A FREQUENCY DOMAIN FINITE ELEMENT MODEL FOR TIDAL CIRCULATION by Joannes J. Westerink, Keith D. Stolzenbach, and Jerome J. Conner Energy Laboratory Report No. MIT-EL 85-006 January 1985 A FREQUENCY DOMAIN FINITE ELEMENT MODEL FOR TIDAL CIRCULATION by Joannes J. Westerink Keith D. Stolzenbach Jerome J. Conner Energy Laboratory and R. M. Parsons Laboratory for Water Resources and Hydrodynamics Department of Civil Engineering Massachusetts Institute of Technology Cambridge, Massachusetts 02139 Sponsored by Northeast Utilities Service Company and New England Power Service Company under the MIT Energy Laboratory Electric Utility Program and by The Sea Grant Office of NOAA U.S. Department of Commerce Energy Laboratory Report No. MIT-EL 85-006 January 1985 ABSTRACT A highly efficient finite element model has been developed for the numerical prediction of depth average circulation within small scale embayments which are often characterized by irregular boundaries and bottom topography. Traditional finite element models use time-stepping and have been plagued with requirements for high eddy viscosity coefficients and small time steps necessary to insure numerical stability, making application to small bays infeasible. These problems are overcome by operating in the frequency domain, an intrinsically more natural solution procedure for a highly periodic process such as tidal forcing. In order to handle non-linearities, an iterative scheme which updates non-linearities as right hand side force loadings must be implemented. Pioneering efforts with the harmonic approach have had shortcomings in either not modeling all physically relevant terms and/or in not gearing towards application to small scale regions. Small embayments are often quite shallow and have rapidly varying depth, making the nonlinear terms in the governing hydrodynamic equations much more significant. This requires that more frequencies be used in order to resolve the tide and account for the greater nonlinear coupling due to bottom friction, convective acceleration and finite amplitude effects. In order to make the process of handling this wide range of frequencies manageable, a hybrid frequency-time domain approach is applied. The iterative scheme revolves around a highly efficient linear core code which can handle a wide range of frequencies. Furthermore, instead of Fourier expanding the nonlinear terms, an efficient least squares error minimization algorithm is used for the discrete spectral analysis of the iteratively updated psuedo-force time history generated by the nonlinearities. With this highly efficient scheme it is now possible to efficiently study both short period and long term residual circulation within small scale embayments. -3- ACKNOWLEDGMENTS This report is part of a research program to develop more efficient and accurate ciculation and dispersion models for coastal waters. The report describes a two-dimensional non-linear frequency domain model (named TEA-NL) for the analysis of circulation in tidal embayments. The work is an extension of a linear model (TEA) described in Westerink, J. J., Connor, J. J., Stolzenbach K. D., Adams, E. E., and Baptista, A. M., "TEA: A Linear Frequency Domain Finite Element Model for Tidal Embayment Analysis," Energy Laboratory Report No. MIT-EL 84-012, February 1984 Also developed as part of this reserch program is a two-dimensional transport model (ELA) which combines Eulerian and Lagrangian techniques and is described in Baptista, A. M., Adams, E. E., and Stolzenbach, K. D., "Eulerian-Lagrangian Analysis of Pollutant Transport in Shallow Water," Energy Laboratory Report No. MIT-EL 84-008, June 1984 (Also published as Technical Report No. 296, R. M. Parsons Laboratory for Water Resources and Hydrodynamics, M.I.T.) Support for this research was provided in part by the Sea Grant Office of NOAA, Department of Commerce, Washington, D.C., and in part by Northeast Utilities Service Company and New England Power Company through the M.I.T. Energy Laboratory Electric Utility Program. -4- TABLE OF CONTENTS Page .......... ABSTRACT ................... ACKNOWLEDGMENTS ................... TABLE OF CONTENTS ....... 4 ...... 5 . .................. LIST OF TABLES ... LIST OF FIGURES 3 . .. .... . .. .... .. . . . .. .. ................... ..... INTRODUCTION ................... 2. DESCRIPTION OF TIDES IN SHALLOW EMBAYMENTS . ......... LINEAR CORE MODEL 5. NONLINEAR MODEL 7. .. .................... . . . . .. . . . REFERENCES .... . ...... . . . .... ....... ............ ........ -5- . . ........ 79 97 104 124 170 193 Description of Bight of Abaco and Its Tides . ....... . .. .... Overtide Computations for the Bight of Abaco Compound Tide Computations for the Bight of Abaco ..... ..... Discussion ................... CONCLUSIONS 79 104 ....... ................... 43 47 56 61 .. ...... ... 24 43 .... ................... ............. 24 29 Harmonic Analysis of Non Linear Pseudo Forcings Iterative Convergence . .................. APPLICATION 6.1 6.2 6.3 6.4 . ......... Weighted Residual Formulation . ....... ...... . ............ Finite Element Method Formulation. . . . . . . . . . . . . . . Frequency Domain Formulation 4. 5.1 5.2 . .. Harmonic Tidal Components in Estuaries NUMERICAL FORMULATION 13 .............. Governing Equations 2.2 3.1 3.2 3.3 6. . .... 2.1 6 8 ....... 1. 3. . . . . 202 205 LIST OF TABLES Page 22 Table 1.1 Wavelengths of a 12.4 Tide in Various Water Depths .. Table 2.1 Astronomical Tides of Importance . ......... . 30 Table 2.2 Major Overtides. . .................. . 31 Table 2.3 Major Compound Tides . ................ Table 2.4a Response-Forcing Table for Overtides as Generated by 31 Finite Amplitude Term at Cycle No. 2 of Iteration. . . 35 . 35 Response-Forcing Table for Compound Tides as Generated by Finite Amplitude Term at Cycle No. 1 of Iteration . 37 Response-Forcing Table for Compound Tides as Generated by Finite Amplitude Term at Cycle No. 2 of Iteration . 37 Table 2.6 Tides of Interest (High Freq. End) . ......... 40 Table 4.1 Sizes and Ranks of Various Matrices. . ........ . 64 Table 4.2 Variation of Convergence with cs 69 Table 4.3 Comparison of Analytical and Numerical Elevations and Velocities for Example Channel Case at Various Locations Table 2.4b Response-Forcing Table for Overtides as Generated by Finite Amplitude Term at Cycle No. 3 of Iteration. . Table 2.5a Table 2.5b Table 5.1 Table 5.2a Table 5.2b . . . . . . . . . . . (a) Linearized Friction Factor X = 0.0000. . ..... . 74 (b) Linearized Friction Factor X = 0.0010. . ..... . 75 (c) Linearized Friction Factor X = 0.0100. . ..... . 76 LSQ Analysis Results Showing Effects of Variation of Number of Frequencies and Time Sampling Points; Example Simulating Overtide Type Frequencies . .... 89 LSQ Analysis Results Showing Effects of Variation of Number of Frequencies and Time Sampling Points; Example Simulating Closely Spaced Compound Tide . . . . . . . . Frequencies. . ............ 92 LSQ Analysis Results Showing Effects of Variation of Number of Frequencies and Time Sampling Points; Example Simulating Closely Spaced Compound Tide . Frequencies . . . . . . . . . . . . . . . . . . . . 93 -6- Page Table 6.1 Table 6.2 Table 6.3a Table 6.3b Table 6.4 Table 6.5a Table 6.5b Reflection and Transmission Coefficients for a Long Wave Passing Over a Step from Depth h i to Depth h 2 for Various Depth Ratios. . . . . . . . . . . . . . . . 107 Summary of Measured Astronomical Tides Along the .. . ................ Open Ocean Boundary .. . ... * 111 Measurement Error for Each Frequency in Terms of Proportional Variance, Vm . . .......... J * 122 Measurement Error for Each Frequency in Terms of .......... Proportional Standard Deviation, S J * 123 Values for Friction Factor cf in Terms of Depth and Bottom Roughness (from Wang and Connor, 1975) . . * 136 Overtide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of . . ............. . Proportional Variance, VP .. . . J * 165 Overtide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of Proportional Standard Deviation, SP . J . ....... * 166 Table 6.6 Tides of Possible Interest for M 2 and N 2 Interaction. . 171 Table 6.7a Compound Tide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of Proportional Variance, V. . . . . . . ............. Table 6.7b Compound Tide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of ......... Proportional Standard Deviation, SP i -7- * 190 * 191 LIST OF FIGURES Page Definition Sketch Showing Typical Elevation Prescribed . and flux prescribed (Q ) Boundaries . . . . . . (FT) 28 Figure 2.2 Schematic of Major Astronomical and Shallow Water Tides 41 Figure 3.1 Schematic of Iterative Non Linear Scheme. . ...... . 60 Figure 4.1 Definition Sketch of Depth Varying Channel Which Illustrates Convergence Problems of Iterative Linear ............. . .. Scheme. . ........... 68 Finite Element Grid Discretization for Closed Ended ......... . . Channel Example Case. . ........ 73 Linear Equation Generated by Least Squares Analysis Procedure . . . . . . . . . . . . . . . . . . . . . . . 83 Figure 2.1 Figure 4.2 Figure 5.1 Figure 5.2 Effects of Variation in Frequency and Time Sampling . Rates for Typical Overtide Frequency Distribution . Figure 5.3 91 .. . 105 Geography of Bight of Abaco, Bahamas. . ...... Figure 6.2 Bathymetry of the Bight of Abaco, Bahamas . ...... Figure 6.3 Field Data for M 2 Astronomical Constituent (after Filloux & Snyder, 1979) (a) Amplitude in centimeters. . ... 109 . . . . . . . . . 112 . . . . . 113 ...... . 114 (b) Phase lag in radians. . ......... Field Data for M 4 Overtide Constituent (after Filloux & Snyder, 1979) (a) Amplitude in centimeters. . ...... . . (b) Phase lag in radians. . ...... Figure 6.5 88 Effects of Variation in Frequency and Time Sampling Rates for Typical Compound Tide Frequency Distribution (maximum period is T = 12.4 hours and maximum synodic . . . . . . . . . . . period is TS = 27 days) . .... Figure 6.1 Figure 6.4 . . . . . . . . . 115 Field Data for M 6 Overtide Constituent (after Filloux & Snyder, 1979) (a) Amplitude in centimeters. . .. . . . . . . . . . . 116 (b) Phase lag in radians. . .... . . . . . . . . . . 117 -8- Figure 6.6 Figure 6.7 Figure 6.8 Figure 6.9 Field Data for N 2 Astronomical Constituent (after Filloux & Snyder, 1979) (a) Amplitude in centimeters . . . . . . . . . . . . . 118 (b) Phase lag in radians . . . . .. 119 .............. . Finite Element Grid Discretization for Bight of Abaco, Bahamas. . . . . . . . . . . . . . . . . . . . . . . .. Results of TEA with Full Non Linear Friction Effects (cf = 0.003) for M 2 Astronomical Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 127 (b) Phase lag in radians . . . . . . . . . . . . . . . 128 Results of TEA with Full Non Linear Friction Effects (cf = 0.006) for M 2 Astronomical Constituent (a) Amplitude in centimeters . . . ... . . . . . .... (b) Phase lag in radians . . . . . . . . . . . . . . Figure 6.10 125 129 . 130 . . . . . 131 (b) Phase lag in radians . . . . . . . . . . . . . . . 132 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) for M 2 Astronomical Constituent (a) Amplitude in centimeters . . . . . . . . Figure 6.11a Trajectory along which M 2 Elevation Amplitudes are Compared for Varying Friction Factor in Figure 6.11b 133 Figure 6.11b Comparison of M 2 Elevation Amplitudes for Varying Friction Factor, cf, along Trajectory S . . . . . . . 134 Figure 6.12 Figure 6.13 Results of TEA with Full Non Linear Friction Effects ...... (cf = 0.009) for Steady State Constituent... Results of TEA with Full Non Linear Friction Effects (cf = 0.009) for M 4 Overtide Constituent (a) Amplitude in centimeters .. . ... ............ (b) Phase lag in radians . . . . ................ Figure 6.14 137 138 139 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) for M 6 Overtide Constituent (a) Amplitude in centimeters (b) Phase lag in radians -9- . ............. . ................ 140 141 Figure 6.15 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for M 2 Astronomical Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 142 . . . . . . . . . . . . 143 Results of TEA with Full Non Linear Friction Effects (c = 0.009) and Finite Amplitude Effects for Steady State Constituent . . . . . . . . . . . . . . . 144 (b) Phase lag in radians . . . Figure 6.16 Figure 6.17 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for M 4 Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 145 . . . . . . . . 146 (b) Phase lag in radians . . . Figure 6.18 . . . . Results of TEA with Full Non Linear Friction Effects (c = 0.009) and Finite Amplitude Effects for M 6 Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 147 . . . . . . . . . 148 (b) Phase lag in radians . . . . . . Figure 6.19 Figure 6.20 Figure 6.21 Figure 6.22 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 2 Astronomical Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 150 (b) Phase lag in radians . . . . . . . . . . . . . . . 151 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for Steady State Constituent. . . 152 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 4 Overtide Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 153 (b) Phase lag in radians . . . . . . . . . . . . . . . 154 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 6 Overtide Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 155 (b) Phase lag in radians . . . . . . . . . . . . . . . -10- 156 Figure 6.23a Velocity Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for Steady State Component . . . . . . . . . ..................... . 158 Figure 6.23b Velocity Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 2 Component at Time of Maximum Ebb for the M 2 Component Relative to the Ocean Boundary. . . . . . . . . . . . . . . . . . 