schievano et al, 2009

6

Click here to load reader

Upload: suruagy

Post on 28-Jan-2016

7 views

Category:

Documents


5 download

DESCRIPTION

Article

TRANSCRIPT

Page 1: Schievano Et Al, 2009

Bioresource Technology 100 (2009) 5777–5782

Contents lists available at ScienceDirect

Bioresource Technology

journal homepage: www.elsevier .com/locate /bior tech

Prediction of biogas potentials using quick laboratory analyses: Upgradingprevious models for application to heterogeneous organic matrices

Andrea Schievano a, Barbara Scaglia a, Giuliana D’Imporzano a, Luca Malagutti b,Annalisa Gozzi a, Fabrizio Adani a,*

a Dipartimento di Produzione Vegetale, Università degli Studi di Milano, Via Celoria 2, 20133 Milano, Italyb Dipartimento di Scienze Animali, Università degli Studi di Milano, Via Celoria 2, 20133 Milano, Italy

a r t i c l e i n f o

Article history:Received 20 April 2009Accepted 28 May 2009Available online 26 June 2009

Keywords:Anaerobic digestionBiogasModel predictionRespirometric indexOrganic wastes

0960-8524/$ - see front matter � 2009 Elsevier Ltd. Adoi:10.1016/j.biortech.2009.05.075

* Corresponding author. Tel.: +39 02 503 16545; faE-mail address: [email protected] (F. Adani).

a b s t r a c t

This study presents an upgrading of the mathematical models to predict anaerobic biogasification poten-tial (ABP) through quick laboratory analyses that have been presented in an earlier study. The aim is towiden the applicability of the models to heterogeneous organic substrates and to improve their reliabilitythrough a deeper statistical approach.

Three multiple-step linear regressions were obtained using biomass oxygen demand in 20 h (OD20)plus the volatile solids content (VS) of 23 new samples of heterogeneous organic matrices, of 46 samplespresented in the earlier work and of the data set comprising all the 69 samples. The two variables chosenwere found to be suitable for very heterogeneous materials. To judge the prediction quality, a validationprocedure was performed with 12 new samples using model efficiency indexes. The proposed model hadgood prediction ability for a large variety of organic substrates, and allows the calculation of the ABPvalue within only 2-day’s laboratory work instead of the 60–90 days required to obtain ABP by anaerobictest.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

With the increasing importance of anaerobic digestion (AD) of or-ganic materials for biogas production in the last few years, especiallyin the European Union, there are many efforts aiming to deepenknowledge on AD and to broaden its application (Schittenhelm,2008). Bio-methane production through AD is a versatile technol-ogy; there is a greater variety of organic materials that can undergothe process as compared to other bio-fuels, such as bio-ethanol andbio-diesel (Börjesson and Mattiasson, 2007).

To achieve the correct approach in operating AD processes, it iscrucial to know the potential production in terms of biogas of thefeeding organic mixture. This will optimize the process and man-agement variables.

In a previous work published in this magazine (Schievano et al.,2008), the anaerobic biogasification potential (ABP) directly de-tected by simulating the AD process on a laboratory scale has pro-ven to be very useful. However, it required a relatively long time toperform (about 60 days). Consequently, in the same study, regres-sion models were provided to predict ABP by quick laboratory anal-yses (Schievano et al., 2008). The most convenient model resultedin a regression equation that considered the volatile solids content

ll rights reserved.

x: +39 02 503 16521.

(VS) of biomass, and the biological oxygen demand for degradingreadily available organic matter (OM) in 20 h (OD20), i.e.ABP = 13.782*VS + 26.161*OD20

1/2 + 997.890 (Schievano et al.,2008). This model took into account the dependence of total biogasproduction on both the quantity of OM (i.e. the VS content) and thequality of OM (i.e. the OD20). Angelidaki and Sanders (2004) re-ported the COD/VS ratio as parameter to predict the quality of theOM. In our work (Schievano et al., 2008), the OD20 was used insteadof the COD as predictor of the quality of the OM, because a biolog-ical oxidation was preferred to a strong chemical oxidation which isnot able to discriminate between degradable and non-degradableOM. The strong correlation between OD20 and the ABP demon-strated that both anaerobic microbial consortia and aerobic micro-bial consortia are influenced by the same two factors: thebiodegradability of the substrate (availability to a microbial com-munity) and the oxidation state of the carbon contained in the OM.

However, the proposed model was limited in that it wascalibrated on relatively homogeneous biomass samples from afull-scale anaerobic digester fed with mixtures of swine manure,energy crops and bio-waste. Other models proposed in the litera-ture were also developed on homogeneous groups of biomass,showing the same limit reported for our model (Bjorndal andMoore, 1985; Buffiere et al., 2006; Chandler et al., 1979; Eleazeret al., 1997; Gunaseelan, 2007; Habig, 1985; Han et al., 1975;Møller et al., 2004; Tong et al., 1990). Namely, most of the authors

Page 2: Schievano Et Al, 2009

5778 A. Schievano et al. / Bioresource Technology 100 (2009) 5777–5782

worked with fiber-containing materials, so that the methane pro-duction of the anaerobic trials was often found to be inversely cor-related with the lignin content (Gunaseelan, 2007, Buffiere et al.,2006), because lignin is hardly degradable anaerobically (Angeli-daki and Sanders, 2004). For example, Buffiere et al., 2006 andGunaseelan (2007) proposed significant correlations between themethane yields and the lignin + cellulose content working on fruit,vegetables and lingo-cellulose materials. Also, in our previouspaper (Schievano et al., 2008), all the considered samples con-tained a certain percentage of lignin and a significant correlationwas found between ABP and lignin content (R2 = 0.559, P < 0.001).

