ENVIRONMENTAL MICROBIOLOGY

Artificial Intelligence for the Evaluation of OperationalParameters Influencing Nitrification and Nitrifiersin an Activated Sludge Process

Oluyemi Olatunji Awolusi1 & Mahmoud Nasr1 & Sheena Kumari1 & Faizal Bux1

Received: 31 August 2015 /Accepted: 8 February 2016# Springer Science+Business Media New York 2016

Abstract Nitrification at a full-scale activated sludge planttreating municipal wastewater was monitored over a periodof 237 days. A combination of fluorescent in situ hybridiza-tion (FISH) and quantitative real-time polymerase chain reac-tion (qPCR) were used for identifying and quantifying thedominant nitrifiers in the plant. Adaptive neuro-fuzzy infer-ence system (ANFIS), Pearson’s correlation coefficient, andquadratic models were employed in evaluating the plant oper-ational conditions that influence the nitrification performance.The ammonia-oxidizing bacteria (AOB) abundance was with-in the range of 1.55 × 108–1.65 × 1010 copies L−1, whileNitrobacter spp. and Nitrospira spp. were 9.32 × 109–1.40 × 1011 copies L−1 and 2.39 × 109–3.76 × 1010

copies L−1, respectively. Specific nitrification rate (qN) wassignificantly affected by temperature (r 0.726, p 0.002), hy-draulic retention time (HRT) (r −0.651, p 0.009), and ammo-nia loading rate (ALR) (r 0.571, p 0.026). Additionally, AOBwas considerably influenced by HRT (r −0.741, p 0.002) andtemperature (r 0.517, p 0.048), while HRT negatively impact-ed Nitrospira spp. (r −0.627, p 0.012). A quadratic combina-tion of HRT and food-to-microorganism (F/M) ratio also im-pacted qN (r2 0.50), AOB (r2 0.61), and Nitrospira spp. (r2

0.72), while Nitrobacter spp. was considerably influenced bya polynomial function of F/M ratio and temperature (r2 0.49).The study demonstrated that ANFIS could be used as a tool to

describe the factors influencing nitrification process at full-scale wastewater treatment plants.

Keywords Adaptive neuro-fuzzy inference system .

Ammonia-oxidizing bacteria . Nitrite-oxidizing bacteria .

Operational parameters . Statistical tools

Introduction

Ammonia toxicity is one of the several forms of nitrogenpollution that exist in aquatic environments [1]. Ammoniacan be harmful to aquatic life and contributes to eutrophicationof water bodies [2]. Moreover, at sufficiently high levels, am-monia can create a large oxygen demand in receiving waters,where the total consumption of oxygen is 4.57 g O2g

−1 N-NH4

+ oxidized [3]. In this context, ammonia removal fromwastewater is one of the primary tasks to protect water re-sources from pollution discharges. Biological nitrification-denitrification is the most commonly used process for remov-ing nitrogen fromwastewater [4]. During nitrification process,ammonia (NH3) is converted to nitrite (NO2

−) by ammonia-oxidizing bacteria (AOB), while nitrite-oxidizing bacteria(NOB) convert the NO2

− to nitrate (NO3−) [5]. In denitrifica-

tion, NO3− is reduced to nitrogen gas (N2) in a four-step pro-

cess, in which NO2−, nitric oxide (NO), and nitrous oxide

(N2O) are electron acceptors in energy generating reactions[6].

Nitrifying bacteria are highly sensitive to changes in envi-ronmental parameters and plant operational conditions, suchas pH, temperature, dissolved oxygen (DO) level, organicloading rate (OLR), ammonia loading rate (ALR), and hy-draulic retention time (HRT) [7]. Neutral to slightly basic pHrange (7.5 to 8.5) has been reported as optimum for efficientnitrification [8]. According to an earlier finding, at neutral pH,

Electronic supplementary material The online version of this article(doi:10.1007/s00248-016-0739-3) contains supplementary material,which is available to authorized users.

* Faizal Buxfaizalb@dut.ac.za

1 Institute forWater andWastewater Technology, Durban University ofTechnology, Durban 4000, South Africa

Microb EcolDOI 10.1007/s00248-016-0739-3

99 % of ammonia was removed. However, this dropped toabout 75 % at a basic pH of 9.7 while acidic pH (4.8) resultedin an impaired removal efficiency of 56 % [7]. Although itwas reported that nitrification would proceed at a temperaturerange of 20 to 37 °C [9], nonetheless niche differentiationexists among the members of NOB group with Nitrobacterhaving preference for relatively low temperatures (24–25 °C)while Nitrospira thrives at higher temperatures of 29–30 °C[10]. Kim et al. [1] demonstrated that raising temperature from20 to 30 °C resulted in a 5.3-fold increase in ammonia oxida-tion and a 2.6-fold increase in nitrite oxidation. Lower nitrifi-cation performance has been observed at higher organic load-ing due to the competition for DO between heterotrophic bac-teria and autotrophic nitrifying organisms in wastewater treat-ment system, [11]. Huang et al. [10] demonstrated that higherDO concentrations were more suitable to Nitrobacter growth,while Nitrospira were selectively enriched when DO concen-trations were less than 1.0 mg/L. The available carbon sub-strate for the unit mass of microorganism (known as F/Mratio) can also impact nitrification. A study by Wu et al. [11]suggested that high F/M ratio should be avoided to minimizeits negative impact on nitrification, and it indicated that F/Mratio between 0.2 and 0.4 was the optimum range for nitrifi-cation. The influence of HRT on nitrification efficiency wasalso observed when HRT decreased from 30 to 5 h with aresultant increase in specific ammonium-oxidizing andnitrate-forming rates [12]. Additionally, the decrease in HRTled to a reduction of AOB population density, whereas theNOB, especially the fast growing Nitrobacter spp., increasedsignificantly [12].

Modeling of a full-scale wastewater treatment plant(WWTP), operated under an uncontrolled environment,requires advanced nonlinear modeling tools to simulatethe complex relationships between inputs and outputs[1]. According to Klemetti [13], due to simultaneousdependence on different factors, competition betweenmicrobial groups is usually nonlinear. Artificial intelli-gence (AI) is an example of a nonlinear system that iscapable of depicting the interactive influence betweenvariables as well as their correlation with the simulationoutput [14]. AI incorporates artificial neural network(ANN), fuzzy inference system (FIS), and adaptive-neuro fuzzy inference system (ANFIS). ANN has theability to model nonlinear systems efficiently owing totheir high accuracy, adequacy, and quite promising ap-plications in engineering [15]. FIS allows a logical data-driven modeling approach, which uses Bif–then^ rulesand logical operators to establish qualitative relation-ships among the variables [16]. ANFIS is a neuro-fuzzy system that has the potential to capture the bene-fits of both ANN and FIS in a single framework [14].Moreover, ANFIS can handle complex and highly non-linear relationships between several parameters, without

the difficult task of dealing with deterministic nonlinearmathematics [17]. Modeling based on ANFIS needs alittle knowledge about the process to track giveninput/output data. In this context, it is expected thatthe effect of system operation on nitrification processcould be studied using ANFIS. To the best of ourknowledge, this is the first study applying ANFIS tech-nique to describe the nitrification performance at a realWWTP subjected to dynamic operational parameters.

