds present in both

the recycle stream and MBR Mixed Liquor (ML). If these desired aims could be

achieved, it will enhance MBR treatment efficiency and improve overall performance of

the novel IMP. In addition, degradation of a portion of these slow- and hard-to-degrade

organic compounds via concentrate recycling (which would otherwise be wasted via the

RO bleed line in an IMP without recycling) will result in a less concentrated RO concen-

trate waste stream. This achievement can also help to reduce the adverse environmentaleffects due to concentrated waste disposal.

A review of literature revealed that an increase in the concentrations of non-reactive

compounds in bioreactors at long SRT can lead to microbial inhibition or toxicity (Brin-

dle and Stephenson, 1996). High concentration of SMP in an MBR was reported to be

inhibitory to the metabolic activity of the activated sludge ( Huang et al., 2000). Similarly,

high levels of total dissolved solids (TDS) have been known to produce an osmotic press-

ure on cells, causing plasmolysis and loss of cell activity. In view of the above, a primary

concern in the operation of the novel IMP will be the inhibition and potential intoxication

of microorganisms, due to potential elevated concentrations of TDS, SMP and wastewater

inherent refractory organic compounds in the MBR ML. Another concern is the deterio-

ration of MBR effluent quality due to the concentrate recycling. These concerns need to

be adequately addressed in order for the proposed novel IMP to be technically feasible.

In view of this, the objective of this research is to study the operating characteristics of

using an IMP to reclaim high quality product water from mixed industrial and domestic

wastewater.

Methods

The schematic diagram of the novel IMP is shown in Figure 1. The MBR was operated

with two submerged Kubota flat sheet microfiltration membranes of 0.45 mm pore size.

The MBR was operated at a constant HRT of 5.5 h and a SRT of 20 days. The total

ML volume in the bioreactor was 16.5 L and 0.825 L of ML was wasted daily in order to

maintain a desired SRT. pH in the bioreactor was maintained within a range of 7.0 to

7.5. The ML dissolved oxygen (DO) level was always above 5 mg/L. Membrane filtered

effluent was intermittently discharged at a constant flux by a suction pump. Each inter-

mittent suction cycle consisted of an 8 mins suction period followed by a 2 mins non-suc-

tion period. The need for a higher suction pressure to maintain a constant permeate flux

gave an indirect indication of membrane fouling. In this study, membrane cleaning was

C.H.L

ewetal.

456

7/30/2019 mbr1

3/9

initiated once suction pressure exceeded 5.6 psi. The fouling layer formed on the mem-

brane surfaces was removed by washing with a sponge and tap water. Chemical cleaning

was carried out once every 34 months by immersing the membranes in sodium hypo-

chlorite solution (0.5 wt.%) for 3 h followed by oxalic acid solution (0.5 wt.%) for

another 3 h. A single thin film GE Osmonics RO membrane element was used for the

RO process. As the RO element can only achieve a limited permeate recovery ratio of

1015%, the RO process was operated in a concentrate recirculation mode. This allowed

the IMP to achieve an overall recovery efficiency of 80%. As bulk of the RO concentrate

stream was recycled back to the effluent tank, the TDS and total organic carbon (TOC)concentrations of the RO feed were higher than the MBR effluent. A concentrate flow

rate equal to 20% of the wastewater influent rate was internally recycled back to the

MBR, while another concentrate stream of the same flow rate was wasted through a

bleed line. Membrane fouling resulted in a higher driving pressure to maintain a constant

permeate flux. When a 2030% increase in the required driving pressure was observed,

chemical cleaning was initiated. A chemical solution of EDTA (0.84 wt.%) and sodium

tripolyphosphate (2.03 wt.%) were used for organic cleaning while a citric acid solution

(2.0 wt.%), with pH adjusted to 4, was used for colloidal and inorganic cleaning.

The experimentation period was divided into four phases. Phase 1 represented the

seeding and start-up of the MBR for 114 days. The seed sludge and wastewater source

for the four phases were collected from a local wastewater treatment plant (WWTP).

Phase 2 represented the operation of the MBR alone at a SRT of 20 days for 37 days.