159 Figure 6.23c Velocity Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 4 Component at Time of Maximum Ebb for the M 2 Component Relative to the Ocean Boundary. . . . . . . . ................. . 160 Figure 6.23d Velocity Results of TEA with Full Non Linear Friction Effects (c = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 6 Component at Time of Maximum Ebb for the M 2 Component Relative to the Ocean Boundary. . . . . . . . ................. 161 Figure 6.24a Continuity Equation Pseudo Forcing Vector Ratios Due to M 2 - N 2 Interaction. . . . . . . . . . . . . . . . 173 Figure 6.24b Momentum Equation Pseudo Forcing Vector Ratios Due to M 2 - N 2 Interaction . . . . . . . . . . . . ....... 174 Figure 6.25 Results of TEA with M 2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for M 2 Astronomical Constituent (a) Amplitude in centimeters. . . . . . ............. . (b) Phase lag in radians. . . . . . . . . Figure 6.26 . . . . . . . ............. . (b) Phase lag in radians. . . . . . . . . . . Figure 6.28 177 Results of TEA with M 2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for N2 Astronomical Constituent (a) Amplitude in centimeters. . Figure 6.27 ..... 176 ...... 178 179 Results of TEA with M 2 -N2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for Steady State Constituent.. . . . . . . ... 180 Results of TEA with M 2 -N2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for MN Compound Constituent . . . . . . . . . . 181 Figure 6.29 Figure 6.30 Figure 6.31 Results of TEA with M2-N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for MN 4 Compound Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 182 (b) Phase lag in radians . . . . . . . . . . . . . .. 183 Results of TEA with M 2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for M 4 Overtide Constituent (a) Amplitude in centimeters . . . . . . . . . . . . . 184 (b) Phase lag in radians . . . . . . . . . . . . . . . 185 Results of TEA with M 2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for 2MN 6 Compound Constituent (a) Amplitude in centimeters . . . . . (b) Phase lag in radians . . . Figure 6.32 . . . . .... 186 . . . . . . . . . . . . 187 Results of TEA with M 2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for M 6 Overtide Constituent (a) Amplitude in centimeters . . . .. (b) Phase lag in radians . . . . . . -12- ............ . . . . . .... . 188 189 CHAPTER 1. INTRODUCTION Recent years have seen the development of numerous coastal circulation models which apply the finite element method. The principal advantage of finite element methods over the more traditional finite difference methods is the greater versatility allowed in grid discretization which is especially important for small scale coastal embayments. This feature permits the convenient fitting of the often irregular boundaries and allows refinement of the grid in such critical areas as high flow and bottom depth gradient regions and/or the narrow mouths connecting these embayments to the open ocean. These circulation models share as a general starting point the well established shallow water equations which are derived by depth averaging the conservation of mass and momentum equations with the application of the hydrostatic and Boussinesq assumptions. Therefore the equations that are applied are based on first principles and require empirical support only for the turbulent exchanges and surface and bottom stresses. Major differences between these models lie in features such as the type of localized expansions used to resolve the spatial dependence of the variables and more importantly the way in which they discretize the time dependence. Traditionally time marching schemes have been applied which are either explicit or implicit. Early time domain models [Grotkop, 1973; Taylor and Davis, 1975; Wang and Connor, 1975; Kawahara, 1978; King et al., 1974] have had severe problems relating to the economy and accuracy of the schemes developed. Economic constraints stem both from the large amount of numerical manipulation required for the schemes and a maximum allowable -13- time step required for the accuracy and/or stability of the computation. For explicit schemes a Courant stability constraint necessitates the maximum time step to decrease along with element size, making it especially infeasible to apply these models to small scale geometries/elements. Furthermore, these early models have been plagued with accuracy problems which relate to short wave length artificial oscillations in elevation and velocity produced by the finite discretization of the domain [Gray, 1980; Sani, et al., 1980]. These models either require artificially high eddy viscosity to damp out this short wave length noise or the numerical scheme is such that it is inherently overdamped. The accuracy problem, however, arises from the fact that not only the numerical noise is damped, but the longer physical waves being simulated are also damped. This aspect of overdamping has drawn strong criticism as to the ability of these models to adequately simulate the physical problem described by the shallow water equations. Efforts to overcome these shortcomings in accuracy and efficiency have been numerous and have had varying degrees of success. Alternatives which have been investigated include different time integration schemes [Gray and Lynch, 1977, 1979; Niemeyer, 1979], mass lumping schemes [Kawahara, 1982] and the examination of the effects of mixed interpolation of elevation and velocity [Walters and Cheng, 1980; Walters and Carey, 1983; Platzman, 1981; Williams and Zienkiewicz, 1981]. Certain investigators paid special attention to the numerically troublesome convective terms by either applying a Petrov-Galerkin weighting scheme (equivalent to upwinding) [Nakazawa, et al., 1980] or by using the method of characteristics in conjunction with finite elements [Benque, et al., 1981]. One of the more promising schemes -14- developed is the use of a wave type equation in conjunction with the fundamental momentum equation as the basis of the finite element formulation [Lynch and Gray, Although some of these alternative 1979]. schemes have been successful at eliminating short wave length noise without damping the longer physical waves and are more efficient than earlier models, all of the above methods still have maximum allowable time steps making them economically unattractive for either long term simulations and/or small scale embayments. A very attractive alternative to time stepping schemes which has recently been employed is the use of harmonic analysis in conjunction with finite elements. Because of the periodic nature of the tidal phenomenon, the harmonic method is an intrinsically more natural solution procedure and was one of the traditional methods for analysis before the advent of finite difference and finite element methods [Dronkers, 1964]. There are no time stepping limitations since this procedure generates a set of quasi-steady (or time independent) equations. In addition, truncation errors and the associated stability problems caused by time stepping are precluded. Furthermore, eliminating the time dependence from the governing equations reduces them from equations of the difficult and time consuming hyperbolic type to that of the elliptic type which are much more readily solved by finite element methods. The harmonic method also offers the potential of economically performing realistic long term simulations in tidal embayments (e.g., up to 30 days) and calculating the associated residual circulation. A possible drawback of the harmonic method is the increased number of frequencies which would be required to model a non-periodic -15- phenomenon such as wind driven circulation. However some winds (steady winds and sea breezes) are periodic and furthermore any wind spectrum can readily be harmonically decomposed. Time domain schemes become economically more attractive when the number of frequencies required to adequately represent the wind spectrum becomes excessive. A more significant difficulty which arises in implementing this frequency domain technique is that non linear terms generate additional responses at frequencies other than the base forcing frequency. Strategies to handle this harmonic coupling produced by the non linear terms (finite amplitude, convective acceleration and bottom friction terms) have consisted of either iterative procedures [Pearson and Winter, 1977; Kawahara, 1978] or some type of perturbation analysis [Askar and Cakmak, 1978; Kawahara et al., 1978; Le Provost et al., 1981]. 1977; Le Provost and Poncet, The iterative scheme applied by Pearson generates a finite spectral series representing the pseudo-forcings due A to all the non linear components of the shallow water equations. linear solution is then used to evaluate the elevation response due to each pseudo forcing component (at pre-determined frequencies). Pseudo forcings are then updated using the updated responses in elevation and velocity until convergence in the elevation is achieved. The full non linear model has only been tested on geometries with a relatively small number of elements while a simplified linear version has been applied to a large grid of a deep bay [Jamart and Winter, 1980]. Kawahara [1977], on the other hand, applied a perturbation analysis which allows grouping of terms in the expanded (by a power series) shallow water equations in order to generate several sets of linearized equations of varying order. In a later paper, Kawahara [1978] applied -16- the periodic Galerkin method for the time dependence which produces a coupled system of non linear simultaneous equations which are then solved iteratively. An important limitation with Kawahara's schemes is that both bottom friction and Coriolis are omitted from the governing equations while keeping the much less important eddy viscosity terms. A further weak point of both Pearson's and Kawahara's work is that they have used a Fourier series expansion in terms of integer values of a base frequency (the M2 tide) to represent the variables. This precludes the possibility of investigating the interaction between the majority of the tidal components (see Table 2.1 for major tidal components). Therefore, such effects as monthly (spring/neap) variation (caused to a large extent by beating effects of closely packed tidal components) can not be looked at and only a major base tide (e.g., M2) and its harmonics may be studied. Le Provost [1981] applies an expansion which considers the interaction of the major closely spaced forcing components (M2 , S2 , N2, K1 ). However, the perturbation analysis and the quasi-linearization for bottom friction which are used only account for the non linear coupling between the major forcing component (M2 ) and its first harmonic (M4). The remaining astronomic constituents of the tide generating potential (S2 , N2 , KI) are treated linearly. The resulting computer code has been applied to the English Channel and Le Provost states that the method is constrained in its application to very shallow waterbodies. Finally, Lynch [1981] has developed a linear harmonic model based on the same re-arrangement of the fundamental equations as an earlier time stepping scheme [Lynch and Gray, 1979] which was found to minimize short wave length noise while retaining the computational accuracy of -17- the longer physical waves. The equations which he uses as the starting point of his finite element scheme are the wave equation (found by substituting the momentum equation into the continuity equation) and the momentum equation. He found that when using similar interpolation orders for elevation and velocity, this scheme will give accurate solutions without noise. As previously mentioned the harmonic method naturally lends itself to the calculation of long term variations and residual circulations. Residual currents may be thought of as a complicated combination of steady and slowly varying currents due to the effects of non linear interactions of both the main and various other astonomic tidal components [Ianniello, 1977]. Due to infeasibility of running long term simulation with time stepping programs, investigators have commonly computed residual circulations either by time averaging the governing equations [Walters and Cheng, 1980; Bonnefille, 1978; Tee, 1981] or by time averaging the results produced by a time domain model over one or several tidal cycles. The former technique places the inability to perform long term computations in a set of additional unknown time averaged terms (tidal-stress terms). relating to both expense and accuracy. The latter technique has problems In order to capture the effects of the variations in tidal forcing (which cause the non linear interaction between tides and produce residual/long term circulations) with sufficient accuracy, the simulation would have to be run over extended periods of time. It would not only be expensive to run a time stepping model for long periods of time but round off errors would also propagate through the solution. Hence it is doubtful that these previous efforts are able to model the effects that drive residual -18- circulation since they are not able to capture the actual physics of the residual currents. As is discussed in Chapter 2, the major variation in the currents of coastal embayments generated by astronomic tidal forcings are well described by considering a one month period. Responses then occur at these astronomical forcing frequencies and their associated higher harmonics (and certain lower frequencies due to the closeness of certain of the components). Hence we have a limited number of frequencies which need to be considered. Therefore it now becomes even clearer that applying the harmonic method is extremely well suited to assess low period fluctuations and residual circulation due to tidal forcing components and