Depending on the location and the type of the biogas plants, thecandidate substrates can be any kind of agro-industrial by-product,food industry residue, ligno-cellulosic material, municipal source-separated bio-waste, sewage sludge, animal manure, and dedicatedcrops. This indicates the high heterogeneity of materials used, thevariability of their chemical composition and the different contentsand qualities of the available OM. This suggests the need to verify orto upgrade our previous model developed on homogeneous matri-ces, so that it can be applicable to heterogeneous biomasses.

To do this, the same experimental scheme previously proposed(Schievano et al., 2008) was applied on a group of heterogeneousbiomasses currently used in AD plants or those recently proposedas possible candidates for this purpose. In addition, using an accu-rate statistical approach and a validation procedure would ensuremore correct and reliable models for the faster prediction of ABP.

2. Experimental section

2.1. Feedstock sample collection

Twenty-three samples of different organic waste materials werecollected, representing a number of potential material types with

Table 1Chemical and biological parameters of the organic matrices studied.

Sample Description TS(g kg�1

w.w.)

VS(g kg�1TS)

TKN(g kg�1TS)

Cellulose(g kg�1 TS)

1 Household organic wastefrom municipality 1

551 ± 11 983 ± 2 20 ± 0 32 ± 4

2 Household organic wastefrom municipality 2

481 ± 10 949 ± 13 32 ± 1 35 ± 32

3 Household organic wastefrom municipality 3

452 ± 9 954 ± 2 28 ± 1 82 ± 26

4 Fruits and vegetableswaste

237 ± 5 915 ± 2 28 ± 1 110 ± 45

5 Meat and fish waste 424 ± 8 960 ± 1 88 ± 2 41 ± 156 Butchery waste 190 ± 4 998 ± 2 21 ± 3 0a

7 Almond skins 682 ± 14 940 ± 2 24 ± 1 236 ± 118 Bakery waste 660 ± 13 984 ± 1 21 ± 2 169 ± 39 Yoghurt waste 144 ± 3 924 ± 1 0.2 ± 0.1 0a

10 Sludge from wastewater-treatment plant 1

33 ± 4 689 ± 2 46 ± 1 44 ± 4

11 Sludge from wastewater-treatment plant 2

221 ± 4 642 ± 3 34 ± 1 93 ± 49

12 Cattle manure 18 ± 1 799 ± 5 10 ± 1 20 ± 513 Swine manure 30 ± 1 602 ± 1 164 ± 3 15 ± 114 Rabbit faeces 371 ± 7 861 ± 1 179 ± 3 244 ± 515 Poultry faeces 235 ± 5 680 ± 1 134 ± 2 129 ± 316 Rice flour 890 ± 18 891 ± 2 27 ± 1 60 ± 317 Maize flour 327 ± 7 969 ± 1 46 ± 1 169 ± 318 Sweet maize grain 267 ± 5 954 ± 2 21 ± 2 289 ± 1119 Maize silage 300 ± 6 915 ± 3 16 ± 1 256 ± 1620 Sweet sorghum silage 200 ± 4 905 ± 3 16 ± 1 363 ± 1421 Olive oil 996 ± 20 1000b 0 0a

22 Sunflower oil 989 ± 20 1000b 0 0a

23 Butter 850 ± 17 1000b 0 0a

a The fiber analysis was not possible to be performed because samples were fluid. Ateristics of the raw materials.

b Ash content below 0.01 g kg�1TS.

different chemical compositions to be used to produce biogas byAD. Samples were also collected according to their availability inthe market, with the aim of applying a positive result on the oper-ational management of biogas plants.

The group of samples was chosen with a very heterogeneousprovenience (Table 1), that is: three different sort-separatedhousehold waste (Sample # 1 and 3) from three different munici-palities, fruits plus vegetables wastes (Sample # 4), meat plus fishwaste (Sample # 5), butchery waste (Sample # 6), almond skins(Sample # 7), bakery waste (Sample # 8), yoghurt waste (Sample# 9), two samples of sewage sludge from two different wastewa-ter-treatment plants (Sample # 10 and 11), cattle and swine man-ures (Sample # 12 and 13), rabbit and poultry faeces (Sample # 14and 15), energetic crops such as rice flour, maize flour, sweet maizegrain, maize silage, sweet sorghum (Sample # 16–20, respectively),olive oil and sunflower oil (Sample # 21 and 22) and butter (Sam-ple # 23).

