
This study intends to investigate the performance of boosted regression tree (BRT) and frequency ratio (FR) models in groundwater potential mapping. For this purpose, location of the springs was determined in the western parts of the Mashhad Plain using national reports and field surveys. In addition, thirteen groundwater conditioning factors were prepared and mapped for the modelling process. Those factor maps are: slope degree, slope aspect, altitude, plan curvature, profile curvature, slope length, topographic wetness index, distance from faults, distance from rivers, river density, fault density, land use, and lithology. Then, frequency ratio and boosted regression tree models were applied and groundwater potential maps (GPMs) were produced. In the last step, validation of the models was carried out implementing receiver operating characteristics (ROC) curve. According to the results, BRT had area under curve of ROC (AUC-ROC) of 87.2%, while it was seen that FR had AUC-ROC of 83.2% that implies acceptable operation of the models. According to the results of this study, topographic wetness index was the most important factor, followed by altitude, and distance from rivers. On the other hand, aspect, and plan curvature were seen to be the least important factors. The methodology implemented in this study could be used for other basins with similar conditions to cope with water resources problem.
Citation: Seyed Mohsen Mousavi, Ali Golkarian, Seyed Amir Naghibi, Bahareh Kalantar, Biswajeet Pradhan. GIS-based Groundwater Spring Potential Mapping Using Data Mining Boosted Regression Tree and Probabilistic Frequency Ratio Models in Iran[J]. AIMS Geosciences, 2017, 3(1): 91-115. doi: 10.3934/geosci.2017.1.91
[1] | Dao Viet Thang, Nguyen Manh Hung, Nguyen Cao Khang, Le Thi Mai Oanh . Structural and multiferroic properties of (Sm, Mn) co-doped BiFeO3 materials. AIMS Materials Science, 2020, 7(2): 160-169. doi: 10.3934/matersci.2020.2.160 |
[2] | Xuan Luc Le, Nguyen Dang Phu, Nguyen Xuan Duong . Enhancement of ferroelectricity in perovskite BaTiO3 epitaxial thin films by sulfurization. AIMS Materials Science, 2024, 11(4): 802-814. doi: 10.3934/matersci.2024039 |
[3] | Z. Aboub, B. Daoudi, A. Boukraa . Theoretical study of Ni doping SrTiO3 using a density functional theory. AIMS Materials Science, 2020, 7(6): 902-910. doi: 10.3934/matersci.2020.6.902 |
[4] | Yu Shiratsuchi, Yoshinori Kotani, Saori Yoshida, Yasunori Yoshikawa, Kentaro Toyoki, Atsushi Kobane, Ryoichi Nakatani, Tetsuya Nakamura . Direct observations of ferromagnetic and antiferromagnetic domains in Pt/Co/Cr2O3/Pt perpendicular exchange biased film. AIMS Materials Science, 2015, 2(4): 484-496. doi: 10.3934/matersci.2015.4.484 |
[5] | Wolfgang Zúñiga-Mera, Sonia Gaona Jurado, Alejandra Isabel Guerrero Duymovic, Claudia Fernanda Villaquirán Raigoza, José Eduardo García . Effect of the incorporation of BiFeO3 on the structural, electrical and magnetic properties of the lead-free Bi0.5Na0.5TiO3. AIMS Materials Science, 2021, 8(5): 792-808. doi: 10.3934/matersci.2021048 |
[6] | Murat Aycibin, Naciye ECE . First-principles calculation of the electronic and optical properties of BiRhO3 compound. AIMS Materials Science, 2017, 4(4): 894-904. doi: 10.3934/matersci.2017.4.894 |
[7] | Evangelos Karagiannis, Dimitra Papadaki, Margarita N. Assimakopoulos . Circular self-cleaning building materials and fabrics using dual doped TiO2 nanomaterials. AIMS Materials Science, 2022, 9(4): 534-553. doi: 10.3934/matersci.2022032 |
[8] | Asima Adak (Maity), Soumya Mukherjee, Mahua Ghosh Chaudhuri, Siddhartha Mukherjee . Phase evolution, microscopic analysis, optical and dielectric property evaluation of Co-doped BaSnO3 by mechanical mix assisted solid state sintering method. AIMS Materials Science, 2016, 3(3): 1281-1293. doi: 10.3934/matersci.2016.3.1281 |
[9] | Yana Fajar Prakasa, Sumari Sumari, Aman Santoso, Muhammad Roy Asrori, Ririn Cahyanti . The performance of radar absorption of MnxFe3–xO4/rGO nanocomposites prepared from iron sand beach and coconut shell waste. AIMS Materials Science, 2023, 10(2): 227-248. doi: 10.3934/matersci.2023013 |
[10] | Leydi J. Cardenas F., Josep Ma. Chimenos, Luis C. Moreno A., Elaine C. Paris, Miryam R. Joya . Enhancing Co3O4 nanoparticles: Investigating the impact of nickel doping and high-temperature annealing on NiCo2O4/CoO heterostructures. AIMS Materials Science, 2023, 10(6): 1090-1104. doi: 10.3934/matersci.2023058 |
This study intends to investigate the performance of boosted regression tree (BRT) and frequency ratio (FR) models in groundwater potential mapping. For this purpose, location of the springs was determined in the western parts of the Mashhad Plain using national reports and field surveys. In addition, thirteen groundwater conditioning factors were prepared and mapped for the modelling process. Those factor maps are: slope degree, slope aspect, altitude, plan curvature, profile curvature, slope length, topographic wetness index, distance from faults, distance from rivers, river density, fault density, land use, and lithology. Then, frequency ratio and boosted regression tree models were applied and groundwater potential maps (GPMs) were produced. In the last step, validation of the models was carried out implementing receiver operating characteristics (ROC) curve. According to the results, BRT had area under curve of ROC (AUC-ROC) of 87.2%, while it was seen that FR had AUC-ROC of 83.2% that implies acceptable operation of the models. According to the results of this study, topographic wetness index was the most important factor, followed by altitude, and distance from rivers. On the other hand, aspect, and plan curvature were seen to be the least important factors. The methodology implemented in this study could be used for other basins with similar conditions to cope with water resources problem.
Growth models have traditionally been used for representing the time course of microbial concentration [1,2,3]. Also, there have been applications for representing the time course of other variables, namely: Colony diameter [4,5,6,7]; tumor size [8]; weight of plant seedlings [9]; dry weight of wheat grains [10]; cumulative epicotyl emergence (germination) of oaks [11]; body weight of Boer goats [12], weight of pigs [13], size of Desinognathus [14], pressure variation [15], among others.
The quality of growth models fitting may be assessed in terms of its capability for representing the slope at the inflection point (k) and the duration of the lag phase (tlag), rather than the squared error (SSE) [16]. The lag phase duration indicates the time that the exponential phase begins. The inflection point characterizes the shape of the curve, and it allows to determine the maximum slope of the growth curve and the lag phase duration via the classical approach. In turn, the maximum growth rate is a measure of the growth speed during the exponential phase. Particularly, in the study of growth of undesired microorganisms on products, the lag phase duration indicates the initial time of the exponential phase of contaminants, and the maximum growth rate indicates its speed of growth [16,17].
One way to have growth models capable of generating the lag phase is through a differential equation, including a multiplicative term that suppresses the vector field for low values of the population size, i.e. during the lapse after the initial time. In the model proposed by Baranyi, Roberts and McClure [1], a time dependent adaptation function called adjustment function (αt) is incorporated into a logistic-type model, yielding a non-autonomous model and achieving satisfactory representation of the lag phase. However, autonomous differential equations also have a significant capability for representing the lag phase and the inflection point, some examples are: The Verhulst's logistic equation [18], the generalized Gompertz function [19], the Putter-Bertalanffy model [20], the generalized logistic model [19], the autonomous neoclassical model of Vadasz, Vadasz [21], and the model of Fujikawa, Kai, Morozumi [16]. Also, they allow performing equilibrium stability analysis [20]. Additional discussion is presented in section 2.
It is worth noticing that Tsoularis and Wallace [19] and Ohnishi, Yamakawa and Akamine [20] determined the analytical expression for the population at the inflection point, and assessed the effect of exponential parameter values on the existence or absence of the inflection point and lag phase, based on the curve of the growth rate versus population size. Tsoularis and Wallace [19] noticed that there is no logarithmic inflection point for certain parameter values. Also, they determined the limit values of the population at the inflection point for parameter values approaching zero or infinite. Ohnishi, Yamakawa and Akamine [20] observed that negative parameter values also lead to inflection point and lag phase, thus representing real life behavior. In contrast to these studies, in this work we consider the generalized Fujikawa's model which included the function fc=(1−Nmin/N)c that characterizes the original Fujikawa's model, the power law function Nα and a saturation term. We assess the effect of exponential parameters on the growth curve, considering separately the population N and the logarithmic population y=ln(N/Nto), with negative and positive parameter values. Indeed: ⅰ) we determine the ranges of parameter values that lead to existence or absence of inflection point, and those for the existence or absence of lag phase; ⅱ) we assess the effect of different combinations of positive and negative parameter ranges on the existence or absence of inflection point and lag phase; ⅲ) we determine the parameter ranges that lead to monotonic decreasing behavior of the slope of the growth rate curve; ⅳ) we propose an approach for the analytical determination of the lag phase duration. Numerical simulations illustrate the approach. In summary, the main contribution of this work is the determination of the capability of a generalized Fujikawa's model to capture all the features of real life monotonic growth behavior, including lag phase and inflection point.
The organization of the study is as follows. Section 2 presents some Preliminaries on the study of growth models. Section 3 presents the proposed model and the properties of the adaptation function. Section 4 presents the features of the regular growth model: Convergence and boundedness properties; determination of the regular inflection point and conditions for its existence; effect of model parameters on the existence of the inflection point and the lag phase; and properties of the regular inflection point, including population size of the inflection point under limiting values of the model parameters, and parameter values that lead to inflection point located at the mean value of the curve. Section 5 presents the features of the logarithmic growth model: Convergence and boundedness properties; determination of the logarithmic inflection point and conditions for its existence; effect of model parameters on the existence of the inflection point and the lag phase; and properties of the logarithmic inflection point, including population size of the logarithmic inflection point under limiting values of the model parameters, and parameter values that lead to inflection point located at the mean value of the curve. In section 6, an approach is proposed for determining the lag phase duration, for each the regular and the logarithmic growth curves. In section 7 the model is applied to experimental data, what includes the arrangement of the model and its properties in terms of the logarithmic measure, and a numerical fitting example. Finally, in section 8 the conclusions are drawn.
In microbial culture studies, a typical monotonic growth curve comprises the following phases: Lag phase, exponential growth phase, and stationary phase, although the lag phase may not appear. Thus, monotonic growth does not comprise decay phase or oscillations [17,21,22]. The lag phase involves the physiological adaptation of microbial cells to the new environment, resulting in low or negligible growth rate. The exponential growth phase involves exponential reproduction of microorganisms. The stationary phase comprises constant population of microorganisms [16,22,23,24]. The maximum growth rate occurs in the exponential growth phase, at the so-called inflection point [17,24,25].
In growth model studies for microbial count data (cfu/mL), the logarithmic measure defined as y=ln(N/Nto) or y=ln(N), is used as dependent variable for curve fitting and determination of the inflection point and lag phase duration [2,26,27].
The inflection point is characterized by the fact that the derivative of the dependent variable is maximum (the growth rate is maximum) [19,21]. The existence of the inflection point requires a convex shape in the growth rate versus population curve [17,25]. In turn, the existence of the lag phase requires the existence of the inflection point, and it is possible that the inflection point exists while the lag phase does not. In autonomous growth models the lag phase is generated if the initial population size (Nto) is sufficiently close to the unstable equilibrium point, thus yielding small value of the initial growth rate (dN/dt|t=to). In this case, the escaping of the population from the neighborhood of the unstable equilibrium point takes more time [17,21].
The lag phase duration can be quantified via the classical approach, or based on the third derivative (maximum acceleration of the growth rate) or the maximum curvature [18,24]. The classical approach is based on the intersection of the tangent line at the inflection point with the horizontal line that meets the initial point [17,24].
