Stoch Environ Res Risk Assess (2015) 29:1–11 DOI 10.1007/s00477-014-0959-3 ORIGINAL PAPER Comparison between Shannon and Tsallis entropies for prediction of shear stress distribution in open channels Hossein Bonakdari • Zohreh Sheikh Mohtaram Tooshmalani • Published online: 21 September 2014 Ó Springer-Verlag Berlin Heidelberg 2014 Abstract The concept of Tsallis entropy was applied to model the probability distribution functions for the shear stress magnitudes in circular channels (with filling ratios of 0.506, 0.666, 0.826), circular with flat bed (filling ratios of 0.333, 0.666), rectangular channel (1.34, 2, 3.94, 7.37 aspect ratios) and compound channel (with relative depths of 0.324, 0.46). The equation for the shear stress distribution was obtained according to the entropy maximization principle, and is able to estimate the shear stress distribution as much on the walls as the channel bed. The approach is also compared with the predictions obtained based on the Shannon entropy concept. By comparing the two prediction models, this study highlights the application of Tsallis entropy to estimate the shear stress distribution of open channels. Although the results of the two models are similar in the circular cross-section, the differences between them are more significant in circular with flat bed and rectangular channels. For a wide range of filling ratio values, experimental data are used to illustrate the accuracy and reliability of the proposed model. Keywords Bed Entropy Open channel Sediment Wall H. Bonakdari (&) Z. Sheikh M. Tooshmalani Department of Civil Engineering, Faculty of Engineering, Razi University, Kermanshah, Iran e-mail: bonakdari@yahoo.com H. Bonakdari Z. Sheikh M. Tooshmalani Water and Wastewater Research Center, Razi University, Kermanshah, Iran 1 Introduction Depending on the cross section shape (Knight et al. 1994), bed roughness (Flintham and Carling 1988) and hydraulics of flow (Ghosh and Roy 1970), the boundary shear stress in any channel cross section is one of the main factors to be considered in different open channel studies such as that caused by flow resistance (Rhodes and Knight 1994), velocity distribution (Tominaga et al. 1989; Wilcock 1996), sediment transport rate (Chien and Wan 1999), channel erosion or deposition bed erosion (Julien 1995), and morphological and geometrical changes in rivers (Khodashenas and Paquier 2002). A considerable number of experimental works carried out in open channel have demonstrated that it is difficult to determine boundary shear stress distribution in an open channel (Lane 1953; Cruff 1965; Myers 1978; Yuen 1989; Olivero et al. 1999; Galip et al. 2006; Lashkar-Ara and Fathi-Moghadam 2009; Kabiri-Samani et al. 2013). To cope with this difficulty, empirical, analytical or simplified computational methods have been carried out by several researchers such as Yang and Lim (1997), Khodashenas and Paquier (1999), Berlamont et al. (2003), Galip et al. (2006), Guo and Julien (2005), Bonakdari et al. (2008), Yang (2010), Bonakdari and Tooshmalani (2010), Javid and Mohammadi (2012) and Huai et al. (2013). Still, despite using sophisticated turbulence models, accurate calculation of the local shear stress value is a difficult task. Depending on Shannon entropy (1948) as essential reference, Chiu (1987) proposed a new approach introducing the maximum entropy and probability concepts in hydraulics to develop new equations for the distribution of both velocity and shear stress in open channels. Based on Chiu’s work, (1991) and other researchers like Cao and Knight (1996), Araujo and Chaudhry (1998) and Sterling 123 2 and Knight (2002) used the Shannon entropy concept to predict shear stress distribution in open channels. Sterling and Knight (2002) developed a new approach to predict the distribution of boundary shear stress in a circular open channel. However, their study showed limitations in reflecting the hydraulic behavior of open channels and was somewhat unsatisfactory in terms of reliability. This might be attributed to the complexity of the model parameters assumption and high sensitivity of the model results to the parameters estimated. Besides Shannon entropy, Tsallis (1988) proposed a generalization of the celebrated Boltzmann–Gibbs entropic measure, which is one of a family of functions for quantifying the diversity, uncertainty or randomness of a system. The Tsallis entropy is a generalization of the Shannon entropy containing an additional parameter which can be used to make it less susceptible to the shape of the probability distribution (Maszczyk and Dush 2008). Singh and Luo (2011) make use of the Tsallis entropy to predict velocity distribution in open channels. The main objective of the present study is to introduce a new method which results from maximizing the Tsallis entropy for predicting the shear stress distribution. A short introduction to the suggested model is given and this relation is used to predict the boundary shear stress in circular, circular with flat bed, rectangular and compound channels. These results are compared with the experimental results to validate this new equation. A comprehensible comparison is made between this suggested equation and the model suggested by Sterling and Knight (2002) that resulted from the Shannon entropy. 