The group of 12 samples used for the validation procedure (Ta-ble 4) was composed by the following matrices: vegetables wastedby a market in Northern Italy (Sample # 24), wheat (Sample # 25),liquid sludge from beer industry (Sample # 26), cake dough wastedby confectionery industry (Sample # 27), wastewater sludge fromolive oil mills (Sample # 28), swine manure (1 week in the collec-tion tank) (Sample # 29), barley straw from beer industry (Sample# 30), the ingestate of a biogas plant formed by a mix of dedicatedcrops, swine manure and industrial by-products (Sample # 31),sludge from a wastewater-treatment plant (Sample # 32), com-posted material from a source-separated organic fraction of muni-cipal solid waste (Sample # 33), the digestate (hydraulic retentiontime of 50 days) of the same biogas plant as Sample # 31 (Sample #34) and molasses from sugar industry (Sample # 35).

Following the same procedures used in our previous work(Schievano et al., 2008), all the samples were dried at 40 �C

Hemicell(g kg�1 TS)

ADL(g kg�1

TS)

CS (g kg�1

TS)ADF(g kg�1

TS)

OD20 (mg O2

g�1TS 20 h�1)ABP (Nlbiogaskg�1TS)

56 ± 78 19 ± 2 893 ± 79 50 ± 3 189 ± 78 781 ± 103

108 ± 38 80 ± 18 777 ± 53 115 ± 26 245 ± 20 782 ± 40

101 ± 63 37 ± 13 781 ± 69 118 ± 23 265 ± 7 777 ± 45

45 ± 18 190 ± 45 655 ± 67 300 ± 5 171 ± 33 667 ± 3

253 ± 20 14 ± 1 693 ± 25 55 ± 15 378 ± 72 980 ± 560a 0a 1000a 0a 139 ± 4 540 ± 0

24 ± 7 195 ± 10 546 ± 17 431 ± 5 98 ± 12 687 ± 144300 ± 2 127 ± 2 404 ± 4 297 ± 2 112 ± 23 731 ± 178

0a 0a 1000a 0a 293 ± 20 783 ± 12050 ± 4 214 ± 3 693 ± 6 257 ± 2 83 ± 3 240 ± 5

15 ± 37 293 ± 36 599 ± 71 386 ± 34 143 ± 81 285 ± 22

0 ± 3 115 ± 4 865 ± 7 135 ± 1 36 ± 7 135 ± 820 ± 0 20 ± 1 946 ± 1 35 ± 0 125 ± 1 387 ± 8

191 ± 4 169 ± 3 395 ± 7 414 ± 4 45 ± 5 351 ± 10164 ± 18 63 ± 3 644 ± 19 192 ± 1 86 ± 11 416 ± 27153 ± 45 80 ± 3 707 ± 46 140 ± 1 106 ± 21 582 ± 11300 ± 2 127 ± 2 404 ± 4 297 ± 2 167 ± 15 690 ± 19184 ± 11 75 ± 1 452 ± 16 364 ± 11 153 ± 23 685 ± 24146 ± 19 60 ± 6 538 ± 25 316 ± 15 184 ± 19 668 ± 21213 ± 11 109 ± 12 315 ± 21 472 ± 7 88 ± 39 594 ± 37

0a 0a 1000a 0a 377 ± 20 1549 ± 60a 0a 1000a 0a 407 ± 144 1523 ± 70a 0a 1000a 0a 243 ± 81 1282 ± 176

bsence of cellulose, hemicellulose and lignin was assumed, because of the charac-

Page 3: Schievano Et Al, 2009

A. Schievano et al. / Bioresource Technology 100 (2009) 5777–5782 5779

overnight, and then again at 80 �C overnight (APHA, 1998) andshredded, if found necessary, in a blender to pass into a 1 mmmesh. Both biological and chemical analyses in this study wereperformed using the same dried and shredded samples.

2.2. Analytical methods

All analytical procedures were applied identically to those re-ported in our previous work (Schievano et al., 2008). In brief, totalsolids (TS) and VS were determined according to the standard pro-cedures (APHA, 1998). Total Kjeldhal Nitrogen (TKN) was deter-mined on fresh material, according to the analytical method forwastewater sludge (IRSA CNR, 1994). Fiber analyses were per-formed for neutral detergent fiber (NDF), acid detergent fiber(ADF), and acid detergent lignin (ADL), according to Van Soestmethod (Van Soest et al., 1991). Cell soluble (CS), ADL (that is ligninplus un-hydrolysable lipid), cellulose (ADF–ADL) and hemicellu-lose (NDF–ADF) were calculated according to Van Soest et al.(1991). All the analyses were performed in duplicate.

2.3. Anaerobic biogasification potential assay

The ABPs of all samples were determined by using the methodof Schievano et al. (2008). In brief, organic matrices were incubatedwith inoculum at a ratio of 1:2 (substrate:inoculum on a TS basis),for 60 days in batch 100-ml serum bottles under mesophilic condi-tions. According to our previous approach (Schievano et al., 2008),the total biogas was reported as parameter to evaluate the organicsubstrates performance under anaerobic condition instead of themethane parameter used by many authors (Angelidaki and Sand-ers, 2004; Gunaseelan, 2007; Møller et al., 2004). Nevertheless, asthe anaerobic tests were performed under standardized conditions(Schievano et al., 2008), methane concentration in the biogas wasconstant at 65 ± 2% v/v. All the tests were performed in duplicate.