In this study, fluorescent in situ hybridization (FISH) andquantitative real-time polymerase chain reaction (qPCR) wereused for detecting and quantifying the dominant nitrifiers at afull-scale WWTP. The effect of operational parameters andenvironmental conditions on nitrification was also investigat-ed. An ANFIS was applied to select and rank the operationalconditions that mostly influence the nitrification process.ANFIS results were further validated with cross-correlationcoefficients and quadratic models.

Materials and Methods

Plant Description

The full-scale WWTP under study is situated in the Midlandsof KwaZulu-Natal province, South Africa. The plant receivesa discharge of 82,880±20,832 m3 day−1 (average dry weatherflow), including 90 % domestic and 10 % industrial wastewa-ters. The plant was designed based on the criteria of aJohannesburg configuration, which allow for nitrification/denitrification biological process [18]. As shown in Fig. 1,the effluent from primary settling tank is distributed to thepre-anoxic and anaerobic tanks. The pre-anoxic basin is fur-ther enriched with return activated sludge from the bottom of afinal settler. The effluent from anaerobic tank is dischargedinto the anoxic unit. An internal recycle is pumped from thelast part of aerobic units to the anoxic zone. The mixed liquor,containing activated sludge, flows from the aerobic zone to asecondary settler, where it is separated under a quiescent con-dition into treated wastewater and return activated sludge.

Sampling Protocol

TheWWTPwas monitored for a period of 237 days, i.e., fromMay to July 2012 and from November 2012 to March 2013.The study period was divided into two phases. Phase 1 repre-sents the winter period during the first 78th days; and phase 2,the summer season from the 79th to the 237th day. Compositesludge samples, from the aerobic chamber, in addition to in-fluent and effluent wastewater samples were collected twiceper month. All samples were kept on ice while in transit to thelab.

O.O. Awolusi et al.

Fluorescent In Situ Hybridization

FISH analysis was performed for initial identification of thenitrifying community according to the Fuchs et al. [19] proto-col with modification. Sample preparation, immobilization onglass slides, dehydration, permeabilization, quality check ofprobes, and hybridization were undertaken as described earlierby Nielsen [20]. The samples were sonicated at 8 W for10 min to disperse the flocs before performing the hybridiza-tion. Oligonucleotide probes targeting betaproteobacterialAOB and genus Nitrobacter spp. and Nitrospira spp., namely,Nso1225 [21], NIT3 [22], and Ntspa662 [12], respectively,were used (Table 1). The probes were labeled with CALFluor Red 590 fluorescence dye (Inqaba BiotechnicalIndustry (Pty), South Africa). The samples were visualizedusing a Zeiss AxioCam HRm (Carl Zeiss, Germany)epifluorescent microscope. Image analysis was carried outusing a Zeiss AxioVision Release 4.6 (12-2006) imagingsoftware.

DNA Extraction and Real-Time Quantitative PCRAmplification

Genomic DNAwas extracted from sludge samples accordingto the Purkhold et al. [23] procedure. Quality and quantity ofthe extracted DNA were ascertained with a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, USA).Individual standard curves were prepared for the different ni-trifiers using purified 16S rRNA gene fragments (target DNA)obtained from PCR-amplified AOB (amoA-1F and amoA-2R), Nitrobacter spp. (FGPS872f and FGPS1269r), andNitrospira spp. (NSR1113F and NSR1264R) as describedby [24, 25] (Table 2). The concentrations were used in calcu-lating their copy by considering their molecular weight andAvogadro’s number. For qPCR standard curves, 10-fold serial

dilutions of the target DNA were prepared from 108 to 101

copy numbers. The real-time PCR quantification was carriedout according to modification of the method described bySteinberg and Regan [26] using Bio-Rad C1000 TouchThermal Cycler-CFX96 Real-Time System (Bio-Rad, USA).For the quantitative real-time PCR, the primers already de-scribed in Table 2 were used. The optimized protocols usedfor quantifying the nitrifiers are shown in Table 3. To confirmamplification of the correct product, the amplicons fromqPCR were electrophoresed in 1.2 % (w/v) agarose gel forthe presence of the expected gene product sizes. The qPCRstandard curve parameters used for the analysis are listed inTable 4.

Analytical Analysis

Concentrations of inorganic nitrogen species and chemicaloxygen demand (COD) were estimated using standardmethods [27]. Temperature, DO, and pH measurements werecarried out using the YSI 556 MPS (Multiprobe System).

Calculations

Operational conditions such as HRT, OLR, ALR, and F/Mratio were calculated as mentioned in Tchobanoglous et al.[28] as follows (Eqs. 1, 2, 3, and 4):

HRT ¼ V

Qð1Þ

OLR ¼ Q� COD

Vð2Þ

ALR ¼ Q� N−NH4þ

Vð3Þ

F=M ¼ Q� COD

MLSS � Vð4Þ

Fig. 1 Schematic diagram of thefull-scale biological treatmentprocess under study

Table 1 rRNA—targetedoligonucleotide probes and theirspecificity

Probe name Target Sequence (5′–3′) FA (%) Reference

NSO 1225 Betaproteobacterial AOB CGCCATTGTATTACGTGTGA 35 [21]

NIT3 Genus Nitrobacter CCTGTGCTCCATGCTCCG 40 [22]

Ntspa 662 Genus Nitrospira GGAATTCCGCGCTCCTCT 35 [12]

FA formamide

Artificial Intelligence for the Evaluation of Operational Parameters

where HRT is the hydraulic retention time; V is the reactorvolume; Q is the flow rate; OLR is the organic loading rate;COD is the chemical oxygen demand; ALR is the ammonialoading rate;N-NH4

+ is the ammonia nitrogen; F/M ratio is thefood-to-microorganisms ratio; and MLSS is the mixed liquorsuspended solids.

Adaptive Neuro-Fuzzy Inference System

Architecture of ANFIS

To present the ANFIS architecture, two fuzzy if–then rulesbased on a first-order Takagi–Sugeno fuzzy model wereconsidered:

Rule-1: If (x is A1) and (y is B1), then (f1 =p1x+q1y+ r1)Rule-2: If (x is A2) and (y is B2), then (f2 =p2x+q2y+ r2)

where x and y are the inputs; A1 and B1 are the fuzzy sets; f1 isthe outputs within the fuzzy region specified by the fuzzy rule;and p1, q1 and r1 are the design parameters determined duringthe training process.

