Professional Documents
Culture Documents
A
common cause of soil and water resources contamination is the subse-
quent release of mixtures of heavy metals to the environment, seriously
threatening the well-being of humans and ecological systems. In this
context, understanding of metal partitioning between solid and aqueous phases is
critical for proper design of control and remediation measures.
It is known that the adsorption properties of a metal in the presence of other
metal(s) are often different from those associated with the adsorption of the indi-
vidual metal (Sheindorf et al., 1981). Sorption and competitive behavior of met-
www.soils.org/publications/sssaj 1231
represents adsorption rate. The Redlich–Peterson isotherm, Eq. qmax K EL,iCi
[3], is a combination of Eq. [1] and [2] and is associated with qe,i N
, Extended Langmuir (Yang, 1987) [9]
three model parameters, that is, KR (L g–1), αR (Lβ mmol–β), 1 K EL,jC j
j 1
and 0 < β < 1. It tends to (a) the Langmuir equation in the limit
ni −1
of low concentrations, or to (b) the Freundlich model in the ⎛ N ⎞
limit of high concentration values. As previously mentioned, it qe,i = K F,iC i ⎜ ∑α ijC j ⎟ , SRS model
has been suggested that the Langmuir model (Barrow, 2008) is ⎝ j=1 ⎠ (Sheindorf et al., 1981) [10]
generally not appropriate to interpret metal sorption in soils.
K R ,i ( C i i )
We prefer to consider it in our analysis as (i) it has been used qe,i = N
( ) , Modified RP
β, j
extensively in sorption studies (e.g., Bolster, 2008; Srivastava et 1+ ∑α R , j C j j
al., 2008; Dişli,2010; Ofomaja et al., 2010), and (ii) it is a limit- j =1
(e.g., Srivastava et al., 2008) [11]
ing form of Eq. [3].
When isotherm models are presented in the form of Eq. [1– Here, qe,i is the amount of metal species i adsorbed per gram
3], C and qe are not independent quantities, because qe is evalu- of soil, and is calculated as
ated according to Eq. [4]. The problem is then recast in terms of
the independently measured quantities C0 and C upon consider-
qei C0,i Ci V [12]
ing the isotherm equations coupled with Eq. [4] (Barrow, 2008; where Ci is the concentration of metal species i in the equilib-
Tellinghuisen and Bolster, 2010) rium solution.
The parameters qm,i and KL,i appearing in the Modified
K FC n
C0 = +C , (Freundlich) [5] Langmuir model are derived from the single component
V Langmuir equation. The parameters ηi take into account metal
competition and are calibrated against the binary system ob-
1 qm K LC
C0 = +C , (Langmuir) [6] servations. The Extended Langmuir model has the same func-
V 1+ K LC tional format of its single component counterpart and the
parameters appearing in Eq. [9] are estimated on the basis of
1 K RC
C0 = +C , (RP) [7] competition experiments. The SRS equation was derived by
V 1+ α RC β assuming that single component behavior is described by the
Optimization routines can then be used to find best esti- Freundlich model, Eq. [1]. The parameters KF,i and ni are the
mates of parameters as explained in the following. Freundlich constant and exponent of a single component iso-
therm of metal species i, respectively. For simplicity and clarity
Binary Component Systems of illustration, we consider in this study the parameters αi,j and
Solutions containing initial concentrations of Cu and Zn αj,i to be symmetrical (i.e., αi,j = 1/αj,i), as originally proposed
were considered according to all combinations of the values by Sheindorf et al. (1981). This implies that different ions com-
C0,i = 50, 100, 200 mg L–1 (corresponding to C0,Cu = 0.79, 1.57, pete only for adsorption sites. When other effects, such as elec-
3.15 mmol L–1 for Cu and C0,Zn = 0.76, 1.53, 3.06 mmol L–1 for trical effects following ion penetration to soil grain surfaces,
Zn) for Bet Dagan soil and C0,i = 100, 300, 500 mg L-1 for Yatir can be considered as relevant, one might fit independently the
soil (corresponding to C0,Cu = 1.57, 4.72, 7.87 mmol L–1 for Cu two competition parameters (e.g., Barrow et al., 2005). The pa-
and C0,Zn = 1.53, 4.59, 7.65 mmol L–1 for Zn). The concentration rameters αR,i and KR,i appearing in the Modified RP model are
values adopted for the experiments performed on Yatir soil were derived from the single component RP equation. The param-
dictated by the large clay content of the soil. The average initial eters i take into account metal competition and are calibrated
value of pH of the competitive solutions analyzed for Bet Dagan against the binary system observations.