The model of Vadasz and Vadasz [17,21] comprises a quadratic term, which defines a stable and an unstable equilibrium point. Certain values of the model parameters, corresponding to a specific region, render the model capable of representing the inflection point and the lag phase. For this parameter region, the value of the stable equilibrium point is higher than the unstable one, both being positive. The stable equilibrium point is the carrying capacity. The state variable behaves monotonically, escaping from the unstable equilibrium point towards the stable one. In addition, the model showed a high capacity to represent the lag phase and the inflection point for experimental data.
The generalized logistic model proposed by Tsoularis and Wallace [19] is:
dNdt=rNα[1−(NNmax)β]γ | (2.1) |
Where r, α, β, γ, Nmax are positive constants, N is the population size, and Nmax is the limit value, usually called carrying capacity. This model combines the function Nα, α>0, with the logistic type model.
dNdt=r[1−(NNmax)β]γ | (2.2) |
Therein, model simulations with α=1.5 and α=3 indicated their capability for representing the lag phase. In addition, several growth models that are capable of representing the lag phase can be expressed in this form, namely: The Korf’s form, the hyperbolic form for regenerative growth, the Smith’s equation and the Richard model [19,25]. Moreover, the Smith’s equation has achieved satisfactory representation of the lag phase for experimental data [9].
The Putter-Bertalanffy model analyzed by [20], can be expressed as
dNdt=rN1−ar[1−(NNmax)γ] | (2.3) |
where r, Nmax and γ are positive constants, and N is the population size. It is a Savageau's model, which is in turn a particular case of the generalized logistic function analyzed by [19]. The possible ranges of (1−ar), either positive, negative or zero, lead to different possibilities of the existence of the inflection point and the lag phase.
The model of Fujikawa, Kai and Morozumi is [16,28]:
dNdt=rN(1−NNmax)(1−NminN)c | (2.4) |
where N is the population size, r, c, Nmin, Nmax are positive constants, Nmax>Nmin and Nto∈(Nmin,Nmax). This model was constructed by adding the function fc=(1−Nmin/N)c to the logistic model
dNdt=rN(1−NNmax) |
The function fc strongly suppresses the growth rate during the lag phase, whereas the term (1−N/Nmax) strongly suppresses the growth rate during the stationary phase. The fc function is not mechanistically inspired. Fujikawa, Kai and Morozumi fitted the new model and the traditional model of Baranyi and Roberts [29] for data of E. coli culture. The tlag of the new model was similar to that of the Baranyi’s model, although slightly less ascertained. In contrast, the slope (k) was more ascertained than that of Baranyi’s model. In the fitting, it was found that a small (Nto−Nmin) value is necessary for the estimated c being less dependent on initial population. To this end, the value Nmin=(1−1×10−6)Nto was used, yielding appropriate representation of the lag phase
We propose the following modification of the Fujikawa’s model:
dNdt=rNα(1−NNmax)γ(1−NminN)c | (3.1) |
where r, α, γ, c, Nmin, Nmax are constants, and r>0, Nmax>Nmin>0; N is the population size, and Nmax is the carrying capacity. In turn, the model (3.1) combines the term fc=(1−Nmin/N)c with the hyperlogistic function:
dNdt=rNα(1−NNmax)γ | (3.2) |
The difference of model (3.1) with the Fujikawa’s model (2.4) is the exponents α and γ. Indeed, Fujikawa’s model can be regarded as a simplified variant corresponding to α=1, γ=1. The model coefficients are assumed constant in order to facilitate the study of model properties. Despite this assumption, the model parameters can be defined as function of growth variables in further studies.
Remark 3.1. The term fc=(1−Nmin/N)c with c>0 slows the growth rate of the population size N during the lag phase, provided small positive (Nto–Nmin). In turn, this is related to the fact that dN/dt|N=Nto is proportional to (Nto–Nmin)c. In contrast, a c<0 value and a positive small (Nto–Nmin) value imply: i) dN/dt|N=Nto values are high, so that the initial time period of the N versus t curve features large slope; ii) (1/N)(dN/dt)|N=Nto values are large, so that the initial time period of the y versus t curve features large slope.
Remark 3.2. In the generalized logistic form of Tsoularis and Wallace [19], Eq (2.1), and in the hyperlogistic form (3.2), the suppression of the growth rate during the lag phase is due to the term Nα, with α>0, whereas in the proposed model (3.1), it is due to the term fc with c>0.
Remark 3.3. The c=0 value implies absence of the fc term in model (3.1), so that it reduces to the hyperlogistic model (3.2), which was studied by Tsoularis & Wallace [19].
Remark 3.4. The shape of the growth curve and the existence of the lag phase are related to the equilibrium points of the system. The term Nα, α>0 generates the equilibrium point Neq=0, whereas the term (1−N/Nmax)γ, γ>0 yields Neq=Nmax; and the term (1−Nmin/N)c, c>0 yields Neq=Nmin. For γ>0, c>0 and Nto>Nmin, the N versus t curve escapes from the equilibrium Neq=Nmin and approaches equilibrium Neq=Nmax. The lag phase occurring at N≈Nmin is related to a trajectory that begins close to Nmin, so that it takes more time near Nmin. Therefore, the existence of the lag phase occurring at N≈Nmin requires: i) the existence of the equilibrium point Neq=Nmin, what in turn implies c>0; ii) a Nto value such that Nto−Nmin is small positive.
At what follows, we state the definitions of the growth rate curve related to the mean between lower and higher populations. The mean between Nmin and Nmax is:
Nmn=Nmin+Nmax2 | (3.3) |
The logarithmic population y, and its minimum and maximum values are defined as:
y=ln(NNto) | (3.4) |
ymin=ln(NminNto) | (3.5) |
ymax=ln(NmaxNto) | (3.6) |
As a consequence of the above definitions, y|t=to=0. The mean between ymin and ymax is:
ym=ymin+ymax2 | (3.7) |
Substituting the definition of ymin and ymax, that is, Eqs (3.5) and (3.6), into ym, gives:
ym=ln(√NminNmaxNto) | (3.8) |
To obtain the value of N corresponding to y=ym, we use the definition y=ln(N/Nto), what gives:
Nmy=√NminNmax | (3.9) |
The adaptation function fc=(1−Nmin/N)c, c>0, has the following properties:
i)fc|N=Nmin=0;fc|N=Nmax=(1−NminNmax)c<1;fc∈(0,1)forN∈(Nmin,Nmax] | (3.10) |
ii)dfcdN=cNminN−2(1−NminN)c−1,hencedfc/dN>0forN∈(Nmin,Nmax],iii)dfcdc=ln(1−Nmin/N)eln(1−Nmin/N)c,so thatdfc/dc<0forN∈(Nmin,Nmax] | (3.11) |
iv)ddc(fc|N=Nmax)<0,so thatascincreases,fc|N=Nmaxdecreases | (3.12) |
Remark 3.5. Property ii implies that fc is monotonically increasing with respect to N, for N∈(Nmin,Nmax]
Remark 3.6. Properties ii, iii and iv imply that fc is lower for high c values and for small positive values of (N−Nmin). As a consequence, the dN/dt value is suppressed for small positive values of (Nto−Nmin).
Property iv follows from:
ddc(fc|N=Nmax)=ln(1−NminNmax)eln(1−Nmin/Nmax)c | (3.13) |
Moreover, fc exhibits an inflection point with the following features:
i)Ifc>1,thenfchas an inflection point atN⋆=(c+1)Nmin/2,so thatN⋆>Nmin | (3.14) |
ii)fc|N=N⋆=(c−1c+1)c,c>1 | (3.15) |
iii)dfcdN|N=N⋆=4c(c+1)(c−1)Nmin(c−1c+1)c,c>1 | (3.16) |
Remark 3.7 The inflection point of fc leads to a steep change of the fc versus N curve; and consequently a steep change of the dN/dt versus t curve for c>1.
Figure 1 shows the fc versus N curve for Nmin=(1−1×10−6)Nto, Nto=1 and several c values.
Proposition 4.1. Boundedness and convergence. Consider the growth model (3.1) with γ>0, real values of α and c, and Nmax>Nmin>0. If Nto∈(Nmin,Nmax), the population size N exhibits the following properties: i) N increases monotonically from N=Nto to Nmax, so that limt→∞N=Nmax; ii) N∈[Nto,Nmax)∀t≥to.
Proof. The above proposition follows from the values of the growth rate dN/dt: ⅰ) dN/dt=0 for N=Nmax; ⅱ) dN/dt>0 for N∈(Nmin,Nmax).
In this study, the regular inflection point (t#, N#) is defined as the point of the N versus t curve where dN/dt is maximum, being dN/dt|N=N# the maximum growth rate.
Proposition 4.2. Consider the model (3.1) with γ>0; real values of α and c; Nmax>Nmin>0; and Nto∈(Nmin,Nmax). With these conditions the regular inflection point population, N#, is provided by:
αN#−γNmax−N#+cNminN#(N#−Nmin)=0 | (4.1) |
Proof. As the regular inflection point corresponds to a maximum of the dN/dt versus t curve, then
d2N/dt2|N=N#=0,dN/dt|N=N#>0 | (4.2) |
The condition Nto∈(Nmin,Nmax) implies that N∈(Nmin,Nmax) ∀t≥to, according to Proposition 4.1. The condition dN/dt|N=N#>0 implies that N#≠Nmax. The above two results imply N#∈(Nmin,Nmax). This result and Eq (4.2) give
ddN(dNdt)|N=N#=0,N#∈(Nmin,Nmax) | (4.3) |
finally, applying condition (4.3) for the growth model (3.1), yields Eq (4.1).
Remark 4.1. The condition (4.3) implies that N# corresponds to a maximum of the dN/dt versus N curve.
Proposition 4.3. Consider growth model (3.1) with real values of α, c and γ, Nmax>Nmin>0 and Nto∈(Nmin,Nmax). If N#<Nto, then N>N#∀t≥to; that is, the N versus t curve exhibits no inflection point. If N#=Nto, then N=N# for t=to; that is, the regular inflection point occurs at t=to, N=Nto.
Proof. Since the N versus t curve is monotonic and N≥Nto ∀t≥to, then N>N#∀t≥to if N#<Nto.
Remark 4.2. Consider growth model (3.1), with γ>0 and real values of α and c, and Nmax>Nmin>0. The existence of the inflection point in the N versus t curve requires: i) the existence of a maximum in the dN/dt versus N curve, which is given by Eq (4.1), subject to N#∈(Nmin,Nmax), and ii) that Nto<N# where Nto>Nmin>0. The existence of the lag phase requires the existence of the inflection point and a small value of dN/dt|t=to. In turn, this implies a positive c value and a small positive (Nto−Nmin) value.
Proposition 4.4. Consider growth model (3.1) with real values of α, c, γ, and Nmax>Nmin>0 and Nto∈(Nmin,Nmax):
i) if
γ>0,andc∈(0,∞) | (4.4) |
holds, then the dN/dt versus N curve exhibits a maximum at N=N#, N#∈(Nmin,Nmax).
ii) the inflection point population, N#, satisfies:
αN−γNmax−N+cNminN(N−Nmin)>0forNmin<N<N#αN−γNmax−N+cNminN(N−Nmin)<0forN#<N≤Nmax | (4.5) |
αN−γNmax−N+cNminN(N−Nmin)=0forN=N# | (4.6) |
iii) if either
i)γ>0,α∈(0,∞)andc∈(−∞, −α(Nmax/Nmin−1)]orii)γ>0,α∈(−∞,0],andc∈(−∞,0] | (4.7) |
holds, the dN/dt versus t curve is monotonic decreasing for N∈(Nmin,Nmax).
Proof. Model (3.1) with Nmax>Nmin>0 subject to parameter regime (4.4) gives:
dNdt|N=Nmin=0;dNdt|N=Nmax=0;dNdt|N∈(Nmin,Nmax)>0 | (4.8) |
Therefore, the dN/dt versus N curve exhibits a maximum at N=N#, with N#∈(Nmin,Nmax).
In addition, we notice from Eq (4.1) that the inflection point population N# is provided by D2|N=N#=0, where
D2=αN+cNminN(N−Nmin)−γNmax−N | (4.9) |
The D2 term exhibits vertical asymptotes at N=Nmin and at N=Nmax. In addition, if parameter regime (4.4) holds, then
limN→N−maxD2=−∞ | (4.10) |
limN→N+minD2=+∞ | (4.11) |
so that there is at least one N value that belongs to range (Nmin,Nmax) and leads to D2=0. Then, there exists a maximum in the dN/dt versus N curve, at N=N#, N# provided by Eq (4.6), and N#∈(Nmin,Nmax).