2.4. Specific oxygen uptake rate assay

The specific oxygen uptake rates (SOUR) were performed as re-ported in Schievano et al. (2008). In brief the measure of the oxy-gen uptake rate during microbial respiration, in degrading theorganic substrates suspended in a continuously agitated watersolution and at 37 �C. To ensure optimal environment for themicrobial growth, the pH was controlled by a buffer solution anda nutrient solution was added.

2.5. Statistical approach

The ABP prediction models were set up by using simple andstepwise linear-regression analyses. Parameters that were not fol-lowing normal distributions according to the Shaphiro–Wilk test(ISO, 1994) were normalized by using the Box–Cox method (Boxand Cox, 1964; Klemm et al., 2002). Only cellulose, hemicellulose,and ADL needed to be normalized, the others being normallydistributed.

According to Scaglia and Adani (2008), the regression modelsobtained were carried out on new series generated, following thejackknife resampling method (Hinkley, 1983; Tukey, 1958), in or-der to avoid getting results that are affected by the features ofthe original sample series. The jackknife method consists in split-ting the original sample series of N elements (e.g., samples of or-ganic matrices) into groups of k-elements. Then, a number ofcombinations of N–k elements are generated by eliminating k(k = 5) different values from the original sample series each time.According to Scaglia and Adani (2008), a total of 250 combinations,each of N–5 samples (new series) were generated to perform linear(simple or multiple stepwise) regressions. Variables included in the

multiple stepwise regressions had a significance P < 0.01. The coef-ficients of the regressions were calculated as the average of dataobtained from the 250 virtual jackknife combinations.

The calibration statistic, that is, coefficient of determination(R2), and the relative standard error of estimate (RSE) (Petiscoet al., 2005) for the regression models were used to test modelperformance.

A validation process of the models proposed was carried outwith new experimental data not included in the calibration pro-cesses derived from 12 new heterogeneous organic matrices.

The validation was carried out evaluating the agreement be-tween measured and estimated values through the quantificationof the following indices: the relative root mean squared error(RRMSE, range: 0% to +1, optimum 0%), the modelling efficiency(EF, range: �1 to 1, optimum = 1), the coefficient of residual mass(CRM, range:�1 to +1; optimum = 0), the slope, and the intercep-tion of the regression equation between the observed and pre-dicted values (�1 to +1, optimum = 1 and �1 to +1,optimum = 0, respectively).

3. Results and discussion

3.1. Chemical and biological parameters studied

The chemical and biological analyses of the samples showed ahigh heterogeneity characterizing the data set (Table 1). For exam-ple, the VS contents varied between 602 g kg�1TS for animal man-ures, and almost 1000 g kg�1TS for oils and fats. The TKN contentvaried in the range 0–179.6 g kg�1 of TS, with null values for veg-etable oils and fats, and the highest values for animal manures.The OD20 and ABP were found in the range of 36–407 mg O2 gTS�1 20 h�1 and 135–1549 Nl kg�1TS, respectively. For bothparameters, the lowest values were found for animal manuresand the highest for fats and oils. Wide intervals were also foundin the results of fiber analyses (Table 1).

To test the heterogeneity of our data set (Data Set # 1, Table 2),we compared it with the same parameters representing other threedata sets of earlier studies (Gunaseelan, 2007; Schievano et al.,2008) reporting good ABP prediction models. In particular,Schievano et al. (2008) considered a group of ingestates and dige-states, sampled at a full-scale biogas plant (Data Set # 2), whileGunaseelan (2007) studied a series of fruits and vegetable wastes(Data Set # 3), and energy crops such as sorghum and Napiergrass(Data Set # 4). Statistical parameters such as the mean (Mean), thestandard deviation (STD) and coefficient of variation (CV), werecalculated and reported for each of the parameter considered andfor each data set (Table 2). Data Set # 1 showed a CV that wasalways much higher than those of the other studies, with theexception of the ABP, VS, and of the OD20, with respect to DataSet # 2 (Schievano et al., 2008). This result confirms the highheterogeneity of the considered data set.

3.2. ABP prediction models by using simple linear regressions

Assuming ABP as a dependent variable, significant linear regres-sions were found with OD20 (R2 = 0.73; P < 0.001), VS (R2 = 0.43;P < 0.001), and ADL (R2 = 0.46; P < 0.001) (Data Set # 1, Table 2).These results partially contrasted with our earlier findings (Schie-vano et al., 2008) in which ABP was well-correlated to more thanonly three parameters: OD20, VS, ADL, TKN, CS, and ADF (Data Set# 2, Table 2) with high significance levels (P < 0.001). Gunaseelan(2007), studying relatively homogeneous organic matrices (DataSet # 3 and # 4, Table 2), found some interesting linear regressionsof the potentials producible methane with VS, ADF, ADL, and thecarbohydrates plus protein content (Data Set # 3 and # 4, Table

Page 4: Schievano Et Al, 2009

Table 2Comparison between the data set of this work and other three from the literature.

