Numerical Evaluation of Fractional Vertical Soil Water Flow Equations

: Signiﬁcant deviations from standard Boltzmann scaling, which corresponds to normal or Fickian diffusion, have been observed in the literature for water movement in porous media. However, as demonstrated by various researchers, the widely used conventional Richards equation cannot mimic anomalous diffusion and ignores the features of natural soils which are heterogeneous. Within this framework, governing equations of transient water ﬂow in porous media in fractional time and multi-dimensional fractional soil space in anisotropic media were recently introduced by the authors by coupling Brooks–Corey constitutive relationships with the fractional continuity and motion equations. In this study, instead of utilizing Brooks–Corey relationships, empirical expressions, obtained by least square ﬁts through hydraulic measurements, were utilized to show the suitability of the proposed fractional approach with other constitutive hydraulic relations in the literature. Next, a ﬁnite difference numerical method was proposed to solve the fractional governing equations. The applicability of the proposed fractional governing equations was investigated numerically in comparison to their conventional counterparts. In practice, cumulative inﬁltration values are observed to deviate from conventional inﬁltration approximation, or the wetting front through time may not be consistent with the traditional estimates of Richards equation. In such cases, fractional governing equations may be a better alternative for mimicking the physical process as they can capture sub-, super-, and normal-diffusive soil water ﬂow processes during inﬁltration.


Introduction
Modeling soil water infiltration is realistically important for several applications in hydrology, meteorology, and environmental sciences since it connects surface and subsurface flow and transport processes. Based on Darcy's law [1] and mass conservation, flow through unsaturated media is described by Richards equation [2] for the one-dimensional water flow in the vertical direction as, ∂θ ∂t Meanwhile, the soil hydraulic head (h) is related to the elevation head (z) and soil capillary pressure head (ψ) by h = ψ(θ) + z, where θ is the volumetric water content, K(θ) is the unsaturated hydraulic conductivity of the soil, h(θ) is the hydraulic head, z is the distance in vertical direction, and t is the time. In the last century, several researchers contributed to the advancement of principles of the governing processes and predictive tools for infiltration dynamics in soils [3][4][5]. Hydraulic functions (e.g., [6][7][8][9]), which relate the hydraulic conductivity to the volumetric water content or to hydraulic head, are needed to solve Equation (1).
Green-Ampt infiltration model and expressed the spatial derivative in the hydraulic flux as a Caputo fractional derivative of the head. Extending the work by Su [13], who studied vertical infiltration into swelling soils by the fractional time derivative, Su [40] presented mass-time and space-time fractional partial differential equations (fPDEs) of water movement for swelling-shrinking soils and non-swelling soils and showed that the fractional infiltration models better fit the field data of Talsma and van der Lelij [41] using a fractional cumulative infiltration model. Su [40] utilized power functions in the fPDEs for diffusivity and hydraulic conductivity which depend on the moisture content.
Recently, starting from Caputo fractional derivative approximation of a function, Kavvas et al. [42] derived the dimensionally consistent equations of continuity and motion for transient soil water flow and soil water flux to obtain the governing equation for transient soil water flow in fractional multidimensional space and fractional time. Closed forms of the governing equations were then obtained from Brooks-Corey relationships [6]. The classical diffusivity expression in the conventional Richards equation for soil water flow depends only on the soil water content, i.e., D = D(θ). Kavvas et al. [42]'s theoretical development not only considers the time-dependent diffusivity that was previously reported [33,35,36] as D = E(θ) t m , but can also address diffusivity that may change in space (see Equation (52) in [42]). Due to fractional powers of the space and time derivatives, the proposed governing equations can capture nonlocal effects in time (time history of the flow) and nonlocal effects in space (long-range dependence in the porous media). In other words, the proposed fractional soil water flow equations have the capacity to account for the effect of the initial conditions on the soil water flow for long periods, and for the effect of the boundary conditions on the flow for long distances by changing the fractional powers [42]. The nonlocality in time can be physically due to mass transfer between relatively immobile and mobile phases, and interaction between segregated regions of high and low permeability [40,43,44]. The nonlocality in space, on other hand, can be physically due to high variation and long spatial autocorrelation of permeability, as is the case of preferential flow paths [40,45,46]. Furthermore, the time and space fractional derivative powers in fractional water movement account for the non-Fickian flow processes, in which the time fractional derivative power corresponds to long time correlations leading to subdiffusive or slow processes, but the space fractional derivative power corresponds to long space correlations leading to super-diffusive processes [40,47]. Lastly, when the powers of fractional time and space derivatives go to unity, the governing equation simplifies to its conventional counterparts.