For the second part, we recall that for model (3.1) with Nto∈(Nmin,Nmax), the inflection point of the N versus t curve satisfies Eq (4.2). As the inflection point corresponds to a maximum of the dN/dt versus t curve, then
ddt(dNdt)>0fort<t#ddt(dNdt)<0fort>t#ddt(dNdt)=0fort=t# | (4.12) |
according to [30]. Equivalently,
(dNdt)⋅ddN(dNdt)>0fort<t#(dNdt)⋅ddN(dNdt)<0fort>t#(dNdt)⋅ddN(dNdt)=0fort=t# | (4.13) |
As dN/dt>0 for N∈(Nmin,Nmax), real values of α, c, γ and Nmax>Nmin>0, then the conditions t<t#, t=t# and t>t# are equivalent to N<N#, N=N# and N>N#, respectively, for N∈(Nmin,Nmax). Therefore, expressions (4.13) lead to
ddN(dNdt)>0forNmin<N<N#ddN(dNdt)<0forN#<N<NmaxddN(dNdt)=0forN=N# | (4.14) |
For model (3.1), and N∈(Nmin,Nmax), we have
ddN(dNdt)=rNα(1−NNmax)γ(1−NminN)c(αN−γNmax−N+cNminN(N−Nmin)) | (4.15) |
This and expressions (4.14) lead to
αN−γNmax−N+cNminN(N−Nmin)>0forNmin<N<N#αN−γNmax−N+cNminN(N−Nmin)<0forN#<N<NmaxαN−γNmax−N+cNminN(N−Nmin)=0forN=N# | (4.16) |
This and condition N∈(Nmin,Nmax) result in second part of Proposition.
For the third part, we notice that if γ>0, and (4.7) holds, then
αN−γNmax−N+cNminN(N−Nmin)<0forN∈(Nmin,Nmax) | (4.17) |
This and Eq (4.15) lead to
ddN(dNdt)<0forN∈(Nmin,Nmax) | (4.18) |
As dN/dt>0 for N∈(Nmin,Nmax), then
(dNdt).ddN(dNdt)<0forN∈(Nmin,Nmax) | (4.19) |
thus d2N/dt2<0 for N∈(Nmin,Nmax). Equivalently, the dN/dt versus t curve is decreasing.
Remark 4.3. In first part of Proposition 4.4 not only the values belonging to the parameter regime (4.4) lead to existence of a maximum in the dN/dt versus N curve. Indeed, certain negative c values with γ>0 also lead to a maximum. The third part of Proposition 4.4 corresponds to the parameter conditions for the absence of a maximum in the dN/dt versus N curve, hence absence of the regular inflection point.
At what follows, the effect of model parameters on the existence of the inflection point and lag phase is assessed, on the basis of the growth rate versus population size. The effect of model parameters on the dN/dt vs N curve and the N vs t growth curve for the model (3.1) is illustrated in Figures 2–4. The following parameter values are used: Nmax=100, γ=1, Nmin=(1−1×10−6)Nto, Nto=1. As the advantage of the proposed model is its capacity to represent the lag phase, it is assumed that γ is positive, and its effect is not assessed. There are four different cases of the α, c values, yielding different possibilities of the existence of the lag phase and the inflection point:
● α>0, c>0: Pⅰ) dN/dt=0 for N=Nmin; Pⅱ) dN/dt=0 for N=Nmax; Pⅲ) dN/dt>0 for N∈(Nmin,Nmax). The first part of Proposition 4.4 applies, so that the inflection point occurs provided Nto<N# and the lag phase existence depends on the (Nto−Nmin) value. Simulations are shown in Figure 2.
● α>0 and c<0: Pⅰ) dN/dt=∞ for N=Nmin, that is, there is a vertical asymptote at N=Nmin; Pⅱ) dN/dt=0 for N=Nmax; Pⅲ) dN/dt>0 for N∈(Nmin,Nmax). Property Pi implies that dN/dt|N=Nto is overly high for low values of (Nto−Nmin), so that there is no lag phase. The occurrence of the inflection point is determined by Eq (4.6) subject to N#∈(Nmin,Nmax), provided Nto<N#. The third part of Proposition 4.4 indicates that for c≤−α(Nmax/Nmin−1) values, there is no inflection point. Moreover, simulations show that for some parameter values there is a local minimum and a local maximum in the range N∈(Nmin,Nmax). As a consequence, there may be inflection point but there is no lag phase. Simulation shown in Figure 3 confirm the above facts, and also shows that the inflection point is owed to the term Nα.
● α<0 and c>0: The properties Pi to Piv of case α>0, c>0 are fulfilled. The first part of Proposition 4.4 applies, so that the inflection point occurs provided Nto<N#, and the lag phase existence depends on the (Nto−Nmin) value. Simulation shown in Figure 4 confirms the above facts, and also show that: ⅰ) the effect of negative values of α on the lag phase duration is low; ⅱ) the maximum growth rate is overly low, so that the slope of the population at the exponential growth phase is overly low. This confirms the importance of the Nα term for the slope of the exponential growth phase.
● α<0 and c<0: Pⅰ) dN/dt=∞ for N=Nmin, that is, there is a vertical asymptote at N=Nmin; Pⅱ) dN/dt=0 for N=Nmax; Pⅲ) dN/dt>0 for N∈(Nmin,Nmax). Proposition 4.4 applies, so that the dN/dt versus N curve is monotonically decreasing for N∈(Nmin,Nmax). Thus, no inflection point occurs, and therefore no lag phase occurs. Simulations in Figure 2 confirm the above properties.
Intermediate cases are:
● α>0 and c=0: The term fc is absent, so that the model becomes dN/dt=rNα(1−N/Nmax)γ, and: Pⅰ) dN/dt>0 for N=Nmin; Pⅱ) dN/dt=0 for N=Nmax; Pⅲ) dN/dt>0 for N∈(Nmin,Nmax); Piv) dN/dt|N=Nto is positive and it is proportional to Nαto. Thus, the dN/dt versus N curve exhibits a maximum with N#∈(Nmin,Nmax), so that there is an inflection point provided Nto<N#, and it is owed to the term Nα with α>0. For general (not small) values of Nto, dN/dt|t=to is not small, so that there is no lag phase. Simulations are shown in Figure 3.
● α=0 and c<0: The term Nα is absent, and the properties are similar to those of case α<0 and c<0: Pⅰ) dN/dt=∞ for N=Nmin, that is, there is a vertical asymptote at N=Nmin; Pⅱ) dN/dt=0 for N=Nmax; Pⅲ) dN/dt>0 for N∈(Nmin,Nmax). According to Proposition 4.4, the dN/dt versus N curve exhibits monotonic decreasing behavior, so that no inflection point occurs, and therefore lag phase neither occurs, what is confirmed by simulation in Figure 3.
● α=0 and c>0: The term Nα is absent, and the properties Pi to Piv of case α<0 and c>0 are fulfilled. The first part of Proposition 4.4 applies, so that an inflection point occurs provided Nto<N# and the lag phase existence depends on the (Nto−Nmin) value. Although simulations are not shown, they indicate that the maximum occurs at low values of the population size (N# is low), and the maximum growth rate is overly low, so that the slope of the growth curve at the exponential growth phase is overly low.
Remark 4.4. The above cases of α and c values indicate that the c parameter has a significant influence on the lag phase duration but the influence of the α value is low. Concerning the slope at the exponential growth phase, the α value has a significant influence, whereas the influence of the c value is low.
Remark 4.5. The parameter region α<0 leads to low slope of the population at the exponential growth phase, whereas c<0 implies that no lag phase occurs. Thus, the condition c>0 is required for representing the lag phase, whereas α>0 is required for higher slopes of the population at the exponential growth phase. This is in accordance with the fact that the formulation of the adaptation function fc by Fujikawa, Kai and Morozumi [16] aimed at suppressing the growth rate during the lag phase, but without affecting the rest of the population growth behavior.
Proposition 4.5. Consider the model (3.1) with γ>0. The population at the inflection point, N#, is defined by:
ˉax2+ˉbx+ˉc=0,x=N#Nmin | (4.20) |
where
ˉa=α+γ;ˉb=(−1)(α(1+NmaxNmin)+γ−c);ˉc=(α−c)NmaxNmin | (4.21) |
And the discriminant is
ˉb2–4ˉaˉc=(α(1−NmaxNmin)+γ−c)2+4γcNmaxNmin | (4.22) |
so that no imaginary solutions are possible for γc≥0. If ˉa>0, c>0, and ˉb2–4ˉaˉc is non-negative, only the positive square root can lead to N#>Nmin.
Proof. The population at the regular inflection point is obtained by applying Eq (4.3) to the growth model (3.1), leading to
α(Nmax−N)(N–Nmin)−γN(N–Nmin)+cNmin(Nmax−N))|N=N#=0 | (4.23) |
Rearranging yields the quadratic Eq (4.20). In order to determine whether the negative square root term can lead to N#>Nmin for ˉa>0, we use Proposition presented in Appendix B. Using the definition of ˉa, ˉb, ˉc, we have: ˉc+ˉa+ˉb=(−1)c(Nmax/Nmin−1). Since (Nmax/Nmin)>1, then ˉc+ˉa+ˉb<0 for c>0, so that the condition ˉc+ˉa+ˉb>0 is not fulfilled for c>0. Consequently, only the positive square root term leads to N#/Nmin>1. End of the proof.
Proposition 4.6. Limit values of N#. Consider model (3.1), with α>0, γ>0, c≥0. The limit values of the regular inflection point, N#, are:
i)limγ→0N#=Nmaxii)limγ→∞N#=Nminiii)limc→0N#=Nmin2(α+γ)[α(1+NmaxNmin)+γ+|(−1)α(NmaxNmin−1)+γ|]iv)limc→∞N#=Nmaxv)limα→0N#=Nmin2γ[γ−c+√(c−γ)2+4cγNmaxNmin]vi)limα→∞N#=Nmax | (4.24) |
Proof. The above limit values are obtained by substituting γ→0, γ→∞, c→0, c→∞, α→0, α→∞ in Eq (4.23), and solving for N#.
Remark 4.6. The above proposition indicates that the limit values of N# for α→0 depends on the values of c and γ; whereas the limit value of N# for c→0 depends on the values of α, γ.
Proposition 4.7. Conditions for N#=Nto. Consider the model (3.1). The population at the regular inflection point, N#, satisfies: N#=Nto for
cγ+α(NmaxNmin−NtoNmin)=(NtoNmin−1)(NtoNmin−αγ+αNmaxNmin) | (4.25) |
Proof. Using N#=Nto in Eq (4.23) and arranging yields Eq (4.25).
Remark 4.7. The above proposition implies that in the N versus t curve, with Nto∈(Nmin,Nmax), the maximum (dN/dt) occurs at t=to, N=Nto, if the α, γ, c values are such that Eq (4.25) holds true.
Proposition 4.8. Conditions for N#=Nmn. Consider model (3.1). The population at the regular inflection point, N#, satisfies N#=Nmn for
cγ+α=12(1+γ−αγ+αNmaxNmin) | (4.26) |
Proof. Using N#=Nmn in Eq (4.23) and arranging yields Eq (4.26).
Remark 4.8. The above proposition implies that in the N versus t curve, with Nto∈(Nmin,Nmn), the maximum dN/dt occurs at N=Nmn where Nmn=(Nmin+Nmax)/2, if the α, γ, c values are such that Eq (4.26) holds true.
Remark 4.9. The determination of the lag time of the regular growth curve via the classical approach requires the determination of the inflection point, as shown in Appendix A. To this end, the population at the inflection point (N#) can be computed using expressions (4.20), (4.21), and the inflection time (t#) must be computed numerically.
Proposition 5.1. Boundedness and convergence. Consider growth model (3.1) with γ>0 and real values of α, c. If Nto∈(Nmin,Nmax), equivalently yto∈(ymin,ymax), then ymin<0<ymax, yto∈(ymin,ymax) and the logarithmic population y exhibits the following features: i) y increases monotonically from y=yto to y=ymax, so that limyt→∞=ymax; ii) y∈[yto,ymax)∀t≥to.
Proof. For y=ln(N/Nto) and the growth model (3.1), with γ>0 and real values of α, c, the specific growth rate dy/dt=(1/N)(dN/dt) fulfills the following properties: ⅰ) (dy/dt)=0 for N=Nmax; ⅱ) (dy/dt)>0 for N∈(Nmin,Nmax).