2 Shear stress distribution model 2.1 Shannon entropy Sterling and Knight (2002) used the Lagrange coefficient to maximize the Shannon entropy and introduce an equation to predict shear stress as follows: y 1 s¼ 1 þ eksmax 1 ; ð1Þ k L where s is the local boundary shear stress, smax the maximum value of boundary shear stress, y the transverse coordinate, L half the wetted perimeter and the Lagrange multipliers, k, can be estimated as: 1 smax eksmax k ¼ ks qgRS ; ð2Þ e max 1 where q is the mass density, g the gravitational acceleration, R the hydraulic radius and S the bed slope of channel. This equation was only used in the circular cross section. 123 Stoch Environ Res Risk Assess (2015) 29:1–11 Fig. 1 Cross section of a circular open channel and its notation The wall and bed relationships in a circular channel with a flat bed, respectively, must be defined as follows: 2ðy yw Þ 1 sw ¼ 1 þ ekw smaxðwÞ 1 yw \y\Pw =2; kw Pw ð3Þ ks 2ðy yw Þ 1 b maxðbÞ 1þ e 1 sb ¼ kb Pb Pw =2\y\Pb =2 þ yw ; ð4Þ where kw and kb are the Lagrange multipliers corresponding to the wall and bed, respectively, Pw and Pb, the wetted perimeter corresponding to the wall and bed of channel, respectively. Figure 1 shows the channel cross section and the notation considered for a circular conduit with a flat bed, where h is the flow height, t the height of sediment in channel bed, Pb the bed wet area, Pw the total wet area on the channel wall and D the channel diameter. To make use of Eqs. (3), (4) and also Eq. (1) the average and maximum shear stress are to be estimated first. Knight (1981) had conducted their tests in a 15 m flume, 0.46 m wide with a constant bed slope of 9.58 9 10-4. The percentage of the shear force carried by the walls or bed was calculated for different aspect ratio (b/h = 1.5, 2, 3, 5, 7, 7.5, 10 and 15). For each depth and roughness the shear stress was measured by a Preston tube or by semi-logarithmic plotting of the velocity profiles. Their proposed empirical method to compute these shear stress values on the bed and wall is as follows [Eqs. (5)–(8)]: smeanðwÞ ¼ 0:01 %SFw ð1 þ Pb =Pw Þ; ð5Þ qgRS smeanðbÞ 1 ¼ ð1 0:01 %SFw Þ 1 þ ; ð6Þ Pb =Pw qgRS Stoch Environ Res Risk Assess (2015) 29:1–11 h i smaxðwÞ ¼ 0:01 %SFw 2:0372ðPb =Pw Þ0:7108 ; qgRS h i smaxðbÞ ¼ ð1 0:01 %SFw Þ 2:1697ðPb =Pw Þ0:3287 ; qgRS 3 ð7Þ ð8Þ where smean(w) and smean(b) are the average and smax(w) and smax(b) the maximum shear stress at the channel wall and bed, respectively and Pb and Pw the wetted perimeters corresponding to the bed and wall, respectively. Knight et al. (1994) proposed an empirical relationship for the prediction of the shear force percentage carried by the walls, as follows [Eq. (9)]: %SFw ¼ Csf expð3:23 logðPb =C2 Pw þ 1Þ þ 4:6052Þ; ð9Þ where Csf = 1 for Pb/Pw \ 6.546, otherwise Csf = 0.5857 (Pb/Pw)0.28471 and in a subcritical flow C2 = 1.5. Thereby, depending on the water depth and the bed slope, the transverse distribution of shear stress can be evaluated for any given channel. 2.2 Tsallis entropy wet area and y is the channel wall path changing from 0 to L. The two other parameters are k0 and k2, which can be evaluated by the following equations (see Appendices 1, 2): h 0 ik 0 k þ k2 smax ½k k ¼ k2 kk ; ð15Þ h 0 ik h 0 ikþ1 0 smax ðk þ 1Þk2 k þ k2 smax k þ k2 smax þ½k kþ1 ¼ ðk þ 1Þk22 kk smean ; ð16Þ where smax is maximum shear stress on the channel wall. According to the Eq. (14), the shear stress on the channel wall and bed can be estimated as follows: 1 0 1 0 k k2w yw k k kw sw ¼ k kw þ ; ð17Þ k2w Pw =2 k2w 1 0 1 0 k k2b yb k k kb sb ¼ kb þ k : ð18Þ k2b Pb k2b 0 Considering a consistent variable s, and based on the Tsallis (1988) idea, the entropy equation is defined as: Z smax 1 HðsÞ ¼ f ðsÞ 1 f ðsÞq1 ds; ð10Þ q1 0 where q is a real number, H(s) a function of Tsallis entropy and f(s) the probability density function. Chiu (1987) and Cao’s (1995) methods of maximizing Lagrange coefficient were used to estimate the probability density f(s), as follows: o 1 f ðsÞ 1 f ðsÞq1 of ðsÞ q 1 ð11Þ o½f ðsÞ o½sf ðsÞ þ k2 ¼ 0: þ k1 of ðsÞ of ðsÞ Simplifying the above equation, f(s) is defined as: 