1. Introduction
Over the past century, the knowledge of velocity vertical distribution in an open channel cross-section received great attention by several research teams due to its importance in the understanding of numerous hydraulic phenomena such as flood control, pollutant dispersion, and sediment transport, and its various applications in the design and planning of hydropower plants and structures or infrastructures interacting with fluid flows [
1,
2,
3,
4,
5,
6,
7,
8]. The sophisticated numerical modeling and advanced experimental techniques allowed for obtaining a good reconstruction of velocity profiles in fully-developed turbulent wide open channel flows, though they have yet not been able to well describe the dip of the maximum velocity below the free surface in narrow open channels due to the presence of secondary currents [
9,
10,
11,
12,
13]. In the latter conditions, in fact, the classical log or power laws, which depict the velocity increasing monotonically with the distance from the bottom, deviate from the experimental results and fail to predict the dip phenomenon [
14]. Coles [
15] was among the pioneers to investigate this deviation, adding a purely empirical correction term, called wake function, to the log law. Later, some researchers tried to adapt Coles’ wake function to experimental data, suggesting different empirical values to estimate the parameters within this law [
10,
11,
16,
17,
18]. In particular, Nezu and Rodi [
10] found Coles’ profile parameter
II to be between 0 and 0.20 by analyzing the longitudinal and vertical velocity components in two-dimensional, fully developed open channel flows over smooth beds. Cardoso et al. [
11] claimed that the wake strength was dependent on the secondary currents, flow history, and inactive turbulence components and yielded a
II value equal to −0.077 for hydraulically smooth conditions. By studying the influence of the suspended sediment on the shape of the velocity distributions, Coleman [
16] obtained an average value of
II equal to 0.19. For fully developed, rectangular, subcritical open channel flows on smooth beds, Kirkgoz [
17] reported a constant value of this parameter of approximately 0.1, considerably smaller than the one given by Coles (
II = 0.55). Kironoto and Graf [
18] showed how the profile parameter was affected by the aspect ratio
Ar = b/D (channel width/flow depth) and the boundary conditions, finding a value of 0.09 for rough plate data with
Ar > 5 and −0.03 for gravel bed data with
Ar < 5.
Other researchers proposed various improvements to the log and wake laws to better predict the velocity-dip phenomenon in open channels flows [
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31].
In 2000, Sarma et al. [
19] introduced a binary velocity profile model, which combines the logarithmic law of the inner region with the parabolic law of the outer region for subcritical and supercritical flows in smooth and rough channels. They found that the junction point of the two curves depended on the width and aspect ratios of the cross-section and was 0.5
D (where
D is the water depth) when there was no dip phenomenon and decreased from 0.5
D to 0 in the presence of the dip phenomenon.
Wang et al. [
20] demonstrated the validity of log and wake laws also for sediment-laden flows and they applied a regression analysis to study the influence of the main factors, such as the Karman constant and the profile parameter, on the vertical velocities. By comparing various experimental literature data, they observed how the maximum velocity was affected by the aspect ratio,
Ar, and, in particular, it was located in the central portion and below the free surface for narrow channels, while it moved towards the sidewall region and near the free surface for wider channels.
A modified log-wake law (MLW-law) was developed by [
21,
22,
23] by introducing a cubic function that well represented the deviation of the experimental data in pipes and boundary layers where a zero-velocity gradient exists at the maximum velocity. Later, the same authors successfully extended the application of the modified law to turbulent open channel flows of the laboratory and field, although it could not be a universal model due to the presence of empirical parameters [
22,
24].
Yang et al. [
25], instead, derived the dip-modified log law (DML-Law) from the analysis of the Reynolds-averaged Navier–Stokes (RANS) equations. This law had the great advantage of containing the unique parameter α for dip correction, even though it was able to accurately reconstruct the velocities along the verticals from the central line to the sidewalls only in uniform open channel flows with smooth boundaries. The extension to flow regimes with rough walls was performed by Absi, who first introduced a simple dip-modified log-wake law in 2009 [
26] and then a full dip-modified log-wake law (fDMLW-law) in 2011, both based on a log-wake-modified eddy viscosity distribution [
27]. The fDMLW-law was further modified by [
28] in 2012 in order to estimate the velocity profile throughout the depth of the channel without relying on numerical integration.
At the same time, Wang and Cheng [
29] proposed a modified streamwise profile velocity similar to the log-wake law but varying periodically in the lateral direction and based on the assumption of zero turbulent shear stress at the maximum velocity location to describe the secondary flows artificially generated with alternate rough and smooth channel bed strips.
On the basis of experimental observations, Bonakdari et al. [
30] developed a sigmoid model to describe the dip phenomenon in the outer region of smooth narrow as well as wide channels, which was a function of the
Ar. Although this model did not overestimate the measured velocities in the central cross-section like those of [
20,
25], it did not satisfy the asymptotic boundary conditions in which the maximum velocity occurred at 0.5
D for
Ar→0 and at 1
D for
Ar→∞. This is due to the fact that the model was obtained to fit the experimental data of ovoid-shaped sewers [
31,
32].