In this study, the logarithmic inflection point (t∗, N∗, y∗) is defined as the point of the y versus t curve where dy/dt is maximum, being [dy/dt]y∗ the maximum specific growth rate; and N∗ the value of N corresponding to y∗, ie N∗=Ntoey∗.
Proposition 5.2. Consider the growth model (3.1) with γ>0; real values of α, c; Nmax>Nmin>0; and Nto∈(Nmin,Nmax), equivalently yto∈(ymin,ymax): The logarithmic inflection point population, N∗, is provided by
α−1N∗−γNmax−N∗+cNminN∗(N∗−Nmin)=0 | (5.1) |
Proof. As the logarithmic inflection point corresponds to a maximum of the dy/dt=(1/N)(dN/dt) versus t curve, then
d2y/dt2|N=N∗=0,dy/dt|N=N∗>0. | (5.2) |
The condition Nto∈(Nmin,Nmax) implies that N∈(Nmin,Nmax)∀t≥t0, equivalently y∈(ymin,ymax) ∀t≥t0, according to Proposition 5.1. The condition dy/dt|N=N∗>0 implies that N∗≠Nmax (equivalently y∗≠ymax). The above two results imply that N∗∈(Nmin,Nmax), equivalently y∗∈(ymin,ymax). This result, Eq (5.2) and using y=ln(N/Nto), yield
ddN(1NdNdt)|N=N∗=0;N∗∈(Nmin,Nmax) | (5.3) |
Applying this condition to the growth model (3.1), yields Eq (5.1).
Remark 5.1. The condition (5.3) implies that N∗ provided by Eq (5.1) corresponds to a maximum of the (1/N)(dN/dt) versus N curve [21].
Proposition 5.3. Consider the logarithmic measure y=ln(N/Nto), being N provided by growth model (3.1) with real values of α, c, γ, and Nmax>Nmin>0; and Nto∈(Nmin,Nmax). If N∗<Nto, (equivalently y∗<yto) then N>N∗∀t≥to; that is, the y versus t curve exhibits no inflection point. If N∗=Nto, (equivalently y∗=yto), then N=N∗ for t=to; that is, the logarithmic inflection point occurs at t=to, N=Nto.
Proof. Recall that the N versus t curve is monotonic and N≥Nto∀t≥to. Then, N>N∗∀t≥to for N∗<Nto. Consequently, y>y∗∀t≥to for y∗<yto.
Remark 5.2. Consider the logarithmic measure y=ln(N/Nto) provided by growth model (3.1), with γ>0, real values of α, c, and Nmax>Nmin>0, and Nto∈(Nmin,Nmax). The existence of the inflection point of the y versus t curve requires: i) the existence of a maximum in the (1/N)(dN/dt) versus N curve which is given by Eq (5.1) subject to N∗∈(Nmin,Nmax), and ii) that Nto<N∗ where Nto>Nmin>0. The existence of the lag phase requires the existence of the inflection point and a small value of (1/N)(dN/dt)|t=to. In turn, this implies a positive c value and a small positive (Nto−Nmin) value.
Proposition 5.4. Consider the logarithmic measure y=ln(N/Nto), being N provided by growth model (3.1) with real values of α, c, γ, and Nmax>Nmin>0; and Nto∈(Nmin,Nmax):
i) If
γ>0,andc∈(0,∞) | (5.4) |
holds, then the (1/N)(dN/dt) versus N curve exhibits a maximum at N=N∗, N∗∈(Nmin,Nmax).
ii) The logarithmic inflection point population N∗, satisfies:
α−1N−γNmax−N+cNminN(N−Nmin)>0forNmin<N<N∗α−1N−γNmax−N+cNminN(N−Nmin)<0forN∗<N≤Nmax | (5.5) |
α−1N−γNmax−N+cNminN(N−Nmin)=0forN=N∗ | (5.6) |
iii) If either
γ>0,α>1,andc∈(−∞, −(α−1)(Nmax/Nmin−1)]orγ>0,α∈(−∞,1]andc∈(−∞,0] | (5.7) |
then (1/N)(dN/dt) versus t curve is monotonic decreasing for N∈(Nmin,Nmax).
Proof. Model (3.1) with Nmax>Nmin>0 subject to parameter regime (5.4) gives:
1NdNdt|N=Nmin=0;1NdNdt|N=Nmax=0;1NdNdt|N∈(Nmin,Nmax)>0 | (5.8) |
Therefore, the (1/N)(dN/dt) versus N curve exhibits a maximum at N=N∗, with N∗∈(Nmin,Nmax). In addition, we notice from Eq (5.1) that the inflection point population N∗ is provided by D2|N=N∗=0, where
D2=(α−1)N+cNminN(N−Nmin)−γNmax−N | (5.9) |
The D2 term exhibits vertical asymptotes at N=Nmin and at N=Nmax. In addition, if parameter regime (5.4) holds, then
limN→N−maxD2=−∞ | (5.10) |
limN→N+minD2=+∞ | (5.11) |
so that there is at least one N value that belongs to range (Nmin,Nmax) and leads to D2=0. Then, there exists a maximum in the (1/N)(dN/dt) versus N curve, at N=N∗, N∗ provided by Eq (5.1), and N∗∈(Nmin,Nmax).
For the second part, we recall that for the model (3.1) with Nto∈(Nmin,Nmax), the inflection point of the y versus t curve satisfies Eq (5.2). As the inflection point corresponds to a maximum of the dy/dt versus t curve, then
ddtdydt>0fort<t∗ddtdydt<0fort>t∗ddtdydt=0fort=t∗ | (5.12) |
according to [30]. As y=ln(N/Nto), then
d2ydt2=(dNdt).ddN(1NdNdt) | (5.13) |
Combining this with Eq (5.12) yields:
dNdt⋅ddN(1NdNdt)>0fort<t∗dNdt⋅ddN(1NdNdt)<0fort>t∗dNdt⋅ddN(1NdNdt)=0fort=t∗ | (5.14) |
As (dN/dt)>0 for N∈(Nmin,Nmax), real values of α, c, γ and Nmax>Nmin>0, then the conditions t<t∗, t=t∗ and t>t∗ are equivalent to N<N∗, N=N∗ and N>N∗, respectively, for N∈(Nmin,Nmax). Therefore, expression (5.14) leads to
ddN(1NdNdt)>0forNmin<N<N#ddN(1NdNdt)<0forN#<N<NmaxddN(1NdNdt)=0forN=N# | (5.15) |
For model (3.1), and N∈(Nmin,Nmax), we have
ddN(1NdNdt)=rNα−1(1−NNmax)γ(1−NminN)c[α−1N−γNmax−N+cNminN(N−Nmin)] | (5.16) |
This and expressions (5.15) lead to
α−1N−γNmax−N+cNminN(N−Nmin)>0forNmin<N<N∗α−1N−γNmax−N+cNminN(N−Nmin)<0forN∗<N<Nmaxα−1N−γNmax−N+cNminN(N−Nmin)=0forN=N∗ | (5.17) |
This and condition N∈(Nmin,Nmax) results in second part of Proposition.
For the third part, we notice that if (5.7) holds, then
α−1N−γNmax−N+cNminN(N−Nmin)<0forN∈(Nmin,Nmax) | (5.18) |
This and Eq (5.16) lead to
ddN(1NdNdt)<0forN∈(Nmin,Nmax) | (5.19) |
As (dN/dt)>0 for N∈(Nmin,Nmax), then
(dNdt).ddN(1NdNdt)<0forN∈(Nmin,Nmax) | (5.20) |
thus d2y/dt2<0 for N∈(Nmin,Nmax). Equivalently, the dy/dt versus t curve is decreasing.
Remark 5.3. In first part of Proposition 5.4, not only the parameter values belonging to parameter regime (5.4) lead to existence of a maximum in the (1/N)(dN/dt) versus N curve. Indeed, certain negative c values with γ>0 also lead to a maximum. The third part of Proposition 5.4 corresponds to absence of a maximum in the (1/N)(dN/dt) versus N curve, therefore absence of the logarithmic inflection point.
At what follows, the effect of model parameters on the existence of the inflection point and lag phase is assessed, on the basis of the specific growth rate versus population size curve.
The effect of model parameters on the (1/N)(dN/dt) vs N curve and the y vs t curve is illustrated in Figures 5–7. The following parameter values are used: Nmax=100, γ=1, Nmin=(1−1×10−6)Nto, Nto=1. As the advantage of the proposed model is its capacity to represent the lag phase, it is assumed that γ is positive and its effect is not assessed. There are four different cases of the α, c values, yielding different possibilities of the existence of the lag phase and the inflection point in the y versus t curve:
● α>1, c>0: Pⅰ) (1/N)(dN/dt)=0 for N=Nmin; Pⅱ) (1/N)(dN/dt)=0 for N=Nmax; Pⅲ) (1/N)(dN/dt)>0 for N∈(Nmin,Nmax). The first part of Proposition 5.4 applies, so that the (1/N)(dN/dt) versus N curve exhibits a maximum at N=N∗, N∗∈(Nmin,Nmax). Thus, the inflection point occurs provided Nto<N∗, and the lag phase existence depends on the (Nto−Nmin) value. Simulation is shown in Figure 5.
● α<1 and c>0: The properties Pi to Piv of case α>1, c>0 are fulfilled. The first part of Proposition 5.4 applies, so that the inflection point occurs provided Nto<N∗, and the lag phase existence depends on the (Nto−Nmin) value. Simulation is shown in Figure 6, confirming the above facts, and also shows that: ⅰ) the effect of α<1 values on the lag phase duration is low; ⅱ) the maximum growth rate is overly low, leading to low values of the slope of the population at the exponential growth phase. This confirms the importance of the Nα term for the slope of the exponential growth phase.
● α>1 and c<0: Pⅰ) (1/N)(dN/dt)=∞ for N=Nmin, that is, there is a vertical asymptote at N=Nmin; Pⅱ) (1/N)(dN/dt)=0 for N=Nmax; Pⅲ) (1/N)(dN/dt)>0 for N∈(Nmin,Nmax). Property Pi implies that (1/N)(dN/dt)|N=Nto is overly high for low values of (Nto−Nmin), so that there is no lag phase. The occurrence of inflection point is determined by Eq (5.6) subject to N∗∈(Nmin,Nmax), and provided Nto<N∗. The third part of Proposition 5.4 indicates that for c≤(−1)(α−1)(Nmax/Nmin−1) values there is no inflection point. Moreover, simulations show that for some parameter values there is a local minimum and a local maximum in the range N∈(Nmin,Nmax). As a consequence, there may be inflection point but there is no lag phase. Simulation in Figure 7 confirms the above facts.
● α<1 and c<0: Pⅰ) (1/N)(dN/dt)=∞ for N=Nmin, that is, there is a vertical asymptote at N=Nmin; Pⅱ) (1/N)(dN/dt)=0 for N=Nmax; Pⅲ) (1/N)(dN/dt)>0 for N∈(Nmin,Nmax). Proposition 5.4 applies, so that the (1/N)(dN/dt) versus N curve is monotonically decreasing for N∈(Nmin,Nmax). Thus, no inflection point occurs, and therefore lag phase neither occurs. Simulation in Figures 5 and 7 confirms the above facts.
Intermediate cases are:
● α>1 and c=0: The term fc is absent, so that the model becomes (1/N)(dN/dt)=rNα−1(1−N/Nmax)γ, and: Pⅰ) (1/N)(dN/dt)>0 for N=Nmin; Pⅱ) (1/N)(dN/dt)=0 for N=Nmax; Pⅲ) (1/N)(dN/dt)>0 for N \in (N_{min}, \, \, N_{max}) ; Piv) \left. (1/N)(dN/dt)\right \rvert_{N = N_{to}} is positive and is proportional to N_{to}^{\alpha - 1} . Thus, the (1/N)(dN/dt) versus N curve exhibits a maximum with N^{\ast} \in (N_{min}, \, \, N_{max}) , so that the inflection point occurs provided N_{to} < N^{\ast} and it is owed to the term N^{\alpha - 1} with \alpha > 1 . For general (not small) values of N_{to} , \left. (1/N)(dN/dt)\right \rvert_{t = to} is not small, so that there is no lag phase. Simulations are shown in Figure 7.