1 q1 q 1 0 f ðsÞ ¼ k þ k2 s ; ð12Þ q 0 1 þ k1 : The cumulative distribution funcin which k ¼ q1 tion of the shear stress is introduced by Eq. (13): Z s FðsÞ ¼ f ðsÞds ¼ y=L: ð13Þ 0 Inserting Eq. (12) into Eq. (13), thus solving the integral and simplifying it, the shear stress function becomes: 2 31k !k 0 0 k 4 k k2 y 5 k þ ; ð14Þ s¼ k2 L k k2 q and q is an exponential parameter in Tsallis where k ¼ q1 relation and has real quantities, L is constant and equals the 0 To calculate kw ; kb ; k2w and k2b, Eqs. (15) and (16) were used. Also, Knight et al. (1994) used Eqs. (5) and (8) to estimate the maximum and average shear stress, smax and smean, respectively, in Eqs. (15) and (16). It should be taken into consideration that the wall average, Eq. (5), and 0 maximum shear stress, Eq. (7), were used to calculate kw 0 and k2w, respectively, and for calculating kb and k2b the bed average, Eq. (6), and maximum shear stress in bed, Eq. (8), respectively, were used. The results of the comparison between the equations [Eqs. (17), (18)] proposed in this study and the extant previous equations [Eqs. (3), (4)] are presented herein using the criteria for root mean square of error (RMSE) and the relative error (RE) as defined below: n X siðpredictedÞ siðmeasuredÞ ; ð19Þ RE ¼ 100 siðmeasuredÞ i¼1 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u n u1 X siðpredictedÞ siðmeasuredÞ 2 RMSE ¼ 100 t ; n i¼1 siðmeasuredÞ ð20Þ where n is the total number of nodes. Considering the error percent, we concluded that both methods show a good performance in predicting the shear stress distribution in a circular channel. 3 Experimental data Knight and Sterling (2000) used the Preston pipe technique to measure shear stress distribution in different depths of a circular channel with a diameter of 244 mm and a wall 123 4 Stoch Environ Res Risk Assess (2015) 29:1–11 Table 1 Summary of main hydraulic parameters in circular channel and circular channel with flat bed, Knight and Sterling (2000) t/D (1) h (mm) (2) (h ? t)/D (3) Pb/Pw (4) S0 (5) Q (l/s) (6) s0 (N/m2) (8) F (9) 0.394 0.4411 0.516 R (9104) (10) 0 81.3 0.333 0 0.001 0 123.5 0.506 0 0.001 11.7 0.493 0.5971 0.505 11.0 0 162.6 0.666 0 0.001 17.3 0.524 0.6897 0.441 13.5 0 201.5 0.826 0 0.001 22.9 0.554 0.7209 0.375 15.0 0.25 20.3 0.333 4.7 0.00862 0.750 1.4905 1.710 4.9 0.25 101.5 0.666 1 0.00862 1.625 4.8040 1.590 34.2 38.9 6.49 U (m/s) s0 (N/m2) Available data for local shear stress References 7.58 0.187 0.084 Bed Tominaga et al. (1989) 15.14 0.192 0.7 Bed and walls Tominaga et al. (1989) 3.9 7 0.338 0.405 0.36 0.164 Bed Walls Knight et al. (1984) Knight et al. (1984) 3.91 18.38 0.495 0.352 Bed Knight et al. (1984) 7.73 22.34 0.464 0.456 Bed Knight et al. (1984) b (cm) h (cm) b/h 40 10.15 3.94 40 19.90 2.01 15.2 15.2 7.6 11.3 2 1.34 38.1 9.75 61.0 7.89 Q (l/s) thickness of 3 mm. Making a sediment layer with the thickness of t over the bed of a circular channel resulted in a circular conduit with a flat bed. This was used for measuring the different ratios of t/D shear stress distribution in different depths. A summary of hydraulic parameters in circular channel and circular channel with flat bed is shown in Table 1, where S0 is the slope of channel length, Q the discharge in channel, U the average velocity, F the Froude number and R the hydraulic radius. The experimental results acquired from the study of Tominaga et al. (1989) and Knight et al. (1984) were made use of in the present study. The most important flow conditions and shear stress measurements are tabulated in Table 2, namely (b = channel width, h = depth of flow, b/h = aspect ratio, Q = discharge, U = mean velocity, smean = average shear stress, smean(w) = average shear stress at wall, and smean(b) = average shear stress at bed). 4 Analysis of results According to its description, the Tsallis entropy becomes a convex function when q [ 0. Thus, q should stay positive to maintain the entropy function convex, so as to achieve the maximum entropy. Based on Singh and Luo (2011) proposition and using both field and experimental data, the feasible range of q proposed was from 0 to 2. A larger q value leads to more difficulty in solving the equations, and the solutions of k values for larger q values are not as stable as for smaller ones. For different values of q, 123 3.39 1.4 1.2 1 τ/ρgRS Table 2 Summary of experimental data of Tominaga et al. (1989) and Knight et al. (1984) 5.36 U (m/s) (7) 0.8 0.6 0.4 0.2 q=1/3 q=2/3 q=3/4 q=4/5 q=5/4 q=3/2 q=2 Experimental results 0 0 0.2 0.4 0.6 0.8 1 y/P Fig. 2 Variation of shear stress distribution with q in circular channel for (h ? t)/D = 0.333, t = 0 different values of Lagrange multipliers are found. Also, the values of