Although Kavvas et al. [42] provided the theoretical derivation of the new generalized governing equations of soil water flow and flux in fractional differential derivative framework, it lacks a methodology to solve the proposed fractional governing equations and an application that explores the significance of fractional powers and coefficients in the governing equations. Within this framework, the purposes of this study are threefold. Firstly, instead of utilizing Brooks-Corey relationships, empirical expressions which were obtained from least square fits through hydraulic measurements for soil water movements [48] were utilized in this study to show the general applicability of the proposed fractional governing equations of soil water flow and soil water flux with not only Brooks-Corey relationships but also other constitutive hydraulic relations in the literature. Secondly, a numerical solution methodology for the fractional vertical soil water flow was presented to solve the developed theory in Kavvas et al. [42]. Lastly, the capabilities of the proposed fractional differentiation approach for vertical soil water flow were investigated in comparison to conventional governing equations. Sub-, super-, and normal-diffusive soil water flow cases were explored numerically within the framework of Zaslavsky [47]'s definition of the transport exponent.

Fractional Derivatives
Unlike integer order derivatives, the fractional derivative of a function depends on its values over an interval [x a , x] and can therefore capture nonlocal effects. The fractional derivative of a function g(x) for order 0 < β < 1 in Caputo sense is defined as follows [49]: where Γ(.) is the gamma function. When the power β goes to unity, the Caputo derivative of g(x) becomes the same as the integer order derivative [50]. Conventional governing equations are based on integer order derivatives that are local. On other hand, local derivatives within an interval [x a , x] contribute to the fractional derivative for that interval with varying weights as defined in Equation (2), which makes the fractional derivative nonlocal. Since the Caputo fractional derivative in Equation (2) can be written for both space and time derivatives, nonlocality both in space and time can be obtained by changing the fractional power β. A detailed review of the concepts of fractional calculus is provided in [50,51]. The fractional differentiation in the Caputo framework was chosen in the derivation of the governing equations of the fractional soil water flow and flux [42] since physically interpretable initial and boundary conditions can be utilized in the Caputo framework, unlike various other fractional differentiation frameworks.

Fractional Soil Water Flow
A generalized approach that can handle both Fickian and non-Fickian behavior through a fractional differentiation framework for soil water infiltration modeling is practical and quite conceivable. The spatial variability of heterogeneous soils can be quantitatively characterized by fractal dimensions, which can capture the geometric complexity of soils and characterize them with a few numbers [10]. Although the full dynamical processes occurring in soils may not be described by the so-called fractal dimension, the fractal geometry has great potential to be utilized in various ways in soil science [52]. Fractional powers (α and β i in Equation (3) below), in a similar manner to the role of fractals in geometry, can be utilized in a fractional governing equation for the infiltration process, by which one can represent nonlocal effects in time and space.
With this motivation, Kavvas et al. [42] developed the equations of continuity and motion for transient soil water flow and soil water flux and combined them to deduce the governing equation for transient soil water flow in multidimensional fractional soil space and fractional time as: where the fractional soil water flux is: Here, 0 < α, β 1 , β 2 , β 3 < 1 are the fractional orders or powers of space and time fractional derivatives, x = (x 1 , x 2 , x 3 ) is the spatial location, and K i is the hydraulic conductivity in the i-spatial direction (i = 1, 2, 3). When fractional orders of space and time derivatives go to unity, the above fractional governing equation for soil water flow becomes the conventional Richards equation.