As shown in Fig. 2, the ANFIS architecture to implementthese two rules has a total of five layers, in which a circleindicates a fixed node, whereas a square indicates an adaptivenode. The functioning of each layer can be described as fol-lows [29]

Layer-1 (input node): Parameters in this layer are referredto Bpremise parameters.^ Every single node generates a fuzzy

membership grade of linguistic label. The membershipfunctions (MFs) of Ai and Bi − 2 are given by Eqs. 5 and6, respectively:

O1i ¼ μAi

xð Þ i ¼ 1; 2 ð5Þ

O1i ¼ μBi−2

yð Þ i ¼ 3; 4 ð6Þ

where x (or y) is the input to node i and Ai (or Bi − 2) is thelinguistic label (small, large, etc.) related to this node. If thebell-shaped MF is generalized, μAi

xð Þ is given by Eq. 7:

μAixð Þ ¼ 1

1þ x−ciai

� �2� �bið7Þ

where ai, bi, and ci are the MF parameters, governing the bell-shaped functions accordingly.

Layer-2 (rule nodes): In the second layer, the nodes arelabeled withM, indicating that they perform as a simple mul-tiplier. The AND/OR operator is used to get one output thatrepresents the antecedent of the fuzzy if–then rule. The out-puts of this layer are defined as firing strengths of the rules.Each node analyzes the firing strength by cross multiplying allthe incoming signals (Eq. 8):

O2i ¼ wi ¼ μAi

xð ÞμBiyð Þ i ¼ 1; 2 ð8Þ

Layer-3 (average nodes): In the third layer, the nodes arelabeled with N, demonstrating that they play a normalization

Table 2 List of primers used andthe optimized annealingtemperature

Target Primer Annealing (°C) References

AOB amoA amoA-1F/amoA-2R 55 [25]

Nitrobacter 16S rDNA FGPS872/FGPS1269 50

Nitrospira 16S rDNA NSR1113F/NSR1264R 65

Table 3 Optimized real-time PCR protocols for quantifying nitrifiers

Real-time PCR step Primer

amoA-1F/amoA-2R FGPS872/FGPS1269 NSR1113F/NSR1264R

1. Initial activation 3:30 min at 95 °C 3:30 min at 95 °C 3:30 min at 95 °C

2. Denaturation 0:30 min at 95 °C 0:30 min at 95 °C 0:30 min at 95 °C

3. Annealing 0:30 min at 54 °C 0:30 min at 50 °C 0:30 min at 65 °C

4. Extension 0:30 min at 72 °C 0:30 min at 72 °C 0:30 min at 72 °C

5. Read fluorescence Read Read Read

6. Go to step 2 for 40 times 40 times 40 times

7. Melt curve 55 to 65 °C, incrementof 0.5 °C every 50 s

55 to 65 °C, incrementof 0.5 °C every 50 s

55 to 65 °C, incrementof 0.5 °C every 50 s

8. Read fluorescence Read Read Read

O.O. Awolusi et al.

role to the firing strengths from the previous layer. Thus, out-puts of this layer are called Bnormalized firing strengths^. Asshown in Eq. 9, the ith node calculates the ratio of the ith rule’sfiring strength to the sum of all rules’ firing strengths:

O3i ¼ wi ¼ wi

w1 þ w2i ¼ 1; 2 ð9Þ

Layer-4 (consequent nodes): In this layer, every node i is anadaptive node with a node function. The output of each nodeis simply the product of the normalized firing strength and afirst-order polynomial (for a first-order Sugeno model).Hence, the outputs of this layer are expressed by Eq. 10:

O4i ¼ wi f i ¼ wi pixþ qiyþ rið Þ i ¼ 1; 2 ð10Þ

where wi is the output of layer-3 and {pi, qi, ri} are conse-quent parameters, pertaining to the first-order polynomial.

Layer-5 (output node): In the fifth layer, there is only onesingle fixed node labeled with ∑. The single node computesthe overall output as the summation of all incoming signals.Thus, the overall output of the model is given by Eq. 11 asfollows:

O5i ¼

X2i¼1

wi f i ¼

X2i¼1

wi f i

!

w1 þ w2ð11Þ

Application of ANFIS

The function exhsrch in Matlab was used to select the set ofinputs that considerably impact the nitrification activity.Theoretically, exhsrch builds an ANFIS model for each com-bination and trains it for one epoch, sequentially reporting theperformance achieved. ANFIS uses a hybrid learning algo-rithm to tune the parameters of a Sugeno-type FIS [30]. Thealgorithm uses a combination of the least-squares and back-propagation gradient descent methods to model a training dataset [29]. The dataset is randomly classified into training(70 %) and checking (30 %) arguments. The training processstops if the designated epoch number is reached or the errorgoal is achieved, whichever comes first. The checking data areused for testing the generalization capability of the ANFIS andfor monitoring how well the model predicts the correspondingdataset output values. Moreover, ANFIS validates modelsusing a checking data set to test for overfitting of the trainingdata. Recently, this technique has been successfully imple-mented in the field of wastewater treatment technology [17,31].

Results and Discussion

Environmental Conditions and System Performance

Influent and effluent wastewater characteristics of the full-scale WWTP over the 237-day monitoring period aredisplayed in Fig. 3. Operational conditions of the treatmentplant were classified into phase 1 and phase 2, correspondingto winter and summer seasons, respectively (Table 5). Thetemperature measurements during winter (16.5±2.1 °C) andsummer (22.4±2.7 °C) recorded were lower than the opti-mum temperature range of 25–30 °C for most strains of nitri-fiers [32]. The winter season (1st–78th day) experienced littleor no rainfall resulting in an average influent flow rate of 61,990 ± 2172 m3 day−1, while it increased to 93,062 ± 18,

Table 4 Description of qPCR standard curve parameters

Parameter Target

AOB Nitrobacter spp. Nitrospira spp.

Efficiency (%) 102.5 ± 2.1 92.75 ± 1.63 107.3 ± 1.9

Slope −3.3 ± 0.05 −3.5 ± 0.05 −3.2 ± 0.04r2 of slope 0.998 ± 0.001 0.99 ± 0.01 0.998 ± 0.04

Intercept 39.8 ± 2.2 35.0 ± 0.59 37.2 ± 0.14

1A

2A

M Nx1w 1w

x y

1B

2B

M Ny2w 2w

x y

11 fw

22 fw

f

Layer-1 Layer-2 Layer-3 Layer-4 Layer-5

∑

Fig. 2 Typical first-order SugenoANFIS architecture

Artificial Intelligence for the Evaluation of Operational Parameters

106 m3 day−1 during the summer (79th–237th day). The in-fluent COD during the summer was 1.3-fold lower than in thewinter due to the dilution effect of increased rainfall. The OLRincreased during summer with increasing flow rate, despitethe lower COD, and reached 4.5±1.8 kg COD m−3 day−1.The deviations in OLRs could be attributed to the variationin the type of influent as a result of commercial and industrialactivities occurring around the treatment plant. Under thesefluctuating conditions, a steady COD removal efficiency of95±3 % was still maintained (Fig. 3). The F/M ratio was0.6±0.1 day−1 in the winter and increased by 48 % duringthe summer (Table 5). There is no ideal F/M ratio that canwork for all activated sludge treatment systems. Previous stud-ies indicate that the recommended range for F/M ratio in

conventional, completely mixed, and high rate activatedsludge processes ranged between 0.2–0.4, 0.2–0.6, and 0.4–1.5 day−1, respectively [33].