these parameters vary significantly with those of q. The shear stress distribution in a circular channel for (h ? t)/D = 0.333, t = 0 was calculated for the values of q = 1/3, 2/3, 3/4, 4/5, 5/4, 3/2, 2, as shown in the Fig. 2. As can be seen in the figure, for q = 5/4, 3/2, the calculated shear stress distribution is different from the experimental results. For the value of q less than 1, the results obtained are closer to the experimental outcomes than for q [ 1. The values of q = 2/3, 3/4, 4/5 deliver the same results and very good estimation of the shear stress. The smallest difference between estimated shear stress and experimental results obtained for these values of q. Table 3 illustrates the RE and RMSE between the introduced model Stoch Environ Res Risk Assess (2015) 29:1–11 5 Table 3 RE and RMSE between the introduced model and the experimental results for different values of q in circular and circular with flat bed cross section q = 1/3 RMSE 4.37 6.46 Circular (h ? t)/D = 0.666, t = 0 Circular (h ? t)/D = 0.826, t = 0 3.96 4.33 Circular (h ? t)/D = 0.333, t = 0.25 Circular (h ? t)/D = 0.666, t = 0.25 RE 6.6 RMSE q = 3/4 RE RMSE q = 4/5 RE RMSE q = 5/4 RE RMSE q = 3/2 RE RMSE q=2 RE RMSE RE 2.31 2.8 2.3 2.8 2.3 2.8 4.02 5 15.27 15 8.08 8.8 3.5 4.4 3.5 4.4 3.5 4.4 3.78 4.4 17.96 18.9 8.97 9.4 6.8 6.3 3.22 3.91 5.6 7.1 2.4 2.6 3.2 3.9 2.4 2.6 3.2 3.9 4.2 12.84 5.6 17.3 14.34 22.13 15.6 27 11.14 13.49 12.5 16.7 4.41 6.4 1.5 2.4 1.5 2.4 1.5 2.4 6.31 7.4 17.06 18 10.37 12 21.32 23.1 3.9 4.6 3.9 4.6 3.9 4.6 9.05 10.4 14.49 15.3 13.41 13.9 10 and the experimental results for different values of q in circular and circular with flat bed cross section in different cases. Based on Singh and Luo’s (2011) proposition and the obtained results (Table 3), the value of q = 3/4 (k = -3) was regarded as constant throughout the modeling in this study. By using the suggested model [Eq. (14)] and the equation of Sterling and Knight (2002) in the present study, the shear stress distribution in the circular channel wall in four different depths was compared with Knight and Sterling (2000) experimental results, as shown in Table 1. In order to compare the results, the local boundary shear stress was normalized by the total shear stress. s0 = qgRS0 (Table 1, column 7). Figures 2, 3, 4 and 5 show the comparisons between the two models and the experimental results, where the horizontal axis shows the dimensionless wet area resulting from dividing y (Fig. 1) into P (wetted perimeter in whole channel), the vertical axis showing the dimensionless shear stress. As seen in Figs. 3, 4, 5 and 6 the predicted values for 0.2 \ y/P \ 0.8 by two models are close to experimental data, whereas the measured values for y/P \ 0.2 and y/P [ 0.8 in (h ? t)/D = 0.333, t = 0, (h ? t)/D = 0.506, t = 0 and (h ? t)/D = 0.666, t = 0 are less than those predicted. Because the boundary shear stress at the free surface is always difficult to measure, on increasing flow depth, the shear stress distribution tended to be uniform. In every situation, the results of the suggested model [Eq. (14)] and those of the Sterling and Knight (2002) model produce similar results. Using the Eqs. (19) and (20), an average percent error of these models was calculated in comparison with the experimental results displayed in Table 4, which illustrates the RE and RMSE of the introduced model and the Sterling and Knight (2002) relation for a circular cross section. Knight and Sterling (2000) filled the bed of a circular channel with sediments and, according to Table 1, used the Preston pipe method to measure the shear stress distribution on a bed slope of S0 = 0.00862 at two different depths, t = 0.25, (h ? t)/D = 0.333 and t = 0.25, (h ? t)/ D = 0.666. The circular channel wall has a steady 1.2 1 0.8 τ/ρgS Circular (h ? t)/D = 0.333, t = 0 Circular (h ? t)/D = 0.506, t = 0 q = 2/3 0.6 Present model 0.4 Sterling and Knight (2002) Experimental results 0.2 0 0 0.2 0.4 0.6 0.8 1 y/P Fig. 3 Shear stress distribution in circular channel for (h ? t)/ D = 0.333, t = 0 1.2 1 0.8 τ/ρgS Case 0.6 Present model 0.4 Sterling and Knight (2002) Experimental results 0.2 0 0 0.2 0.4 0.6 0.8 1 y/P Fig. 4 Shear stress distribution in circular channel for (h ? t)/ D = 0.506, t = 0 curvature. As the channel bed gets filled, the wall curvature, not being fixed, leads to the wall breaking down, creating a flat bed shaped cross section. As a result, the shear stress distribution in a circular channel with flat bed is estimated by using a rectangular channel pattern. Estimating it in the channel wall and bed, the Eqs. (17) and 123 Stoch Environ Res Risk Assess (2015) 29:1–11 1.2 1.2 1 1 0.8 0.8 τ/ρgS τ/ρgS 6 0.6 Present model 0.6 Present