soil was 5.2, resulting in a final average pH value of 5.1.The average As already noted for the single component isotherms, Ci
initial value of pH of the competitive solutions analyzed for the and qe,i are correlated by virtue of Eq. [12]. This problem is
Yatir soil was 4.8, resulting in a final average pH value of 5.7. handled (Barrow et al., 2005) by solving the following system
In their competitive Ni–Cd sorption experiments, Liao and of equations: sorption equations (in the format of Eq. [8–11],
Selim (2009) added to the system a concentration of the compet- depending on the model selected) and related to the two ion
ing ion of 0.047, 0.235, 0.766 mmol L–1. These concentrations species involved in the system, and equations linking sorption to
correspond to C0 = 5.3, 26.4, 86.1 mg L–1 for Cd and C0 = 2.8, change in concentration (in the format of Eq. [12]). This leads,
13.8, 45.0 mg L–1 for Ni. for N = 2, to the following equations, relating the independently
Experimental data were interpreted according to the follow- measured quantities C0,i and Ci
ing four isotherm models:
qm ,i K L,1 Ci i
qm ,i K L,i Ci i C0,i Ci
qe,i 2 (i = 1,2), Modified
N
1 K L, j C j j
, Modified Langmuir
(Bellot and Condoret, 1993) [8]
1
V
j 1
K L, j C j j
Langmuir [13]
j 1
www.soils.org/publications/sssaj 1233
N 1
KIC = NLL N P ln C ln Q . [26] IC k p M k
exp
2 2
p Mk C* NM 1
[27]
These allow discrimination among different competing IC i p M i
exp
models on the basis of their quality of fit to observations, number
i 1 2
of parameters, and quality of the available data and parameter Here, ΔICk = ICk − ICmin, ICk being any of the considered mod-
estimates. Here, Q is the Cramer–Rao lower-bound approxima- el information criteria Eq. [23] to Eq. [26] and ICmin = min {ICk} its
tion for the covariance matrix of the parameter estimates, AIC is minimum over all considered models; p(Mk) is the prior probability
due to Akaike (1974), AICc to Hurvich and Tsai (1989), BIC to associated with each model. In the following, because no prior infor-
Schwartz (1978), and KIC to Kashyap (1982). All of these iden- mation is available, we set p(Mk) = 1/NM (i.e., an equal prior probabil-
tification criteria support the principle of parsimony, in the sense ity is assigned to each model). Adopting model identification criteria
that when everything else is equal, one should prefer adoption of and posterior model probabilities Eq. [27] can lead to ranking a set of
the model with the smallest number of parameters. candidate models in terms of their associated posterior probabilities
KIC balances parsimony with the expected information and allows discrimination among models in a relative sense.
content by means of |Q| and favors the model that is least prob- The traditional approach has been to calibrate a model against
able (in an average sense) of being incorrect (Ye et al., 2008). On observed state variables by means of a suitable inverse methodolo-
the other hand, Tsai and Li (2008, 2010) and Li and Tsai (2009) gy. In addition to estimating model parameters, the key advantage
suggest that KIC can potentially favor models with large param- of the theoretical framework we adopt here is to allow (i) ranking
eter estimation uncertainty following unreliable estimation of of models according to their individual skill to predict observed
Q, potentially leading to controversial results. An extensive dis- system states, and (ii) assigning posterior probability weights to
cussion of these model selection criteria is presented in Ye et al. each tested model, thus quantifying prediction uncertainty.
(2008, 2010 and references therein).
The posterior probability associated with model Mk (k = RESULTS AND DISCUSSION
1,..., NM) is then calculated as (Ye et al., 2008) Variance Function
Because of the different laboratory conditions, we consider
separately the set of experiments performed in this work and the
experiments performed by Liao and Selim (2009). Figure 1 re-
ports the experimental estimates of measurement error variance
(symbols) together with the estimated variance functions (solid
curves) for Bet Dagan and Yatir soils (Fig. 1a) and Windsor
(mixed, mesic Typic Udipsamments), Olivier (fine-silty, mixed,
active, thermic Aquic Fraglossudalfs), and Webster (fine-loamy,
mixed, superactive, mesic Typic Endoaquolls) soils (Fig. 1b). The
corresponding R2 regression values are reported for complete-
ness. Even though a general trend for i 2 is detectable, the deter-
mination of the variance functions is affected by some ambiguity,
as also reflected by the relatively low R2 values.