For convenience in the rest of the manuscript, we have replaced x 3 with z for the vertical dimension, β 3 with β as the fractional power for space, q 3 with q as the fractional soil water flux, and K 3 with K as the hydraulic conductivity in the vertical direction. Then, the fractional governing equation for the vertical direction is simplified to: and Kavvas et al. [42] then obtained the closed form of the above equations by introducing the Brooks-Corey relationships [6]. In this study, instead of Brooks-Corey relationships, empirical expressions, obtained by least square fits through hydraulic measurements for Yolo light clay and sand (as provided in Haverkamp et al. [48]) were used. Further details are provided in the Numerical Applications section.

Numerical Solution
Following Murio [53], the first-order approximation of Caputo's fractional time derivative of a function f can be approximated by where the time interval [0, T] is divided into N subintervals of equal increment dt = T/N using the nodes t n where n = 0, 1, 2, . . . , N, and f n Dividing the space interval [0, a] into M subintervals of equal increments dz = a/M using the nodes z j where j = 0, 1, 2, . . . , M, the Caputo fractional space derivative at a > 0 can be approximated in second order by [54] D β where 0 < β ≤1 is the fractional order of the space derivative and f z j is the first-order derivative at z j . Utilizing Equation (5), the fractional soil water flow equation given by Equation (5) can be written as (∂z) β can be calculated by Equation (8).

Numerical Applications
Haverkamp et al. [48] compared an implicit numerical solution of Richards equation with Philip's quasi-analytical solution [5,55] for soil water movements in Yolo light clay and sand ( Figure 4 and Table 3 in [48]). Here, we utilize the same infiltration problems to estimate the water front through time to be able to compare the solutions of the conventional Richards approach with the numerical solution of the case β = α = 1 of the proposed fractional vertical soil water flow equations. In other words, Table 3 in Haverkamp et al. [48] is actually used to provide a baseline for the numerical solution of the case β = α= 1 of the proposed fractional vertical soil water flow equations. After verifying that the proposed fractional model works for the standard Richards equation by setting β = α = 1, fractional powers are altered to investigate the corresponding anomalous diffusion behavior within the framework of Zaslavsky [47]'s definition of the transport exponent, which depends on the powers of the space and time fractional derivatives of the governing diffusion equation.
The first example problem for infiltration through Yolo light clay is an experiment that was used by Philip [5,55], for which the hydraulic conductivity and water content relationships are: Ks = 4.428.10 −2 cm/h, and θ s = 0.495, θ r = 0.124 where the subscript s refers to saturation and r refers to residual. The initial and boundary conditions for the first problem are: For the second problem, i.e., the infiltration through sand, the hydraulic conductivity and water content relationships, as provided in Haverkamp et al. [48] are: where K s = 34 cm/h, θ s = 0.287, θ r = 0.075. These relationships are based on experimental measurements [56,57]. The initial and boundary conditions for the second problem are:

Results and Discussion
Utilizing the numerical examples for infiltration through the Yolo light clay and sand [48], one of the purposes of this study was to explore the capabilities of fractional governing equations to simulate vertical soil water flow compared to their integer order conventional counterparts. When the fractional powers of space and time derivatives of the fractional vertical soil water equation become one, the solution should converge to the conventional Richards equation (Equation (1)). Figure 1 depicts the comparison between the water content profiles of the Philip solution [5,55], h-implicit solution [48], and the proposed fractional governing equations when space and time fractional derivative powers are unity, i.e., β = α = 1. The fractional approach, within this framework, when solved numerically by the scheme in Equation (9) Table 3 in [48].
We now explore the effects of the space and time fractional derivative powers on the vertical water movement as proposed in Equation (5) above. Comparisons among water content profiles when space and time fractional derivative powers are fractional are depicted in Figure 2 for Yolo light clay and in Figure 3 for sand. The transport exponent µ quantifies the competing time and space fractional derivative powers or the competing suband super-diffusivity [47]: µ < 1 is for sub-diffusion, µ = 1 for classical or normal diffusion, and µ > 1 is for super-diffusion. The transport exponent is defined as µ = 2β/α where β is the power of time derivative and α is the total power of the space derivative for the diffusive term [47]. To be consistent with our notations, β in [47] is replaced with α, and α in [47] is replaced with 2β so that the transport exponent becomes µ = 2α/2β = α/β in our notation.