Parameter Mean ± STD CV (%) R2 P

Data Set 1: This work (N = 23)ABPa 701 ± 363 51.8 – –VSb 890 ± 121 13.6 0.43 <0.001TKNb 42 ± 51 122.3 0.18 <0.05OD20

c 180 ± 106 59.1 0.73 <0.001Celluloseb 98 ± 111 112.7 0.17 <0.05Hemicell.b 89 ± 92 103.6 0.07 >0.05ADLb 82 ± 82 99.6 0.46 <0.001ADFb 190 ± 161 84.7 0.23 <0.05CSb 731 ± 219 30.0 0.17 <0.05

Data Set 2: Schievano et al. (2008) (N = 46)ABPa 362 ± 191 52.7 – –VSb 782 ± 84 10.7 0.81 <0.001TKNb 63 ± 2 38.7 0.56 <0.001OD20

c 124 ± 69 55.2 0.70 <0.001Celluloseb 85 ± 34 40.4 n.s.d >0.05Hemicell.b 57 ± 42 72.6 n.s. >0.05ADLb 168 ± 82 48.7 0.56 <0.001ADFb 252 ± 105 41.6 0.47 <0.001CSb 690 ± 111 16.1 0.59 <0.001

Data Set 3: Gunaseelan (2007) (N = 17)ABPe 360 ± 80 22.4 – –VSb 923 ± 41 4.5 0.13 n.r.f

Celluloseb 155 ± 67 43.7 0.3 n.r.ADLb 94 ± 40 42.3 0.49 n.r.ADFb 262 ± 91 34.5 0.6 n.r.Carbohydr. + Protb. 645 ± 108 16.8 0.34 n.r.

Data Set 4: Gunaseelan (2007) (N = 7)ABPe 390 ± 100 25.5 – –VSb 912 ± 86 9.5 0.70 n.r.Celluloseb 293 ± 123 41.9 0.18 n.r.ADLb 74 ± 51 69.5 0.37 n.r.ADFb 441 ± 190 43.0 0.68 n.r.

a Nl kg�1TS.b g kg�1TS.c mg O2 g�1TS 20 h�1.d Not significant.e Nl kg�1VS.f Not reported.

5780 A. Schievano et al. / Bioresource Technology 100 (2009) 5777–5782

2). We concluded that homogeneity versus non-homogeneity (Ta-ble 2) of the data set influenced the dependence of the ABP onother parameters.

The works cited earlier suggested that potential biogas produc-tion depended on either quantitative (VS and TKN) or qualitativeparameters (e.g., carbohydrates plus proteins, ADF, cellulose, hemi-cellulose, ADL, and CS contents), according to other literature(Chandler et al., 1979; Habig, 1985; Hashimoto, 1986; Robbinset al., 1979). Furthermore, the quality of the OM depends on twomain factors. The first one is the availability of the OM to microbialcommunities: for instance, fat and cellulosic materials are moreresistant to hydrolysis than carbohydrates or proteins (Angelidaki

Table 3Linear regressions proposed for predicting anaerobic biogasification potential (ABP) from

Model Equation Variables involved

1a ABP = 2.41OD20 + 0.96VS � 589.69 OD20d

VSd

2b ABP = 26.15OD201/2 + 1.38VS � 997.59 OD20

1/2

VS3c ABP = 1.93OD20 + 1.27VS � 843.50 OD20

VS

a Data from this work.b Schievano et al. (2008).c Data from this work plus data from Schievano et al. (2008).d Volatile solids (g kg�1TS) and oxygen demand in 20 h (mg O2 g�1TS 20 h�1).

and Sanders, 2004). The second factor is the biochemical natureof the biodegradable OM (i.e. the oxidation state of the carbon):for instance, 1 g of carbohydrates degraded does not give the sameamount of biogas that 1 g of lipids (Angelidaki and Sanders, 2004).As the correlation between ABP andOD20 was strong (Table 2),these two factors were found to influence both anaerobic and aer-obic microbial consortia.

In the present study, where more heterogeneous materials wereconsidered (Table 2), quantitative parameters (VS), again describedABP, but to a lesser extent than in earlier studies, while almost allqualitative parameters did not do so. Lignin (ADL) was still a rele-vant parameter, as it represents the non-degradable fraction of thetotal OM. Comparing results from Data Set # 1 with those of DataSet # 2, it can be observed that the only parameter that really re-tained the ability to predict ABP was the measurement of the oxy-gen uptake rate, that is, OD20. The OD20 is a parameter able to giveboth quantitative (O2 uptake depends on the organic matter con-tained in the dry matter) and qualitative information about thebiomass (O2 uptake depends on the kind of organic matter con-tained in the dry matter), but at a higher level than the otherparameters described earlier. To understand this, we need to con-sider that both the oxygen uptake in the respirometric test, and themethane produced in the ABP test, depend on both the aforemen-tioned factors: the OM availability to microbes (biodegradability)and the oxidation state of the carbon forming the substrates. Onthe other hand, chemical parameters such as CS, cellulose, hemicel-lulose, ADL, and ADF, contain only partial information about eithertheir biological availability or the oxidation state of the total car-bon. For example, a substrate rich in fats (Samples # 5, 6, 21, 22,and 23) would show higher ABP and also OD20, because its oxida-tion state is lower than a substrate rich in carbohydrates, proteins,or cellulose (Samples # 1, 2, 3, 17, etc.), but no difference would benoticed in terms of VS or CS content.