exponent of the Freundlich model, n, for Cu (n = 0.26 or 0.23, re- unweighted calibrations in Bet Dagan soil and n = 0.29 or 0.30 for
spectively for weighted and unweighted calibrations in Bet Dagan the Yatir soil). The maximum sorption capacities, qm, calculated
soil and n = 0.18 or 0.27 for Yatir soil) is always smaller than that of according to the Langmuir model, are 2.80 or 3.49 (for weighted
Zn for both soils (n = 0.39 or 0.36, respectively for weighted and and unweighted calibrations, respectively) and 3.68 or 4.31 (for
Fig. 3. Experimental equilibrium concentration data and calibrated models for single component adsorption on Yatir soil: (a) Cu, weighted
calibration; (b) Cu, unweighted calibration; (c) Zn, weighted calibration; (d) Zn, unweighted calibration. Error bars of ±2σi are reported.
www.soils.org/publications/sssaj 1235
Fig. 4. Comparison between parameters values obtained when single component systems isotherm equations are interpreted by weighted and
unweighted calibration. Symbols indicate different interpretive models: (open diamond) Freundlich model, (open circle) Langmuir model, and
(open triangle) RP model. Different colors indicate different metals and soils; (a) Cu: solid and empty symbols indicate Bet Dagan and Yatir soils,
respectively; (b) Zn: solid and empty symbols indicate Bet Dagan and Yatir soils, respectively; (c) Cd: black symbols indicate Windsor soil, empty
symbols represent Olivier soil and gray symbols indicate Webster soil; (d) Ni: black symbols indicate Windsor soil, empty symbols represent
Olivier soil and gray symbols indicate Webster soil.
weighted and unweighted calibrations, respectively) g kg–1 for Cu panion information related to the experiments of Liao and Selim
and Zn ions, respectively, in Bet Dagan soil. Yatir soil is associated (2009) and related to sorption of Cd and Ni on Windsor soil is
with qm = 11.14 or 15.59 and 11.69 or 15.48 g kg–1, respectively, reported in Fig. 4c and 4d (see also below for additional comments).
for Cu and Zn ions and for weighted or unweighted calibration. The complete list of the parameter estimates is provided in Table
The maximum sorption capacity of Yatir soil is about three to four A1 of the Appendix together with the square root of the estimation
times larger than that of the Bet Dagan soil sample, consistently error variance (RSD). Our results show that the largest differences be-
with the different CEC of the two soils (see Table 1). tween parameter estimates can be observed for KL (Langmuir model).
The scatter plots in Fig. 4a and 4b compare parameter es- We further note that these estimates are associated with large relative
timates obtained by unweighted calibrations and those related uncertainty, as evidenced by the relatively large RSD.
to weighted calibrations for Cu and Zn, respectively. The com-
Table 2. Single-component posterior model weights (%) and corresponding model ranks (in parentheses) for Bet Dagan and Yatir soils.
Model identification Cu Zn
Soil
criterion Freundlich [5] Langmuir [6] RP† [7] Freundlich [5] Langmuir [6] RP [7] †
Bet Dagan‡ AICc 94.41(1) 0.05(3) 5.55(2) 94.20(1) 1.55(3) 4.25(2)
KIC 88.51(1) 3.69(3) 7.80(2) 26.29(2) 2.95(3) 70.76(1)
Yatir‡ AICc 90.51(1) 0.00 9.49(2) 85.92(1) 0.00 14.08(2)
KIC 1.13(2) 0.00 98.87(1) 8.39(2) 0.00 91.62(1)
Bet Dagan§ AICc 24.49(2) 0.00(3) 75.51(1) 54.47(1) 41.16(2) 4.37(3)
KIC 1.06(2) 0.00(3) 98.94(1) 11.96(3) 12.96(2) 75.08(1)
Yatir§ AICc 98.94(1) 0.01(3) 1.05(2) 97.44(1) 0.26(3) 2.30(2)
KIC 3.93(2) 0.00(3) 96.06(1) 1.93(2) 0.01(3) 98.06(1)
† Redlich–Peterson model.
‡ Weighted calibration.
§ Unweighted calibration.