Although the influent ammonia during summer was 1.2-fold lower than in winter, the respective ALR increased by18.8 % resulting from the summer rainfall (Table 5). Theammonia removal efficiency improved from 60±18 to 83± 13 % although the ALR rose from 121 ± 22 to 144±29 g N-NH4

+m−3 day−1 through winter to summer, respec-tively. Nitrification rate followed the same trend of ammoniaremoval, where it also increased by 60 % and recorded 119± 30 g N-NH4

+m−3 day−1 during the summer season. Theincrease in ammonia removal efficiency and nitrification ratewith ALR indicated that ammonia concentration was not the

86

88

90

92

94

96

98

100

0

200

400

600

800

1000

1200

1400

1600

CO

D r

emoval

(%

)

Infl

uen

t an

d e

fflu

ent

CO

D (

mg L

-1)

Influent COD

Effluent COD

COD removal

0

10

20

30

40

50

60

70

80

90

100

0

5

10

15

20

25

30

35

40

45

Am

monia

rem

oval

(%

)

Infl

uen

t an

d e

fflu

ent

N-N

H4

+(m

g L

-1)

0

10

20

30

40

50

60

70

80

90

100

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

0 20 40 60 80 100 120 140 160 180 200 220 240

Nit

rate

rem

oval

(%

)

Infl

uen

t an

d e

fflu

ent

nit

rate

(m

g L

-1)

Time (d)

Influent nitrate

Effluent nitrate

Nitrate removal

Influent N-NH4+

Effluent N-NH4+

Ammonia removal (%)

Fig. 3 Influent and effluentcharacteristics of the full-scalewastewater treatment plant understudy

O.O. Awolusi et al.

nitrification limiting factor in this plant. This suggested thatother operational conditions had higher influence on the nitri-fication process rather than ALR. Results from our study werelower than the 97–99.9% ammonia removal that was obtainedby Campos et al. [3] when the nitrifying activated sludge unitwas subjected to high nitrogen loading rates (up to 7500 g N-NH4

+m−3 day−1). The higher nitrification performance of theplant could be due to operation of the unit under a controlledenvironment, where the nitrifying activated sludge was fedwith synthetic wastewater containing ammonia and other nu-trient sources with regulated pH. The influent nitrate concen-tration during the study period varied between 0.25 and3.68 mg L−1 with an average value of 1.51± 0.77 mg L−1

(Fig. 3). Nitrate removal efficiencies exhibited 64.6 ± 27.6and 88.0±26.3 % during winter and summer seasons, respec-tively. The concentration of effluent nitrate during the winterwas 3.6-fold higher than summer. This might be an indicationof the increased NOB/AOB ratio during the winter that pro-motes nitrate accumulation and thus a higher effluent nitrateconcentration in winter [6, 34]

The DO levels throughout the aeration tank varied between0.24 and 1.27 mg L−1, with an average value of 0.63

±0.22 mg L−1. The DO concentration in our study was lowerthan the optimum value of 1.7 mg L−1 for a complete nitrifi-cation process [35]. The pH in the feed over the entire sam-pling period was relatively stable with an average value of 7.2±0.1, which is close to the optimum for nitrifiers (7.5–8.0)[35]. Ruiz et al. [35] reported that at the range of pH 6.45–8.95, as observed here, a complete nitrification to nitrate oc-curs, while at pH lower than 6.45 and higher than 8.95 wouldresult in complete inhibition of nitrification.

Identification and Quantification of Dominant Nitrifiers

The betaproteobacterial AOB, Nitrobacter spp., andNitrospira spp. were detected in abundance throughout thesampling period (Fig. 4). The AOB population abundancewas quantified using the primer set targeting the ammoniamonooxygenase (amoA) gene locus, whereas Nitrobacterand Nitrospira 16S rDNAs were targeted for the NOB. Thespecificity of the primers was confirmed by gel electrophore-sis with the expected base pair length (AOB 490 bp;Nitrobacter spp. 386 bp;Nitrospira spp. 151 bp). The efficien-cies of the qPCR runs were between 90 and 110 %, and thestandard curves were linear over six orders of magnitude(r2 >0.99).

The AOB abundance was within the range of 1.55×108–1.65 × 1010 copies L−1 MLSS, whereas those of theNitrobacter spp. and Nitrospira spp. were found to be in therange of 9.32 × 109–1.40 × 1011 copies L−1 MLSS and2.39 × 109–3.76 × 1010 copies L−1 MLSS, respectively(Fig. 5). The Nitrobacter spp. abundance was 29-fold higherthan that of AOB throughout the study period showing a cleardominance of NOB in the selectedWWTP. The average AOBto NOB ratio varied from 0.03:1 (winter) to 0.15:1 (summer),which was lower than the theoretical ratio of 2:1 reported forgood nitrification [36]. Due to the prevailing limiting oxygenlevels (0.63±0.22 mg O2L

−1) in the reactor, there was a pos-sibility of nitrite loop. This phenomenon usually occurs whendenitrifiers reduce nitrate to nitrite supplying additional nitritefor NOB, leading to a higher NOB/AOB ratio than theoreti-cally expected [36]. Additionally, Liu [37] noted that underlong-term low DO, the oxygen affinity of NOB increases

Table 5 Operational conditions and ANFIS model parameters of thefull-scale wastewater treatment plant under study

Parameters Phase 1 (winter) Phase 2 (summer)

Rainfall (mm) 26.0 ± 18.6 116.8 ± 32.0

Temperature (°C) 16.5 ± 2.1 22.4 ± 2.7

HRT (h) 6.3 ± 0.2 4.3 ± 1.0

OLR (kg COD m−3 day−1) 4.0 ± 1.1 4.5 ± 1.8

ALR (g N-NH4+m−3 day−1) 121 ± 22 144 ± 29

F/M (g COD g−1 MLSS day−1) 0.6 ± 0.1 0.9 ± 0.3

qN (mg N-NH4+ g−1 MLSS day−1) 12.3 ± 6.1 26.6 ± 10.7

AOB (copies × 109 L−1) 1.00 ± 0.86 7.35± 5.75

Nitrobacter spp.(copies × 109 L−1)