Only a complete analysis of the chemical composition (i.e. cel-lulose, hemicellulose, lignin, lipids, proteins, carbohydrates, fattyacids, and alcohols) and of the availability to microbial communi-ties (i.e. the soluble versus insoluble fractions of the OM and micro-structural parameters) could give such information, but thanks to atime-consuming laboratory work. The OD20 alone, instead, wasable to describe ABP independently on the heterogeneous versushomogeneous organic matrices.

The OD20 is a test performed in a short time (20 h). Neverthe-less, it was able to predict biogas producible in a long-time test(about 60 days). Moreover, the ABP is affected by the whole con-tent of degradable OM, while the OD20 depends on the readilyavailable OM, that is, the dissolved organic matter (DOM)(D’Imporzano and Adani, 2007). Earlier studies showed that theDOM content of a biomass was related to the availability of theinsoluble organic matter (D’Imporzano and Adani, 2007; Said-Pullicino and Gigliotti, 2007), so that the measurement of the oxy-gen uptake of the DOM allowed a description of the availability of

quick analyses.

R2 Intercept Slope P RSE

0.81 �589.69 2.41 <0.001 23.50.96

0.88 �997.59 26.15 <0.001 18.71.38

0.85 �843.50 1.93 <0.001 26.51.27

Page 5: Schievano Et Al, 2009

Table 4Potential biogas production (ABP) determined on 12 new samples used for thevalidation procedure.

Samples VSa

(g kg�1TS)OD20

a (mg O2

g�1TS 20 h�1)ABPa(Nlkg�1TS)

ABPb

(Nlkg�1TS)

ABPc

(Nlkg�1TS)

24 Vegetablewaste

950 ± 10 288 ± 8 646 ± 22 761 919

25 Wheat 960 ± 10 214 ± 15 667 ± 22 712 78926 Liquid Beer

Yeast920 ± 0 244 ± 12 472 ± 3 672 796

27 Waste CakeDough

990 ± 1 351 ± 18 992 ± 42 850 1091

28 Sludge fromolive oil mill

960 ± 3 180 ± 16 728 ± 33 670 723

29 Fresh swinemanure

770 ± 8 172 ± 9 354 ± 3 403 466

30 Barley straw 970 ± 1 134 ± 35 391 ± 8 643 64731 Ingestate full-

scale plant930 ± 15 170 ± 22 650 ± 12 625 666

32 Sludge fromwastewaterplant

720 ± 6 86 ± 14 310 ± 32 237 237

33 Compost fromOFMSW d

600 ± 5 10 ± 6 50 ± 9 �88 �62

34 Digestate full-scale plant

710 ± 3 64 ± 9 110 ± 13 190 182

35 Molasses 990 ± 1 169 ± 10 509 ± 7 709 740

a Experimental data.b Calculated by using Model 2.c Calculated by using Model 3.d Organic fraction of municipal solid waste.

0 250 500 750 1000-250

0

250

500

750

1000

ABP measured

AB

P ca

lcul

ated

Fig. 1. Validation of the proposed models: experimental ABP vs calculated ABP andtheir linear trends, using model 2 (j, - - -) and model 3 (., ). (Nl kg�1TS).

A. Schievano et al. / Bioresource Technology 100 (2009) 5777–5782 5781

the insoluble substrate. Similar results were obtained by studyingpotential biogas production for municipal solid waste through thedetection of the OUR in short-term aerobic test (Müller et al.,1998).

3.3. ABP prediction models by using multiple linear regressions andvalidation process

Simple linear regressions presented in the aforementioned sec-tion did not allow a good prediction of the ABP as they consideredonly quantitative or qualitative aspects. Consequently, we movedto consider more parameters together, to predict ABP in an im-proved manner. Following the procedure of our earlier work(Schievano et al., 2008), multiple stepwise linear regressions wereperformed with ABP as the dependent variable, introducing inde-

Table 5Validation indexes calculated for the three proposed models.

Parameter Min Max Best mod 1 mod 2

RRMSEa 0 +1 0 45.99 27.36EFb �1 1.00 1.00 0.23 0.73CRMc �1 +1 0.00 �0.37 �0.09Sloped �1 +1 1.00 0.72 0.84Intercepte �1 +1 0.00 5.30 44.84

a Root mean squared error.b Modeling efficiency.c Coefficient of residual mass.d Slope of the regression equation between the observed and predicted values.e Intercept of the regression equation between the observed and predicted values.f Pala et al. (1996).g Donatelli et al. (1997).h Pannkuk et al. (1998).i Sutherland et al. (1995).j Adani et al. (2004).k Rinaldi (2001).l Confalonieri et al. (2006).

m Confalonieri and Bocchi (2005).

pendent variables (all parameters studied) step-by-step andexcluding the non-significant ones. The VS content was the onlyvariable that together with the OD20 contributed to a good regres-sion (Model 1, Table 3). The R2 coefficient was found to be of 0.81(P < 0.001), and the RSE was of 23.5%. The VS and OD20, lumped to-gether, may be thought as a variable which take into account at thesame time the amount, the biodegradability and the quality (oxi-dation state of the carbon) of the available OM.