Fig. 5. Experimental equilibrium concentration data and calibrated models for single component adsorption of Cd and Ni onto Windsor soil: (a)
weighted calibration; (b) unweighted calibration. The width of error bars of ±2σ are of the same size of the symbols.
www.soils.org/publications/sssaj 1237
Table 3. Single-component posterior model weights (%) and corresponding model ranks (in parentheses) for Windsor, Olivier, and
Webster soils.
Cd Ni
Soil Model identification criterion
Freundlich [5] Langmuir [6] RP† [7] Freundlich [5] Langmuir [6] RP [7] †
Windsor‡ AICc 74.93(1) 0.00 25.07(2) 99.98(1) 0.02(2) 0.00
KIC 4.70(2) 0.00 95.30(1) 98.75(1) 1.25(2) 0.00
Olivier‡ AICc 2.90(2) 0.00 97.10(1) 100.00(1) 0.00 0.00
KIC 0.05(2) 0.00 99.95(1) 100.00(1) 0.00 0.00
Webster‡ AICc 100.00(1) 0.00 0.00 100.00(1) 0.00 0.00
KIC 100.00(1) 0.00 0.00 100.00(1) 0.00 0.00
Windsor§ AICc 98.45(1) 0.71(3) 0.84(2) 0.51(3) 2.24(2) 97.25(1)
KIC 65.58(1) 0.47(3) 33.95(2) 0.22(3) 58.12(1) 41.66(2)
Olivier§ AICc 97.34(1) 2.61(2) 0.05(3) 100.00(1) 0.00 0.00
KIC 94.52(1) 2.53(3) 2.95(2) 99.99(1) 0.01 0.00
Webster§ AICc 82.39(1) 17.40(2) 0.21(3) 1.00(2) 98.34(1) 0.66(3)
KIC 73.12(1) 15.44(2) 11.44(3) 0.72(3) 70.50(1) 28.78(2)
† Redlich–Peterson model.
‡ Weighted calibration.
§ Unweighted calibration.
the Freundlich model is always preferred to describe Ni sorption a non-negligible probability (larger than 15%) to interpret Cd
for all the soils tested. sorption on Webster soil. Note that this is consistent with the
It is interesting to observe that the unweighted calibration observation that in this case the largest measured concentrations,
procedure considers the Langmuir model to be associated with a which are close to saturation, have the largest impact on the cali-
relatively large probability (larger than 50%) to be able to char- bration results.
acterize Ni sorption onto both Windsor and Webster soils and
Fig. 6. Experimental equilibrium concentration data and calibrated models for competitive adsorption of Cu and Zn on Bet Dagan soil: (a)
Cu equilibrium concentration for initial concentration of Zn = 100 mg L–1, weighted calibration; (b) Cu equilibrium concentration for initial
concentration Zn = 100 mg L–1, unweighted calibration; (c) Zn equilibrium concentration for initial concentration of Cu = 100 mg L–1, weighted
calibration; (d) Zn equilibrium concentration for initial concentration of Cu = 100 mg L–1, unweighted calibration. Error bars of ± 2σ are reported.
Binary Systems
er than unity, indicating that the presence of Zn ions has only a
Figures 6a and 6b depict equilibrium experimental concen- marginal effect on the sorption of Cu.
trations of Cu on Bet Dagan soil for a selected initial concentra- Figure 8a compares estimated parameter values obtained
tion of Zn ions and model predictions obtained through weight- through unweighted and weighted calibrations for Bet Dagan and
ed and unweighted calibrations. Figures 6c and 6d report the Yatir soil. A complete list of parameter estimate is reported in Table
corresponding depiction for Zn equilibrium concentrations for A4 (Appendix) together with the associated standard deviations.
a given initial concentrations of Cu. Corresponding experiments Posterior model weights and corresponding model ranks
performed on Yatir soil samples are shown in Fig. 7. Results relat- are presented in Table 4. AICc and KIC clearly identify the
ed to different values of initial ion concentrations display a very modified RP as the best model to interpret Bet Dagan experi-
similar behavior (not shown). ments for both the weighted and unweighted calibrations. This
With reference to Bet Dagan samples, while the two cations is consistent with the preference toward the RP model displayed
analyzed display almost the same affinity to the soil when consid- in most of the single component systems.