34.8 ± 19.0 50.8 ± 46.1

Nitrospira spp. (copies × 109 L−1) 3.32 ± 0.60 11.9 ± 11.2

HRT hydraulic retention time, OLR organic loading rate, ALR ammonialoading rate, F/M food to microorganisms ratio, qN specific nitrificationrate, AOB ammonia-oxidizing bacteria

Fig. 4 Micrograph of sampleshybridized with Cal 590 labeled.a NSO 1225 (AOB); b NIT3(Nitrobacter spp.); c Ntspa 662(Nitrospira spp.) oligonucleotideprobes

Artificial Intelligence for the Evaluation of Operational Parameters

significantly, which makes NOB a better competitor for oxy-gen compared to AOB. This high NOB/AOB ratio resulted insub-optimal ammonia transformation in this study. Whereasearlier studies have recorded about 99 % ammonia removal[7], in this study the highest NH3 removal that was recordedwas 83±13 %.

A higher Nitrobacter/Nitrospira ratio of 7.4: 1.0 wasalso recorded. This could possibly be explained basedon the earlier observations of Wagner et al. [38] andNogueira and Melo [39]. They noted an irreversibleprevalence of Nitrobacter spp. over Nitrospira spp. inWWTP after a history of spike in nitrite concentration,even after subsequent reduction in nitrite concentration.Nitrobacter usually exhibit inhibitory effect on thegrowth of Nitrospira once it dominates. Furthermore,Fukushima et al. [40] reported that Nitrobacter spp.can be selected over Nitrospira spp. in plants withlow inorganic carbon in addition to low nitrite concen-tration. However, there have been no studies on thedistribution of Nitrospira and Nitrobacter in DO-limiting condition over an extended period of time ata full-scale WWTP. Despite several other reports thatNitrospira spp. often are the dominant NOB in activatedsludge systems [41], the result from this study indicatesthat the knowledge about nitrifying bacteria populationsat full-scale level still needs further investigations.

Effect of Operational Conditions on Specific NitrificationRate

In this investigation, it was observed that the efficiency andeffectiveness of a nitrifying activated sludge system dependedon several factors. The specific nitrification rate (qN) showed astrong positive correlation to temperature (r 0.726, p 0.002).The figures associated with all the regression analysis is givenin Supplementary Material Figure S1, S2, S3, S4, S5, S6, andS7. The qN noticeably increased by 2.2-fold and exhibited26.6±10.7 mg N-NH4

+g−1 MLSS day−1 during the summerseason when the temperature was elevated to 22.4±2.7 °C.The ANFIS model indicated that temperature exhibited theleast error, demonstrating its relevance with respect to qN(Fig. 6a). The considerable impact of temperature on qN couldbe attributed to the high seasonal variation observed duringthe monitored period. This observation was in agreement withprevious studies, which stated that increasing temperaturecould enhance the rate of nitrification and nitrifier growth[42]. Xu et al. [43] reported that growth rate of nitrifiers in-creased exponentially at a temperature range of 10–25 °C,reaching a constant and optimal growth rate between 25 and35 °C; however, at 40 °C the growth rate diminisheddrastically.

According to the ANFIS results in Fig. 6a, HRT has thesecond rank after temperature regarding the operational

0

2E+10

4E+10

6E+10

8E+10

1E+11

1.2E+11

1.4E+11

1.6E+11

,B

OA

retcabortiN

and

arepsortiN

(co

pie

s.L

-1)

AOB

Nitrobacter

Nitrospira

0

20

40

60

80

100

120

140

160

0 20 40 60 80 100 120 140 160 180 200 220 240

dna

BO

A/B

ON

retcabortiN

/arepsorti

N

Time (d)

NOB/AOB

Nitrobacter/Nitrospira

Fig. 5 Temporal changes indominant nitrifiers of the full-scale wastewater treatment plantunder study

O.O. Awolusi et al.

conditions affecting qN. The current study witnessed a signif-icant increase in qN with a decrease in HRT (r −0.651, p0.009). This result was in agreement with a study by Li et al.[12], where qN increased from 320 to 450 mg N-NH4

+ g−1

MLSS day−1 (elevated by 41 %) when the HRT decreasedfrom 10 to 5 h, showing that the decline in HRT led to anenhancement in ammonia oxidation activity. On the contrary,other studies have reported that lower HRT results in increas-ing loading rates, which negatively affects the nitrifiers due totheir competition with heterotrophic bacteria for substrates(oxygen and ammonia) [44]. In our study, the negative trendbetween qN and HRT can be linked to the seasonal change,which was the key factor in the development of nitrification,since qN increased to 26.6±10.7mgN-NH4

+g−1MLSS day−1

in summer season when temperature increased to 22.4±2.7 °C, while HRT declined to 4.3±1.0 h.

In the current study, qN showed a significant positive cor-relation with ALR (r 0.571, p 0.026) (Supplementary TableS1, S2, S3, S4, S5, S6 and S7). However, it was found thathigh concentration of ammonia in the influent can negativelyaffect nitrification due to substrate inhibition by free ammonia[45]. This discrepancy could be attributed to the fact that theALR of 120–140 g N-NH4

+m−3 day−1observed in our studywas still lower than the inhibitory limits reported inprevious studies [3, 45]. For example, Campos et al.[3] investigated the possibilities of obtaining a full am-monia oxidation at increasing ALR from 500 to7500 g N-NH4

+m−3 day−1. Additionally, Kim et al.

Fig. 6 Effect of operatingconditions on qN (the left-mostinput variable is the most relevantwith respect to qN). a Every inputvariable’s influence on qN; b alltwo-input variable combinationsand their influence on qN

Artificial Intelligence for the Evaluation of Operational Parameters

[45] found that nitrification efficiency increased to100 % with N-NH4

+ loading of 700 g m−3 day−1 at18 °C, and leachate was completely nitrified up to aload of 1500 g N-NH4

+m−3 day−1 at 28 °C.Other operational conditions such as DO showed no sig-

nificant effect on qN (r−0.141, p 0.617). This observation wasin agreement with a study by Kim et al. [45], who noted thatDOwas not a limiting factor for nitrification. Additionally, therole of pH in our study was not considerable when comparedto other environmental conditions (p 0.332) due to its narrowrange of variation. The pH range observed in this study (6.97–7.47) was within the optimal range for the metabolism andgrowth of autotrophic nitrifiers [6].