their non linear interaction. It is not only computationally convenient to do so but also allows the calculation to be done in a manner which is based on the same first principles with which we presently perform short term circulation computations and furthermore allows all the significant effects to be caught. In summarizing the many advantages of the harmonic method, we note that: (i) no time stepping constraints due to small element sizes are required; (ii) it is well suited for the highly periodic tidal computations in estuaries; (iii) the results may be stored in a much more economical and convenient form for applications with a transport model; (iv) it allows computations of long term residual currents; and (v) there are no cold start problems which time domain approaches often have. -19- In spite of its many unique features, being exploited to its full potential. the harmonic method is far from Pioneering efforts which have applied this technique have had shortcomings in not modeling all physically relevant terms, by not allowing for full interaction between the major tidal components and in not gearing towards application of the method to residual circulation computation and to small scale geometries. Even though application of the method to small scale estuaries increases the computational effort due to the often shallower depths which increases the significance of non linearities, it is nonetheless in these small scale regions that the effort associated with time domain approaches becomes entirely insurmountable due to the exessively small time steps needed. We conclude that there is a definite need for the development of improved strategies for computing tidally induced circulation within coastal embayments. The present research addresses this issue with the development of a general harmonic finite element model which allows the in-depth study of the many complex non linear interactions which occur in shallow waterbodies. This includes not only the investigation of the coupling occurring between a given astronomical tide and the harmonics it generates through the non linear terms in the governing equations but also the complicated interactions between the various astronomical tides Among the nonlinear harmonic responses and their associated harmonics. to be investigated are steady and other very long period residual circulations which are generated. A direct iteration scheme will be used to handle the non linear terms in the governing equations. However, inherent to any solution scheme which iteratively updates the non linearities as right hand side -20- the limitation that the relative magnitude of the right hand loadings is side non linear terms must be small compared to the left hand side linear terms (Ketter and Prawel, 1969). The implication of this is the non linear solution must be a perturbed linear solution. that The significance of the non linear friction term far exceeds that of the other non linear terms for the case of tidal estuaries. In order to minimize the importance of non linear friction as a right hand side term, a close approximation of the linear part of the friction term (a major part of the friction term is linear as is shown in be included on the left hand side of the equations. Chapter 2) will This greatly enhances iterative stability and allows computations to be performed for very shallow estuaries. Furthermore there may be theoretical limitations for iterative schemes relating to the relative size (with respect to wavelength) of the estuary. Lamb (1932) shows that when solving for the case of an open ended canal, the solution obtained by treating the finite amplitude term by successive approximation will be unstable if 2n () small, where x = the size of the canal and X = wavelength. (. ) is not Even though the same difficulty does not necessarily occur for the case of a closed ended canal, the criterion may be viewed as being indicative of potential instabilities. Table 1.1 shows the wavelengths associated with a 12.4 hour tide in various water depths without considering bottom friction. Since it is the intent to apply the method to estuaries and bays and not to coastal seas, the stability criterion will be small and this potential instability will not be a limitation of the method. In the following chapters the solution strategies applied are discussed in detail. Chapter 2 describes the governing equations used -21- Table 1.1 Wavelengths of a 12.4 Tide in Various Water Depths x h (m) (km) 2 200 10 442 100 1400 -22- and furthermore describes important tides and the types of interactions that may occur. In Chapter 3 the governing partial differential equations are discretized such that numerous sets of linear algebraic equations are generated. These sets of equations are in the frequency domain and are coupled through the non linearities. Chapter 4 examines various strategies to optimally solve each of these linear sets of equations and Chapter 5 discusses the details of the implementation of the fully non linear scheme. In Chapter 6, the program TEA-NL (Non Linear Tidal Embayment Analysis), developed with the methods described, is applied to an example case. -23- DESCRIPTION OF TIDES IN SHALLOW EMBAYMENTS CHAPTER 2. 2.1 Governing Equations The equations which are used to describe tidal wave propagation may be readily developed by depth averaging the Navier-Stokes and continuity equations with the assumptions of hydrostatic pressure distribution, constant density fluid, constant pressure at the air-water interface and negligible momentum dispersion (or eddy viscosity). (Dronkers, 1964): equations are n9t + [u(h + n)],x + [v(h + n)] y - fv + gn gt ,x u v ,t + fu - + g~ ,y The resulting (2.1) 0 + vu ) = 0 C /p(h + r) + Tb/p(h + n) + (uu ,y ,x x x (2.2a) ) = 0 (2.2b) y /p(h + n) + Tb/p(h + n) + (uv y ,x + vv ,y where: u(x,y,t) v(x,y,t) components of water velocity in x and y directions, respectively r(x,y,t) = surface elevation relative to mean sea level (MSL) t = time h = depth to MSL g = acceleration due to gravity p = water density f = the Coriolis factor = the applied surface stresses x, x y b T ,' b = the bottom stresses -24- . I~----------- Bottom stresses are quantified as [Daily and Harleman, 1966]: b S/0 2 1/2 c f (u xb x /0 (2.3a) 2 +v2) 1/2 +v) v = C (2.3b) where Cf = friction factor = 1/8 fDW Darcy-We isbach g/c2 Chezy 2 ng (2.4) Manning h The fully non linear friction terms may be approximated by linearized friction terms as follows: blin /p= [h/(h+n)] b/P b [h/(h+n)] b/ y = b,lin ln/p y = X (2.5a) XV (2.5b) where X = cfU = U = representative flow velocity linearized friction coefficient (2.6) For tidal flow with only one frequency present and the linearization being performed on an equivalent work over a tidal cycle basis, X has the following form [Ippen, 1966]: 8 max 3n (2.7) f where Umax max representative maximum velocity during a cycle -25- Although the linear friction term does not characterize the fully non linear one in spreading energy to other frequencies, it can approximate the magnitude of the actual non linear friction term quite reasonably. Linearized friction is helpful in the fully non linear scheme as an iterative stabilizer. Wind stresses may be approximated by the empirical formulas [Van Dorn, 1953; Wu, 1969]: S= x s y where: U10 air D U2 cos 0 w 10 (2.8a) p ir air U2 sin e w 10 (2.8b) D air = = wind speed pair = density of air cD = wind drag coefficient S w = wind approach angle The applied wind stress term in the governing equations may be simplified by assuming that finite amplitude effects will be unimportant and approximating 1/(h+n) as 1/h, a very reasonable assumption when considering the empirical nature of the formula used to represent this term and the inherent limitations of a depth averaged model in simulating wind-driven circulation. Finally, the Coriolis parameter is calculated as: f = (2.9) 20 sin6 where 4 = degrees latitude and 0 = radian frequency of the Earth. The boundary conditions associated with the governing equations are elevation prescribed and normal flux (or velocity) prescribed conditions which are respectively expressed as: -26- = n (x,y,t) on P (2.10) Qn(x,y,t) = Qn(x,y,t) on TQ (2.11) (x,y,t) and where P = elevation prescribed boundary r = flux prescribed boundary As is shown in Figure 2.1, elevation prescribed boundaries are usually associated with open ocean boundaries and flux prescribed boundaries are usually associated with land (including rivers) boundaries. Since the governing equations (2.1 and 2.2) are based on first principles (with the exception of surface stresses), their ability to model the circulation in estuaries depends on whether the assumptions made in their derivation are satisfied. The depth averaging of the equations precludes the accurate modeling of strongly stratified estuaries and/or wind induced circulation in deep waterbodies. The constant density assumption does not allow density driven currents to be simulated and the hydrostatic pressure assumption rules out the modeling of short or intermediate length waves. The importance of eddy viscosity has been deemed as negligible in most estuaries [Dronkers, 1964] and hence only empirical support is required for the surface stress terms. Since we have confidence that Eqs. 2.1 and 2.2 accurately describe tidal circulation in well mixed estuaries, the task at hand is to implement a numerical scheme which allows these complicated partial differential equations to be solved accurately for arbitrarily shaped estuaries over extended periods, such that the effects of the variations -27- River Inflow Q TIDAL EMBAYMENT r OCEAN Figure 2.1 Definition Sketch Showing Typical Elevation Prescribed (P ) and flux prescribed (rQ) Boundaries -28- in tidal forcing can be included. As we have seen, the harmonic method in conjunction with finite elements is excellently suited for this purpose. 2.2 Harmonic Tidal Components in Estuaries Prior to proposing a general harmonic finite element scheme which allows the treatment of the more cumbersome non linear terms in the shallow water equations, it will be beneficial to discuss in more detail the nature of the non linear responses which arise in shallow estuaries. Tides are the result of complex gravitational interactions between the moon, sun and oceans. Tides in the open ocean may be described by the superpositioning of a series of harmonic components which are unaffected by the non linear terms in the governing equations. In shallow water, however, non linear effects such as the finite amplitude of the tide (compared to the overall depth), bottom friction and convective acceleration become significant, and as a consequence the astronomical tides present (see Table 2.1) generate shallow water tides which cannot be ignored [Dronkers, 1964]. Shallow water tides may be classified as either overtides or compound tides. Overtides correspond to the generation of a tidal response through the non linearities by one astronomical component. The frequencies associated with the overtides are exact multiples of the frequencies of the astronomical tidal components which generate them (see Table 2.2 for overtides of importance). Compound tides are the result of the non linear interaction between two astronomical constituents, and their associated frequencies correspond to sums and differences of frequencies of the astronomical tide (see Table 2.3 for -29- Table 2.1 Astronomical Tides of Importance Tide -1 f(hr - ) Mm MSf Mf 0.00131 0.00151 0.00305 763 = 32 days 661 = 27.5 days 328 = 13.6 days M2 0.08054 12 hr 25 min S2 0.08333 12 hr 0 min 46 N2 0.079051 12 hr 39 min 19 K2 0.083449 11 hr 59 min 13 L2 0.082080 12 hr 11 min 4 K1 0.04178 23 hr 56 min 58 01 0.04198 23 hr 49 min 41 P1 0.04158 24 hr 03 min 19 T(hr) P /P (%) 1M LOW PERIOD 1.2 9.1 17.2 SEMI-DIURNAL 100 DIURNAL -30- _~I~_ Table 2.2 Major Overtides f(hr- 1 ) Tide Freq. Comp. M4 2wM2 0.161074 6.21 M6 3 wM2 0.241611 4.14 M8 4n2 0.322148 3.10 S4 20 0.166667 6.00 S6 302 0.250000 4.00 Table 2.3 Tide Major Compound Tides Freq. Comp. MS4 T(hr) M2 + wS2 f(hr-1) T(hr) 0.163870 6.10 2MS 6 2M2 + 2 0.244407 4.09 2SM 6 2 + +S2 2 0.247204 4.05 2MS 2 2 - WS2 0.077740 12.86 wM2 + wN2 0.159588 6.27 + N2 0.240125 4.16 - 2 ON2 0.082022 12.19 MN 2MN6 2MN 2MN 2 2w 2 2o2 . 