Later, Lassabatere et al. [
33] derived a new law for the streamwise velocity profile in the outer region of the central section of open channels by integrating the Reynolds-averaged Navier–Stokes equations with the analytical modeling of the vertical component and by adopting a negligible viscosity. This law, tested on several experimental datasets that include rough and smooth flow regimes, satisfied both the asymptotic boundary conditions, unlike the previous models.
In recent decades, apart from the log law and the log-wake law, the power law was proposed to describe the velocity distribution in most pipe flows, boundary layers and wall jets, especially in the overlapping region of the inner law (i.e., the law of the wall) and the outer law (i.e., the velocity defect law) [
34]. Already in the late fifties, various researchers [
35,
36,
37] had shown how the power velocity profile was preferable to the logarithmic velocity distribution. In particular, it could be applied to different flow regimes and Reynolds numbers [
34], it seemed better at incorporating the effects of sediment without singularities near the bed and/or discontinuities at the axes and planes of symmetry [
34,
38], and it better agreed with the pipe flow data [
37,
39].
In 2007, Cheng [
40] derived the power law as a first order approximation to the log law and was able to reconstruct the velocity profile in the overlap layer between the inner and outer regions of open channel flows. He found that its exponent was a function of the Reynolds number, the relative roughness height, and the friction factor, and that it exceeded the value of 1/6 in the presence of large-scale boundary roughness, as already observed by [
34].
Afzal et al. [
41,
42] proposed the envelope of the skin friction power law to study the velocity distribution in fully developed turbulent pipe and channel flows. Although the results of this model were not very different from the ones obtained with the log law for high Reynolds numbers, they were able to better reconstruct the velocity profiles for low Reynolds numbers.
In 2009, Castro-Orgaz [
43] derived an analytical approach by integrating the von Karman momentum equation using a power velocity distribution inside the boundary layer flow in the turbulent rough regime. The analytical solutions proved to be a close fit to the experimental data of flows over uncontrolled spillway crests followed by a steep chute of constant slope.
The aforementioned deterministic approaches based on theoretical predictions or empirical evaluations of variables and parameters are often questionable and do not account for the randomness included in time-averaged streamwise velocities. To overcome these limitations, the informational theory, using Shannon’s entropy [
44] and the principle of maximum entropy (POME) [
45,
46,
47], was applied first by Chiu [
48] in order to discuss the uncertainties associated with the flow field. This theory was also used to study different hydraulic phenomena, such as the distribution of the boundary shear stresses [
49,
50,
51,
52] and suspended sediment concentrations [
53,
54,
55,
56,
57] in open channel flows. The entropy law has the advantages of satisfactorily predicting the velocity near the bed and of relying on the estimation of a single discriminating parameter, depending on the ratio between the mean and maximum velocity of the investigated cross-section. This parameter seems to keep constant over the entire section despite the varying water discharge in [
58,
59,
60,
61] and over the entire reach for rivers with the same morphological characteristics in [
62]. As a consequence of its application, such uniformity induces the simplification of the numerical modeling and the reduction of the sampling time during the water discharge measurement in rivers by predicting the value of the mean velocity from the knowledge of the maximum velocity only [
61]. However, the entropy model deviates from the near-free-surface velocities in narrow open channels when the maximum velocity occurs below the water surface due to the presence of secondary currents and wall effects, as widely observed for the classical logarithmic and power laws [
63,
64,
65]. The present paper proposes a new dip-modified entropy law in order to reconstruct the entire velocity profile and predict the dip phenomenon for low values of the aspect ratio. This law was first validated using measurements collected in this research on a rectangular laboratory channel, in conditions of steady flow and smooth boundaries and for different water discharges, bottom slopes, and aspect ratios, and was then tested on a set of literature data. A detailed error analysis was applied on the observed and calculated velocities to demonstrate the good performance of the modified model compared with the original one.
4. Discussion of Results
The reliability of the proposed model is demonstrated by the comparison of the velocities computed through the entropy wake law and original entropy law with the ones observed in the open channel cross-sections for each value of water discharge, aspect ratio, and bottom slope (
Table 2), reported in
Figure 5.
As one can see from
Figure 5, high values of the determination coefficient and of Pearson’s correlation coefficient underline a perfectly positive linear relationship between the calculated and the observed data and less error variance in the latter specified by the proposed model. The velocities are reproduced fairly well by the proposed law for the investigated range of water discharges, bottom slopes, and aspect ratios, showing an error within ±10% for the 99.5% of the observed velocities, while in the case of the classical model, only 66.3% of the observed velocities fall within the 10% error band.
In addition, a detailed error analysis using different statistical indices, such as the root mean square error (RMSE), the RMSE observations standard deviation ratio (RSR), the mean absolute error (MAE), the percentage of bias (PBIAS), and the Nash–Sutcliffe efficiency (NSE), was considered to quantify the good performance of the new law [
73]:
where
ucom and
uobs are the computed and observed time-averaged velocities along the central vertical of the investigated cross-section, respectively.