Phase 3 represented a transition period of 46 days, whereby the MBR was integrated with

the RO process with a concentrate recycle stream. Phase 4 represented the stabilization

and operation of the novel IMP at a SRT of 20 days for 147 days. Samples taken from

IMP for analyses include the wastewater, ML supernatant, MBR effluent, RO concentrate

and RO permeate. TOC tests (TOC Analyzer, Shimadzu) were conducted for all the

samples. TDS levels (TDS Meter, Hanna Instruments) were measured for the wastewater,

ML, RO concentrate and RO permeate. Ammonia nitrogen, NH3-N, in the wastewater

and MBR effluent were both measured using the automated phenate method. Assessment

of microbial viability was carried out based on the specific oxygen uptake rate (sOUR).

TOC, MLSS, MLVSS and OUR were measured in accordance with Standard Methods

(APHA, 2000). Primary settled effluent was collected from the WWTP and the waste-

water has a 60% industrial and 40% municipal wastewater composition. The wastewater

characteristics are shown in Table 1. As indicated by the standard deviation (SD), the

wastewater strength in terms of TOC was highly variable, as compared with the NH 3-N

and TDS concentrations, which were relatively stable.

Figure 1 Schematic flow diagram of the IMP system

C.H.L

ewetal.

457

7/30/2019 mbr1

4/9

Results and discussion

To investigate the effects of concentrate recycling on MBR performance, experimentalresults obtained during Phases 2 and 4 were analyzed and compared. A time period of

2 SRT was given for the stabilization of the MBR during Phase 4, allowing the biomass

to be acclimated to the conditions of concentrate recycling. Only experimental data col-

lected after this time period were analyzed. Throughout both phases, nitrification was

complete and no NH3-N was detected in the MBR effluent. Table 2 summarizes the mean

values and SD of the MBR TOC mass removal rates, ML and effluent characteristics

obtained in Phases 2 and 4. Table 3 summarizes the mean values and SD of both the RO

concentrate and RO permeate characteristics obtained in Phase 4. The performance of the

MBR during Phases 2 and 4 was assessed and compared based on TOC mass removal

rate, given by Eqs. (1) and (2).

M2 Cww Qww 2 Ceff Qeff 1

M4 Cww Qww CR QR 2 Ceff Qeff 2

where M2 is the MBR TOC mass removal rate during Phase 2 (g/d), M4 the MBR TOC

mass removal rate during Phase 4 (g/d), Cww the wastewater TOC concentration (g/L),

Ceff the MBR effluent TOC concentration (g/L), CR the TOC concentration (g/L) of the

RO concentrate recycle stream, Qww the wastewater influent flow rate (L/d), Qeff the

MBR effluent flow rate (L/d), and QR the RO concentrate recycle stream flow rate (L/d).

Effect of concentrate recycling on MBR mass removal rate

As shown in Table 2, the average MBR TOC mass removal rate obtained in Phase 4 was

higher than that of Phase 2. Similar observation could also be seen from Figure 2 where

the profiles of TOC mass removal rates during Phases 2 and 4 were plotted over a range

of wastewater TOC concentrations. It is noted that the trend line associated with Phase 4was consistently above that of Phase 2 over the entire range of wastewater concentrations

tested. This finding indicated that concentrate recycling resulted in a higher TOC mass

Table 1 Characteristics of mixed municipal and industrial wastewater

Wastewater characteristics

Parameters Mean value 6 Standard deviation

TOC, mg/L 110.61 ^ 53.19NH3-N, mg/L 37.20 ^ 5.49TDS, ppm 945 ^ 89

Table 2 Comparison of MBR TOC mass removal rate, ML and effluent characteristics during Phases 2

and 4

Parameters Mean value 6 SD

Phase 2 Phase 4

MLSS, g/L 10.12 ^ 0.92 12.48 ^ 2.19MLVSS, g/L 8.20 ^ 0.61 10.00 ^ 1.81sOUR, mgO2/gVSS.hr 7.86 ^ 0.97 6.48 ^ 2.45

ML DOC, mg/L 29.75^

2.22 30.42^

6.90ML TDS, ppm 922 ^ 73 1,282 ^ 191MBR effluent TOC, mg/L 14.53 ^ 1.25 20.47 ^ 3.68MBR TOC mass removal rate, g/d 5.72 ^ 2.54 6.44 ^ 2.09

C.H.L

ewetal.