2. .. -31- compound tides of importance). Residual circulation corresponds to both steady state (zero frequency) circulation which results from an overtide type interaction or very low period compound tides produced by frequencies of two major closely spaced tides. Let us examine how the non linearities interact with the astronomical forcing tide to produce overtides and compound tides. We shall do this by following an iterative type procedure which is indicative of the overall non linear iterative numerical scheme that The non linearities are treated as right hand has been implemented. side force loadings of sets of equations produced by harmonically separating the (time domain) governing equations. Details concerning this harmonic decomposition of the governing equations are given in Chapter 3. For the present analysis, we shall simply state that a non linear forcing at a given frequency will generate a response at additional frequencies. In general the variables describing tidal motion may be expressed as a harmonic series of the form (for a one-dimensional tidal wave): where: (x,y,t) = _ J(x,y) cos(wjt + * ) (2.12) u(x,y,t) = uj(x,y) cos(w t + (2.13) j) wj = frequency of the jth harmonic component 'n = amplitude of elevation of the jth harmonic uj = amplitude of velocity of the jth harmonic j = phase shift for elevation of the jth harmonic *j = phase shift for velocity of the jth harmonic By definition these series representations are exact for the case of linear deep water tides where only astronomical species exist. -32- They are ~W~CIX~_ EDIIYLil~ -LU~-~--~. -~I~ II approximate UI*E*?6~(1LYL for the case of non linear shallow water tides and the accuracy with which they represent the tides depends on how many terms in the series are considered. The general type of terms produced by substituting our series representations into the non linear finite amplitude term may be studied by only considering a two term series. (un),x f {[u [; cos(wt + cos(wit + This yields: ,) + um cos(&mt + ) + r m cos(Wmt + ,m)1 m) 9x (2.14) This may readily be expanded to: S(u),x (Uln),x cos(W t + I) cos(W t + c9) + (uIm),x cos( Lt + * ) cos( mt + m) + ) Vm ) cos(Wmt + m) + (Um1 + (um We note that Eq. ,x cos(wmt + Jm) cos(t ) ,x cos(mt + (2.15) 2.15 contains both terms which involve only one frequency and mixed terms which involve two frequencies. The former describes the effect of finite amplitude produced by a single tidal component while the latter describes the interaction of two tidal components. The mixed term is the representative term in Eq. 2.15 and may be further expanded through a trigonometric identity as follows: (U1m),x cos(Wt + =2 (Ufm),x ) ) co(Wmt cos[( - ~,)t + (' - m) + Wm)t + (0 + rm ) ]l + cos[w + I COE it -33- (2.16) We conclude that the non linear finite amplitude term may be represented by a harmonic expansion with associated frequencies equal to the sums and differences of all the possible combinations of frequencies present in the expansion for responses. We now apply the previous expansions to examine how overtides are generated through the finite amplitude term. We start out our iterative investigation by assuming that the non linearities are non-existent and that only one astronomical species of frequency wl exists in our estuary. If we treat the finite amplitude term as a forcing term in the continuity equation, we may deduce from Eqs. 2.14 - 2.16 that a tide at frequency wl produces a forcing at both steady state (or zero frequency) and at 2wl. Both these forcings in turn generate responses at their associated frequencies. Hence after one cycle of our iteration we have responses at frequencies 0, wl and 2wl. Each of these harmonic responses will again generate two forcings equal to the sum and difference of the associated frequency. forcings at 0, 201 and 4wl. At Cycle 2 this generates Furthermore the various harmonic repsonses will interact to generate mixed forcings at frequencies equal to the sums and differences of the mixed frequencies. generates forcings at wl, 201 and 3wI. For Cycle 2 this Table 2.4a summarizes the response-forcing interaction which are produced at Cycle 2 by the finite amplitude term when only one astronomical tidal forcing frequency is present. Table 2.4b shows the interaction occuring at Cycle 3 and the process is repeated until a sufficient number of frequencies has been found. These integer multiples of the base frequency wl then are the overtides associated with the astronomical tide at frequency wl. For the tidal problem, the energy transferred to each successive harmonic -34- __IIILILU l__ylLI_1Ls~.~llY--..C^~___ Table 2.4a Response-Forcing Table for Overtides as Generated Finite Amplitude Term at Cycle No. 2 of Iteration Response Frequency 211 0 0 1 2w1 2w1 w1 3w 0 l 4w1 0 1 Table 2.4b __ Response-Forcing Table for Overtides as Generated by Finite Amplitude Term at Cycle No. 3 of Iteration Response Frequency 0 1 0 w1 20 1 2"1 3w1 2wl 3w1 41 0 . 1 4w 1 0 1 3 2w1 15w .1 6w1 301 0 -35- I L -1 11^11. will decrease substantially. Hence only a limited number of overtide frequencies need be considered in order to accurately model the non linear interactions occurring. In fact, for the finite amplitude term, the most significant harmonics are the zero frequency and the 2w~, although an entire series of overtides (3wl, 4wl, etc.) exist. We note that each of these response-forcing tables is indicative of the amount of energy being transferred to the various harmonics, in that the more cycles it takes for a non linear frequency to show up, the less important it is. The convective acceleration terms will produce the same overtide frequencies as the finite amplitude terms, even though the energy will be distributed somewhat differently to each of the various harmonics due to there being different phase shifts for the terms. When two or more astronomical forcing tides exist we can again deduce from our previous expansion that the finite amplitude and convective terms will produce not only the harmonic overtides associated with each individual astronomical tide, but also tidal species with frequencies equal to the sums and differences of combinations of the astronomical frequencies present and their associated compound tides and overtides. Table 2.5a and b illustrate the frequencies present at the first and second cycle. We note that the cycle at which a frequency shows up is indicative of its relative importance. Furthermore we note that all frequencies which appear at a particular cycle are not necessarily equally important as can be seen by examining the individual expansions. The non linear friction term differs somewhat from the finite amplitude and convective terms in that all frequencies generated by the -36- LILI___LL1IILU__Y___LI~-XXI Table 2.5a Response-Forcing Table for Compound Tides as Generated by Finite Amplitude Term at Cycle No. 1 of Iteration Table 2.5b Response-Forcing Table for Compound Tides as Generated by Finite Amplitude Term at Cycle No. 2 of Iteration -37- non linearity are present at the very first cycle. Subsequent cycles only update the magnitude of the forcings and the associated responses. Let us examine the bottom friction term for a one dimensional case. b 7 x P h h S-q,cf u u (2.17) This equation may be approximated by Taylor series expanding the finite depth ratio term to get: b h h+ x u u cfpUnU (2.18) cf1 uI Assuming only one astronomical tide exists we have at the first cycle of the iteration a response of = u (2.19) u cos(w,t) Substituting into the dominating term of our approximation for friction and performing a Fourier expansion we find that the non linear forcing will be: 2 cffulu - cos 5 cfu (0.8488 cos olt + 0.1698 cos 3w1 t - 0.0242 1t + ... ) (2.20) Hence there are forcing terms at all the odd harmonics of the base frequency wl. We note that the major forcing component remains at the astronomical frequency, whereas both the finite amplitude and convective acceleration terms distributed forcing to frequencies other than that of the responses that were generating the forcing. Hence after the first cycle we expect responses at frequencies wl, 3w1, 5w1, -38- etc. It is quite simple to demonstrate that at the second and subsequent cycles, when all these odd harmonics are included in the expansion for velocity, the term cfulul will still only produce forcings at the same odd The precise coefficients of the harmonics as the first cycle. approximating Fourier series will be case dependent. However since the major forcing is at the base frequency, the major reponse generated should be at the same base frequency. For this case of dominant response frequency remaining the same, the Fourier series expansion representing all the interacting frequencies should be quite similar (for the first few loading terms at least) to the very first expansion equation (2.20). It is also the case for compound tides that when one frequency component dominates, the Fourier series approximating cfu uI will be similar to the one generated if only the dominant terms were considered. Finally we note that the (n/h)cflulu term will generate even harmonics (including zero frequency) in the same way that cf uIu generates odd harmonics. It is obvious that the forcings and hence responses due to this second order term are of smaller magnitude than that of the main term. Table 2.6 shows a summary of the astronomical, overtide and compound tidal constituents (at the high frequency end) which could be important in shallow estuaries. Table 2.6 in graphical form. Figure 2.2 shows the information from It is emphasized that these constituents are case dependent and are affected by such factors as the relative importance of the various astronomical tides for the bay under consideration and/or important. which non linear terms in the equations are most We note that all these frequencies modulate over a period of up to 208 days or possibly longer if other components are considered. -39- Table 2.6 f T Af T = = = = Tides of Interest (High Freq. End) frequency of tide period of tide freq. increment between consecutive tides 1/Af = synodic period (length of record required to distinquish between tidal constituents) -40- ~L-LI-L-^LL .--I~Xr S -~~L-1_.ICI__ ~_ 6 H, 0I I S4 **2SM6 "*.* -* .*..__ ...- * MN 4 * **** * * ** * K2 2MN 2 & L2 ....... M2 N2 2MS 2 U1 p, MSf |I I isl So' C -41- -- ---- steady--- .- ,--( I M4 II I It is clear that it is not possible to include in a general sense all of these tidal components with either a perturbation analysis (in which the modeler determines a priori which terms and interactions are to be investigated) or a scheme which uses a Fourier expansion in terms of integer values of a base frequency (allowing only overtides of one major frequency to be studied). The most general and computationally convenient scheme which allows investigation of any number of these closely spaced tidal components is an iterative scheme in which the non linear interactions are treated as force loadings which are somehow harmonically analyzed. chapters discuss in detail the numerical The following techniques applied in the iterative frequency domain scheme used to develop the numerical code. -42- _ CHAPTER 3. 3.1 ~__I~ NUMERICAL FORMULATION Weighted Residual Formulation The spatial dependence of our governing equations will be resolved through the use of the finite element method. In order to apply the finite element method a weighted residual formulation [Connor and Brebbia, 1976] must be established which shall be used as the basis of our finite element formulation. As is shown in Chapter 4, in order to avoid matrix rank problems the algebraic matrix equations produced with our finite element scheme must be solved by substituting the finite element momentum equation into the finite element continuity equation. The resulting formulation is similar to that of Lynch [1980] with the exception that the finite element discretization is continuity and momentum equations are merged. applied before the These manipulations, however, dictate that the continuity equation be used in order to establish the symmetrical weak weighted residual form. It may readily be shown that formulating the fundamental weak form in this fashion leads to elevation prescribed boundary conditions, j*, being essential and normal flux prescribed boundary conditions, Qn*, being natural. Specifically applying Galerkin's method to establish the fundamental weak form, the error in the continuity equation is weighted by the variation in elevation, 6r, and is integrated over the interior domain, Q. Furthermore the natural boundary error must be accounted for by weighting it with 8n and integrating it over the natural boundary SN . It is required that the combined integrated and properly -43- __~TI_*_____*CC3_1___*_A~-. weighted interior and natural boundary errors vanish and the following expression results: ff + Q* l i 18n dO + f T-Q + [v(h+-)] + [u(h+n)] ft) d = 0 (3.1) 0 theorem in order to eliminate the derivatives on the Applying Gauss' flux terms yields: ff {, 9 + 8r - u(h+n)(6r) f fu(h+n)a r where anx, nx ,x - v(h+)(p) + v(h+j)a 1 ny I do 6n dV + (-Q r n + Qn ) 5 n dP = 0 (3.2) any are the direction cosines on the boundary. However, the normal flux may be expressed as: Q n = anx Q + any Qy nx x ny (3.3) y where Q = u(h+) (3.4a) Q = v(h+) (3.4b) Using the previous relationships for flux simplifies Eq. ff In I - u(h+n)(8n) + f Qn 6 r dr + 3.2 to: I dO - v(h+r)(8 ) y f (-Qn+ r., n)' dP = 0 Furthermore the entire boundary, r, is divided into an essential boundary, r E , and a natural boundary r N . -44- This then leads to: (3.5) I___- .ii._ I___WIIPI__ _e II1L_1LIIII____Yli } dQ - v(h+n)(8n) ff {rn 8n - u(h+n)(8r) + f QnSndr + f Q 8dr r nr r E N Qn 6dr+ o ( Q&n 6dr- (3.6) N N The natural boundary integrals of normal flux cancel and the essential boundary integral of normal flux vanishes due to the selection of elevation as an essential boundary condition. This implies that exactly satisfied on PE and therefore the variation, An, "* is is by With these simplifications, Eq. 3.6 leads to definition zero on r E . the symmetrical weighted residual weak form: ff {n - v(h+n)( n) IdQ + f Qn tn - u(h+r)(~i) dTr - 0 (3.7) N We now proceed with the weighted residual method by establishing the weighted residual form of the momentum equations. In order to allow for the possibility of establishing symmetrical final system matrices each of the momentum equations is multiplied through by depth, h. The weighted residual forms for this modified form of the momentum equations is then obtained by weighting the associated errors with residual velocities and integrating over the interior domain, resulting in: ffhu + gh x + h(uu ff fhv + gh, + h(uv - fhv - ,x + vu ,y s 2 2)1/2 u x/p + (h/h+n) cf(u2+ )I 6u dQ = - fhu - T/o + (h/h+n) cf(u2+ ,x + vv )1 6v dQ ,y = -45- (3.8a) 0 0 2 )/2 (3.8b) The weighted residual equations which will be the basis of the finite element formulation have now been established. The final weighted residual equations include first order spatial derivatives of all the variables, n, u, and v. continuity requireumen Hence the associated functional imposed on these variables is that they be continuous over the domain. Furthermore there are first order derivatives taken of the weighting function 6T, again requiring continuity over the