Referring to the ANFIS model (Fig. 6a), the trainingand checking errors were comparable, which impliesthat no overfitting occurred. This means that selectionof more than one input can be explored to rebuild theANFIS model. The plot in Fig. 6b showed all two-inputvariable combinations and their influence on qN. It wasfound that HRT and F/M ratio (the left-most input var-iable) formed the optimal combination of two input at-tributes. Additionally, it was observed that the minimaltraining and checking errors reduced significantly fromthat of the best one-input model, indicating that thecombination of HRT and F/M ratio improved the pre-diction performance. A quadratic model was developedto confirm the ANFIS results by estimating qN overindependent variables (HRT and F/M ratio). The poly-nomial equation (Eq. 12), including constant, linear, in-teraction, and squared terms, provided a determinationof coefficient (r2 value) with the experimental data of0.50:

qN ¼ Aþ B HRTð Þ þ C F=Mð Þ þ D HRT� F=Mð Þþ E HRTð Þ2 þ F F=Mð Þ2 ð12Þ

A=−76.2257; B=24.3626; C=149.0787; D=−17.1593;E=−1.8146; F=−41.3636

where qN is expressed inmilligrams of N-NH4+ per gram of

MLSS per day; HRT in hours; and F/M ratio in per day

Effect of Operational Conditions on AOB

The AOB was found to be HRT-dependent with an r value of−0.741 (p 0.002). The high correlation between HRT andAOB was in accordance with the ANFIS results, where HRTexhibited the least training error among other operational con-ditions (Fig. 7a). It was found that AOB exhibited a negativecorrelation with the current range of HRT (4.3–6.3 h). Similarresults were reported by Li et al. [12] where an enhancementin AOB community was observed when HRT declined from 7to 5 h. They also noted that AOB had a positive correlation

when HRT became higher than 10 h [12]. Our results depictedthat the AOB was considerably correlated with temperature (r0.517 and p 0.048), indicating that temperature had a positiveimpact on the AOB community. The positive correlation be-tween temperature and AOB was previously illustrated byPark et al. [46], who found that low temperature could notonly decrease the attached biomass and activity of AOB butcould also produce a change in the composition of the AOBspecies. According to ANFIS results, the two variables, HRTand temperature, were the most relevant parameters with re-spect to AOB (Fig. 7a). Our results suggested that an increasein AOB during the summer season resulted from an increasein temperature in line with a decrease in HRT.

As noticed from the ANFIS model (Fig. 7b), interac-tion of HRT with other environmental factors provided areliable assessment of the plant performance. This mightbe due to the fact that the current study was based onfull-scale observations where several environmental pa-rameters were interacting together in a dynamic manner.The influence of two environmental parameters indicat-ed that the combination of HRT and temperature exhib-ited lower training error than either HRT or temperatureby 64 and 70 %, respectively. These results further con-firm our hypothesis that the AOB increased in summerseason due to the impact of both HRT and temperature.

Comparable to the qN results, the ANFIS model indicatedthat the combination of HRT and F/M ratio could be the mostrelevant input to the AOB (output) (Fig. 7b). Subsequently,HRTand F/M ratio was employed in a polynomial function ofdegree 2 to determine their quadratic regression (Eq. 13). Theestimated coefficient of determination showed an r2 value of0.614.

AOB ¼ ðAþ B HRTð Þ þ C F=Mð Þ þ D HRT � F=Mð ÞþE HRTð Þ2 þ F F

.M

� �2Þ � 1011

ð13Þ

A = 1.0114; B = −0.2456; C = −0.6173; D = 0.0567;E=0.0161; F=0.1696

where, AOB is expressed in copies per liter; HRT in hours;and F/M ratio in per day

Effect of Operational Conditions on NOB (Nitrobacterand Nitrospira)

The presence of NOB is necessary for WWTPs to achievecomplete nitrification. In this investigation, Nitrobacter spp.indicated no significant correlation with the operational con-ditions (p > 0.1). This suggested that the abundance ofNitrobacter could tolerate seasonal and environmental varia-tions. The ANFIS model indicated that the F/M ratio had thehighest impact on Nitrobacter abundance (Fig. 8a). Thestrength of the relationship between the F/M ratio andNitrobacter spp. was further estimated by Pearson correlation

O.O. Awolusi et al.

coefficient which showed an r value of 0.359 (p>0.1). Theprominence of F/M ratio was previously reported, where someorganic compounds in the wastewater positively affected theactivity of NOB [45]. The ANFIS model was also used toidentify relationships between Nitrobacter spp. abundance

and the combination of two operational parameters (Fig. 8b).The model indicated that F/M ratio and temperature form theoptimal combination of two input attributes. The polynomialequation of their quadratic interaction is presented in Eq. 14(r2 value 0.49):

Nitrobacter ¼ Aþ B F=Mð Þ þ C Tð Þ þ D F=M� Tð Þ þ E F=Mð Þ2 þ F Tð Þ2� �

� 1011 ð14Þ

A = −0.7143; B = −3.4550; C = 0.2579; D = −0.0050;E=1.9650; F=−0.0062

whereNitrobacter is expressed in copies per liter; F/M ratioin per day; and T in degree Celsius.

Cross-correlation coefficients indicated that increase inNitrospira spp. was significantly affected by a decline in

HRT (r −0.627 and p 0.012). Similarly, the ANFIS modelshowed that the input HRT exhibited the least training error(Fig. 9a), which was in accordance with the p values.Additionally, HRT showed inverse correlation withNitrobacter spp. (r −0.364 and p>0.1). Similarly, Li et al.[12] reported that a short HRT favored the relative growth of

Fig. 7 Effect of operatingconditions on AOB (the left-mostinput variable is the most relevantwith respect to AOB). a Everyinput variable’s influence onAOB; b all two-input variablecombinations and their influenceon AOB

Artificial Intelligence for the Evaluation of Operational Parameters

NOBs, particularly the fast-growing Nitrobacter spp., in theconventional activated sludge system. As observed from theANFIS results (Fig. 9b), the combination of HRT and F/Mratio exhibited a 4-fold lowering of the training error when

compared to HRT individually. Therefore, the quadratic poly-nomial formula indicating the combinatory effect of HRT andF/M ratio on Nitrospira can be presented by Eq. 15 (r2 value0.716):

Nitrospira ¼ Aþ B HRTð Þ þ C F=Mð Þ þ D HRT� F.M

� �þ E HRTð Þ2 þ F F

.M

� �2� �� 1011 ð15Þ

A = 1.9980; B = −0.7133; C = 0.1599; D = −0.0062;E=0.0636; F=−0.1640

where Nitrospira is expressed in copies per liter; HRT inhours; and F/M ratio in per day

In a similar study, Huang, Gedalanga, and Olson [10] in-vestigated the impact of environmental variables onNitrobacter and Nitrospira. They observed that Nitrobacterpopulations were negatively correlated to temperature (r

Fig. 8 Effect of operatingconditions on Nitrobacter (theleft-most input variable is themost relevant with respect toNitrobacter). a Every inputvariable’s influence onNitrobacter; b all two-inputvariable combinations and theirinfluence on Nitrobacter

O.O. Awolusi et al.