As stated in Section 1, in our earlier work (Schievano et al.,2008), a similar model was set up working on a more homoge-neous data set (Data Set # 2, Table 2). During that study, the modelwas not correctly evaluated by a statistical approach. Therefore, inthis study, that model (Model 2, Table 3) was reconsidered andevaluated by a cross-validation with the jackknife method(N = 46, k = 5), choosing average coefficients between the 250 ran-domized series of 41 samples. The obtained R2 and RSE were 0.88and 18.7%, respectively.

Attempting to widen the applicability of the model, a new mod-el was created, considering together the 23 samples of the presentstudy and the 46 samples of the earlier work (Schievano et al.,2008). Applying the jackknife method (N = 69, k = 5), the equationwas chosen as average of the 250 regressions, obtained from therandomized series of 64 samples (Model 3, Table 3). The obtainedR2 and RSE were 0.83 and 26.5%, respectively.

The three models obtained were validated on a group of 12 newsamples of heterogeneous organic substrates (Table 4). Based on

mod 3 Typical values for biological models reported in literature

29.56 9–32f; 4.8–21g; 17h; 18i; 8.9–40j; 41k

0.68 0.38–0.85g; �1.01 to �0.94j; 0.60l; 0.70l

�0.17 0.002–0.110g; �0.16–0.38j; 0.11m

0.82 0.7–1.0f; 0–1.06g; 0.92h; 0.25– 2.24j

20.07 �117 to 1644f; 0.21–2.74g; �606 to 820j

Page 6: Schievano Et Al, 2009

5782 A. Schievano et al. / Bioresource Technology 100 (2009) 5777–5782

validation indexes reported in Table 5, Models 3 and 2 predictedABP experimental data better than Model 1, probably because ofthe larger number of data. The ABP was then calculated by usingModels 2 and 3 (Table 4, Fig. 1). Both models gave similar results,even if Model 2 performed slightly better. The similarity betweenModels 2 and 3 was also an interesting result (Fig. 1). The model(Model 2) obtained from a data set of homogeneous substrates(Data set # 2) did not sensibly change after adding a data set (Dataset # 1) of heterogeneous substrates. This finding demonstrates theversatility of using the VS and the OD20 as parameters for the pre-diction of ABP.

4. Conclusions

Three regression models were proposed to predict ABP of het-erogeneous organic matrices. Two independent variables, that is,VS and OD20 obtainable by much quicker laboratory analyses,were used. Using the most reliable of these three models, thatis, Model 2, it was possible to obtain a good assessment of theABP of heterogeneous organic substrates within 2 days of labora-tory work instead of 60–90 days required to perform the ABPtest.

References

Angelidaki, I., Sanders, W., 2004. Assessment of the anaerobic biodegradability ofmacropollutants. Reviews in Environ. Sci. Bio/Technol. 3, 117–129.

APHA – American Public Health Association, 1998. Standard Methods for theExamination of Water and Wastewater, 20th ed. APHA, Washington, DC.

Bjorndal, K.A., Moore, J.E., 1985. Prediction of fermentability of biomass feedstocksfrom chemical characteristics. In: Smith, W.H. (Ed.), Biomass EnergyDevelopment. Plenum Press, New York, pp. 447–454.

Börjesson, P., Mattiasson, B., 2007. Biogas as a resource-efficient vehicle fuel. TrendsBiotechnol. 26 (1), 7–13.

Box, G.E.P., Cox, D.R., 1964. An analysis of transformation. J. Royal Stat. Soc. 26 (B),211–243.

Buffiere, P., Loisel, D., Bernet, N., Delgenes, J.-P., 2006. Towards new indicators forthe prediction of solid waste anaerobic digestion properties. Water Sci. Technol.53 (8), 233–241.

Chandler, J.A., Jewell, W.J., Gossett, J.M., Van Soest, P.J., Robertson, J.B., 1979.Predicting methane fermentation biodegradability. Biotechnol. Bioeng. Symp.10, 93–107.

Confalonieri, R., Bocchi, S., 2005. Evaluation of CropSyst for simulating the yield offlooded rice in northern Italy. Eur. J. Agron. 23, 315–326.

Confalonieri, R., Gusberti, D., Bocchi, S., Acutis, M., 2006. The CropSyst model tosimulate the N balance of rice for alternative management. Agron. Sustain. Dev.26, 241–249.

D’Imporzano, G., Adani, F., 2007. The contribution of water soluble and waterinsoluble organic fractions to oxygen uptake rate during high rate composting.Biodegradation 18, 103–113.