458

7/30/2019 mbr1

5/9

removal rate which in turn suggested that in addition to the residual readily degradable

organic compounds, a fraction of the slow- and-hard-to-degrade organic compounds

present in the recycle stream could be biodegraded in the MBR. The above claim can

further be supported by the results presented in Table 2 whereby it could be seen that

concentrate recycling did not result in any significant changes to the ML dissolved

organic carbon (DOC) concentration. The higher ML DOC SD was attributed to the

highly variable wastewater strength experienced in Phase 4. There was also an observed

increase in ML TDS concentration, although it did not reach an inhibitory level. This

phenomenon could be attributed to the high TDS concentration present in the recycle

stream (Table 3).

Effect of concentrate recycling on MBR biomass concentration

As shown in Table 2, the average MLSS and MLVSS concentrations were higher in

Phase 4 than in Phase 2. The higher SD could be attributed to the highly variable waste-

water strength experienced during Phase 4. These observations could be better illustrated

in Figure 3, which shows the MLVSS concentrations over a range of wastewater TOC

concentrations observed in Phases 2 and 4. For both phases, the trend lines obtained indi-

cated a corresponding increase in MLVSS concentration with the wastewater TOC con-

centration. The increase in biomass concentration with substrate concentration indicated

that a higher substrate concentration could support a higher level of growth which in turn

suggested that there was insignificant inhibitory effect associated with concentrate recy-

cling. It should be highlighted that the trend line of Phase 4 was consistently above that

of Phase 2. This finding could be attributed to the higher growth rate associated with

Phase 4 that in turn facilitated a higher mass removal rate during phase 4. These findings

also suggested that the higher TOC mass removal rate was likely attributed to biodegra-

dation. That is, a higher biomass concentration was attainable due to the uptake and

assimilation of additional substrates available in the concentrate recycle stream.

Table 3 RO concentrate and permeate characteristics during Phase 4

Parameters Mean value 6 SD

RO concentrate RO permeate

TOC, mg/L 56.43 ^ 11.53 0.811 ^ 0.477TDS, ppm 3,650 ^ 488 140 ^ 35

Figure 2 Comparison of MBR TOC mass removal rate during Phases 2 and 4 at different wastewater TOC

concentrations

C.H.L

ewetal.

459

7/30/2019 mbr1

6/9

Biodegradation batch tests

To further validate the observation that a fraction of the slow- and hard-to-degrade

organic compounds present in the concentrate recycle stream could be degraded by an

acclimated biomass, four biodegradability batch tests were conducted. The type of sub-

strate and source of biomass used for each batch test are summarized in Table 4. Residual

DOC of each batch test was monitored during a 5 h time period. The four batch tests

were operated at F/M ratios of around 0.5g TOC/g VSS.day. ML from the MBR and

WWTP were centrifuged and the resulting supernatant was discarded. The biomass wasthen washed twice by resuspension with phosphate buffer solution. This preparation pro-

cedure allowed compounds adsorbed onto the biomass to be removed. Autoclaving of

biomass for batch test B was carried out at 121 8C for 20 mins. After the addition of bio-

mass to their respective substrates in conical flasks, the flasks were placed on a shaker

operated at an rpm of 250. The DO level in the flasks was found to be above 4.0 mg/L.

Samples were taken intermittently from the four batch tests by withdrawing 10 mL of

ML from each flask before passing them through 0.45 mm filter papers. The filtrates were

then measured for residual DOC.

Figure 4 compares the variations of residual DOC with time for the four batch tests.

As shown in Figure 4(a), Batch test D, which used wastewater as the substrate, showed a

rapid decrease in DOC during the first hour, followed by a more gradual decrease. This

finding would suggest that bulk of the wastewater could be readily biodegraded. In con-

trast, Batch test A, which used RO concentrate as the substrate, showed only a relatively

small decrease in DOC. The much slower rate of residual DOC decrease observed in

Batch test A, as compared with Batch test D, indicated that the bulk of the constituents in

the RO concentrate were slow- and hard-to-biodegrade in nature.

Figure 4(b) compares the experimental results obtained from Batch tests A and B. The

aim of this comparison was to investigate the roles of biomass adsorption and biodegra-

dation. As the biomass used in Batch test B was washed and autoclaved, any decrease in

residual DOC would be attributed to bioadsorption of concentrated constituents. It is

noted from Figure 4(b) that Batch test B had a smaller decrease in DOC compared with

Table 4 Substrate type and biomass source for Batch tests A, B, C and D

Batch test A B C D

Substrate type RO concentrate RO concentrate RO concentrate WastewaterBiomass source MBR MBR (autoclaved) WWTP MBR

Figure 3 Comparison of MLVSS concentrations during Phases 2 and 4 at different wastewater TOC

concentrations

C.H.L

ewetal.