model 0.4 0.4 Sterling and Knight (2002) Sterling and Knight (2002) Experimental results 0.2 0.2 0 Experimental results 0 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 y/P Fig. 5 Shear stress distribution in circular channel for (h ? t)/ D = 0.666, t = 0 0.8 1 Fig. 7 Shear stress distribution in circular channel with flat bed for (h ? t)/D = 0.333, t = 0.25 1.2 1.2 1 1 0.8 τ/ρgS 0.8 τ/ρgS 0.6 y/P 0.6 Present model 0.4 0.6 Present model 0.4 Sterling and Knight (2002) Sterling and Knight (2002) 0.2 Experimental results Experimental results 0.2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 y/P 0 0 0.2 0.4 0.6 0.8 1 y/P Fig. 6 Shear stress distribution in circular channel for (h ? t)/ D = 0.826, t = 0 Table 4 Comparison of the experimental results in a circular channel with the results obtained by the proposed model and the Sterling and Knight equation Case Present model Eq. Sterling and Knight (2002) RE RE RMSE RMSE (h ? t)/D = 0.333, t = 0 2.3 2.8 3.6 4.6 (h ? t)/D = 0.506, t = 0 3.5 4.4 2.1 2.5 (h ? t)/D = 0.666, t = 0 (h ? t)/D = 0.826, t = 0 2.4 2.6 3.2 3.9 2.2 3.8 4.3 4.4 (18), respectively, were used. The same was done to esti0 0 mate kw ; kb ; k2w and k2b. To predict shear stress distribution, the Eqs. (3) and (4) were solved by using the Sterling and Knight (2002) equation. The shear stress distribution at two different depths in circular channels with flat bed for 123 Fig. 8 Shear stress distribution in circular channel with flat bed for (h ? t)/D = 0.666, t = 0.25 the suggested model and that of Sterling and Knight (2002) is shown in Figs. 7 and 8. As seen in Fig. 7, the results in a channel bed of the suggested model correspond to the experimental results, but the Sterling and Knight relationship predicts a larger shear stress in comparison with the experimental results. Both methods showed different results from the experimental ones in the channel wall. The discrepancy ought to be related to the small flow depth (2 cm) in this section. Figure 8 shows the shear stress distribution in t = 0.25, (h ? t)/D = 0.666. As shown, the suggested model results are close to the experimental ones, but the Sterling and Knight relation predicts a smaller shear stress at the channel wall and a larger one on the channel bed in comparison with the experimental results. The average RE percentage and RMSE of both models at two different flow depths on wall and bed were estimated by the Eq. (19) and are presented in Table 5. It is seen that the average error percent for the Sterling and Knight method is twice as large as that for the suggested method. Based on these results we can conclude that the method suggested in this study Stoch Environ Res Risk Assess (2015) 29:1–11 Case Present model Wall RE Eq. Sterling and Knight (2002) Bed RMSE RE Wall Bed RMSE RE RMSE RE RMSE (h ? t)/D = 0.333, t = 0.25 6.6 7.3 1.5 2.4 11.5 12.1 2.4 2.5 (h ? t)/D = 0.666, t = 0.25 3.5 5.1 3.9 4.6 9.2 9.8 6.8 8.1 1.4 1.4 1.2 1.2 1 1 0.8 0.8 τ/ρgRS τ/ρgRS Table 5 Comparison of the experimental results in a circular channel with flat bed with the results obtained by the proposed model and the Sterling and Knight equation 7 0.6 0.6 Present model 0.4 Sterling and Knight (2002) Experiment Tominaga et al. (1989) 0.2 0.4 Present model 0.2 Sterling and Knight (2002) Experiment Knight et al. (1984) Experiment Knight et al. (1984) 0 0 0 0.2 0.4 0.8 1 0 y/P (a) b/h=2 0.2 0.4 1.4 1.4 1.2 1.2 1 1 0.8 0.8 0.6 0.6 0.8 1 0.8 1 y/P (c) b/h=7.37 τ/ρgS τ/ρgRS 0.6 0.6 Present model 0.4 Experiment Tominaga et al. (1989) 0.2 Present model 0.4 Sterling and Knight (2002) Sterling and Knight (2002) 0.2 Experiment Knight et al. (1984) Experiment Knight et al. (1984) 0 0 0 0.2 (b) b/h=3.94 0.4 0.6 0.8 y/P 1 0 0.2 (d) b/h=1.34 0.4 0.6 y/P Fig. 9 Shear stress distribution in rectangular channel with different aspect ratios performs better when compared with the Sterling and Knight (2002) model. The experimental data of Knight et al. (1984) and Tominaga et al. (1989) were used to validate the proposed equation. Figure 9a–d show the shear stress distribution in a rectangular channel with different b/h ratios. The straight lines represent the shear stress value obtained by the proposed equation, the dash lines are for Eqs. (3) and (4) and the points are the experimental data. The Sterling and Knight model tends to overestimate the bed shear stress; this overestimation being the most important near the corners in all cases. The bed shear stress calculated by the Sterling and Knight model is nearly uniform for b/h = 2 and 1.34. For all aspect ratios the bed shear stress calculated by the proposed model reaches its maximum value in the center. The Sterling and Knight model indicates slightly higher peak