For the case when the power of the time fractional derivative is one (α = 1), the wetting front moves down faster as the power of the space fractional derivative decreases from 1 (β < 1 in Figure 2a for Yolo light clay and Figure 3a for sand). The wetting front for the lower moisture content moves down even faster than that of the higher moisture content. In light of the above transport exponent definition, the cases (when β < 1) in Figure 2a for Yolo light clay and Figure 3a for sand correspond to super-diffusion since the transport exponent, µ, is greater than 1 (µ > 1) since α = 1, β < 1. This finding is consistent with [40], who stated that the space fractional derivatives cause super-diffusion, mainly by the flow processes in the media with higher velocity flow paths of long spatial correlation.
For the case of when the power of the space fractional derivative is one (β = 1), the wetting front slows down as the power of the time fractional derivative decreases from 1 (Figure 2b for Yolo light clay and Figure 3b for sand). The wetting front for the higher moisture content slows down even more than that of the lower moisture content. These cases (when α < 1) in Figure 2b for Yolo light clay and Figure 3b for sand correspond to sub-diffusion since the transport exponent, µ, is less than 1 (µ < 1) since α < 1, β = 1. This finding is also consistent with [40], who stated that the time fractional derivatives result in sub-diffusion, mainly by partitioning of water parcels on sticky porous surfaces.
When the powers of time and space fractional derivatives are equal and decrease from one, the effects of time and space fractional powers, as discussed above, are superimposed (Figure 2c for Yolo light clay and Figure 3c for sand). The transport exponent µ = α/β becomes one, showing overall normal diffusion in theory. However, as the time and space fractional powers decrease from 1, the movement of the wetting front for the higher moisture content slows down and that of the wetting front for the lower moisture content moves down faster.  Table 3 in [48].   For the case when the power of the time fractional derivative is one (α = 1), the wetting front moves down faster as the power of the space fractional derivative decreases from 1 (β < 1 in Figure 2a for Yolo light clay and Figure 3a for sand). The wetting front for the lower moisture content moves down even faster than that of the higher moisture content. In light of the above transport exponent definition, the cases (when β < 1) in Figure  2a for Yolo light clay and Figure 3a for sand correspond to super-diffusion since the transport exponent, µ, is greater than 1 (µ > 1) since α = 1, β < 1. This finding is consistent with [40], who stated that the space fractional derivatives cause super-diffusion, mainly For the case when the power of the time fractional derivative is 0.7 (α = 0.7), the wetting front moves down faster as the power of the space fractional derivative decreases from 1 (β < 1 in Figure 2d for Yolo light clay). The wetting front for the lower moisture content moves down even faster than that of the higher moisture content. On other hand, for the case when the power of the space fractional derivative is 0.7 (β = 0.7), the wetting front slows down as the power of the time fractional derivative decreases from 1 (Figure 3d for sand). The wetting front for the higher moisture content slows down even more than that of the lower moisture content.
After investigating water content profiles through time, we now examine the performances of cumulative infiltration approximations by a conventional approach and a fractional approach. As opposed to Philip [58]'s approximate description of the cumulative infiltration for conventional vertical infiltration into rigid media I = At + St 0.5 (22) where A is the final infiltration rate and S is the sorptivity. On other hand, Su [40] conjectured a cumulative infiltration model for water movement in soils as where 0 < α ≤2 is the order of the time fractional derivative, and 0 < λ ≤ 2 is the order of the mass (for swelling soils) or the space (for non-swelling soils) fractional derivative of the diffusion term. When Equation (23) was fitted to the field data of cumulative infiltration versus time as reported by Figure 1 in [41], parameters α, λ, A, and S were determined by Su [40] as α = 0.3445, λ = 1.9523, A = 1.30 mm/day, and S = 48.64 mm/day α/(2λ−1) . Compared to Equation (22) for the conventional vertical infiltration (see Figure 1b in [13]), Equation (23) provides a considerably better fit to the field data of [41], confirming the superior performance of fractional cumulative infiltration models. The transport exponent for field data of [41], µ = 0.353, corresponds to sub-diffusion [40]. Lastly, the cumulative infiltration values were calculated using the fractional model for eight different time instances for Yolo light clay (t = 10 5 , 2.5 × 10 5 , 5 × 10 5 , 10 6 , 1.5 × 10 6 , 2 × 10 6 , 2.5 × 10 6 , 3 × 10 6 s) and for sand (t = 0.1, 0.2, 0.3, . . . , 0.8 h) when α = β = 0.9, 0.8, 0.7. The conventional and fractional cumulative infiltration approximations in Equations (22) and (23) require the estimation of two parameters, the