460

7/30/2019 mbr1

7/9

that attained by Batch test A. This finding suggested that the reduction in DOC observed

in Batch test A was attributed to both biomass adsorption and biodegradation. In contrast,

the reduction in DOC associated with Batch test B was mainly attributed to biomass

adsorption.

Figure 4(c) compares the experimental results obtained from Batch tests A and C. It is

noted from Figures 4(b) and 4(c) that the profile of residual DOC associated with Batchtest C was similar to that of Batch test B. This finding suggested that there was minimal

or no observable biodegradation occurring in Batch test C, and that biomass adsorption

was the main mechanism responsible for the decrease in DOC. Although live biomass

was used for both tests, biodegradation only occurred in Batch test A where the biomass

had been previously exposed and acclimated to conditions of concentrate recycling.

Batch test C, which used non-acclimated biomass, did not show any observable biodegra-

dation. This finding indicated the importance of biomass acclimation to achieve biodegra-

dation of slow- and hard-to-degrade RO concentrate constituents, which is crucial to the

success of the novel IMP.

Effect of concentrate recycling on microbial viability and activity

Assessment of microbial viability and activity was carried out based on sOUR measure-ments. As shown in Table 2, the mean sOUR value obtained from Phase 4 was slightly

less than that of Phase 2. The higher SD again could be explained by the highly variable

wastewater strength experienced in Phase 4. It is also noted from Figure 5 that sOUR

increased with F/M ratio. This finding suggested that the microorganisms could accom-

modate the increase in substrate loading and the increase in loading led to a correspond-

ing increase in oxygen uptake. The food in all F/M ratios for Phase 4 was calculated

based on the weighted concentration of the wastewater and the concentrate recycle

stream. It is also noted that the trend line associated with Phase 4 was below that of

Phase 2. This finding suggested that Phase 4 experienced a suppression of respiratory

activities over the same range of F/M ratios compared with Phase 2. This phenomenon

could be attributed to the consequence of concentrate recycling. SMP and wastewater

inherent refractory compounds recycled back to the MBR had an inhibitory effect on the

microorganisms. However, this inhibition was not severe and the microorganisms

remained viable. They accommodated the increase in F/M ratio via a corresponding

increase in sOUR (during phase 4).

Figure 4 Comparisons of the variation of residual DOC with time for Batch tests A, B, C and D

C.H.L

ewetal.

461

7/30/2019 mbr1

8/9

Effect of concentrate recycling on MBR effluent quality

As shown in Table 2, the average MBR effluent TOC concentration associated with

Phase 4 was higher than that of Phase 2. This observation can be better illustrated in

Figure 6, which shows the relationship between effluent TOC concentrations and F/M

ratios for Phases 2 and 4. It is noted from this figure that the MBR effluent TOC associ-

ated with Phase 2 deteriorated slightly as F/M ratio increased. This finding reflected the

robustness of the MBR in terms of its ability to produce effluent of consistent qualityeven with varying wastewater strength. However, such a trend was not observed when

recycling of a concentrate stream was carried out. The deterioration of the effluent quality

was more significant when F/M ratio was increased. However, it should be pointed out

that the increase in F/M ratio during Phase 4 was attributed to both an increase in the

wastewater concentration, as well as an increase in the concentration of the recycle

stream. Although operating the MBR with a concentrate recycle stream resulted in an

overall worsening of the effluent quality, this drawback could be overcome by operating

the MBR at a lower F/M ratio. For example, with an F/M below 0.03g TOC/g VSS.day,

the effluent quality associated with Phase 4 was not significantly different from that

obtained from Phase 2 whereby no concentrate recycling was implemented.

Figure 5 Comparison of sOUR during Phases 2 and 4 at different F/M ratios

Figure 6 Comparisons of MBR effluent TOC trend during Phases 2 and 4 at different F/M ratio

C.H.L

ewetal.

462

7/30/2019 mbr1

9/9

Performance of the novel IMP

The performance of the novel IMP was assessed based on the overall TOC mass removal

efficiency given by Eq. (3).

MIMP Cww Qww 2 Cperm Qperm 2 CR QB

Cww Qww 100% 3

where MIMP is the overall IMP TOC mass removal efficiency (%), Cperm the RO permeate

TOC concentration (g/L), CR the RO concentrate TOC concentration (g/L), Qperm the RO

permeate flow rate (L/d), QB the RO concentrate bleed rate (L/d), and Cww and Qww as

previously defined. The mean value and SD of MIMP were found to be 88.94 ^ 3.80%.