Donatelli, M., Stockle, C.O., Ceotto, E., Rinaldi, M., 1997. Evaluation of CropSyst forcropping systems at two locations of northern and southern Italy. Eur. J. Agron.6, 35–45.

Eleazer, W.E., Odle, W.S., Wang, Y.S., Barlaz, M.A., 1997. Biodegradability ofmunicipal solid waste components in laboratory-scale landfills. Environ. Sci.Technol. 31, 911–917.

Gunaseelan, V.N., 2007. Regression models of ultimate methane yields of fruits andvegetable solid wastes, sorghum and napiergrass on chemical composition.Bioresour. Technol. 98, 1270–1277.

Habig, C., 1985. Influences of substrate composition on biogas yields ofmethanogenic digesters. Biomass 8, 245–253.

Han, Y.W., Lee, J.S., Anderson, A.W., 1975. Chemical composition and digestibility ofryegrass straw. J. Agric. Food Chem. 23, 928–931.

Hashimoto, A.G., 1986. Pretreatment of wheat straw for fermentation to methane.Biotechnol. Bioeng. 28, 1857–1866.

Hinkley, D.W., 1983. Jakknife methods. Encycl. Stat. Sci. 4, 280–287.IRSA CNR, 1994. Metodi analitici per le acque, Quaderni, N. 100. Istituto Poligrafico e

Zecca dello Stato Roma, Italy.ISO, 1994. Accuracy (Trueness and Precision) of Measurement Methods and Results.

ISO, pp. 5725–5732.Klemm, D.J., Blocksom, K.A., Thoney, W.T., Fulk, F.A., Herlihy, A.T., Kaufmann, P.R.,

Cormier, S.M., 2002. Method development and use of macroinvertebrates asindicators of ecological conditions for streams in the mid-Atlantic highlandsregion. Environ. Monit. Assess. 78, 169–212.

Møller, H.B., Sommer, S.G., Ahring, B.K., 2004. Methane productivity of manure,straw and solid fractions of manure. Biomass Bioenerg. 26, 485–495.

Müller, W., Fricke, K., Vogtmann, H., 1998. Biodegradation of organic matter duringmechanical biological treatment of MSW. Compost Sci. Util. 6 (3), 42–52.

Pala, M., Stockle, C.O., Harris, H.C., 1996. Simulation of durum wheat (Triticumturgidum ssp. Durum) growth under different water and nitrogen regimes in amediterranean environment using. CropSyst. Agric. Syst. 51, 147–163.

Pannkuk, C.D., Stockle, C.O., Papendik, R.I., 1998. Evaluating CropSyst simulations ofwheat management in a wheat-follow region of the US Pacific Northwest. Agric.Syst. 57, 121–134.

Petisco, C., Garcia-Criado, B., Vazquez de Aldana, B.R., Zabalgogeazcoa, I., Mediavilla,S., Garcia-Ciudad, A., 2005. Use of near-infrared reflectance spectroscopy inpredicting nitrogen, phosphorus and calcium contents in heterogeneous woodyplant species. Anal. Bioanal. Chem. 382, 458–465.

Rinaldi, M., 2001. Durum wheat simulation in southern Italy using CERES-WHEATmodel. I. Calibration and validation. In: M. Bindi et al. (Eds.), 2nd InternationalSymposium on Modelling Cropping Systems, Florence, Italy, July 16–18. CNR,Inst. for Biometeorol., Florence, pp. 81–82.

Robbins, J.E., Armold, M.T., Lacher, S.L., 1979. Methane production from cattle wasteand delignified straw. Appl. Environ. Microbiol. 38, 175–177.

Said-Pullicino, D., Gigliotti, G., 2007. Oxidative biodegradation of dissolved organicmatter during composting. Chemosphere 68, 1030–1040.

Scaglia, B., Adani, F., 2008. An index for quantifying the aerobic reactivity ofmunicipal solid wastes and derived waste products. Sci. Total Environ. 394,183–191.

Schievano, A., Pognani, M., D’Imporzano, G., Adani, F., 2008. Predicting anaerobicbiogasification potential of ingestates and digestates of a full-scale biogas plantusing chemical and biological parameters. Bioresour. Technol. 99, 8112–8117.

Schittenhelm, S., 2008. Chemical composition and methane yield of maize hybridswith contrasting maturity. Eur. J. Agron. 29, 72–79.

Sutherland, J.P., Bayliss, A.J., Braxton, D.S., 1995. Predictive modelling of growth ofEscherichia coli O157:H7: the effects of temperature, pH and sodium chloride.Int. J. Food Microbiol. 25, 29–49.

Tong, X., Smith, L.H., Mc Carty, P.L., 1990. Methane fermentation of selectedlignocellulosic materials. Biomass 21, 239–255.

Tukey, J.W., 1958. Bias and confidence in not quite large samples. Ann. Stat. 29,164–175.

Van Soest, P.J., Robertson, J.B., Lewis, B.A., 1991. Methods for dietary fibers, neutraldetergent fiber and non-starch polysaccharides in relation to animal nutrition. J.Dairy Sci. 74, 3583–3597.