shear stresses. As seen in Fig. 9a, the results indicate that the proposed equation predictions agree with the laboratory measurements, but the Sterling and Knight relationship predicts a larger shear stress in comparison with the experimental results for b/h = 2. The average RE at channel bed found between the proposed equation and Sterling and Knight relation and the experiments of Tominaga et al. (1989) is 3.6 and 10.3, respectively, and the RMSE is 5.8 and 17.7, respectively (Table 5). In Fig. 9b, the boundary shear stress at the bed predicted by the two models are compared with experimental data for b/h = 3.94. Reasonably good 123 8 Table 6 Comparison of the experimental results in a rectangular channel with the results obtained by the proposed model and the Sterling and Knight equation Stoch Environ Res Risk Assess (2015) 29:1–11 Case Present model Wall RE Eq. Sterling and Knight (2002) Bed RMSE RE Wall RMSE Bed RE RMSE RE RMSE b/h = 2, Tominaga et al. (1989) 6.2 7.3 3.6 5.8 13.4 19.5 10.3 17.7 b/h = 2, Knight et al. (1984) – – 5.2 7.0 – – 9.8 12.8 b/h = 3.94, Tominaga et al. (1989) – – 7.1 10.2 – – 11.9 17.1 b/h = 3.94, Knight et al. (1984) – – 7.1 7.6 – – 11.7 14.3 b/h = 7.37, Knight et al. (1984) – – 1.4 1.5 – – 2.9 3.9 b/h = 1.34, Knight et al. (1984) 4.7 5.5 – – 21.8 28.6 – – 1.8 Present model 1.6 Sterling and Knight (2002) 1.4 Experiment Rajaratnam and Ahmadi (1981) τ/ρgRS 1.2 1 0.8 0.6 0.4 Fig. 10 Compound channel cross section characteristics (Rajaratnam and Ahmadi 1981) 0.2 0 0 0.2 0.4 (a) h/H=0.324 0.8 1 0.8 1 1.8 Present model 1.6 Sterling and Knight (2002) 1.4 Experiment Rajaratnam and Ahmadi (1981) 1.2 τ/ρgRS agreement is given by the proposed model at bed corners. It is also interesting to note that there is no significant difference in the shear stress distribution given by the two models for b/h = 7.37 except in bed corners (see Fig. 9c). For b/h = 1.34 (Fig. 9d) very good agreement is given by the proposed model. The average RE between the proposed equation and Sterling and Knight relationship and the experiments of Knight et al. (1984) at channel wall is 4.7 and 21.8, respectively, and the RMSE is 5.5 and 28.6, respectively (Table 6). 0.6 y/P 1 0.8 0.6 0.4 0.2 0 0 0.2 (b) h/H=0.46 0.4 0.6 y/P 5 Shear stress in a compound channel Fig. 11 Shear stress distribution in compound channel The shear stress distribution at the compound channel boundaries, shown in Fig. 10, is calculated for two different heights of h/H = 0.324 and 0.46 (where h = floodplain depth and H = flow depth) with the aid of the Sterling and Knight (2002) method and the proposed model (see Fig. 11a, b). The results were compared by using the laboratory outcomes of Rajaratnam and Ahmadi (1981). In general, the two models did not match the experimental data and recorded high errors. It should be mentioned that the proposed equation had higher errors in predicting the shear stress distribution in comparison with Sterling and Knight method, as shown in Table 7. 123 In compound channels, the flow behavior is much more complex than in simple geometry due to the presence of secondary currents and vortices occurring in the cross section which develops along the main channel–floodplain interface. The two methods give very similar results, with no good predicted results relative to the measured values on the left side of the main channel wall. From the Fig. 7 a reverse prediction is observed, a lesser prediction at low range of the y/P values and higher estimation at the parts near the bed. On the bed of the main channel, the Sterling and Knight method showed better agreement with the Stoch Environ Res Risk Assess (2015) 29:1–11 Table 7 Comparison of the experimental results in a compound channel with the results obtained by the proposed model and the Sterling and Knight equation Border 9 Present model Eq. Sterling and Knight (2002) h/H = 0.324 h/H = 0.46 h/H = 0.324 h/H = 0.46 RE RMSE RE RMSE RE RMSE RE RMSE Left side wall 14.83 21 26.88 43.6 14.83 21 26.88 43.6 Bed 11.95 15.04 15.32 19.48 5.54 8.8 7.13 9.8 1.27 2.9 Main channel Right side wall 3.72 6.3 1.27 2.9 10.63 12.87 Floodplain Bed 10.63 14.49 8.36 11 16.17 22.19 1.12 10.15 Wall 10.76 9 4.97 6.9 10.76 9 4.97 6.9 experimental results as compared to the proposed model. On the right side of main channel wall, the results derived from the proposed method were closer to the experimental results in comparison with the Sterling and Knight method. Although, with an increment of flow height in the main channel, the difference between the two methods with laboratory outcomes decreases. There is a lateral transfer of momentum from the fast flowing main channel to the floodplain at their interface. This phenomenon causes the