−0.49 and p<0.001), while the Nitrospira abundance showeda strong positive correlation to temperature (r 0.59 andp<0.0001). Additionally, Nitrobacter populations were sig-nificantly and positively correlated to DO (r 0.38 andp<0.01). However, Nitrospira abundance showed a signifi-cantly negative correlation to DO (r −0.46 and p< 0.01).Moreover, HRT showed a significant impact on Nitrobacterspp. (r 0.334 and p<0.05). When comparing our study withthe latter investigation [10], both studies used the cross-correlation coefficients to determine the significant impact ofoperational parameters on Nitrobacter and Nitrospira.However, our study further applied artificial modeling tech-nique in confirming the r results and in determining the opti-mum combination of two input variables.

Conclusions

The AI approach succeeded in describing the effect of operat-ing condition on nitrification process. Results from the ANFISmodel were in accordance with Spearman’s correlation coef-ficients, and it was concluded that:

& The Nitrobacter spp. abundance was 29-fold higher thanthat of AOB throughout the study period, showing a cleardominance of NOB.

& The qN was noticeably increased by 2.2-fold and exhibited26.6±10.7 mg N-NH4

+g−1 MLSS day−1 when the tem-perature elevated from 16.5±2.1 to 22.4±2.7 °C (r 0.726,p 0.002).

Fig. 9 Effect of operatingconditions on Nitrospira (the left-most input variable is the mostrelevant with respect toNitrospira). a Every inputvariable’s influence onNitrospira; b all two-inputvariable combinations and theirinfluence on Nitrospira

Artificial Intelligence for the Evaluation of Operational Parameters

& The qN was also significantly affected by HRT (r −0.651,p 0.009) and ALR (r 0.571, p 0.026).

& HRT and F/M ratio formed the optimal combination oftwo inputs affecting the qN, and their quadratic equationshowed an r2 value of 0.50.

& AOB increased in the summer season, when temperaturewas 1.4-fold higher than during winter (r 0.517, p 0.048),and HRT decreased by 31 % as a result of rainfall (r−0.741, p 0.002).

& No single input has a significant effect on Nitrobacterspp., indicating that the abundance of Nitrobacter couldtolerate seasonal and environmental variations.

& Nitrospira spp. increased by 3.6 times whenHRT declinedduring the summer season (r −0.627, p 0.012).

& A polynomial function of 2nd degree for HRT, F/M ratio,and AOB; F/M ratio, temperature, and Nitrobacter; andHRT, F/M ratio, and Nitrospira showed r2 values of 0.61,0.49, and 0.72, respectively.

Acknowledgments The authors hereby acknowledge the DurbanUniversity of Technology and National Research Foundation for provid-ing financial assistance.

References

1. Kim YM, Cho HU, Lee DS, Park D, Park JM (2011) Influence ofoperational parameters on nitrogen removal efficiency and micro-bial communities in a full-scale activated sludge process. Water Res45:5785–5795. doi:10.1016/j.watres.2011.08.063

2. Angel R, Asaf L, Ronen Z, Nejidat A (2010) Nitrogen transforma-tions and diversity of ammonia-oxidizing bacteria in a desertephemeral stream receiving untreated wastewater. Microb Ecol59:46–58

3. Campos J, Garrido-Fermindez J, Mendez R, Lema J (1999)Nitrification at high ammonia loading rates in an activated sludgeunit. Bioresour Technol 68:141–148

4. Nasr M, Moustafa M, Seif H, El Kobrosy G (2011) Modelling andsimulation of German BIOGEST/EL-AGAMY wastewater treat-ment plants—Egypt using GPS-X simulator. Alexandria Eng J50:351–357

5. Third K, Paxman J, Schmid M, Strous M, Jetten M, Cord-RuwischR (2005) Enrichment of anammox from activated sludge and itsapplication in the CANON process. Microb Ecol 49:236–244

6. Zhang L, Jiang J, Yang J, Hira D, Furukawa K (2012) High ratenitrogen removal by the CANON process at ambient temperature.Water Sci Technol : J Int Assoc Water Pollut Res 65:1826–1833

7. Hu J, Li D, Liu Q, Tao Y, He X,Wang X, Li X, Gao P (2009) Effectof organic carbon on nitrification efficiency and community com-position of nitrifying biofilm. J Environ Sci 21:387–394. doi:10.1016/S1001-0742(08)62281-0

8. Fulweiler RW, Emery HE, Heis EM, Berounsky VM (2011)Assessing the role of pH in determining water column nitrificationrates in a coastal system. Estuar Coasts 34:1095–1102

9. Stark J (1996) Modeling the temperature response of nitrification.Biogeochemistry 35:433–445

10. Huang H, Gedalanga P, Olson BH (2010) Distribution ofnitrobacter and nitrospira communities in an aerobic activatedsludge bioreactor and their contributions to nitrite oxidation.WEFTEC

11. Wu YJ, Whang LM, Chang MY, Fukushima T, Lee YC, Cheng SS,Hsu SF, Chang CH, Shen W, Yang CY, Fu R, Tsai TY (2013)Impact of food to microorganism (F/M) ratio and colloidal chemicaloxygen demand on nitrification performance of a full-scale mem-brane bioreactor treating thin film transistor liquid crystal displaywastewater. Bioresour Technol 141:35–40. doi:10.1016/j.biortech.2013.02.108

12. Li H, Zhang Y, Yang M, Kamagata Y (2013) Effects of hydraulicretention time on nitrification activities and population dynamics ofa conventional activated sludge system. Front Environ Sci EngChina 7:43–48. doi:10.1007/s11783-012-0397-8

13. Klemetti R (2010) Modelling the start-up of biological phosphorusremoval at low temperature. Aato University

14. Nasr M, Mahmoud AED, Fawzy M, Radwan A (2015) Artificialintelligence modeling of cadmium(II) biosorption using rice straw.Appl Water Sci. doi:10.1007/s13201-015-0295x

15. Nasr M, Moustafa M, Seif H, El Kobrosy G (2012) Application ofartificial neural network (ANN) for the prediction of EL-AGAMYwastewater treatment plant performance-EGYPT. Alexandria Eng J51:37–43

16. Nasr M, Moustafa M, Seif H, El-Kobrosy G (2014) Application offuzzy logic control for Benchmark simulation model.1. SustainEnviron Res 24:235–243

17. Nasr M, Ateia M, Hassan K (2015) Artificial intelligence forgreywater treatment using electrocoagulation process. Sep SciTechnol. doi:10.1080/01496395.2015.1062399

18. Daims H,WagnerM (2010) Themicrobiology of nitrogen removal.In: Seviour RJ, Nielsen PH (eds) Microbial ecology of activatedsludge. IWA, London, pp 259–280