● \alpha = 1 and c < 0 : The term N^{\alpha-1} is absent, and the properties are similar to those of case \alpha < 1 and c < 0 : Pⅰ) (1/N)(dN/dt) = \infty for N = N_{min} , that is, there is a vertical asymptote at N = N_{min} ; Pⅱ) (1/N)(dN/dt) = 0 for N = N_{max} ; Pⅲ) (1/N)(dN/dt) > 0 for N \in (N_{min}, \, \, N_{max}) . According to the third part of Proposition 5.4, the (1/N)(dN/dt) versus N curve exhibits monotonic decreasing behavior, so that no inflection point occurs, and therefore lag phase neither occurs.
● \alpha = 1 and c > 0 : The term N^{\alpha-1} is absent, and the properties Pi to Piv of case \alpha < 1 and c > 0 , are fulfilled. The first part of Proposition 5.4 applies, so that the inflection point occurs provided N_{to} < N^{\ast} ; and the lag phase existence depends on the (N_{to} - N_{min}) value. Simulation in Figure 5 confirms the above properties and: ⅰ) the maximum occurs at low values of the population size ( N^{\ast} is low); ⅱ) the maximum growth rate is overly low, so that the slope of the growth curve at the exponential growth phase is overly low.
Remark 5.4. The above cases of \alpha and c values indicate that the c parameter has a significant influence on the lag phase duration but the influence of the \alpha value is low. Concerning the slope at the exponential growth phase, the \alpha value has a significant influence, whereas the influence of the c value is low.
Remark 5.5. The parameter region \alpha < 1 leads to low slope of the population at the exponential growth phase, whereas c < 0 implies that no lag phase occurs. Thus, the condition \alpha > 1 is required for higher slopes of the population at the exponential growth phase, whereas c > 0 is required for representing the lag phase.
Proposition 5.5. Consider y = \ln(N/N_{to}) , and the model (3.1) with \gamma > 0 . The population at the logarithmic inflection point, N^{\ast} , is defined by
\begin{eqnarray} & & \bar{a} x^2 + \bar{b} x + \bar{c} = 0, \, \, x = \frac{N^{*}}{N_{min}} \end{eqnarray} | (5.21) |
where
\begin{eqnarray} \begin{array}{l} \bar{a} = \alpha - 1 + \gamma, \bar{b} = - \left ( \Big (\alpha - 1 \Big ) \left ( 1 + {\frac{N_{max}}{N_{min}} }\right ) + \gamma - c \right ), \bar{c} = \left ( \alpha - 1 - c \right ) {\frac{N_{max}}{N_{min}}} \end{array} \end{eqnarray} | (5.22) |
and the discriminant is
\begin{eqnarray} & & \bar{b}^2 – 4 \bar{a} \bar{c} = \left ( \Big (\alpha - 1 \Big ) \left ( 1 - \frac{N_{max}}{N_{min}} \right ) + \gamma - c \right )^2 + 4 \gamma c \frac{N_{max}}{N_{min}} \end{eqnarray} | (5.23) |
so that no imaginary solutions are possible for \gamma c \geq 0 . If \bar{a} > 0 , c > 0 and \bar{b}^2 – 4 \bar{a} \bar{c} is non-negative, only the positive square root leads to N^{\ast} > N_{min} .
Proof. The population at the logarithmic inflection point, N^{\ast} , is obtained by applying Eq (5.3) to the growth model (3.1), leading to
\begin{eqnarray} ( (\alpha -1) (N_{max} - N) (N-N_{min}) - \gamma N (N – N_{min}) + c N_{min} (N_{max} - N) ) \rvert _{N = N^{\ast}} = 0 \end{eqnarray} | (5.24) |
Rearranging yields the quadratic Eq (5.21). In order to determine whether the negative square root term can lead to N^{\ast} > N_{min} for \bar{a} > 0 , c > 0 we use Proposition presented in Appendix B. Using the definition of \bar{a} , \bar{b} , \bar{c} , we have \bar{c} + \bar{a} + \bar{b} = (-1) c (N_{max}/N_{min}-1) . Since (N_{max}/N_{min}) > 1 , then \bar{c} + \bar{a} + \bar{b} < 0 for c > 0 . Hence, the condition \bar{c} + \bar{a} + \bar{b} > 0 is not fulfilled for c > 0 . Consequently, only the positive square root term leads to N^{\ast}/N_{min} > 1 .
Proposition 5.6. Limit values of N^{\ast} . Consider the model (3.1) with \alpha - 1 > 0 , \gamma > 0 , c \geq 0 . The limit values of the population at the logarithmic inflection point, N^{\ast} , are:
\begin{eqnarray} \begin{array}{llll} i)\; \; \; { \lim\limits_{\gamma \rightarrow 0} N^{\ast} } = N_{max} \\ ii)\; \; {\lim\limits_{\gamma \rightarrow \infty} N^{\ast}} = N_{min} \\ iii)\; {\lim\limits_{c \rightarrow 0} N^{\ast} } = \frac{N_{min}}{2 (\alpha -1 + \gamma)} \left [ (\alpha - 1) \left ( 1 + \frac{N_{max}}{N_{min}} \right ) + \gamma + \left | (-1) (\alpha - 1) \left ( \frac{N_{max}}{N_{min}} - 1 \right ) + \gamma \right | \right ] \\ iv)\; \; {\lim\limits_{c \rightarrow \infty} N^{\ast}} = N_{max} \\ v)\; \; \; {\lim\limits_{ (\alpha -1) \rightarrow 0} N^{\ast}} = \frac{N_{min}}{2 \gamma} \left ( \gamma - c + \sqrt { \left ( c - \gamma \right )^2 + 4 c \gamma \frac{N_{max}}{N_{min}} } \right ) \\ vi)\; \; {\lim\limits_{ (\alpha-1) \rightarrow \infty } N^{\ast} } = N_{max} \end{array} \end{eqnarray} | (5.25) |
Proof. The above limit values of N^{\ast} are obtained by substituting \gamma \rightarrow 0 , \gamma \rightarrow \infty , c \rightarrow 0 , c \rightarrow \infty , (\alpha -1) \rightarrow 0 , (\alpha - 1) \rightarrow \infty in Eq (5.24) and solving for N^{\ast} .
Remark 5.6. The limit values of y^{\ast} can be obtained straightforwardly by using y^{\ast} = \ln(N^{\ast}/N_{to}) and the definitions (3.5), (3.6).
Proposition 5.7. Conditions for N^{\ast} = N_{to} . Consider the model (3.1). The population at the logarithmic inflection point, N^{\ast} , satisfies: N^{\ast} = N_{to} for
\begin{eqnarray} & & \frac{c}{\gamma + \alpha -1} \left ( \frac{N_{max}}{N_{min}} - \frac{N_{to}}{N_{min}} \right ) = \left ( \frac{N_{to}}{N_{min}} – 1 \right ) \left ( \frac{N_{to}}{N_{min}} - \frac{\alpha-1}{\gamma + \alpha - 1} \frac{N_{max}}{N_{min}} \right ) \end{eqnarray} | (5.26) |
Proof. Using N^{\ast} = N_{to} in Eq (5.24) and arranging yields Eq (5.26).
Remark 5.7. The above proposition implies that in the y versus t curve, with N_{to} \in (N_{min}, \, \, N_{max}) , the maximum dy/dt occurs at t = t_o , N = N_{to} , if the \alpha , \gamma , c values are such that Eq (5.26) holds true.
Proposition 5.8. Conditions for N^{\ast} = N_{my} (equivalently, y^{\ast} = y_{m} ). Consider model (3.1). The population at the logarithmic inflection point, N^{\ast} , satisfies N^{\ast} = N_{my} for
\begin{eqnarray} & & \frac{c}{\gamma + \alpha -1} = 1 - \frac{\alpha -1}{\gamma + \alpha - 1} \sqrt{ \frac{N_{max}}{N_{min}}} \end{eqnarray} | (5.27) |
Proof. Using N^{\ast} = N_{my} , N_{my} = (N_{min} N_{max})^{1/2} in Eq (5.24) and arranging yields Eq (5.27).
Remark 5.8. The above proposition implies that in the y versus t curve, with N_{to} \in (N_{min}, \, \, N_{my}) , the maximum dy/dt occurs at N = N_{my} ; where N_{my} = (N_{min} N_{max})^{1/2} , if the \alpha , \gamma , c values are such that Eq (5.27) holds true.
Remark 5.9. The determination of the lag time of the logarithmic growth curve via the classical approach requires the determination of the inflection point, as shown in Appendix A. To this end, the population at the inflection point (N^{\ast}) can be computed using expressions (5.21), (5.22), and the inflection time (t^{\ast}) must be computed numerically.
In this section, we propose an approach for the analytical determination of the lag time, for each the regular and the logarithmic measures. Also, a numerical example is provided, using the proposed approach and the classical approach.
The classical measure of the lag time requires the determination of the inflection point, as shown in Appendix A. In turn, the regular and the logarithmic population at the inflection point ( N^{\#} and N^{\ast} ) can be computed using the expressions (4.20), (4.21), (5.21), (5.22). The inflection times (t^{\#}, t^{\ast}) must be computed numerically.
The proposed approach is based on an application of the classical approach to a simplified growth model whose lag time value is quite similar to that of the original model. The simplified model is derived from proposed model (3.1) so as to fulfill the following properties: Rⅰ) its lag time value is quite similar to that of the original N versus t curve provided by model (3.1); Rⅱ) it allows to obtain an analytical expression of the population N ; Rⅲ) it allows to apply the classical t_{lag} approach. Thus, given a model simplified on the basis of the above facts, the t_{lag} value of the simplified model can be considered as a t_{lag} measure of the original model.
Disregarding the saturation term (1-N/N_{N_{max}}) simplifies the determination of the lag phase duration, as was used by Baranyi, Roberts and McClure [1]. For the proposed model (3.1), with given values of \alpha , c , \gamma , remarks 11, 20 indicate that: i) the effect of parameter \alpha on the lag phase and the lag time is low; ii) during the lag phase and the early exponential growth phase, N \approx N_{to} , so that the effect of simplification (1-N/N_{max})^{\gamma} = (1-N_{to}/N_{max})^{\gamma} on the lag time is low. Thus, a simplified form that accomplishes requirements Ri to Riii is:
\begin{eqnarray} & & \frac{dN_{s2}}{dt} = r N_{s2}^{k1} \left( 1 - \frac{N_{to}}{N_{max}} \right) ^{\gamma} \left( 1 - \frac{N_{min}}{N_{s2}} \right) ^c \end{eqnarray} | (6.1) |
\begin{eqnarray} & & \mbox{with} \, \, \, k_1 \in \left \{ c-k_2, \, \, c, \, \, 2, \, \, 3 \right \} \end{eqnarray} | (6.2) |
where k_2 is a positive integer, and the k_1 option that is the nearest to the \alpha value must be chosen. The N_{s2} model is obtained from the proposed model (3.1) with the simplification \alpha = k_1 and (1-N/N_{max})^{\gamma} = (1-N_{to}/N_{max})^{\gamma} . The N_{s2} curve has no inflection point, and it does not converge towards an equilibrium point, but its lag phase and the early part of the exponential growth phase are similar to those of the model (3.1). In the particular case k_1 = c , the analytical solution of Eq (6.1) is:
\begin{eqnarray} & & \frac{1}{1-c} \left[ (N_{s2} - N_{min})^{1-c} - (N_{to} - N_{min})^{1-c}\right] = r \left( 1 - \frac{N_{to}}{N_{max}} \right) ^{\gamma} (t - t_o), \, \, \, c \neq 1 \end{eqnarray} | (6.3) |
The application of the classical t_{lag} approach is different for the regular and the logarithmic growth curves.