high velocity gradient between the floodplain and main channel. Therefore, the two methods predict the shear stress less than the measured data at the main channel interface with the floodplain. While, with increment of flow height in main channel interface with the floodplain, more fitness between the Sterling and Knight equation and experimental data is obtained. On channel wall of the floodplain, the two methods give very similar results, and with the increment of flow height, the difference between the methods and experimental results is reduced. corner region. In rectangular channels, the bed shear stress calculated by the Shannon relation is nearly uniform for b/h = 2 and 1.34, while for compound channels the Tsallis model ought to be extended further. This study illustrates how the differences between the two models are distinguishable case by case in open channels. Acknowledgments The authors would like to express their appreciation to the anonymous reviewers for their helpful comments and to Ellen Vuosalo Tavakoli for the painstaking editing of the English text. Appendix 1 This appendix presents the way Eq. (15) is derived. The first constraint applied concerns the general probability of the variable s, i.e., Z smax f ðsÞds ¼ 1; ð21Þ 0 6 Conclusion The Shannon and Tsallis entropy concepts based on the probability theory were applied to predict the distribution of the boundary shear stress. These computational methods are described and the results compared with experimental data. The comparison between the two models for computing the boundary shear stress distribution in circular, circular with flat bed and rectangular channels showed that these two methods produce similar results for the circular cross section, with no significant difference between the two models. The bed shear stress calculated by both models in circular with flat bed and rectangular channels for all aspect ratios reaches its maximum value in the center; however, the Tsallis model provides more reliable results in comparison with those got through the Shannon relation. The model based on Shannon entropy is not effective in predicting the distribution of boundary shear stress in the by substitution of the probability density function [Eq. (12)] into Eq. (21), one obtains: 1 Z smax iq1 q1h 0 k þ k2 s ds ¼ 1; ð22Þ q 0 integrating between 0 and smax gives: # k h ik smax 1 1 0 k þ k2 s ¼ 1; k k2 ð23Þ 0 the definite integral over that interval 0 and smax is given by: k h ik 1k 1 0 1 1 0 k þ k2 smax ½k k ¼ 1; ð24Þ k k2 k k2 thus: h 0 ik 0 k þ k2 smax ½k k ¼ k2 kk : ð25Þ 123 10 Stoch Environ Res Risk Assess (2015) 29:1–11 thus: Appendix 2 This appendix presents the way Eq. (16) is derived. The second constraint concerns the continuity of the variable s, i.e.: 1 Z smax iq1 q1h 0 k þ k2 s s ds ¼ smean ; ð26Þ q 0 this can be written in the form: 1Z 1 iq1 q 1 q1 smax h 0 s k þ k2 s ds ¼ smean ; q 0 by applying integration by parts one obtains: 1 q q1 q 1 q1 q1 1 0 smax k þ k2 smax q q k2 3 q Z smax h i q1 7 0 q1 1 k þ k2 s ds7 5 ¼ smean ; q k2 0 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} after integrating between 0 and smax we will have: smax 2q1 q1 1 q1 0 q1 k þ k2 s ; A¼ q k22 2q 1 0 ð27Þ ð28Þ ð29Þ ð30Þ the definite integral over that interval 0 and smax is given by: 2q1 q1 1 q1 0 q 1 1 q 1 0 2q1 q1 A¼ k ðk Þ q1 ; þ k s 2 max 2 q k2 2q 1 q k22 2q 1 ð31Þ by inserting Eq. (31) into Eq. (28) one obtains: 1 q q1 q 1 q1 q1 1 0 smax k þ k2 smax q q k2 2q1 0 q1 1 q1 q1 k þ k s 2 max q k22 2q 1 # q 1 1 q 1 0 2q1 ðk Þ q1 ¼ smean ; þ q k22 2q 1 ð32Þ this can be written in the form: k ik 1k 1 1 h 0 ikþ1 1 smax h 0 k k þ k2 smax þ k s 2 max k k k22 k þ 1 k2 k 0 1 1 1 ½k kþ1 ¼ smean ; þ ð33Þ k k22 k þ 1 123 ð34Þ References A to solve Eq. (28), part A can be expressed as: q Z smax iq1 q 1 1 h 0 k þ k2 s ds; A¼ q k2 0 ik 1 1 h 0 ikþ1 smax h 0 k þ k2 smax k þ k2 smax 2 k2 k2 k þ 1 0 kþ1 1 1 ½k þ 2 ¼ kk smean : k2 k þ 1 Araujo JC, Chaudhry FH (1998) Experimental evaluation of 2-D entropy model for open channel flow. J Hydraul Eng 124(10):1064–1067 Berlamont JE, Trouw K, Luyckx G (2003) Shear stress distribution in partially filled pipes. J Hydraul Eng 129(9):697–705 Bonakdari H, Tooshmalani M (2010) Numerical study of the effect of roughness of solid walls on velocity fields and shear stress in rectangular open channel flow. In: 10th International symposium on stochastic hydraulics, water 2010 symposium, Quebec City, Canada Bonakdari H, Larrarte F, Joannis C (2008) Study of shear stress in narrow channels: application to sewers. J Urban Water 5(1):15–20 Cao S (1995) Regime theory and geometric model for stable channels. PhD Thesis, School of Civil Engineering, University of Birmingham