domain, while no derivatives are taken of the weighting functions 6u and 6v, requiring only that they be finite over the domain. Since the non-linearities will be treated in an iterative manner Eqs. (3.7) and (3.8) are re-arranged such that non linear terms appear on the right hand side where they can be conveniently updated as pseudo force loadings to a linear problem with each iteration. All boundary and non-variable loadings are also placed on the right-hand sides. Finally, in order to enhance iterative stabilty, a linearized friction term is included on both sides of the momentum equation. These modifications result in the following equations: ff {n &r - uh(6n) fQ } dQ = - vh(6) 6n dr + ff {un(6) + vyn() } dQ (3.9) N ff { hu + ghj ,t - fhv + Xu ,x ff{Ts/p + (X - (h/h+)cf(u x } 6u dQ = + v2)/2u - h(uu -46- ,x - vu )}6u dQ ,y (3.10a) ff { hvt 0 x 3.2 + ghy - fhu + Xv } 6v dQ - ,y f,9x Finite Element Method Formulation In order to generate a system of algebraic equations from integral equations, the finite element method is applied to the final form of the This involves dividing the global domain, weighted residual equations. Q, into element subdomains, Qe, and representing the variables within each element by polynomial expansions. Contributions from all elements are summed and inter-element functional continuity requirements are taken into account in order to generate a global system of equations. To satisfy the minimum functional continuity requirements on the variables linear or higher order expansions must be used for the finite For the development of this method it was felt element approximations. that the simplicity of linear expansions outweighed the improved accuracy achieved (for the same number of total nodes) of higher order elements. Therefore the simplest possible element, the linear triangle, was selected and the variables are expressed within each element as: S= tl1nI1 + 2"2 +4 2 u2 ++ u = 1u 1 + v = ( 1V 1 + 2 v2 3 13 = (n u3 = u(n) 3v 3 = 3 + where: -47- (n) ()v (3.11) (3.12a) (3.12b) 2 (n) 1 (3.13) are the nodal elevation values for element n. v1 U1 u( n ) and u2 v(n) u3 v2 (3.14a,b) v3 are the nodal velocity values in each coordinate direction for element n. Finally, -- (3.15) 2 62] [ l - are the normalized element coordinates for each element. The same linear expansions are used for the weighting functions which is expressed as: (3.16) )(n) 6 = 4 6u = , 6 u(n) (3.17a) 6v = , 6v ( n) (3.17b) 6 Furthermore, linear element expansions were selected for mean water level depths h. In this manner elevations, velocities and depths are -48- all defined at the nodes such that inter-element fluxes are both Hence: continuous and cleary defined. = h 4 (3.18) h(n) Finally, for reasons discussed in Chapter 5 it is desirable to have nodal values for both the linearized friction factor X and our friction factor cf, again requiring linear expansions: and X = cf = X( n) (3.19) (n) (3.20) _ Substituting the finite element expansions into the final weighted residual form of the continuity equation (Eq. 3.7) for each element sub-domain, Pe, and summing over all elements within the domain, Q, leads to: T (n) 8(n) Sel - ,t h 4 u - -,x -- (n ) (n) (n) n) (n) (n) (n) v 4, d, h dQ e 6 (n)d r +(n) * JQ n _ . rT N e n u dr + ff (n) 4 4r (n) (n) (n) Sn.. +4)v 4)9 4 -,x- - - - (n) ,9 ( lpdQl e (3.21) Re-arranging somewhat gives: T S6 T [(n[ ) 9e el =(-fQ4 TN e T T dQ)n(n) 0f 4 h(n)4 dQ)u(n) d )v(n) X1 e e T (n) (n) T +4 4) tx 4) I 4u ( dr) + (n) T 6 h(n) (n) T e -49- (n) (n) }dQ] 4 v (3.22) U(n) Letting E L u(n) (3.23) v(n) 6u(n) , 6U(n) and (3.24) Eq. 3.22 may expressed (n)be therefore as: Eq. 3.22 may therefore be expressed as: (n) ) el (n) -D (n) U (n) Sp(n),lin + p(n) ,nl - el In) (3.25) where the element sub-matrices are defined as: S (3.26) T d fT 4 h e -E _ h (n ) ff j _ -,y SdQ d (3.27) e P(n),lin f 4T Q dr T) rN (3.28) n Ne P(n),nl ff (4 -) -G - (n) 4u- (n) - T +4S -,y 6 r (n) 4)vv(n))dQ )dQ (3.29) e into global matrices leads to the Loading the element sub-matrices following global equations: 6 T [M , - D= -U =T) - t - -P lin nl1 + P ] T) -T -50- (3.30) where 'n = global elevation vector (1 elevation per node) U = global velocity vector (2 velocities per node) = global continuity equation coefficient matrix = global derivative matrix = global load vector for flux prescribed boundaries = global load vector for finite amplitude effects 11 D lin pnl -71 leads to a final set of Allowing for an arbitrary variation in 68q non linear algebraic equations which minimize the error incurred in the continuity equation due to the finite representation of the spatial variables: M r,t - DU lin + Pnn1 - Plin = (3.31) Substituting the finite element expansion into the weighted residual forms of the momentum equation (Eqs. 3.10a,b) for each element and summing over all elements within the domain, Q, sub-domain, Qe, leads to: (n) _(n) [ff{sh elQ X u9 + u $u (n) (n) h(n) ( (n) (n) h(n)(n) ( + 0 h _fd 4 h n) v _9 dQ -- e s ff {9 ff{ Q 4 6u(n)} d Q + ff Q h . Sh) 4u - (n) 4, u -,x- (n) I{A(n) 46 u(n)ldQ fric-u (n) h + -.. for the x-momentum equation and -51- ..v (n)d , u(n) ,y- }Su -- (n) dQ ] (3.32a) Sf S[ff{$ h el 9 e f L (n) v(n) v v + 4 n ) (n) (n) (n)+ g (n) _6v(n) -f h ()un+ _ ( n (n) ) (n dO s f {-~. p _ 6v(n)}do + ff { A-v(n)}dQ ) fric-v e e (h - ff Q e (n) () v 6 u( (n) for the y-momentum equation. h + () (n) v(n) v (n) },v (n ) dQ (3.32b) The right-hand side load terms representing the difference between linear and non linear friction are now denoted as: (n) fric-u A v2 cf(u 2 c (u 2+ f h h+ = (n) fric-v frn)==[ch+n c fv(u 2 v ) 1/2 2)1/2 + u1 (3.33a) v (3.33b) and have not been expanded in terms of the finite element approximations used since they require a numerical integration scheme. Re-arranging Eqs. 3-32a,b leads to: ,T Su(n)[ -U - - el _ h(n) i , dQ)u(n) + (fT, -, t -±- Qe e h(n)_ dQ)v(n) + g(f4T T +(_-ff X(n)_ dQ)u(n) - Sp h(n) - ,x dQ)n(n) e e s d) STTd) f x(n) + (ffrA() (ff±T T-=dQ) e SP -( ffT d fric-u e h(n)_ u(n) (n) T e e -52- h(n) (n)}do)] (3.34a) ___IIILYIYY__JYI__~_L-I~~~IWICiYIZI and 6 v(n) el t- Tfr 9 h(n)f ) h (n)4 dQ)v (n -f.T dC)v(n) + ( - -,t9-- e +(ffT h(n) (n) g(ffT + + g(ffh d()u( e (ff T e (+ +(J e p e -(ff{ P) dP) l(n) T h(n) (n4) cfric-v (n)S,x- T dQ) - (n) T h(n) -..- - ,y - (n)}dQ) (3.34b) e Adding together expressions 3.34a and 3.34b (which does not change these equations due to the arbitraryness of 6u(n) and 6v(n)) and recalling expressions 3.23, 3.24 and 3.27 we have: e _( n) n) [M(n) u(n) + (n) U( +! + g D(n) i (n) = el + p(n) ,nl S fric _(n),nl (3.35) -cony where the element sub-matrices are defined as: 0 effT Sh(n)4) dQ M(n) --U e f 0 II fT M(n) =F T h(n) (3.36) Qe (n)X dQ ff -) -53- (3.37) h(n)X dQ _fffT dQ e (n) C h(n) jf±T (3.38) ) C dQ Qe S x dQ ff e .(n) Sff e T (n) ,nl ff -4T -A-fric (3.39) s - d - dQ p (n) (3.40) dQ n) fric-v e u(n)+ h(n) 4 u (n) ff(4T - - --,x- T h(n) v( n ) u(n))dQ ii )do e p(n),nl -conv (3.41) ( T T ( n 4)u (n)4) v (n)+ 44h (n)4 v h (n) y v )d v (n))dQ ,y - , x- Again loading element sub-matrices into global matrices leads to the following set of global equations: + M U + M U + g DT 6UT [MU -U ,t -F -W -C -54- nl A-fri c nl cony (3.42) where _ = global momentum equation mass matrix M = global linearized frictional distribution matrix M = global Coriolis matrix P = global wind stress loading vector nl P - = global load vector containing difference between linearized friction and full nonlinear friction nl P -cony = global load vector containing convective acceleration effects (non linear) -C Allowing for arbitrary variation in 6U results in the following final set of non linear algebraic equations which minimize the error incurred in the momentum equation due to the finite representation of the spatial variables: M -U Eqs. U + M ,t T U + M U + g D -C -F = P + P W - A-fric - P (3.43) cony 3.31 and 3.43 are still differentially time dependent which will be resolved by applying the harmonic analysis procedure presented in the next section. With the exception of the non-linear frictional difference load vector, (n),nl , all the previous element matrices P-f *-d-fric and vectors may be readily developed in an analytical fashion. The procedure is very straight-forward and the resulting element matrices and vectors are presented in Appendix A. The matrices M matrix, MC, , M and M are all symmetric while the Coriolis is skew symmetric. In the development of the prescribed flux load vector (calculated only on natural boundaries TN) a linear varying normal flux is assumed. For the wind stress load vector PM wind shear stress is considered constant over each element. -55- Since it is not possible to analytically evaluate the integrals in the friction difference load vector, a numerical integration procedure is applied. Allowing for the fact that velocity, elevation and depth all vary linearly, at least a cubic integration formula is required for the non-linear friction contribution. 3.3 Frequency Domain Formulation The finite element technique has been used to resolve the spatial dependence in the governing equations and has thus reduced the non linear partial differential form of the governing equations to the following set of non linear differentially time dependent algebraic equations: - DU M -n - t M U +M tUF tC lin = nl (3.44) Pin + -71 --n U + M U + gD - P (3.45) + P _FU where nl -U (3.46) -A-fric --conv Both the variables r and U and the loadings P lin and P -U are time dependent vectors. Pn P In addition all terms are linear with the exception of the right hand side pseudo forcing terms pnland Pnl which contain non linear combinations of the variables -U n velocity and elevation. The differential time dependence in Eqs. 3.44 and 3.45 will be resolved by a scheme which reduces them to sets of harmonic equations -56- which are coupled through the non linear terms. The non linear coupling will be treated by an iterative updating scheme which shall be discussed The reduction of Eqs. 3.44 and 3.45 from the time in more detail later. domain to the frequency domain assumes that the responses of the system in elevation and velocity may be expressed as a harmonic series of the form: N (t) Re (3.47) e = J=1 Nf = U(t) i t (3.48) ) Re{ 7 U. e j=1 - It is noted that the complex quantities denoted by magnitude and a phase shift. ^ include both a Furthermore it is assumed that both the linear and the non linear load vectors may be represented by similar harmonic series: = P(t) Re{ Nf A Y P io t (3.49) e j=1 Substituting the harmonic series representations of the variables into Eqs. 3.44 and 3.45 and taking appropriate time derivatives leads to: Nf A M ( Y iw = J= 1= N f iW t e j=1 -rj N ijt lin - y P Nf ^ ) - D( 7 Ue J=1 e + i jt ) iWtj f ,n y P J=1 -J -57- = e (3.50) and Nf A i Nf iWJ t Nf N N N .e + gDT( e PW j) pUnl e J+ (3.51) j j=1 =1 = j=1 ) ijt nl f it it f t J=l j=1 J=1 ij e ( ! )+ e U 7 + e U ( 7Ji t These equations may be re-arranged as: Nf j=1 (iw M - fA - A A T -J -iUj_ + =-J F A ^ U + gD (3.52) 0 ei P - (1w !U U + M U + M i& t An - DUj + P j j Ain A A -P iWj t nl - WJ -P -U )e = 0 (3.53) Due to the orthogonality properties of sinusoidal functions (and hence complex exponentials), each of the expressions within the brackets must equal zero. This leads to Nf sets of time independent linear equation of the following type: A iW A Anlin Pi - =D -U. M = A ij 5i!M - A + U +M! + U +M! 4nl (3.54) P -i A U ^ T A + + g D = W + ^nl P (3.55) Note that the natural flux prescribed boundary conditions are Alin The essential boundary incorporated in the load vectors Pj . condition may also be expressed as a harmonic series and hence Nf A j j=1 1 iwj t e . ij e rNj N -58- t (3.56) which leads to a set of boundary conditions associated with each of the set of equations (3.54-3.55) of the form: l N = (3.57) 'n The iterative solution strategy starts out with the assumption that the non linear loadings are zero. Each of these Nf sets of equations are then solved for the boundary loadings imposed. Time histories may then be generated for velocity and elevation with Eqs. 3.47 and 3.48. This in turn allows time histories of the non linear psuedo loading vectors Pnl(t) and Pnl(t) to be produced. n --U As was assumed earlier, these time domain pseudo forcings may now be approximated as harmonic series. Hence the total non linear loadings for continuity and momentum are distributed to all or some of the Nf sets of frequency domain equations. Now each of these Nf sets of equations the entire procedure is is solved again and repeated until convergence is reached. schematic of the iterative scheme is shown in Figure 3.1. A Strategies of how to optimally harmonically decompose the nonlinear load vectors and details such as the number of frequencies required for the harmonic series representation used will be discussed in Chapter 5. Each of the Nf sets of equations are linear at each cycle of the iteration although they are coupled through the non linear loadings. As is clear from Figure 3.1 solving each set of linear equations is the heart of our overall non linear solution scheme. In the next chapter we shall examine methods to solve these linear sets of equations in an optimal manner. -59- BOUNDARY CONDITIONS n AND Q FOR ALL FREQUENCIES wj j = 1,M I: ,- LINEAR SOLUTION AT GIVEN Wj A lin ^* SP l An ^* ,(Q )+ RESULTS IN U SELECT NEXT AND 'j I CHECK CONVERGENCE GENERATE TIME DOMAIN RESPONSE HISTORY BY SUPERPOSITIONING OF SOLUTIONS Uj AND nj U(t) AND n(t) GENERATE TIME HISTORY LOADING Pn(t) USING TIME HISTORY RESPONSE HARMONIC ANALYSIS OF Pnl(t) TO GENERATE FREQ. DOMAIN LOADING P J = 1,M SELECT SET OF w j's FOR WHICH LINEAR CORE IS RUN Figure 3.1. Schematic of Iterative Non Linear Scheme -60- CHAPTER 4. LINEAR CORE MODEL In Chapter 3 the finite element method was applied in order to resolve the spatial