ered separately, in a competitive system they are associated with With reference to the Yatir soil experiments, we observed
different dynamics. It is noted that equilibrium concentration of that the sorptive behavior of Cu is only slightly influenced by the
Cu onto Bet Dagan soil is increased by about 10% on average presence of Zn. We calculate an average increase of about 20%
in the competitive systems in comparison with the behavior of in the presence of the different applied Zn initial concentration
a single element system for the same initial concentration ad- when compared to the single metal experiments. On the other
opted (considering an average condition over the different initial hand, Zn equilibrium concentration increases by 60% on aver-
concentrations of Zn applied during the competitive sorption age in the presence of the applied initial Cu concentrations. This
experiments, i.e., 50, 100, 200 mg L–1). This increase in Cu equi- suggests that Zn sorption tends to be inhibited by the presence of
librium concentration causes a corresponding decrease in the Cu, consistently with the estimated value of the SRS coefficient
amount of metal sorbed (see Eq. [4]). The presence of competing αij (subscript i and j indicating Cu and Zn, respectively) that is
Cu ions significantly affects Zn sorption, increasing Zn equilib- significantly smaller than unity. All model information criteria and
rium concentration by about 30% on average over the various posterior model weights (see Table 4) are in agreement favoring
initial concentrations of Cu applied to the system. Consistently the Extended Langmuir as the best performing model within the
with this observation, the estimated value of the SRS coefficient weighted calibrations. With reference to the unweighted calibra-
αij (subscript i and j indicating Cu and Zn, respectively) is small- tions, the Modified Langmuir model is associated with the highest
www.soils.org/publications/sssaj 1239
Table 5. Binary-component posterior model weights (%) and
corresponding model ranks (in parentheses) for Windsor,
Olivier, and Webster soils.
variance estimation error depends on the concentration, and with the Langmuir model for some combinations of soil and ions
(b) an unweighted calibration, where the variance estimation (Zn ions on Bet Dagan soil, Cd and Ni ions on Webster soil, Ni
error is considered to be constant (and generally unknown). ions on Windsor soil). As there are no sound theoretical bases for
In addition to calibrating model parameters, our analysis al- this model (Barrow, 2008), this result indicates that a weighted
lows (a) ranking of models according to their individual skill calibration should be preferred for this kind of analysis.
to predict observed system states, and (b) assigning posterior In the binary component systems, a weighted calibration
probability weights to each tested model, thus quantifying pre- identifies the Extended Langmuir (for all soils tested with the
diction uncertainty. exception of Bet Dagan) or the Modified RP model (for the Bet
Even as these two procedures render similar estimated parame- Dagan soil) as the best performing model with posterior prob-
ters, the results associated with model discrimination criteria depend ability weight close or equal to 100%. Following an unweighted
on whether a weighted or unweighted calibration is performed. calibration, selection of one model excluding the contribution of
In the single component systems analyzed, a weighted cali- all others is in general ambiguous for three out of five soils ana-
bration leads to ranking the Freundlich or RP model as first or lyzed. The Modified Langmuir (62%), Modified RP (22%), and
second best model, while the posterior probability associated with Extended Langmuir (15%) have non-negligible weights for Yatir
the Langmuir model is always close to zero. This result supports soil. The Extended Langmuir model and the SRS model are the
the finding of Barrow (2008) who shows that the Langmuir model two most highly ranked models for Windsor and Olivier soils.
is not justified on theoretical bases to describe sorption of ions on
soils. We also note that in several cases the probability weights as- ACKNOWLEDGMENTS
sociated with Freundlich and RP models cannot be considered We are grateful to H.M. Selim and L. Liao for sharing with us their
valuable dataset. The financial support of the European Commission
as negligible. In these cases, a single interpretive model might
(Contract no. PITN-GA-2008-212298), the Politecnico di Milano
not succeed in providing a complete uncertainty quantification, (Project GEMINO, Progetti di ricerca 5 per mille junior), and the
so that a multi-model analysis should be preferred. Unweighted Sussman Family Center for the Study of the Environmental Sciences
calibrations result in non-negligible probability weights associated is gratefully acknowledged.
www.soils.org/publications/sssaj 1241
APPENDIX
Table A1. Parameter estimates and associated relative standard deviation (RSD) for single-component isotherms.