The RO permeate quality is given in Table 3. Thus, it has been demonstrated that thenovel IMP could achieve good TOC mass removal efficiency and it is able to produce

water of good quality for reuse purposes.

Conclusions

The novel IMP with concentrate recycling could deliver good performance in terms of

overall TOC mass removal efficiency. A portion of the slow- and hard-to-degrade organic

constituents present in the recycle stream could be degraded by an acclimated biomass

operating at a SRT of 20 days. The availability of additional substrates in the concentrate

recycle stream could support a higher biomass growth that led to a higher biomass con-

centration. However, the recycled concentrate constituents had an inhibitory effect on the

microorganisms, resulting in a slight suppression of respiratory activities. Nonetheless,

the inhibition was not severe and the microorganisms remained viable and responded tothe increase in F/M ratio by a higher specific oxygen uptake rate. Experimental results

suggested that operating the MBR (in the novel IMP) at a F/M ratio below 0.03 g TOC/g

VSS.day could yield an effluent quality comparable to that achievable without imple-

menting concentrate recycling. On the whole, the novel IMP had been demonstrated to

achieve good overall TOC mass removal efficiency and was able to produce water of

good quality for reuse purposes.

References

APHA (2000). Standard Methods for the Examination of Water and Wastewater, APHA/AWWA/WEF,

Washington DC, USA.

Brindle, K. and Stephenson, T. (1996). The application of membrane biological reactors for the treatment of

wastewater. Biotechnology and Bioengineering, 49, 601610.Cote, P., Buisson, H., Pound, C. and Arakaki, G. (1997). Immersed membrane activated sludge for the reuse

of municipal wastewater. Desalination, 113, 189196.

Duncan, J.B. and David, C.S. (1999). A review of soluble microbial products (SMP) in wastewater treatment

systems. Water Research, 33(14), 30633082.

Huang, X., Liu, R. and Qian, Y. (2000). Behaviour of soluble microbial products in a membrane bioreactor.

Process Biochemistry, 36, 401406.

Rosenberger, S., Kruger, U., Witzig, R., Manz, W., Szewzyk, U. and Kraume, M. (2002). Performance of a

bioreactor with submerged membranes for aerobic treatment of municipal wastewater. Water Research,

36, 413420.

Shin, H.-S. and Kang, S.-T. (2003). Characteristics and fates of soluble microbial products in ceramic

membrane bioreactor at various sludge retention times. Water Research, 37, 121127.

Yamamoto, K. (2001). Membrane bioreactor: an advanced wastewater treatment/reclamation technology and

its function in excess-sludge minimization. In: Advances in Water and Wastewater Treatment Technology ,

Elsevier, Amsterdam, pp 229237.

C.H.L

ewetal.

463

http://dx.doi.org/10.1016/S0011-9164(97)00128-8http://dx.doi.org/10.1016/S0011-9164(97)00128-8http://dx.doi.org/10.1016/S0011-9164(97)00128-8http://dx.doi.org/10.1016/S0043-1354(99)00022-6http://dx.doi.org/10.1016/S0043-1354(99)00022-6http://dx.doi.org/10.1016/S0032-9592(00)00206-5http://dx.doi.org/10.1016/S0043-1354(01)00223-8http://dx.doi.org/10.1016/S0043-1354(01)00223-8http://dx.doi.org/10.1016/S0043-1354(02)00249-Xhttp://dx.doi.org/10.1016/S0043-1354(02)00249-Xhttp://dx.doi.org/10.1016/S0043-1354(02)00249-Xhttp://dx.doi.org/10.1016/S0043-1354(02)00249-Xhttp://dx.doi.org/10.1016/S0043-1354(02)00249-Xhttp://dx.doi.org/10.1016/S0043-1354(01)00223-8http://dx.doi.org/10.1016/S0043-1354(01)00223-8http://dx.doi.org/10.1016/S0032-9592(00)00206-5http://dx.doi.org/10.1016/S0043-1354(99)00022-6http://dx.doi.org/10.1016/S0043-1354(99)00022-6http://dx.doi.org/10.1016/S0011-9164(97)00128-8http://dx.doi.org/10.1016/S0011-9164(97)00128-8