To apply the classical t_{lag} approach on the N_{s2} versus t curve, we notice that this curve has no inflection point. Thus, we choose other point, located in the exponential growth phase, whose N value, denoted as N_{s2}^p equals the inflection point of the simplified N versus t curve provided by
\begin{eqnarray} & & \frac{dN_{s1}}{dt} = r N_{s1}^{k1} \left( 1 - \frac{N_{s1}}{N_{max}} \right) ^{\gamma} \left( 1 - \frac{N_{min}}{N_{s1}} \right) ^c \end{eqnarray} | (6.4) |
so that N_{s2}^p = N_{s1}^{\#} . This curve involves the simplification \alpha = k_1 , k_1 satisfying (6.2). In accordance with the classical t_{lag} approach, the intersection of the line that is tangent to the N_{s2} versus t curve at N_{s2} = N_{s1}^{\#} , with the line N_{s2} = N_{to} yields the t_{lag} value. The N_{s1}^{\#} value can be obtained via Eqs (4.20) and (4.21), with \alpha = k_1 . In the particular case k_1 = c , the N_{s1} ^{\#} value is obtained by applying Eq (4.3) to model (6.4) with k_1 = c :
\begin{eqnarray} & & N_{s1}^{\#} = \frac{cN_{max} + \gamma N_{min}}{c+\gamma} \end{eqnarray} | (6.5) |
The time for N_{s2} = N_{s1} ^{\#} (denoted t_s ) is obtained from the analytical solution of Eq (6.1), and using N_{s2} = N_{s1} ^{\#} . In the particular case k_1 = c , t_s is obtained using Eq (6.3):
\begin{eqnarray} & & t_s = t_0 + \frac{1}{r \left( 1 - \frac{N_{to}}{N_{max}} \right) ^{\gamma} (1-c) } \left[ (N_{s1}^{\#}- N_{min})^{1-c} - (N_{to}- N_{min})^{1-c}\right] \end{eqnarray} | (6.6) |
The t_{lag} can be computed from
\begin{eqnarray} & & t_{lag} = t_s - \frac{(N_{s1}^{\#} - N_{to})}{\left . \frac{dN_{s2}}{dt} \right \rvert _{N_{s2} = N_{s1}^{\#}}} \end{eqnarray} | (6.7) |
where (dN_{s2}/dt) at N_{s2} = N_{s1}^{\#} can be computed using Eq (6.1).
In summary, the application of the proposed approach for the regular growth curve involves the following steps:
● Recall the parameters of model (3.1) or obtain them by fitting to experimental data
● Select the k_1 option (6.2) that is the nearest to the \alpha value
● Determine the analytical expression of the N_{s2} curve by using Eq (6.1), and in the case k_1 = c use Eq (6.3)
● Determine N_{s1}^{\#} value using Eqs (4.20) and (4.21), with \alpha = k_1 ; in the particular case k_1 = c , use Eq (6.5)
● Compute the time for N_{s2} = N_{s1}^{\#} (that is, t_s ), using the analytical expression for N_{s2} .
● Compute the t_{lag} value using Eq (6.7).
As a numerical example we consider model (3.1), with r = 1 , N_{min} = (1- 1 \times 10^{-6})N_{to} , N_{to} = 1 , \gamma = 1 ; \alpha = 0.7 and c values c = 0.74 , c = 1.1 , c = 1.2 . We apply the aforementioned procedure, choosing k_1 = c , and the results are shown in Figure 8. Upper left: The time course of N for c = 0.74 , c = 1.1 , c = 1.2 (solid, dash-dotted and dotted lines), with the inflection points (black squares) and the lag time points (gray squares) determined via the classical approach. Each of the remaining figures comprises: The N time course (thin solid line) with the inflection point (black square) and the lag time point (gray square) determined via the classical approach; the N_{s1} time course (dash-dot line); and the N_{s2} time course (thick solid line) with the point N_{s2} = N_{s1} ^{\#} (black point), and the lag time point (gray point) determined via the proposed approach. The c values used are c = 0.74 (upper right), c = 1.1 (lower left), and c = 1.2 (lower right).
To apply the classical t_{lag} approach on the y_{s2} = \ln (N_{s2}/N_{to}) versus t curve, we notice that this curve has no inflection point. Thus, we choose other point, located in the exponential growth phase, whose y_{s2} value, denoted as y_{s2}^p , results from the inflection point of the simplified y versus t curve provided by:
\begin{eqnarray} & & y_{s1} = \ln (N_{s1}/N_{to}) \end{eqnarray} | (6.8) |
\begin{eqnarray} & & \frac{dN_{s1}}{dt} = r N_{s1}^{k1} \left( 1 - \frac{N_{s1}}{N_{max}} \right) ^{\gamma} \left( 1 - \frac{N_{min}}{N_{s1}} \right) ^c \end{eqnarray} | (6.9) |
so that y_{s2}^p = y_{s1}^{\ast} . This curve involves the simplification \alpha = k_1 , being k_1 chosen from options in (6.2). In accordance with the classical t_{lag} approach, the intersection of the line that is tangent to the y_{s2} versus t curve at y_{s2} = y_{s1}^{\ast} , with the line y_{s2} = y_{s2} \rvert_{t = to} = 0 yields the t_{lag} value.
The y_{s1}^{\ast} value and the corresponding N_{s1}^{\ast} value (which satisfies y_{s1}^{\ast} = \ln (N_{s1}^{\ast}/N_{to}) ) are obtained from Eqs (5.21) and (5.22) with \alpha = k_1 . The time for y_{s2} = y_{s1}^{\ast} , equivalently N_{s2} = N_{s1}^{\ast} , denoted as t_s , is obtained from the analytical solution of Eq (6.1) using N_{s2} = N_{s1}^{\ast} . In the particular case k_1 = c , the time t_s is obtained using Eq (6.3) with N_{s2} = N_{s1}^{\ast} :
\begin{eqnarray} & & t_s = t_0 + \frac{1}{r \left( 1 - \frac{N_{to}}{N_{max}} \right) ^{\gamma} (-c+1) } \left[ (N_{s1}^{\ast}- N_{min})^{-c+1} - (N_{to}- N_{min})^{-c+1}\right] \end{eqnarray} | (6.10) |
The t_{lag} value can be computed from
\begin{eqnarray} & & t_{lag} = t_s - \frac{\ln (N_{s1}^{\ast}/N_{to})}{ \left . \frac{dy_{s2}}{dt} \right \rvert _{N_{s2} = N_{s1}^{\ast}}} \end{eqnarray} | (6.11) |
where \left. dy_{s2}/dt \right \rvert_{Ns2 = N_{s1}^{\ast}} can be computed using dy_{s2}/dt = (1/N_{s2})dN_{s2}/dt , being dN_{s2}/dt provided by Eq (6.1) with N_{s2} = N_{s1}^{\ast} .
In summary, the application of the proposed approach for he logarithmic growth curve involves the following steps:
● Recall the parameters of model model (3.1) or obtain them by fitting to experimental data
● Select the k_1 option (6.2) that is the nearest to the \alpha value
● Determine the analytical expression of the N_{s2} curve by using Eq (6.1), and in the case k_1 = c use Eq (6.3).
● Determine the y_{s1}^{\ast} = \ln (N_{s1}^{\ast}/N_{to}) value, using Eqs (5.21) and (5.22) with \alpha = k_1 .
● Compute the time for y_{s2} = y_{s1}^{\ast} (that is, t_s ), using the analytical expression for N_{s2} with N_{s2} = N_{s1}^{\ast} .
● Compute the t_{lag} value, using Eq (6.11)
As a numerical example, we consider model (3.1), with r = 1 , N_{min} = (1- 1 \times 10^{-6})N_{to} , N_{to} = 1 , \gamma = 1 ; \alpha = 0.7 , and c values c = 0.74 , c = 1.1 , c = 1.2 . We apply the aforementioned procedure, choosing k_1 = c , as shown in Figure 9. Upper left: Time course of y for c values c = 0.74 , c = 1.1 , c = 1.2 (solid, dash-dotted and dotted lines), with the inflection points (black squares) and lag time points (gray squares) determined via the classical approach. Each of the remaining figures comprise: The y time course (thin solid line) with the inflection point (black square) and the lag time point (gray square) determined via the classical approach; the y_{s1} = \ln (N_{s1} / N_{to}) time course (dash-dot line); and the y_{s2} = \ln (N_{s2} / N_{to}) time course (thick solid line) with the point y_{s2} = y_{s1} ^{\ast} (black point) and the lag time point (gray point) determined via the proposed approach. The c values used are c = 0.74 (upper right), c = 1.1 (lower left), and c = 1.2 (lower right).
In this section, the proposed model (3.1) is applied to experimental microbial growth curves.
The differential Eq (3.1), the expression for the population at the inflection point N^{\ast} and the expression for the lag time ( t_{lag} ) are rewritten in terms of the logarithmic measure y = ln(N/N_{a}) as follows, being N_a a user defined positive constant. Using model (3.1), the above definition of y instead of definition (3.4), and the definition
\begin{eqnarray} & & y_{min} = ln (N_{min}/N_a), \, \, y_{max} = ln (N_{max}/N_a), \end{eqnarray} | (7.1) |
one obtains:
\begin{eqnarray} & & \frac{dy}{dt} = \bar{r} e^{(\alpha - 1) y } \left( 1 - e^{(y - y_{max})} \right)^{\gamma} \left( 1 - e^{-(y - y_{min})} \right)^{c} \\ & & \bar{r} = r N_{a}^{\alpha - 1}, \end{eqnarray} | (7.2) |
and its parameters are \bar{r} , \alpha , \gamma , c , y_{min} , y_{max} . Using definition (7.1), the term N_{max}/N_{min} appearing in definitions (5.22) can be expressed as
\begin{eqnarray} & & \frac{N_{max}}{N_{min}} = e^{y_{max} - y_{min}} \end{eqnarray} | (7.3) |
Definition y = ln(N/N_{a}) implies y^{\ast} = ln(N^{\ast} / N_{a}) . Using definition (7.1) and Eq (5.21), one further obtains y^{\ast} = ln (e^{y_{min}} x) , being x obtained from Eq (5.21). The t^{\ast} value can be computed numerically by using Eq (7.2), so as to obtain y = y^{\ast} . The lag time can be computed via the classical approach with
\begin{eqnarray} & & t_{lag} = t^{\ast} + \frac{ ( y_{to} - y^{\ast} ) }{\frac{dy}{dt} \rvert _{y = y^{\ast}}} \end{eqnarray} | (7.4) |
being (dy/dt) \rvert_{y^{\ast}} obtained using Eq (7.2).
Remark 7.1. In summary, the model properties in terms of the logarithmic measure includes: i) the logarithmic population at the inflection point (y^{\ast}) , which can be computed using expression y^{\ast} = ln(e^{y_{min}} x) , being x obtained from Eqs (5.21) and (5.22) with (N_{max}/N_{min}) = e^{y_{max} - y_{min}} ; ii) the inflection time (t^{\ast}) , which can be computed numerically using Eq (7.2); and iii) the lag time t_{lag} which can be computed by means of Eq (7.4), being (dy/dt) \rvert_{y^{\ast}} obtained using Eq (7.2). In this way, the computation of y^{\ast} , t^{\ast} , t_{lag} neither requires N^{\ast} nor N_{a} .
The model of Baranyi and Roberts [31] is:
\begin{eqnarray} & & y = Ln(N), \end{eqnarray} | (7.5) |
\begin{eqnarray} & & y (t) = y_o + \mu_{max} A_t - \frac{1}{m} Ln \left( 1 + \frac{e^{m \, \mu_{max} A_t}-1}{e^{m(y_{max}-y_o)}} \right), \end{eqnarray} | (7.6) |
\begin{eqnarray} & & A_t = t + \frac{1}{v} Ln \left( \frac{e^{-vt} + q_o}{1 + q_o}\right), \end{eqnarray} | (7.7) |
its parameters are \mu_{max} , m , y_{max} , q_o , v ; and the lag time is t_{lag} = (1/v) Ln (1+1/q_o) .
The Gompertz model is [32]:
\begin{eqnarray} & & y = y_a + (y_{max} - y_a) exp \left (-exp \left( 1 + \frac{e^1 \mu_{max}}{y_{max} - y_a} (\lambda - t) \right) \right) \end{eqnarray} | (7.8) |
its parameters are \mu_{max} , y_{max} , \lambda , y_a ; and the lag time is t_{lag} = \lambda .
In order to assess the capability of the proposed model to simulate growth curves, it and also the model of Baranyi and Roberts and the Gompertz model were fitted to growth records of Salmonellae. The datasets were taken from [1], corresponding to constant temperature and pH. The proposed model, the inflection point and the lag time expressed in terms of the logarithmic measure y are stated in subsection 7.1, whereas the model of Baranyi and Roberts and the Gompertz model are stated in subsection 7.2. The value N_a = 1 is used for the proposed model, in order to compare the simulation of y = ln (N) for the three models.