Bet Dagan (Cu) Yatir (Cu) Windsor (Cd) Olivier (Cd) Webster (Cd)
Model Parameter
Value RSD Value RSD Value RSD Value RSD Value RSD
KF 0.99 0.04 4.22 0.11 5.95 0.13 23.33 0.06 21.54 0.26
Freundlich† [5]
n 0.26 0.05 0.18 0.12 0.69 0.05 0.61 0.03 0.55 0.16
KL 0.28 0.41 0.40 0.89 5.86 0.34 26.39 0.55 33.81 0.21
Langmuir† [6]
qm 2.86 0.09 11.14 0.09 3.95 0.23 8.12 0.18 8.21 0.12
KR 3.89 0.14 18.97 2.10 29.48 0.12 356.38 0.19 248.72 0.09
RP† [7] α 3.28 0.16 4.03 2.39 5.21 0.24 21.47 0.09 19.71 0.72
β 0.79 0.01 0.84 0.07 0.65 0.23 0.59 0.19 0.80 0.37
KF 1.10 0.04 2.61 0.12 5.61 0.02 19.27 0.04 19.42 0.05
Freundlich‡ [5]
n 0.23 0.04 0.27 0.08 0.66 0.03 0.55 0.03 0.51 0.04
KL 0.09 0.30 0.01 0.49 2.31 0.25 127.08 0.29 14.63 0.15
Langmuir‡ [6]
qm 3.49 0.05 15.59 0.09 6.42 0.14 9.81 0.20 10.46 0.06
KR 3.20 0.47 6.82 4.63 43.12 0.59 0.89 0.27 157.25 0.32
RP‡§ [7] α 2.63 0.52 2.50 4.87 6.91 0.63 10.70 0.16 13.44 0.28
β 0.79 0.01 0.74 0.06 0.44 0.22 10.99 0.07 0.94 0.29
Bet Dagan (Zn) Yatir (Zn) Windsor (Ni) Olivier (Ni) Webster (Ni)
Value RSD Value RSD Value RSD Value RSD Value RSD
KF 0.58 0.13 2.28 0.11 2.97 0.08 14.78 0.04 39.24 0.23
Freundlich† [5]
n 0.39 0.09 0.29 0.08 0.58 0.03 0.60 0.01 0.57 0.11
KL 0.07 0.35 0.08 0.53 12.76 0.31 18.85 0.41 25.75 1.66
Langmuir† [6]
qm 3.68 0.12 11.69 0.11 1.66 0.22 6.38 0.24 16.97 1.93
KR 0.83 1.31 5.65 1.36 36.95 0.78 99.18 0.18 578.33 0.11
RP† [7] α 0.94 1.91 2.08 1.61 13.37 0.96 11.11 1.13 48.31 0.95
β 0.70 0.20 0.74 0.07 0.56 0.14 0.87 0.64 0.89 0.34
KF 0.66 0.17 2.08 0.21 2.49 0.06 12.91 0.02 23.02 0.14
Freundlich‡ [5]
n 0.36 0.11 0.30 0.12 0.48 0.11 0.55 0.02 0.42 0.11
KL 0.04 0.22 0.01 0.48 5.97 0.13 7.62 0.23 459.30 0.21
Langmuir‡ [6]
qm 4.31 0.06 15.48 0.10 2.30 0.04 8.82 0.11 49.14 0.19
KR 0.77 1.95 4.32 5.46 17.10 0.24 75.15 0.47 1.00 0.06
RP† [7] α 0.85 2.59 1.91 5.97 7.14 0.22 7.20 0.33 49.96 0.07
β 0.70 0.18 0.71 0.12 0.87 0.14 0.85 0.40 9.29 0.02
† Weighted calibration.
‡ Unweighted calibration.
§ Redlich–Peterson model.