dependence of the governing partial differential equations and reduce them to a set of non linear algebraic equations in space with the differential time dependence left unresolved. The assumption of harmonic forcings on the system and harmonically decomposible pseudo forcings due to the non linearities allowed the elimination of the time dependence from this set of equations and thus produced numerous sets of equations in the frequency domain. Furthermore the concurrent assumption of an iterative type solution scheme which updates the non linear pseudo forcings yielded sets of linear algebraic equations of the following type for each required frequency: A A is M r - D U A = (4.1) P iw M U + M U + M U + g D n P (4.2) A The continuity loading vector P can include both contributions from a flux loading at frequency w and from the component at frequency W of a harmonically decomposed finite amplitude pseudo forcing. Similarly the momentum loading vector PU can include contributions from a harmonic type wind loading at frequency w and the components of the non linear pseudo loadings at frequency w due to convective acceleration and the difference between a linearized friction and the full non linear -61- friction. As was noted in Chapter 3 these equations form the core of our fully non linear scheme. As was discussed in Chapter 2, numerous frequencies are needed to accurately simulate shallow water tides with a frequency domain approach. The required frequencies represent either the astronomical tides of importance which are present or their associated non linear Hence the linear core equations need to be over and compound tides. solved for each of these frequencies. model is iterative, In addition, the fully non linear which means that all (or at least most) of these frequencies need to be solved for at each cycle of the iteration until convergence is reached. Therefore it is important that the linear core solution strategy be not only accurate and free of spurious oscillations but also very efficient. The method selected to solve Eqs. 4.1 and 4.2 should also take into account the wide range of frequencies that may be required, from the zero frequency and low frequency astronomical tides and residuals generated up to the very high frequency harmonics. Finally, the method applied should take into account the physical characteristics of typical tidal embayments and the nature of the tidal flows within these Let us now examine some of the various possible ways in embayments. which Eqs. 4.1 and 4.2 may be solved. One possible solution method is to substitute for elevation into the momentum equation and obtain a final equation with velocity as the basic variable. Solving for elevation with the continuity equation yields: A S= 1 i -1 A w MT-n(P A + =D U) -62- (4.3) Substituting Eq. 4.3 into Eq. 4.2 and re-arranging leads to: 2 (w M U + iM -- + iM+ T-1 g DM = D) U-= U _ DT gD -1 M-- (4.4) P - Hence Eq. 4.4 is solved for velocity U which may then be backsubstituted into Eq. 4.3 in order to obtain elevation. This strategy, however, fails due to a rank problem with the total left-hand side system matrix generated in Eq. the momentum mass matrix, 4.4. Due to the w2 factor multiplying , and the w factors multiplying the linearized friction and Coriolis matrices, M and MC, in Eq. 4.4, the contribution of the gDTM matrices. D matrix dominates the sum of these For higher harmonics, the effect of &2M ;;U could disappear entirely due to the round off accuracy of the computer. Table 4.1 shows that the rank of the g DTM1 D matrix is N, making the rank of the total = =T) = system matrix also N. Clearly we can not solve for 2N velocities with a system of 2N equations of rank N. This then indicates that the appropriate manner in which to solve Eqs. 2.1 and 2.2 is to substitute for velocity in the continuity equation which produces a wave-like equation with n as the basic variable. Let us now go into some of the various ways that Eqs. 2.1 and 2.2 can be solved if velocity is substituted into the continuity equation. The first technique examined involves solving for velocity with the momentum equation such that only the mass matrix M need be inverted. Note that if the mass lumping procedures are applied, the matrix M = U simplifies to a diagonal matrix which makes the required inversion extremely economical. More details on the mechanics and implications of lumping procedures are discussed later in this chapter. Eq. 4.2 for U as indicated: -63- Hence solving Sizes and Ranks of Various Matrices N = Number of Node Points Table 4.1 Size Matrix U M ' M- 1 D M- 1 D n M- 1 nT -64- Rank 2N x 2N 2N NxN N 2N x 2N N N xN N ^ - 1 -1 iw U - M U A P - (M +M -U T^ A ) U - g DT1 (4.5) =C The above procedure bars zero frequency cases from being solved due to the 11w term. This problem may readily be corrected by adding the U to both sides of Eq. 4.2, where cs is an arbitrary non-zero term c constant. Therefore Eq. 4.2 is now solved for U as: ^ U TU (i 1 1 -1 + cs t-U A - (M + M - c -C -F T M )U - g D U A (4.6) such that the zero frequency case is not excluded from investigation. Substituting for U using Eq. 4.6 into the continuity equation (4.1) and re-arranging produces: (0 2 - iwc)M -1 - gD -DM -1 DT A P - ( (F -U = +- M _-C A + cs)P -cn (4.7) -1 T gD MU D , and the additional Now both the dominating component, component, A 1 - (i, _)UI )M , of the left hand size matrix are of rank N. (w2 - iwc s---n Hence the rank of the left-hand side matrix is sufficient to solve for N unknown elevations. However, since Eq. 4.7 contains the variable U on the right hand side of the equation an iterative scheme is required to solve Eqs. 4.6 and 4.7. This linear type iteration is distinct from the iteration scheme discussed in previous chapters which updated the non linear terms in the governing equations. However both types of iterations could proceed simultaneously. Placing the very small (w hand side, - iWC )M term in Eq. 4.7 on the right- leads to the following attractive iterative linear core -65- solution scheme: gDM -1 T D ^k+l 2 ( +DM =U 1 Ak+l - (M + F +-C P -U -1 ^ -TP + cs -c M ) (4.8) sU U T k JU- (F (iW+ c ) -U-( ^k n + (i - ic) c - -gD Cs M )U U + ^k+1 n The superscript k refers to the cycle of the iteration. iterative solution procedure is ^1 convergence is (4.9) The linear started by initially evaluating the ^o o right-hand side of Eq. 4.8 using r = 0 and U then evaluating U 1 0, solving for ^1 I , using Eq. 4.9 and continuing the iteration until achieved. As previously mentioned this linear iterative scheme has some very desirable features. First of all, the system matrix, gD -1 T 1D , that needs to be solved is both real and symmetric thus saving in computational expense and storage. Furthermore, the system matrix does not have frequency, w, embedded in it which eliminates the need to re-set and re-solve a system matrix for each of the many frequencies required in the fully non linear scheme. The advantages of this linear iterative scheme, however, by the severe iterative stability problems which can occur, leading to divergence or extremely slow convergence. are offset either The convergence criterion for this linear core model is: Ronv ^k ^k+1 k ^k^k ^ -- ^k- = c 2 2 gh (1 + )gh + -66- (c s --- h f)(a + c s (4.10) (4.10) where Rconv = local convergence criterion and I = element dimension. The actual convergence rate may be checked by examining log(l/Rconv) which indicates improving every cycle. how many decimal places the solution is For example, Rconv equals 0.1 when the solution is improving by one decimal place every cycle of the iteration. The smaller Rconv the more rapid the convergence while for Rconv > 1 the iteration diverges. It is stressed that Rconv is a local convergence criterion and must be satisfied everywhere within the domain. Typical iterative stability problems can be demonstrated by examining a zero frequency flow in a depth varying channel with a constant linear friction factor and no Coriolis effects (see Figure 4.1). For this case the convergence criterion reduces to: R cony 1= -. (4.11) he Note that convergence is dependent on both the selection of a value for c s and the location in the channel. Table 4.2 shows values of Rconv for various selections of cs at different locations in the channel in terms of v, the ratio of maximum to minimum depth in the channel. Only by setting c s = X/hmin will convergence be guaranteed everywhere in the channel regardless of the depth variation. For example, for a depth variation of y = 10, the maximum value of Rconv obtained at any point in the channel by using c s = X/hmin, cs = X/havg and cs X/hmax are respectively 0.9, 4.5 and 9. = Even though the optimal selection of c s = X/hmin produces a stable convergent scheme, it very slow due to the value of Rconv being close to 1. -67- is M.S.L. min max h avg h depth ratio = y - max h. min Figure 4.1 Definition Sketch of Depth Varying Channel Which Illustrates Convergence Problems of Iterative Linear Scheme -68- > 1 - Table 4.2 Variation of Convergence with c s R cony c shallow end average depth deep end hx0 min 2 avg 1 hmax I1 1+1 Y -69- 0 The convergence criterion shows that for cases in which the depth X, vary substantially within the domain, or the linear friction factor, Even though stability problems will arise with this particular scheme. for some cases it convergence is possible to adjust the global factor c s is guaranLeed everywhere within the domain, it is such that a cumbersome procedure to find this value and will result in at best very It is clear from examining either Table 4.2 or slow convergence rates. Eq. 4.11, that optimal convergence (i.e. Rconv = 0) is achieved by selecting local values of c s equal to X/h. is that instead of adding c M What this implies U using a global value for c s to the right and left hand sides of Eq. 4.2, local (or element) matrices S(n) (n) which use the local optimal value for cs should be used to establish a global matrix which is added to the left and right hand sides of Eq. 4.2. However, it is noted that this optimal global matrix is exactly M. What this indicates is that the frictional distribution matrix should be kept on the left-hand side when solving Eq. 2.2 for U. Based on examination of values of Rconv and running computer simulations which apply the linear iterative scheme discussed above for a variety of embayments, this iterative scheme proved impractical for use as the linear core solver. other schemes to solve Eq. Furthermore, examination of various 2.1 and 2.2 led to the conclusion that a direct one pass non-iterative solution technique for the linear core was optimal. As we shall see, this optimal method yields a final system matrix which is complex and non-symmetric. Furthermore, it has frequency embedded into it requiring that at each cycle of the non linear iteration the system matrix must be re-set and re-solved. -70- However, the overall amount of computational effort is substantially less for the one pass linear solution when considering the number of iterations required for the other linear solution methods. This direct linear scheme was implemented as the linear core solver for the overall non linear code. For the direct solution scheme, the momentum equation is now solved for U as follows: A A U - -1 ^ T M (P - g D - TOT -U A (4.12) ) where: TO = (4.13) MF +) (iwM+ Substituting for U into our continuity equation produces: (i ^ -1 IT D M + g D M = = TOT ) A r - P -n ^ -1 P + DM (4.14) = :TOT -U As previously mentioned the total left hand system matrix in Eq. 4.14 is complex, non-symmetric (since the Coriolis matrix M MTOT ) and has frequency embedded into it. is contained in Due to storage limitations it is preferable to re-set and resolve the system matrix for each frequency at each cycle of the iteration rather than storing the matrix produced for each frequency at the first cycle and using it in subsequent cycles of the iteration. Finally in order to make this technique viable, the lumping procedure has been applied not only to the mass matrix MU, but also to the linearized frictional distribution matrix M and the Coriolis M. matrix, ;;c M This now allows the required inversion of ;O -71- to be performed economically. also lumped. For the sake of consistency the matrix M was The lumping for the symmetric mass and friction matrices involves combining all terms on a row onto the diagonal and for the skew symmetric Coriolis matrix combining rows onto off-diagonal terms (in order to retain the skew symmetric natural of the matrix). TOT will be tri-diagonal. Hence matrix The lumping procedure in effect amounts to a slight re-distribution of mass and the linear friction and Coriolis Model results have been shown to be forcings between neighboring nodes. quite insensitive to these lumpings. The linear core model has been verified against the analytical solution for a tidal wave entering a rectangular channel closed at one end both with and without bottom friction damping (Ippen, 1966). The example channel used for this simulation was 25 km long and 4 km wide and had a depth of 10 m. shown in Figure 4.2. The grid representing this channel is A constant element size of 1 km was used yielding 5 nodes across the channel width. A 12.4 hour forcing tide of 1 m in amplitude at the open (ocean) end was used. The linear core model was run for a no bottom friction case, a lightly damped case (with a linearized friction factor, X, equal to 0.001 m/sec) and a heavily damped case (X = 0.01 m/sec). Results of the linear core model for all three cases are shown and compared to the corresponding exact analytical solution (at various locations) in Table 4.3. Agreement between analytical values and numerical predictions is excellent for both elevation amplitude and phase. For the undamped and lightly damped cases, the linear core model slightly overpredicts elevation amplitudes by an average -72- 1 4 i~~.tLL /V A\-.... 