final infiltration rate A, and the sorptivity S. Using the cumulative infiltration estimates of the first and last time instances, A and S can be calculated for the two models and the performance of both models can be evaluated based on the cumulative infiltration approximations for the remaining six time instances. Tables 1 and 2 provide the percent difference between the approximated and simulated cumulative infiltration values (I(simulated) − I(approximated))/I(simulated) for Yolo light clay and sand, respectively. Since the parameters A and S are estimated based on the first and last time instances, approximations at these times are same as the simulated values, i.e., the percent difference values are 0. The percent difference values are less than 2.5% for Yolo light clay and 0.5% for sand when the fractional cumulative infiltration approximation (Equation (23)) is used. The percent difference values are less than 8% for Yolo light clay and 3% for sand when the conventional cumulative infiltration approximation (Equation (22)) is used. Hence, the fractional cumulative infiltration approximation (Equation (23)) performed better than the conventional cumulative infiltration model (Equation (22)) when the powers of time and space fractional derivatives are fractional.  In practice, as in the case of [41], the observed cumulative infiltration values can deviate from the conventional cumulative infiltration approximation (Equation (22)), or the wetting front through time may not be consistent with the traditional estimates of Richards equation. In such cases the fractional governing equations may be a better alternative to mimic the physical process, and the fractional approximation (Equation (23)) can be a better alternative for the estimation of the cumulative infiltration.

Conclusions
In this study, empirical expressions obtained by least square fits through hydraulic measurements for soil water movements (Haverkamp et al. [48]) were utilized to show the general applicability of the proposed fractional governing equations of soil water flow and soil water flux [42] with not only Brooks-Corey relationships but also other constitutive hydraulic relations in the literature. A numerical solution methodology for the fractional vertical soil water flow was presented to solve the fractional governing equations of vertical soil water flow. Finally, the modeling capabilities of the fractional governing equations of vertical soil water flow were investigated numerically for infiltration problems through Yolo light clay and sand. It was demonstrated for both flow media that when the powers of space and time fractional derivatives are one, the fractional approach provides solutions that are same as the conventional Richards equation approach. Furthermore, the numerical investigation demonstrated that: (a) When the power of the time fractional derivative is one, the wetting front moves down faster (super-diffusive behavior) as the power of the space fractional derivative decreases from 1. The wetting front for the lower moisture content moves down even faster than that corresponding to the higher moisture content. (b) When the power of the space fractional derivative is one, the wetting front slows down (sub-diffusive behavior) as the power of the time fractional derivative decreases from one. Additionally, the wetting front for the higher moisture content slows down even more than that of the lower moisture content. (c) When the powers of time and space fractional derivatives are equal, the effects of time and space fractional powers are superimposed. The transport exponent µ becomes one, showing overall normal diffusion in theory. However, as the time and space fractional powers decrease from one to zero, the wetting front for the higher moisture content slows down (sub-diffusive behavior) while the wetting front for the lower moisture content moves down faster (super-diffusive behavior). To our knowledge, this combined sub-and super-diffusive behavior with a resultant normal diffusion has been reported for the first time and should be investigated further in future studies. (d) The fractional cumulative infiltration approximation (Equation (23)) performs better than the conventional cumulative infiltration model (Equation (22)) when the powers of time and space fractional derivatives are fractional.
To conclude, as in the case of [41], the observed cumulative infiltration values in practice can deviate from the conventional cumulative infiltration approximation. In such cases where the wetting front through time may not be consistent with the traditional estimates of the Richards equation, the fractional governing equations may be a better alternative to mimic the physical process of vertical soil water flow, and the fractional approximation can be a better alternative for the estimation of the cumulative infiltration. Although the numerical study herein provides helpful insights to the capabilities of the proposed fractional approach in terms of the fractional powers of the space and time derivatives, further research is needed to combine experimental and field studies with the proposed theory of fractional soil water movement.