–––––––––––––––––––––––––––––––––––– % –––––––––––––––––––––––––––––––––––––
Freundlich† [5] AIC 58.26 43.91 67.86 0.05 100.00 63.88 33.37 0.23 100.00 100.00
AICC 94.41 90.51 98.45 2.90 100.00 94.20 85.92 0.51 100.00 100.00
BIC 57.61 43.24 65.58 0.05 100.00 63.27 32.77 0.22 100.00 100.00
KIC 88.51 1.13 1.46 0.00 100.00 26.29 8.39 0.03 100.00 100.00
Langmuir† [6] AIC 0.03 0.00 0.49 0.00 0.00 1.05 0.00 44.20 0.00 0.00
AICC 0.05 0.00 0.71 0.00 0.00 1.55 0.00 97.25 0.00 0.00
BIC 0.03 0.00 0.47 0.00 0.00 1.04 0.00 41.66 0.00 0.00
KIC 3.69 0.00 0.73 0.00 0.00 2.95 0.00 30.06 0.00 0.00
RP†‡ [7] AIC 41.71 56.09 31.65 99.95 0.00 35.07 66.63 55.57 0.00 0.00
AICC 5.55 9.49 0.84 97.10 0.00 4.25 14.08 2.24 0.00 0.00
BIC 42.37 56.76 33.95 99.95 0.00 35.69 67.23 58.12 0.00 0.00
KIC 7.80 98.87 97.81 100.00 0.00 70.76 91.62 69.91 0.00 0.00
Freundlich§ [5] AIC 2.59 88.53 5.19 94.80 73.96 36.58 77.49 98.87 99.99 0.74
AICC 24.49 98.94 74.93 97.34 82.39 54.47 97.44 99.98 100.00 1.00
BIC 2.53 88.25 4.70 94.52 73.12 36.22 77.01 98.75 99.99 0.72
KIC 1.06 3.93 0.20 7.07 0.94 11.96 1.93 5.87 79.46 0.01
Langmuir§ [6] AIC 0.00 0.01 0.00 2.54 15.62 27.64 0.21 0.00 0.00 72.57
AICC 0.00 0.01 0.00 2.61 17.40 41.16 0.26 0.00 0.00 98.34
BIC 0.00 0.01 0.00 2.53 15.44 27.37 0.21 0.00 0.00 70.50
KIC 0.00 0.00 0.00 15.80 12.57 12.96 0.01 0.00 2.62 14.42
RP§ [7] AIC 97.41 11.46 94.81 2.66 10.43 35.79 22.31 1.13 0.01 26.70
AICC 75.51 1.05 25.07 0.05 0.21 4.37 2.30 0.02 0.00 0.66
BIC 97.47 11.74 95.30 2.95 11.44 36.41 22.78 1.25 0.01 28.78
KIC 98.94 96.06 99.80 77.14 86.50 75.08 98.06 94.13 17.92 85.57
† Weighted calibration.
‡ Redlich–Peterson model.
§ Unweighted calibration.
Modified Langmuir† ηi 0.30 0.73 0.48 1.53 1.24 0.11 0.78 0.16 0.44 0.12
[13] ηj 0.49 0.77 1.00 0.48 0.62 0.09 0.64 0.17 1.58 0.10
Ki 0.05 0.50 0.32 0.64 3.32 0.13 8.37 0.13 44.04 0.13
Extended Langmuir†
Kj 0.02 0.50 0.02 0.61 5.67 0.13 14.60 0.14 12.21 0.12
[14]
qm 5.36 0.13 12.45 0.07 4.27 0.07 11.75 0.06 13.58 0.05
SRS†‡ [15] αij 0.18 0.27 0.03 0.37 1.20 0.21 1.55 0.19 0.24 0.18
ηi 0.33 0.17 0.19 0.30 0.85 0.08 0.14 0.10 0.35 0.13
Modified RP† [16]
ηj 0.18 0.18 0.68 0.45 0.36 0.06 0.02 0.10 0.58 0.13
Modified Langmuir§ ηi 0.15 2.06 0.50 0.18 1.75 0.09 0.90 0.10 0.75 0.29
[13] ηj 0.18 2.04 2.03 0.18 0.84 0.05 0.67 0.12 0.62 0.23
Ki 0.04 0.57 0.03 0.43 1.08 0.12 1.64 0.14 10.76 0.14
Extended Langmuir§
Kj 0.02 0.54 0.01 0.38 2.96 0.14 3.69 0.15 3.25 0.13
[14]
qm 5.36 0.12 15.75 0.07 5.33 0.06 20.32 0.08 20.50 0.07
SRS§ [15] αij 0.40 0.31 0.12 0.20 1.25 0.13 2.06 0.13 0.27 0.16
ηi 0.37 0.16 0.35 0.13 0.75 0.08 0.85 0.08 0.69 0.27
Modified RP§ [16]
ηj 0.20 0.14 0.94 0.15 0.85 0.04 0.36 0.06 0.60 0.20
† Weighted calibration.
‡ Sheindorf–Rebhun–Sheintuch model.
§ Unweighted calibration.
www.soils.org/publications/sssaj 1243
Table A5. Model information criteria and posterior model weights for binary-component isotherms.