Model simulation and parameter estimation were performed using Matlab software (the Math Works Inc., Natick, Mass.). The differential Eq (7.2) was numerically integrated using the ode45 routine. Parameter estimation was carried out by minimizing the sum of the squared deviations between the values calculated by the model and the experimental data. The R^2 value was used to evaluate the goodness of fit (see [33]), whereas the Akaike's Information Criterion (AIC) allows to compare the model acceptability on the basis of both fitting capability and complexity, being the complexity related to number of parameters [34,35]:
\begin{eqnarray} & & AIC = N Ln \left ( \frac{RSS}{N} \right ) + 2 (N_{prm} + 1) + \frac{2(N_{prm} + 1) (N_{prm} + 2)}{N - N_{prm} -2} \end{eqnarray} | (7.9) |
where N is the number of data points, N_{prm} is the number of model parameters, and RSS is the sum of the squared deviations between the values calculated by the model and the experimental data. The model with the lowest AIC value is the most acceptable. The fitting results are:
\begin{eqnarray} & & Generalized~~ Fujikawa's~~ model: \\ & & \quad \bar{r} = 0.92614, \, \, \gamma = 0.4745, \, \, \alpha = 0.9647, \, \, c = 0.8210, \, \, y_{min} = 7.7897, \, \, y_{max} = 20.15, \\ & & \quad y^{\ast} = 10.977, \, \, t^{\ast} = 8.715 h, \, \, t_{lag} = 3.492 h, \, \, R^2 = 0.9919, \, \, AIC = -11.35, \\ & & Model ~~of ~~Baranyi~~ and~~ Roberts: \\ & & \quad \mu_{max} = 0.5518, \, \, m = 1.1051, \, \, y_{max} = 20.31, \, \, q_o = 0.0004574, \, \, v = 2.7169 \\ & & \quad t_{lag} = 2.8305 h, \, \, R^2 = 0.9922, \, \, AIC = -16.4288, \\ & & Gompertz ~~model: \\ & & \quad \mu_{max} = 0.69007, \, \, y_{max} = 21.013, \, \, \lambda = 3.7876 h, \, \, Y_a = 7.5373 \\ & & \quad R^2 = 0.9883, \, \, AIC = -11.4761, \end{eqnarray} |
There is a good agreement between model simulation and experimental measurements (Figure 10), with a relatively high and similar R^2 value for all the models. The estimated \alpha , \gamma , c , are positive, and \alpha is close to 1.0 . The lag time t_{lag} values were not quite similar. In terms of the AIC value, the Baranyi model is the most acceptable. This is partially related to the number of parameters, indeed N_{prm} = 6 , N_{prm} = 5 , N_{prm} = 4 for the proposed model, the Baranyi's model and the Gompertz model, respectively. This worsens the AIC value of the proposed model in comparison with that of Baranyi's model. The fitting of the three models exhibit disagreement for the lag phase points (first seven points). In view of the above limitations, the fitting of the models to the lag phase points was improved, although at the cost of lowering the overall R^2 value; \alpha = 1 was used in the proposed model, what is a common value, see [16,28]; and the R^2 for the lag phase (R^2_{lag}) was calculated. The fitting results are:
\begin{array}{l} Generalized\;\;Fujikawa's\;\;model\;\;(\alpha = 1):\\ \quad \bar r = 0.61107, {\mkern 1mu} {\mkern 1mu} \gamma = 1.1423, {\mkern 1mu} {\mkern 1mu} c = 0.88021, {\mkern 1mu} {\mkern 1mu} {y_{min}} = 7.7785, {\mkern 1mu} {\mkern 1mu} {y_{max}} = 20.2143, {\mkern 1mu} {\mkern 1mu} {y^ * } = 13.866, \\ \quad {t^ * } = 14.94h, {\mkern 1mu} {\mkern 1mu} {t_{lag}} = 4.9822h, {\mkern 1mu} {\mkern 1mu} {R^2} = 0.98845, {\mkern 1mu} {\mkern 1mu} AIC = - 4.9322, {\mkern 1mu} {\mkern 1mu} R_{lag}^2 = 0.9663\\ Modelof\;\;Baranyiand\;\;Roberts:\\ \quad {\mu _{max}} = 0.6114, {\mkern 1mu} {\mkern 1mu} m = 0.92099, {\mkern 1mu} {\mkern 1mu} {y_{max}} = 20.1474, {\mkern 1mu} {\mkern 1mu} {q_o} = 0.047274, {\mkern 1mu} {\mkern 1mu} v = 0.69237\\ \quad {t_{lag}} = 4.4745h, {\mkern 1mu} {\mkern 1mu} {R^2} = 0.9886, {\mkern 1mu} {\mkern 1mu} AIC = - 8.1242, {\mkern 1mu} {\mkern 1mu} R_{lag}^2 = 0.96475\\ Gompertz\;\;model:\\ \quad {\mu _{max}} = 0.69168, {\mkern 1mu} {\mkern 1mu} {y_{max}} = 20.9808, {\mkern 1mu} {\mkern 1mu} \lambda = 4.00h, {\mkern 1mu} {\mkern 1mu} {Y_a} = 7.3672\\ \quad {R^2} = 0.98695, {\mkern 1mu} {\mkern 1mu} AIC = - 9.0126, {\mkern 1mu} {\mkern 1mu} R_{lag}^2 = 0.84902 \end{array} |
The three models achieved satisfactory agreement between simulation and experimental data, with a high R^2 value (Figure 11). In terms of the AIC value, the Gompertz model is the most acceptable, but it exhibits significantly flawed fitting of several data points. The Baranyi model exhibits the second best AIC value. Both the Baranyi's model and the proposed model achieved proper fitting of the lag phase points, and the highest R^2_{lag} was achieved by the proposed model.
The obtained results confirm the capability of the proposed model to simulate monotonic growth curves, mainly those involving lag phase.
The studied generalized form of the Fujikawa’s growth model is autonomous, exhibits monotonic behavior of the population size and is capable of generating the common features of monotonic growth of microorganisms along time, including growth with decreasing slope without inflection point and growth with inflection point and lag phase. Indeed, either a large or a small lag time can be combined with either a large or a small slope of the exponential growth phase. Thus, the model is highly flexible and is suitable for different applications. Although Fujikawa, Kai and Morozumi [16,28] only employed positive values of \alpha , c , negative values lead to features of the y versus t curve that are possible in real situations.
A deep insight on the growth curves is obtained by analyzing the dN/dt versus N curve at N = N_{min} , N = N_{max} and for N \in (N_{min}, \, \, N_{max}) , in addition to assessing the presence of a maximum (which corresponds to an inflection point) or a minimum. A minimum and a maximum occur for \alpha > 0 and c < 0 , what is not common in growth curves.
The study of the effect of \alpha and c values on the existence of the regular inflection point and lag phase indicates that: ⅰ) if both are positive, the inflection point occurs provided adequate N_{to} ; ⅱ) if either \alpha or c is negative, the occurrence of the inflection point is possible; ⅲ) if both parameters are negative, there is no inflection point nor lag phase. The existence of lag phase depends on the value of the growth rate at initial time, what is proportional to (N_{to} - N_{min}) .
The \alpha and c parameters also affect the lag time magnitude and the slope of the exponential growth phase: ⅰ) the \alpha value has a significant influence on the slope at the exponential growth phase, whereas the influence of the c value is low; ⅱ) the c parameter has a significant influence on the lag phase duration but the influence of the \alpha value is low. For adequate representation of a growth curve that exhibits lag phase and steep exponential growth phase, it is required that both the \alpha and the c values be positive.
The limit values of the inflection point population N^{\#} confirms it great flexibility, due to the parameters \alpha , c , \gamma .
The numerical fitting example indicated the capability of the proposed model for fitting to experimental data of microbial growth. Indeed, the R^2 value is high and similar to that of the Baranyi model.
A Rincón was supported by Universidad Católica de Manizales.
The authors declare that there is no conflict of interests regarding the publications of this paper.
For the regular measure (N), the tangent line at the inflection point (t^{\#}, N^{\#}) is
\begin{eqnarray} & & N = \left. \frac{dN}{dt} \right \rvert_{t = t^{\#}} t + N^{\#} - \left. \frac{dN}{dt} \right \rvert_{t = t^{\#}} t^{\#} \end{eqnarray} | (A.1) |
The lag phase duration (t_{lag}) is obtained from the intersection of this line with the horizontal line N = N_{to}, what gives
\begin{eqnarray} & & t_{lag} = t^{\#} – \frac{(N^{\#}-N_{to})}{\left. dN/dt\right \rvert_{t = t^{\#}}} \end{eqnarray} | (A.2) |
For the logarithmic measure y = \ln(N/N_{to}), the tangent line at the inflection point (t^{\ast}, y^{\ast}) is
\begin{eqnarray} & & y = \left. \frac{dy}{dt} \right \rvert_{t^{\ast}} t + y^{\ast} - \left. \frac{dy}{dt} \right \rvert_{t = t^{\ast}} t^{\ast} \end{eqnarray} | (A.3) |
where
\begin{equation} \left. \frac{dy}{dt} \right \rvert_{t = t^{\ast}} = \left. \frac{1}{N} \frac{dN}{dt} \right \rvert_{N = N^{\ast}} \end{equation} | (A.4) |
The lag phase duration (t_{lag}) is obtained from the intersection of this line with the horizontal line y = y_{to} = 0, what gives
\begin{eqnarray} & & t_{lag} = t^{\ast} – \frac{y^{\ast}}{\left. dy/dt\right \rvert_{t = t^{\ast}}} = t^{\ast} – \frac{\ln(N^{\ast}/N_{to})}{\left. dy/dt\right \rvert_{N = N^{\ast}}} \end{eqnarray} | (A.5) |
Proposition A.1. Consider the quadratic algebraic equation
\begin{equation} \bar{a} x^2 + \bar{b} x + \bar{c} = 0 \label{ec:quadratic_gral} \end{equation} | (A.6) |
being \bar{b}^2 – 4 \bar{a} \bar{c} non-negative. If \bar{a}>0, the solution with negative square root requires \bar{c} + \bar{a} + \bar{b} >0 for obtaining x>1.
Proof Consider the quadratic Eq (A.6), with \bar{a}>0, and the solution comprising the negative square root term. Accomplishment of condition x>1 implies
\begin{eqnarray} & & - \sqrt{\bar{b}^2-4\bar{a} \bar{c}} \gt 2 \bar{a} + \bar{b} \nonumber\\ & & \sqrt{\bar{b}^2-4\bar{a} \bar{c}} \lt (-1) (2 \bar{a} + \bar{b}) \label{ec:Appb_scnd_sqrroottermineq} \end{eqnarray} | (A.7) |
This expression implies that 2 \bar{a} + \bar{b} < 0. Taking squares to Eq (A.7) and arranging yields -\bar{c} < \bar{a} + \bar{b}. Therefore, the solution with negative square root term, requires the condition \bar{c} + \bar{a} + \bar{b} >0 for achieving x>1.