Cao S, Knight DW (1996) Shannon’s entropy-based bank profile equation of threshold channels. In: Tickle KS, Goulter JC, Xu C, Wasimi SA, Bouchart F (eds) Stochastic hydraulics’96, proceedings of the 7th IAHR international symposium on stochastic hydraulics, Central Queensland University, Australia, July. Balkema, Rotterdam, pp 169–175 Chien N, Wan ZH (1999) Mechanics of sediment transport. ASCE Press, New York Chiu CL (1987) Entropy and probability concepts in hydraulics. J Hydraul Eng 114(7):738–756 Chiu CL (1991) Application of entropy concept in open channel flows. J Hydraul Eng 117(5):615–628 Cruff RW (1965) Cross channel transfer of linear momentum in smooth rectangular channels. Water Supply Paper 1592-B. USGS-Center, Washington, DC, pp 1–26 Flintham TP, Carling PA (1988) The prediction of mean bed and wall boundary shear in uniform and compositely rough channels. In: Whith WR (ed) Proceedings of the 2nd international conference on river regime. Wiley, Chichester, pp 267–287 Galip S, Neslihan S, Recep Y (2006) Boundary shear stress analysis in smooth rectangular channels. Can J Civ Eng 33(3):336–342 Ghosh SN, Roy N (1970) Boundary shear distribution in compound channel flow. J Hydraul Div ASCE 96(4):967–994 Guo J, Julien PY (2005) Shear stress in smooth rectangular openchannel flow. J Hydraul Eng 131(1):30–37 Huai W, Chen G, Zeng Y (2013) Predicting apparent shear stress in prismatic compound open channels using artificial neural networks. J Hydroinform 15(1):138–146 Javid S, Mohammadi M (2012) Boundary shear stress in a trapezoidal channel. Int J Eng Trans A 25(4):323–332 Julien PY (1995) Erosion and sedimentation. Cambridge University Press, Cambridge Kabiri-Samani A, Farshi F, Chamani MR (2013) Boundary shear stress in smooth trapezoidal open channel flows. J Hydraul Eng 139(2):205–212 Stoch Environ Res Risk Assess (2015) 29:1–11 Khodashenas SR, Paquier A (1999) A geometrical method for computing the distribution of boundary shear stress across irregular straight open channels. J Hydraul Res 37(3):381–388 Khodashenas SR, Paquier A (2002) River bed deformation calculated from boundary shear stress. J Hydraul Res 40(5):603–609 Knight DW (1981) Boundary shear in smooth and rough channels. J Hydraul Div ASCE 107(7):839–851 Knight D, Sterling M (2000) Boundary shear in circular pipes partially full. J Hydraul Eng 126(4):263–275 Knight DW, Demetriou JD, Hamed ME (1984) Boundary shear stress in smooth rectangular channel. J Hydraul Eng 110(4):405–422 Knight DW, Yuen KWH, Al Hamid AAI (1994) Boundary shear stress distributions in open channel flow. In: Beven K, Chatwin P, Millbank J (eds) Physical mechanisms of mixing and transport in the environment. Wiley, New York, pp 51–87 Lane EW (1953) Progress report on studies on the design of stable channels by the Bureau of Reclamation. Proc ASCE 79(280):1–30 Lashkar-Ara B, Fathi-Moghadam M (2009) Wall and shear forces in open channel. Res J Phys 4(1):1–10 Maszczyk T, Dush W (2008) Comparison of Shannon, Renyi and Tsallis entropy used in decision trees. Lect Notes Comput Sci 5097(2008):643–651 Myers WRC (1978) Momentum transfer in a compound channel. J Hydraul Res 16(2):139–150 Olivero M, Aguirre-Pey J, Moncada A (1999) Shear stress distributions in rectangular channels. In: 28th IAHR congress, Graz, Australia Rajaratnam N, Ahmadi R (1981) Hydraulics of channels with floodplains. J Hydraul Res 19(1):43–60 11 Rhodes DG, Knight DW (1994) Distribution of shear force on the boundary of a smooth rectangular duct. J Hydraul Eng 120(7):787–807 Shannon CE (1948) A mathematical theory of communication. Bell Syst Tech J 27:379–423, 623–656 Singh VP, Luo H (2011) Entropy theory for distribution of onedimensional velocity in open channels. J Hydrol Eng 16(9):725–735 Sterling M, Knight DW (2002) An attempt at using the entropy approach to predict the transverse distribution of boundary shear stress in open channel flow. Stoch Environ Res Risk Assess 16:127–142 Tominaga A, Nezu I, Ezaki K, Nakagawa H (1989) Three dimensional turbulent structure in straight open channel flows. J Hydraul Res 27:149–173 Tsallis C (1988) Possible generalization of Boltzmann–Gibbs statistics. J Stat Phys 52:479–487 Wilcock PR (1996) Estimating local bed shear stress from velocity observations. Water Resour Res 32(11):3361–3366 Yang SQ (2010) Depth-averaged shear stress and velocity in openchannel flows. J Hydraul Eng 136(11):952–958 Yang SQ, Lim SY (1997) Mechanism of energy transportation and turbulent flow in a 3D channel. J Hydraul Eng 123(8):684–692 Yuen KWH (1989) A study of boundary shear stress, flow resistance and momentum transfer in open channels with simple and trapezoidal cross section. PhD Thesis, University of Birmingham 123