NLL and model information criteria Posterior model weights, %
Model
Model identification Bet Dagan Yatir Windsor Olivier Webster Bet Dagan Yatir Windsor Olivier Webster
criterion
Cu–Zn Cu–Zn Cd–Ni Cd–Ni Cd–Ni Cu–Zn Cu–Zn Cd–Ni Cd–Ni Cd–Ni
AIC 250.34 547.58 1826.18 7233.43 12,530.30 0.00 0.00 0.00 0.00 0.00
Modified
Langmuir† AICC 250.34 547.58 1826.18 7233.43 12,530.30 0.00 0.00 0.00 0.00 0.00
[13]
BIC 252.12 549.36 1830.08 7239.40 12,534.20 0.00 0.00 0.00 0.00 0.00
KIC 258.41 551.25 1836.44 7235.21 12,540.30 0.00 0.00 0.00 0.00 0.00
AIC 119.06 337.58 639.18 9144.43 7,939.25 0.00 99.74 100.00 100.00 100.00
Extended
Langmuir† AICC 120.06 338.58 639.18 9144.43 7,939.25 0.00 99.57 100.00 100.00 100.00
[14]
BIC 121.73 340.25 645.03 9146.42 7,945.22 0.00 99.37 100.00 100.00 100.00
KIC 138.21 349.49 647.22 9147.09 7,937.98 0.00 98.47 100.00 100.00 100.00
AIC 124.85 349.48 1292.18 13140.40 12,238.30 0.00 0.26 0.00 0.00 0.00
SRS†‡ [15] AICC 124.85 349.48 1292.18 13140.40 12,238.30 0.00 0.43 0.00 0.00 0.00
BIC 125.74 350.37 1294.13 13144.40 12,240.20 0.00 0.63 0.00 0.00 0.00
KIC 130.02 357.82 1295.02 13161.80 12,244.70 0.00 1.53 0.00 0.00 0.00
AIC 94.23 430.28 828.18 7233.43 16,010.30 100.00 1.53 0.00 0.00 0.00
Modified RP†
AICC 94.23 430.28 828.18 7233.43 16,010.30 100.00 0.00 0.00 0.00 0.00
[16]
BIC 96.01 432.06 832.08 7239.40 16,014.20 100.00 0.00 0.00 0.00 0.00
KIC 105.90 437.53 841.54 7235.21 16,022.10 100.00 0.00 0.00 0.00 0.00
AIC 134.03 148.53 –286.84 –234.14 –190.50 0.00 56.96 0.00 0.00 0.00
Modified
Langmuir§ AICC 134.03 148.53 –286.84 –234.14 –190.50 0.00 62.58 0.00 0.00 0.00
[13]
BIC 135.81 150.31 –282.94 –230.16 –186.52 0.00 62.05 0.00 0.00 0.00
KIC 141.13 154.51 –276.29 –223.72 –182.69 0.02 92.41 0.00 0.00 0.00
AIC 121.70 150.35 –325.40 –328.97 –343.35 0.98 22.85 24.03 60.51 100.00
Extended
Langmuir§ AICC 122.70 151.35 –325.40 –328.97 –343.35 0.60 15.23 24.03 60.51 100.00
[14]
BIC 124.37 153.03 –319.54 –323.01 –337.38 0.63 15.95 4.30 17.33 99.98
KIC 141.96 172.76 –312.02 –319.51 –338.35 0.02 0.01 0.25 5.18 100.00
AIC 134.02 3966.84 –327.70 –328.12 –321.81 0.00 0.00 75.97 39.49 0.00
SRS§ [15] AICC 134.02 3966.84 –327.70 –328.12 –321.81 0.00 0.00 75.97 39.49 0.00
BIC 134.91 3967.73 –325.75 –326.13 –319.82 0.00 0.00 95.70 82.67 0.02
KIC 137.33 3973.42 –323.98 –325.33 –315.37 0.16 0.00 99.75 94.82 0.00
AIC 112.47 150.60 –289.58 828.18 –198.85 99.02 20.20 0.00 0.00 0.00
Modified RP§
AICC 112.47 150.60 –289.58 828.18 –198.85 99.40 22.19 0.00 0.00 0.00
[16]
BIC 114.25 152.38 –285.68 832.08 –194.88 99.37 22.00 0.00 0.00 0.00
KIC 124.44 159.51 –276.87 841.54 –190.43 99.80 7.58 0.00 0.00 0.00
† Weighted calibration.
‡ Sheindorf–Rebhun–Sheintuch model.
§ Unweighted calibration.
www.soils.org/publications/sssaj 1245