[1] | Altenburger R, Ait-Aissa S, Antczak P, et al. (2015) Future water quality monitoring—adapting tools to deal with mixtures of pollutants in water resource management. Sci Total Environ 512: 540-551. |
[2] | Chezgi J, Pourghasemi HR, Naghibi SA, et al. (2015) Assessment of a spatial multi-criteria evaluation to site selection underground dams in the Alborz Province, Iran. Geocarto Int 31: 628-646. |
[3] | Oh HJ, Kim YS, Choi JK, et al. (2011) GIS mapping of regional probabilistic groundwater potential in the area of Pohang City, Korea. J Hydrol 399: 158-172. |
[4] | Nampak H, Pradhan B, Manap MA (2014) Application of GIS based data driven evidential belief function model to predict groundwater potential zonation. J Hydrol 513: 283-300. |
[5] | Tweed SO, Leblanc M, Webb JA, et al. (2007) Remote sensing and GIS for mapping groundwater recharge and discharge areas in salinity prone catchments, southeastern Australia. Hydrogeol J 15: 75-96. |
[6] | Ozdemir A (2011a) Using a binary logistic regression method and GIS for evaluating and mapping the groundwater spring potential in the Sultan Mountains (Aksehir, Turkey). J Hydrol 405: 123-136. |
[7] | Ozdemir A (2011b) GIS-based groundwater spring potential mapping in the Sultan Mountains (Konya, Turkey) using frequency ratio, weights of evidence and logistic regression methods and their comparison. J Hydrol 411: 290-308. |
[8] | Manap MA, Nampak H, Pradhan B, et al. (2014) Application of probabilistic-based frequency ratio model in groundwater potential mapping using remote sensing data and GIS. Arabian J Geosci 7: 711-724. |
[9] | Pourtaghi ZS and Pourghasemi HR (2014) GIS-based groundwater spring potential assessment and mapping in the Birjand Township, southern Khorasan Province, Iran. Hydrogeology J 22:643-662. |
[10] | Naghibi SA, Pourghasemi HR, Pourtaghi ZS, et al. (2015) Groundwater qanat potential mapping using frequency ratio and Shannon's entropy models in the Moghan watershed, Iran. Earth Sci Inform 8: 171-186. |
[11] | Pourghasemi HR, Beheshtirad M (2015) Assessment of a data-driven evidential belief function model and GIS for groundwater potential mapping in the Koohrang Watershed, Iran. Geocarto Int 30: 662-685. |
[12] | Naghibi SA, Pourghasemi HR (2015) A comparative assessment between three machine learning models and their performance comparison by bivariate and multivariate statistical methods in groundwater potential mapping. Water Res Manage 29: 5217-5236. |
[13] | Al-Abadi A, M Pradhan B, Shahid S (2015) Prediction of groundwater flowing well zone at An-Najif Province, central Iraq using evidential belief functions model and GIS. Environ Monit Assess 188:549. |
[14] | Naghibi SA, Pourghasemi HR, Dixon B (2016) GIS-based groundwater potential mapping using boosted regression tree, classification and regression tree, and random forest machine learning models in Iran. Environ Monit Assess 188: 1-27. |
[15] | Naghibi S A, Dashtpagerdi MM (2016) Evaluation of four supervised learning methods for groundwater spring potential mapping in Khalkhal region (Iran) using GIS-based features. Hydrogeol J 1-21. |
[16] | Zabihi M, Pourghasemi, HR, Pourtaghi ZS, et al. (2016) GIS-based multivariate adaptive regression spline and random forest models for groundwater potential mapping in Iran. Environ Earth Sci 75: 1-19. |
[17] | Naghibi SA, Pourghasemi HR. , Abbaspour K (2017). A comparison between ten advanced and soft computing models for groundwater qanat potential assessment in Iran using R and GIS. Theoretical and Applied Climatology. |
[18] | Mennis J and Guo D (2009) Spatial data mining and geographic knowledge discovery—an introduction. Comput Environ Urban Syst 33:403-408. |
[19] | Tien Bui D, Le K-T, Nguyen V (2016a) Tropical forest fire susceptibility mapping at the Cat Ba National Park Area, Hai Phong City, Vietnam, using GIS-based Kernel logistic regression. Remote Sens 8: 347. |
[20] | Pacheco, Fernando, Ana Alencoão (2002) "Occurrence of springs in massifs of crystalline rocks, northern Portugal." Hydrogeol J 10.2: 239-253. |
[21] | Pacheco FAL and Vander Weijden CH (2014). Modeling rock weathering in small watersheds. J Hydrol 513C 13-27. |
[22] | Iranian Department of Water Resources Management (2014) weather and climate report, Tehran province. Available from: http://www.thrw.ir/. |
[23] | Dehnavi A, Aghdam IN, Pradhan B, et al. (2015) A new hybrid model using step-wise weight assessment ratio analysis (SWARA) technique and adaptive neuro-fuzzy inference system (ANFIS) for regional landslide hazard assessment in Iran. Catena 135: 122-148. |
[24] | Conforti M, Pascale S, Robustelli G, et al. (2014) Evaluation of prediction capability of the artificial neural networks for mapping landslide susceptibility in the Turbolo River catchment (northern Calabria, Italy). Catena 113: 236-250. |
[25] | Oh H-J and Pradhan B (2011) Application of a neuro-fuzzy model to landslide-susceptibility mapping for shallow landslides in a tropical hilly area. Comput Geosci 37: 1264-1276. |
[26] | Wilson JP and Gallant JG (2000) Terrain Analysis Principles and Applications. John Wiley and Sons, Inc. , New-York, 479. |
[27] | Moore ID and Burch GJ (1986) Sediment Transport Capacity of Sheet and Rill Flow: Application of Unit Stream Power Theory. Water Resour Res 22: 1350-1360. |
[28] | Pourghasemi H, Pradhan B, Gokceoglu C, et al. (2013) A comparative assessment of prediction capabilities of Dempster-Shafer and weights-of-evidence models in landslide susceptibility mapping using GIS. Geomatics Nat Hazards Risk 4: 93-118. |
[29] | Moore ID, Grayson RB, Ladson AR (1991) Digital terrain modelling: A review of hydrological, geomorphological, and biological applications. Hydrol Process 5: 3-30. |
[30] | Ayalew L and Yamagishi H (2005) The application of GIS-based logistic regression for land-slide susceptibility mapping in the Kakuda-Yahiko Mountains, Central Japan. Geomorphol 65: 15-31. |
[31] | Pradhan B, Abokharima MH, Jebur MN, et al. (2014) Land subsidence susceptibility mapping at Kinta Valley (Malaysia) using the evidential belief function model in GIS. Nat Hazards. |
[32] | Iranian forest, rangeland and watershed management organization. 2014. Available from: http://www.frw.org.ir/00/En/default.aspx. |
[33] | Geology Survey of Iran (GSI) (1997) Geology map of the Chaharmahal-e-Bakhtiari Province. http://www.gsi.ir/Main/Lang_en/index.html.AccessedSeptember2000. |
[34] | Dormann CF, Elith J, Bacher S, et al. (2013) Collinearity: a review of methods to deal with it and a simula-tion study evaluating their performance. Ecogr 36: 27-46 |
[35] | Hair JF, Black WC, Babin BJ, et al. (2009) Multivariate data analysis. Prentice Hall, New York. |
[36] | Keith TZ (2006) Multiple regressions and beyond. Pearson, Boston |
[37] | Lee S and Talib JA (2005) Probabilistic landslide susceptibility and factor effect analysis. Environ Geol 47: 982-990 |
[38] | Yilmaz I (2007) GIS based susceptibility mapping of karst depression in gypsum: a case study from Sivas basin (Turkey). Eng Geol 90: 89-103 |
[39] | Bonham-Carter G (1994) Geographic information systems for geoscientists modelling with GIS. Pergamon. |
[40] | Lee S and Pradhan B (2006) Probabilistic landslide hazards and riskmapping on Penang Island, Malaysia. Earth Sys Sci 115: 661-667 |
[41] | Lombardo L, Cama M, Conoscenti C, et al. (2015) Binary logistic regression versus stochastic gradient boosted decision trees in assessing landslide susceptibility for multiple-occurring landslide events: application to the 2009 storm event in Messina (Sicily, southern Italy). Nat Hazards 79: 1621-1648. |
[42] | Abeare SM (2009) Comparisons of boosted regression tree, glm and gam performance in the standardization of yellowfin tuna catch-rate data from the gulf of mexico lonline fishery. |
[43] | Friedman J (2001) Greedy boosting approximation: a gradient boosting machine. Ann Stat 29: 1189-1232. |
[44] | Schapire RE (2003) The boosting approach to machine learning: an overview. Nonlinear Est Class 171: 149-171. |
[45] | Sutton CD (2005) 11-Classification and Regression Trees, Bagging, and Boosting. Handbook of statistics 24: 303-329 |
[46] | Kuhn M, Wing J, Weston S, et al. (2015) Package "caret". Available from: http://caret.r-forge.r-project.org. |
[47] | Elith J, Leathwick JR, Hastie T (2008) A working guide to boosted regression trees. J Animal Ecol 77: 802-813 |
[48] | Leathwick JR, Elith J, Francis MP, et al. (2006). Variation in demersal fish species richness in the oceans surrounding New Zealand: An analysis using boosted regression trees. Mar Ecol Prog Ser 321: 267-281. |
[49] | Schillaci C, Lombardo L, Saia S, et al. (2017) Modelling the topsoil carbon stock of agricultural lands with the Stochastic Gradient Treeboost in a semi-arid Mediterranean region. Geoderma 286: 35-45. |
[50] | R Development Core Team (2006) R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0. Available from: http://www.R-project.org |
[51] | Ridgeway G (2015) Package "gbm. " |
[52] | Guzzetti F, Reichenbach P, Ardizzone F, et al. (2006) Estimating the quality of landslide susceptibility models. Geomorphol 81: 166-184 |
[53] | Umar Z, Pradhan B, Ahmad A, et al. (2014) Earthquake induced landslide susceptibility mapping using an integrated ensemble frequency ratio and logistic regression models in West Sumatera Province, Indonesia. Catena 118: 124-135 |
[54] | Swets JA (1988) Measuring the accuracy of diagnostic systems. Sci 240: 1285-1293 |
[55] | Negnevitsky M (2005) Artificial Intelligence: a guide to intelligent systems. Inf & Comput Sci 48: 284-300. |
[56] | Frattini P, Crosta G, Carrara A (2010). Techniques for evaluating the performance of landslide susceptibility models. Eng Geol 111: 62-72. |
[57] | Lombardo L, Cama M, Maerker M., et al. (2014). A test of transferability for landslides susceptibility models under extreme climatic events: Application to the Messina 2009 disaster. Natural Hazards 74: 1951-1989. |
[58] | Heckman T, Gegg K, Gegg A, et al. (2014) Sample size matters: investigating the effect of sample size on a logistic regression susceptibility model for debris flow. Nat. Hazards Earth Syst. Sci 14: 259-278. |
[59] | Youssef AM, Pourghasemi HR, Pourtaghi ZS, et al. (2015) Landslide susceptibility mapping using random forest, boosted regression tree, classification and regression tree, and general linear models and comparison of their performance at Wadi Tayyah Basin, Asir Region, Saudi Arabia. Landslides 1-18 |
[60] | Carty DM (2011) An analysis of boosted regression trees to predict the strength properties of wood composites. Arch Oral Biol 60: 45-54. |
1. | Soumya G. Nair, Jyotirmayee Satapathy, N. Pavan Kumar, Influence of synthesis, dopants, and structure on electrical properties of bismuth ferrite (\hbox {BiFeO}_{3}), 2020, 126, 0947-8396, 10.1007/s00339-020-04027-x | |
2. | Dao Viet Thang, Nguyen Manh Hung, Nguyen Cao Khang, Le Thi Mai Oanh, Structural and multiferroic properties of (Sm, Mn) co-doped BiFeO3 materials, 2020, 7, 2372-0484, 160, 10.3934/matersci.2020.2.160 | |
3. | A.T. Apostolov, I.N. Apostolova, J.M. Wesselinowa, Magnetic field effect on the dielectric properties of rare earth doped multiferroic BiFeO3, 2020, 513, 03048853, 167101, 10.1016/j.jmmm.2020.167101 | |
4. | S. R. Dhanya, Jyotirmayee Satapathy, Pavan Kumar, A review on the structural and magnetic properties of differently doped bismuth-ferrite multiferroics, 2021, 27, 0889-311X, 178, 10.1080/0889311X.2021.2020262 | |
5. | Mekonnen Tefera Kebede, Sheela Devi, Babita Tripathi, Sunil Chauhan, Venus Dillu, Structural transition and enhanced magnetic, optical and photocatalytic properties of novel Ce–Ni co-doped BiFeO3 nanoparticles, 2022, 152, 13698001, 107086, 10.1016/j.mssp.2022.107086 | |
6. | Dao Viet Thang, Nguyen Manh Hung, Du Thi Xuan Thao, Le Thi Mai Oanh, Do Danh Bich, Nguyen Cao Khang, Van Quang Nguyen, Nguyen Van Minh, Structural, Electrical, and Magnetic Properties of \text{Bi}_{0.90}\text{Nd}_{0.10}\text{Fe}_{0.98}TM_{0.02}\text{O}_{3} (TM= \text{Mn}, Co, Ni, and Cu) Materials, 2019, 10, 1949-307X, 1, 10.1109/LMAG.2019.2896847 | |
7. | Ibtissem Ragbaoui, Samia Aydi, Souad Chkoundali, Marwa Enneffati, Abdelhedi Aydi, Effect of substitution on the structural, electrical properties, and dielectric relaxor behavior in lead-free BiFeO3-based ceramics, 2024, 14, 2046-2069, 1330, 10.1039/D3RA06962A |