't I . ' ~t i I V '1/VV 1/1/~.\~;\; ~-~r Ir I ~ C- t.S VV \~~V\~ ~~.'"' -- , - 1---- ;n L//VI -~------,---- i LI IVik 1 x (kin) Figure 4.2 Finite Element Grid Discretization for Closed Ended Channel Example Case ' Table 4.3 Comparison of Analytical and Numerical Elevations and Velocities for Example Channel Case at Various Locations (a) Linearized Friction Factor X = 0.0000 Elevations Numerical Analytical x Amplitude Phase Amplitude (m) (m) (rad) (m) Phase (rad) 0 1.00000 0.00000 1.00000 0.00000 6000 1.02796 0.00000 1.02800 0.00000 13000 1.05113 0.00000 1.05119 0.00000 19000 1.06273 0.00000 1.06277 0.00000 25000 1.06661 0.00000 1.06657 0.00000 Velocities Numerical Analytical x Amplitude (m/sec) Phase (rad) Amplitude (m/sec) Phase (rad) 0.36747 0.00000 0.36847 1.57080 6000 0.28178 0.00000 0.28408 1.57080 13000 0.18015 0.00000 0.18181 1.57080 19000 0.08997 0.00000 0.09260 1.57080 25000 0.00000 0.00103 1.57080 (m) 0 -74- Table 4.3 Comparison of Analytical and Numerical Elevations and Velocities for Example Channel Case at Various Locations (b) Linearized Friction Factor X = 0.0010 Elevations Analytical Numerical x Amplitude Phase (m) (m) (rad) Amplitude (m) Phase (rad) 1.00000 0.00000 1.00000 0.00000 6000 1.02744 -0.02026 1.02748 -0.02029 13000 1.05039 -0.03637 1.05044 -0.03641 19000 1.06194 -0.04422 1.06198 -0.04425 25000 1.06581 -0.04680 1.06578 -0.04678 0 Velocities Numerical Analytical x Amplitude (m/sec) Phase (rad) Amplitude (m/sec) Phase (rad) 0.36721 1.53906 0.36821 1.53905 6000 0.28158 1.53267 0.28388 1.53274 13000 0.17915 1.52744 0.18167 1.52749 19000 0.08990 1.52485 0.09253 1.52485 25000 0.00000 0.00103 1.52589 (m) 0 -75- Table 4.3 Comparison of Analytical and Numerical Elevations and Velocities for Example Channel Case at Various Locations (c) Linearized Friction Factor X = 0.0100 Elevations Numerical Analytical x Phase (rad) Amplitude (m) Phase (rad) Amplitude 1.00000 0.00000 1.00000 0.00000 6000 0.98152 -0.18880 0.98144 -0.18904 13000 0.98420 -0.34690 0.98409 -0.34718 19000 0.99167 -0.42508 0.99152 -0.42532 25000 0.99507 -0.45097 0.99483 -0.45070 (m) 0 (m) Velocities Numerical Analytical x (m) 0 Amplitude (m/sec) Phase (rad) Amplitude (m/sec) Phase (rad) 0.34439 1.27029 0.34530 1.27017 1.20652 0.26540 1.20734 6000 0.26328 13000 0.16729 1.15433 0.16963 1.15483 19000 0.08393 1.12844 0.08637 1.12843 25000 0.00000 0.00096 1.13888 , -76- of about 0.00005 m which equals approximately 0.005 percent of the amplitude at each point. For the heavily damped case, the numerical predictions for elevation amplitude are slightly below exact values. The predictions for velocity amplitude and phase are also excellent although errors are somewhat larger for velocity than for elevation. We note that this is consistent with the fact that velocities are The linear core model consistently computed as derivatives of elevation. overpredicts velocity amplitudes by about 0.002 m/sec. Although the absolute error for velocity is about the same throughout the channel, the relative error (defined as a percentage of the exact velocity amplitude at a point) becomes large at the closed end of the channel We note that this slight due to velocities decreasing to zero. overprediction corresponds to a small amount of leakage through the closed end of the channel. This is due to the fact that normal flux is treated as a natural boundary condition and will only be satisfied exactly (no leakage) in the limit as the grid is refined. However, in terms of the velocity at the entrance, the error between exact and Furthermore we note that the predicted velocities is less than 1%. finite element method is an error minimization procedure. The errors depend on the degree of spatial discretization and solutions will improve with increased grid refinement. The numerically predicted values shown in Table 4.3 are values for the channel centerline. Node to node oscillations across the width of the channel were extremely small. The character of the maximum node to node oscillation remained about the same regardless of the amount of damping. For elevation, node to node oscillations for this case were somewhat less than the discrepancy which existed between the -77- exact and numerical solution (typical maximum difference in elevation amplitude across the channel was 0.00002 m which is about 0.002% of the amplitude at each point). For velocity the node to node oscillations were slightly greater than the discrepancy between the exact and numerical solution (typical maximum difference in elevation amplitude across the channel was 0.004 m/sec). For general two- dimensional flows these node to node oscillations will increase somewhat. However it is estimated that even under severe depth and geometry changes the elevation amplitude oscillations will typically remain less than 1% with corresponding velocity amplitude oscillations increasing to several percent. We conclude that the linear core model accurately simulates the linearized governing equations at very low node to node oscillation levels. Now that an accurate linear core model has been developed, we are ready to proceed with the more complex task of incorporating the non linear terms in the governing equations into our computations. -78- NON LINEAR MODEL CHAPTER 5. The iterative solution technique used in the development of the fully non linear model was described in Section 3.3. Chapter 4 examined strategies for the optimal solution of each of the linear sets of equations produced with this iterative scheme. Attention will now be focused on the details regarding the implementation of the fully non linear scheme. Of primary importance is the selection of a harmonic analysis procedure for the pseudo loadings generated by the non linear terms in the governing equations. In addition, issues such as iterative stability and convergence rates will be addressed in this chapter. 5.1 Harmonic Analysis of Non Linear Pseudo Forcings The selection of a harmonic analysis procedure for the non linear pseudo forcing vectors is of vital importance for the efficiency, accuracy and generality of the fully non linear model. The efficiency of the model is strongly influenced by the number of time history points required for the harmonic analysis procedure, since the procedure must be applied at every node in the grid at each iteration. The accuracy and generality of the harmonic analysis relate to the type and the detail of harmonic information extracted from a given time history record. A variety of standard Fourier harmonic analysis procedures can be applied to convert time history loadings to frequency domain loadings. All standard Fourier procedures operate with integer multiples of some base frequency. Therefore, Fourier analysis is quite satisfactory when -79- However, as examining one major astronomical tide and its overtides. was noted in Chapter 2, tidal harmonics are not limited to frequencies which are integer multiples of some base frequency. Tidal energy exists throughout a wide range of frequencies which may be extremely closely spaced and are, in general, irregularly distributed. Hence, in order to obtain sufficient frequency resolution when Fourier analyzing the non linear pseudo forcing time histories, an extremely small base frequency step is required. Associated with this very small frequency step is a very large total number of frequencies being processed, most of which In Fourier analysis procedures, time have no associated tidal energy. history record lengths and the total number of time sampling points increase inversely with respect to the frequency step. Hence, the finer the desired frequency resolution, the larger the number of time history data points which need to be generated. Even application of the very efficient Fast Fourier Transform algorithm would be impractical due to the excessive amount of numerical operations required to obtain the frequency resolution needed to separate important tidal components [Oppenheim and Schafer, 1975; Newland, 1980]. A very attractive alternative to standard Fourier analysis procedures is the least squares harmonic analysis method. This method consists simply of a common least squares error minimization procedure which uses a harmonic series as the fitting function. This harmonic series only contains frequencies which are known to exist in the time history record. The method is able to extract extremely closely and irregularly spaced frequency information, yet it only requires a number of time history sampling points equal to twice the number of frequencies contained in the time history record. -80- The almost infinite frequency resolution and the extremely low number of required time history sampling points make the least squares method the optimal choice for Furthermore, the analysis of tidal records. since there are no set requirements for record length and time sampling intervals, this method is ideally suited for the analysis of field data [Munk and Hasselman, 1964; Filloux and Snyder, 1979a; Speer, 1984]. The method is even better qualified for the analysis of analytically generated harmonic response and non-linear pseudo forcing histories since these signals are guaranteed to contain only the exact predictable harmonics associated with a given set of astronomical forcing frequencies (i.e., there is no energy due to non-tidal forcings). In order to harmonically decompose a time history record with values f(ti) at time sampling points ti; i frequency content wj; J = 1,M, - 1,N and with known the following harmonic series is used for the least squares procedure: M g(t) = faj cosw t + b sinj j=1 (5.1) ti where aj, bj are the unknown harmonics coefficients. The squared error between the sampling points and the fitting function is: N E If(ti) = - g(ti ) 2 (5.2) Hence: N E = S r7 i=1 2 M (a cos t + b sinwt j=1 -81- i ) - f(t (5.3) The error minimization is accomplished by setting equal to zero the partial derivatives of E with respect to each of the coefficients aj and bj. Hence: N M (a coswjLi + b sinw t) [ = E -_ f(ti)]cosw t = 0 j=1,N (5.4a) t = 0 j=1,N (5.4b) i=1 J=1 j M N [ 7 (a cosW ti T = 8 j i=1 J=1 + b sinw ti) - f(t )]sinj These equations lead to a complete set of 2M simultaneous linear equations: LSQ a = (5.5) SLS Q where: is the least squares a is the vector of unknown coefficients -LSQ is S -LSQ (LSQ) matrix M the LSQ signal vector Equation 5.5 is shown in expanded form in Figure 5.1. Steady state components in the signal being analyzed simply correspond to a frequency equal to zero in the harmonic analysis series (Eq. 5.1). When setting up the least squares matrix, the zero frequency component will generate a row and a column of zeros. These should obviously be eliminated when generating the LSQ matrix. This then results in a 2M-1 x 2M-1 matrix when zero frequency is included as an analysis frequency. The harmonic least squares method does not have the matrix ill conditioning problems often associated with the least squares procedure -82- i i cos 2 w t sinw t Ycosltisinl t1 i t t Tsinwtcosolt YcoswMtisinwl t sinwMtisin lti Icosw Tcos m cosw Isin2 wlti i i i rf(ti)coslti i Yf(ti)sinw lt 2 sinblticos $t t i t Ycosw tisinMt i Ysinwltsin NMt i Icoswlticos Figure 5.1 Tcos I i ... ,sinMt cosw Mt i f( t Tsin2 Nti i ff(ti)sinwMti ti YsinwMticoswMti i i Linear equation generated by least squares analysis procedure )coswMti applied with polynomial fitting functions. As is seen by examining Figure 5.1, the LSQ matrix is diagonally dominant due to the squaring of the diagonal summation terms, while terms in the matrix are still of the same order of magnitude due to the nature of the sine and cosine functions. The LSQ matrix, MLSQ, need only be generated once in order to analyze any of a number of time history signals with the same frequency content. Hence, SQ is set and LDLT decomposed only once before the cycling begins in the overall non linear iterative scheme. Upon each cycle of the iteration the right hand side of Eq. 5.5, SLSQ which contains the information on the actual time history signal being harmonically analyzed, is re-set. The vector a, which contains the harmonic coefficients being sought, may then be solved for by a forward and backward substitution procedure. Each pair of coefficients ai, bi, contained in vector a for every analysis frequency, are readily converted to the complex form required for the frequency domain pseudo loading vectors. Setting up the signal vector, SLSQ, involves approximately 8MN operations and solving the vector a requires roughly 4M2 operations since the matrix ~SQ is already in decomposed form. These operations must be performed for every nodal point at every cycle of the iteration. In addition, the nodal time history values, f(ti), must be generated at every node for each of the N sampling points. The effort required for this is dependent on the node to element ratio of the grid and on which non linear terms are included in the overall non linear analysis. -84- ~1M~___ ___I__ For a time history signal for which the entire frequency content is both known and used in the harmonic fitting series, the number of time sampling points required to find the precise signal equals twice the number of frequencies in the signal. Together with the orthogonality properties of sinusoidal functions, this requirement insures that all the equations in the system of equations (5.5) will be linearly independent. The fact that the reproduction of the signal is precise may be inferred by examining Figure 5.1 which shows that if f(ti) is substituted by the original harmonic generating signal, becomes an identity. this equation We note that although there is no noise in the signal, there is still an inherent