The RMSE and the MAE, depending on the presence of outliers and the shape of the error distribution, have the advantage of showing the difference between the predicted and the observed values in the same units. Values of the RMSE and MAE close to zero highlight higher model accuracy. The RMSE can be substituted by its standardized version, RSR, which includes the standard deviation of the measured data as the scaling/normalization factor, and still incorporates the benefits of the RMSE itself [
74].
The PBIAS measures the average tendency of the calculated data to be higher or lower than the observed ones [
75]. The null value of PBIAS indicates an accurate model simulation, while its positive and negative values indicate a model over- and underestimation, respectively. The NSE is a dimensionless technique, which determines the relative magnitude of the residual variance when compared with the observed data variance [
76]. The values of the NSE range from −∞ to 1.0, with NSE = 1 as the optimal value. When the NSE is between 0 and 1, the levels of performance are acceptable, while with the NSE lower or equal to 0, the performance is unacceptable. These last three indices allow assessing the accuracy of the proposed model according to the four categories defined in
Table 3 [
77].
Table 4 shows the comparison between the suggested statistics of the velocities obtained from the entropy wake law and the ones evaluated through the classical entropy law. As seen in the table, the model performance goes from satisfactory to very good. This underlies how the new formula provides a very accurate and reliable estimation of the velocities along the entire vertical from the bottom to the free surface.
A further validation of the proposed model was carried out using the literature data of [
16,
78], and [
79]. The authors collected point streamwise velocities both with clear water and with sediment-laden flows in rectangular flumes with varying slope, water discharge, flow depth, and aspect ratio in order to analyze the influence of the suspended sediment on the shape of the velocity profiles. In this work, only the runs conducted in the condition of clear water are used (
Table 5). Coleman [
16] measured the velocity distributions by a Pitot-static tube, while Lyn [
78] as well as Muste and Patel [
79] used laser Doppler velocimetry.
Table 6 underlines the very good performance of the proposed new law also for the literature data, while the classical entropy model is often unsatisfactory.
Figure 6 and
Figure 7 report the predicted velocity profiles through the entropy wake and classical entropy laws with the experimental data collected in this research and in the literature. As noted in the figures, although both models agree very well with the velocities near the channel bed, the original entropy model is not able to describe the flow field in the outer region close to the free surface. In particular, the latter tends to overestimate the measured data and this deviation increases with a decreasing aspect ratio. This condition could be especially due to a greater effect of the secondary currents and the sidewalls on the streamwise velocity profile.
The proposed entropy wake law, instead, thanks to the presence of additional terms, reproduces perfectly the dip phenomenon and gives a good description of the velocity distribution over the entire water column, including the inner and outer regions. Only a light underestimation of the observed velocities was noted for lower values of the
Ar. This suggests investigating the dependence of the coefficient
α on the aspect ratio. Therefore,
Figure 8 reports the relationship between
α and
Ar based on the experimental data. It is possible to see from the figure that the data interpolation shows how
α could be a quadratic function of the aspect ratio equal to
Figure 8 displays also how the coefficient
α tends to decrease for the condition of a wider channel when the dip phenomenon disappears, while it increases when the influence of the secondary flows is strong. However, the effect of the secondary currents on the value of the coefficient
α cannot be described only by the aspect ratio. Therefore, a more detailed investigation of flow conditions and channel characteristics is needed to validate the obtained empirical relationship.
5. Conclusions
A new wake law for the estimation of streamwise velocity profiles in steady open channel flows was derived from Shannon’s informational theory together with the principle of maximum entropy. The proposed law includes three additional terms: the first similar to Coles’ function; the second linearly proportional to the logarithmic distance from the free surface; and the third depending on the cubic correction near the maximum velocity. Although the entropy wake law presents a more complex equation compared with that of the classical entropy profile, it depends on a single coefficient, α, which seems to keep constant in all three terms.
A set of laboratory measurements was carried out in a rectangular straight flume, with a smooth bed and sidewalls, for different values of water discharge, aspect ratio, and bottom slope, in order to validate the developed formula. A detailed error analysis demonstrated the very good performance of the modified entropy law in predicting the flow field over the entire water column, including the inner and outer regions. In addition, a further validation of the proposed model was carried out using the literature data and it confirmed the high accuracy of the new law.
The comparison with the streamwise velocity profiles calculated through the classical entropy model showed how the velocity distributions obtained from the proposed formula are able to reproduce the dip phenomenon and the velocity negative gradient near the free surface, as well as to match the experimental and literature data perfectly. Actually, a light underestimation of the entropy wake law was observed for low values of the aspect ratio, which led to examine the dependence of the coefficient α on the Ar. Such a study, although still in a preliminary phase, demonstrated that the presence of secondary currents and the effects of sidewalls affect the parameter α. However, a deeper analysis of this parameter is needed in the future. The high precision of the new model in describing correctly the streamwise velocity distributions of a smooth and rectangular cross-section suggests that it could also be used to investigate flows in open channels with different shapes and boundary roughness conditions.