19. Fuchs BM, Pernthaler J, Amann R (2007) Single cell identificationby fluorescence in situ hybridization. In: Reddy CA, Beveridge TJ,Breznak JA,Marzluf G, Schmidt TM, Snyder LR (eds)Methods forgeneral and molecular microbiology, 3rd edn. ASM Press,Washington D.C., pp 886–896

20. Nielsen JL (2009) Protocol for fluorescence in situ hybridization(FISH) with rRNA-targeted oligonucleotides. In: Nielsen PH,Daims H, Lemmer H (eds) FISH handbook for biological wastewa-ter treatment. IWA Publishing, London, pp 73–84

21. Bassin JP, Kleerebezem R, Muyzer G, Rosado AS, van LoosdrechtMC, Dezotti M (2012) Effect of different salt adaptation strategieson the microbial diversity, activity, and settling of nitrifying sludgein sequencing batch reactors. Appl Microbiol Biotechnol 93:1281–1294. doi:10.1007/s00253-011-3428-7

22. Wu P, JI X, Song X, Shen Y (2014)Mechanism of efficient nutrientremoval and microbial analysis of a combined anaerobic baffledreactor–membrane bioreactor process. Int J Environ Sci Technol11:1611–1618

23. Purkhold U, Pommerening-Röser A, Juretschko S, Schmid MC,Koops HP, Wagner M (2000) Phylogeny of all recognized speciesof ammonia oxidizers based on comparative 16S rRNA and amoAsequence analysis: implications for molecular diversity surveys.Appl Environ Microbiol 66:5368–5382

24. Lienen T, Kleybocker A, Verstraete W, Wurdemann H (2014)Moderate temperature increase leads to disintegration of floatingsludge and lower abundance of the filamentous bacteriumMicrothrix parvicella in anaerobic digesters. Water Res 65:203–212. doi:10.1016/j.watres.2014.07.019

25. Geets J, de Cooman M, Wittebolle L, Heylen K, Vanparys B, DeVos P, Verstraete W, Boon N (2007) Real-time PCR assay for thesimultaneous quantification of nitrifying and denitrifying bacteria inactivated sludge. Appl Microbiol Biotechnol 75:211–221. doi:10.1007/s00253-006-0805-8

26. Steinberg LM, Regan JM (2009) mcrA-targeted real-time quantita-tive PCR method to examine methanogen communities. ApplEnviron Microbiol 75:4435–4442. doi:10.1128/AEM.02858-08

O.O. Awolusi et al.

27. APHA (1998) Standard Methods for the Examination of Water andWastewater American Public Health Association

28. Tchobanoglous G, Burton FL, Stensel HD (2003) Wastewater en-gineering: treatment and reuse. Metcalf and Eddy Inc. McGraw-Hill, New York

29. Jang JS (1993) ANFIS: adaptive-Network-based Fuzzy InferenceSystems. IEEE Trans Syst Man Cybern 23:665–685

30. Azar A (2011) Adaptive neuro-fuzzy system as a novel approachfor predicting post-dialysis urea rebound. Int J Intell Syst TechnolAppl 10:302–330

31. Fawzy M, Nasr M, Abdel-Gaber A, Fadly S (2015) Biosorption ofCr(VI) from aqueous solution using agricultural wastes, with artifi-cial intelligence approach. Sep Sci Technol. doi:10.1080/01496395.2015.1115068

32. Watson S, Valois F, Waterbury J (1981) The familyNitrobacteraceae. In: Starr MP, Stolp H, Truper HG, Balows A,Schlegel HG (eds) The prokaryotes. Springer, Berlin, pp 1005–1022

33. Tchobanoglous G, Burton F (1991) Wastewater engineering treat-ment, disposal and reuse. McGraw-Hill, New York

34. Mozumder MS, Picioreanu C, van Loosdrecht MC, Volcke EI(2013) Effect of heterotrophic growth on autotrophic nitrogen re-moval in a granular sludge reactor. Environ Technol 35:1027–1037

35. Ruiz G, Jeison D, Chamy R (2003) Nitrification with high nitriteaccumulation for the treatment of wastewater with high ammoniaconcentration. Water Res 37:1371–1377

36. Winkler MK, Bassin JP, Kleerebezem R, Sorokin DY, vanLoosdrecht MC (2012) Unravelling the reasons for disproportionin the ratio of AOB and NOB in aerobic granular sludge. ApplMicrobiol Biotechnol 94:1657–1666. doi:10.1007/s00253-012-4126-9

37. Liu G (2012) Nitrification performance of activated sludge underlow dissolved oxygen conditions. PhD, Missouri University ofScience and Technology

38. Wagner M, Loy A, Nogueira R, Purkhold U, Lee N, Daims H(2002) Microbial community composition and function in waste-water treatment plants. Antonie Van Leeuwenhoek 81:665–680

39. Nogueira R, Melo LF (2006) Competition between Nitrospira spp.and Nitrobacter spp. in nitrite-oxidizing bioreactors. BiotechnolBioeng 95:169–175. doi:10.1002/bit.21004

40. Fukushima T, Whang LM, Chiang TY, Lin YH, Chevalier LR,Chen MC,Wu YJ (2013) Nitrifying bacterial community structuresand their nitrification performance under sufficient and limited in-organic carbon conditions. Appl Microbiol Biotechnol 97:6513–6523. doi:10.1007/s00253-012-4436-y

41. Whang LM, Chien IC, Yuan SL, Wu YJ (2009) Nitrifying commu-nity structures and nitrification performance of full-scale municipaland swine wastewater treatment plants. Chemosphere 75:234–242

42. Tarre S, Green M (2004) High-rate nitrification at low pH insuspended and attached biomass reactors. Appl EnvironMicrobiol 70:6481–6487

43. Xu M, Schnorr J, Keibler B, Simon HM (2012) Comparative anal-ysis of 16S rRNA and amoA genes from archaea selected withorganic and inorganic amendments in enrichment culture. ApplEnviron Microbiol 78:2137–2146. doi:10.1128/AEM.06845-11

44. Nogueira R, Melo L, Purkhold U, Wuertz S, Wagner M (2002)Nitrifying and heterotrophic population dynamics in biofilm reac-tors: effects of hydraulic retention time and the presence of organiccarbon. Water Res 36:469–481

45. Kim D, Lee D, Keller J (2006) Effect of temperature and free am-monia on nitrification and nitrite accumulation in landfill leachateand analysis of its nitrifying bacterial community by FISH.Bioresour Technol 97:459–468

46. Park J, Byun I, Park S, Park D (2008) Nitrifying bacterial commu-nities and its activities in aerobic biofilm reactors under differenttemperature conditions. Korean J Chem Eng 25:1448–1455

Artificial Intelligence for the Evaluation of Operational Parameters