round-off accuracy for the elements in the LSQ matrix and signal vector which places requirements on the time sampling procedures. Let us now examine some of the sampling criteria which allow the LSQ method to be optimally applied. In order to avoid duplicity of information in the elements of the LSQ system of equations, time sampling points should be contained within the overall period of the signal being sampled. However, to allow maximum dissimilarity in each equation, sampling points should be spread throughout the period of modulation of the signal. modulation of a signal is The period of obtained by examining its frequency content and selecting the maximum value of all periods or synodic periods of the harmonic components contained in the signal. The synodic period describes the period of beating of two closely spaced frequencies and is calculated as: T S 2 1 (5.6) 2 -85- where wl, w2 = adjacent frequencies in radians/sec. It is especially important to sample throughout the modulation period of the signal when analyzing records with extremely closely spaced frequencies in order to avoid round-off accuracy problems which could lead to singularity of the matrix M SQ . Hence, when sampling a signal which contains only two very closely spaced frequencies, the four sampling points should be spread over the synodic period which will be much greater than the periods of the individual components. Unlike field sampling procedures, the numerical generation of time sampling points is only affected by the number of points generated and is unaffected by the period of time over which they are generated. If the procedures discussed are adhered to, harmonic signals generated with a typical tidal frequency content (e.g. frequencies of Table 2.6) can be exactly (to within machine accuracy) recuperated while using only twice as many time sampling points as frequencies contained in the signal. The pseudo forcing signals generated by the non linear terms are such that energy is transferred indefinitely to over and compound tidal frequencies. The amount or order of the energy transfer to each frequency is roughly described by the various levels of response-forcing tables in Chapter 2. The number of frequencies appearing, as each subseqent table is generated, increases exponentially. However, the amount of energy transferred to new frequencies becomes increasingly insignificant as they appear. Since it is numerically impossible (and meaningless) to consider all the frequencies that energy is spread to, the harmonic series representing the pseudo forcing must be truncated, which establishes an order of accuracy for the harmonic analysis. -86- This _ ~_~__~4~~ accuracy should be compared with the order of spatial accuracy being achieved by the finite element method. Truncating the series has the effect that no interaction is between the truncated harmonics and those being considered. This will therefore affect the response at the harmonic being analyzed. if allowed However, we have consistently selected as analysis frequencies all those which correspond to a harmonic non linear pseudo forcing above some given threshold, the effect of there being no feedback from the frequencies not considered into the analysis frequencies introduces no more error into the overall computation than the neglect of these frequencies in the first place. We shall now consider the effects the truncations have on the LSQ analysis procedure. Let us examine some results of numerical experiments which are illustrative of the behavior of the LSQ procedure when fewer frequencies are present in the LSQ analysis series than are present in the signal being analyzed. Figure 5.2a shows an input signal with seven equally spaced harmonics with the same input amplitude at each harmonic. When analyzing this input signal with all seven frequencies present in the manner previously prescribed (i.e., 14 time sampling points spread over the maximum modulation), we find that the input signal is exactly recuperated (to accuracy of 14 digits) as is shown in Case 1 of Table 5.1. However, if only four sampling frequencies are included in the analysis series, and 8 time sampling points are used (while still sampling over the entire modulation period), severe aliasing occurs at the higher sampling frequencies. Increasing the number of time sampling points by only 2 corrects the aliasing problem and accurate amplitudes -87- 2.0 1.0. 0 -111111 2 3 1 F quency Frequency 4w 1 5W 6w i (a) Input Signal 2.0. 1.0 0 WF 2eI 3w Frequency (b) Results for LSQ Analysis with 4 Frequencies and 8 Time Sampling Points 1.0- I~-I~ G 0 W 3w1 2wI Frequency (c) Results for LSQ Analysis with 4 Frequencies and 10 Time Sampling Points Figure 5.2 Effects of Variation in Frequency and Time Sampling Rates for Typical Overtide Frequency Distribution -88- I Table 5.1 Analysis Case Frequencies I X . LSQ Analysis Results Showing Effects of Variation of Number of Frequencies and Time Sampling Points; Example Simulating Overtide Type Frequencies Resulting Coefficients for Frequencies N* At** (hrs) 1 2 3 4 5 6 1 1 - 7 14 5.57 1.000000 0.000000 1.000000 0.000000 1.000000 0.000000 1.000000 0.000000 1.000000 0.000000 1.000000 0.000000 2 1 - 4 8 9.75 1.000980 0.000000 1.002718 -0.000161 2.001712 0.005177 1.998599 0.004993 - - 3 1 - 4 10 7.80 1.000048 0.000000 1.000252 -0.000410 1.000820 -0.000818 1.002468 -0.001220 - - 4 1 - 4 14 5.57 0.999265 0.000000 0.998617 -0.000639 0.998885 -0.001276 0.999371 -0.001911 - - 5 1 - 4 56 1.395 1.003836 0.000000 1.007666 0.000410 1.007645 0.000821 1.007611 0.001232 - - 6 1 - 4 224 0.3482 0.999219 0.000000 0.998439 -0.000023 0.998439 -0.000046 0.998440 -0.000070 * N = number of time history sampling points **At = time step between consecutive time history sampling points 7 1.000000 0.000000 - and phases are recuperated from the signal for all the frequencies However, included in the sampling series. increasing the number of time sampling points further does not increase the accuracy obtained whatsoever. 5.1. These effects are summarized in both Figure 5.2 and Table The extent of aliasing depends on how many frequencies are neglected and their associated energy level (i.e., neglecting higher harmonics with small associated amplitudes produces less aliasing). Finally, we note that in the example considered, which is representative of an overtide frequency distribution, there was an increase in economy in neglecting higher harmonics in the LSQ analysis since both the number of frequencies and the number of time sampling points decreased. Care must be taken that responses are not calculated for frequencies which have aliased pseudo forcing amplitudes since this can lead to instabilities in the overall iterative process. Let us now consider an input signal which includes closely spaced frequencies and hence is more representative of compound tidal frequencies. Figure 5.3a shows an input signal with four groups of clustered frequencies. Again, if we sample all 10 frequencies present and adhere to the prescribed sampling procedures (20 time sampling points distributed over the full modulation), we can exactly recuperate the input signal, as is shown in Case 1 of Table 5.2a. However, if only the first seven frequencies are used in the LSQ analysis series (and the last cluster is neglected), the number of time sampling points must be increased by more than 4 times over what it was when all ten frequencies were present. Table 5.2a shows that severe errors occur (e.g., more than 1000 fold increases in amplitude) if the time sampling point density is not sufficiently high. This type of error leads to -90- -LLsrr ~.~u~'-----CI-?~--~-?iliryL"1~- 0!0 0.5 .5 10 Frequency (hr- 1 ) 2 0 (a) Input Signal 2.0 1.01 0.0 0.5 1.5 1.0 -1 Frequency (hr ) 2.0 (b) Results for LSQ Analysis with 7 Frequencies (one entire cluster neglected) and 80 Time Sampling Points 2.0. 1.0 0.0 0.5 1.5 1.0 -1 Frequency (hr ) 2.0 (c) Results for LSQ Analysis with 9 Frequencies (one frequency neglected from within cluster) and 160 Time Sampling Points Figure 5.3 Effects of Variation in Frequency and Time Sampling Rates for Typical Compound Tide Frequency Distribution (maximum period is T = 12.4 hours and maximum synodic period is TS = 27 days) -91- Table 5.2a Analysis Case Frequencies LSQ Analysis Results Showing Effects of Variation of Number of Frequencies and Time Sampling Points; Example Simulating Closely Spaced Compound Tide Frequencies Resulting Coefficients for Frequencies N* At** (hrs) 1 2 3 4 1.0000 0.0000 1.0000 0.0000 5 1 1 - 10 20 38.4 1.0000 0.0000 2 1 - 7 14 54.9 3.5800 1.6209 0.0000 -0.0109 0.9163 1.9452 0.1232 -0.3419 3 1 - 7 16 48.0 1.0543 0.0000 3.0844 3644. 0.0855 -137. 4 1 - 7 40 19.2 1.0589 3.9483 -0.5780 0.0000 -2.0861 1.4183 5 1 - 7 80 6 1 - 7 160 1.0000 0.0000 1626. -94. 1.0000 0.0000 6 1.0000 0.0000 7 1.0000 0.0000 1.3661 1.9080 -1.3900 0.0769 -0.2909 -1.0710 0.5364 -1623. -0.3611 89. -361. 1383. 8 9 10 1.0000 0.0000 1.0000 0.0000 1.0000 0.0000 - - - - - - 0.8362 0.5070 0.9887 0.0312 1.0152 0.0746 1.0601 0.1003 - - - 9.6 1.0271 1.0903 1.0743 1.0735 0.0000 -0.0547 -0.0425 -0.0632 0.9905 0.0158 1.0097 0.0606 1.0490 0.0922 - - - 4.8 1.0120 0.0000 0.9925 0.0176 0.9813 0.0603 1.0056 0.1110 - - - 1.0207 0.0072 1.0193 0.0068 1.0320 0.0004 * N = number of time history sampling points **At = time step between consecutive time history sampling points I,I4 . I I P I # 0 Table 5.2b Analysis Case Frequencies LSQ Analysis Results Showing Effects of Variation of Number of Frequencies and Time Sampling Points; Example Simulating Closely Spaced Compound Tide Frequencies N* At** (hrs) 1 2 Resulting Coefficients for Frequencies ' 3 4 5 6 7 8 0.1001 0.0092 0.0996 0.0068 - 0.7510 0.2290 1.0618 1.0762 0.0157 -0.0147 - 0.9977 0.0006 0.9961 0.0042 - 1.0644 1.0548 0.0219 -0.0059 - 1.0046 0.0239 0.5231 0.3497 - 1.1610 0.8487 0.9925 0.3483 -0.2594 -0.0128 1.0498 0.0529 1.0562 0.0231 - 0.9954 0.0121 0.9989 0.0252 1.0128 0.0374 1.0499 0.0535 1.0572 0.0239 - 0.9982 0.0059 0.9987 0.0150 1.0062 0.0228 1-6, 8-10 18 48.0 0.1908 0.0000 8 1-6, 8-10 40 19.2 0.9809 1.4910 0.0000 -0.3460 9 1-6, 8-10 42 18.3 1.9852 0.0000 10 1-6, 8-10 80 9.6 1.0028 1.0035 1.0037 1.0044 0.0000 -0.0157 -0.0149 -0.0048 11 1-6, 8-10 82 9.4 0.9973 1.0561 0.8064 0.0000 -0.0213 -0.3497 12 1-6, 8-10 160 4.8 1.0039 1.0053 1.0050 0.9992 0.0000 -0.0126 -0.0118 -0.0063 13 1-6, 8-10 162 4.7 1.0039 1.0061 1.0058 0.9994 0.0000 -0.0133 -0.0124 -0.0073 1.0468 0.0843 10 0.9891 1.0170 0.0448 -0.1295 7 0.1101 0.1948 9 0.9872 0.0623 1.0015 0.9923 0.0066 -0.0471 1.1660 1.7550 0.2095 -0.3962 * N = number of time history sampling points **At = time step between consecutive time history sampling points 0.1064 0.1601 -0.0155 -0.1720 1.0134 0.0216 0.0259 0.1286 1.0146 0.0235 1.0020 0.0293 0.9993 0.9661 -0.0036 -0.0886 1.0019 0.0499 1.0122 0.0305 1.0241 0.0356 exponential growth of response in the full non linear iterative scheme and causes complete iterative instability in a matter of several cycles. Again the accuracy of the LSQ analysis doesn't increase with further increases in the number of time sampling points beyond a certain number. Comparing the computational economy of using all the signal frequencies versus using less sampling frequencies and more time sampling points, shows that for this case it is far better to include all the signal frequencies. When using all 10 signal frequencies and 20 time points, 2000 operations are required, while when using only 7 frequencies and the 80 time sampling points, 4676 operations are required. These numbers are only indicative of the computational effort required for the actual LSQ analysis per nodal point (i.e., the conversion of a time history signal to a set of amplitudes in the frequency domain for the pseudo forcing at a node) and do not include the effort required to generate the time history sampling points. This effort could very well be even greater than the increased effort associated with the LSQ analysis. In general, signals with very closely spaced frequencies are much more sensitive to exclusion of frequencies in the LSQ analysis series than signals with a widely spaced frequency content. The net increase in the number of required time sampling points is very case specific and depends on the closeness of frequencies in the signal, the number of dropped frequencies and their associated magnitude. Often we may want to increase the number of sampling frequencies, even though they have magnitude which fall outside of the range of interest, simply in order to decrease the number of time sampling points. -94- There will be an -UI~~-IYYIYIOIF~-Lq~ L~-- optimal balance between the number of sampling frequencies and the number of time sampling points for every case. We note that storage requirements are also effected by this balance between number of time sampling points and number of sampling frequencies and should be taken into consideration. Finally the effect of excluding only one frequency contained in a given frequency cluster from the analysis series is shown in Figure 5.3c and Table 5.2b. We note that frequencies within the cluster from which the frequency is excluded are most sensitive to the exclusion. not illustrated in Table 5.2b, can be shown that if it Although the requirement of spreading time sampling points throughout the modulation period is not met, energy from the missing frequency will be lumped to the other frequencies in the cluster in a constructive or destructive manner, depending on where in the modulation period the time sampling points lie. Table 5.2b also illustrates a convenient methodology for determining whether or not a certain number of time sampling points is sufficient for the number of frequencies and characteristics of the signal being analyzed. By simply adding two time sampling points and accordingly adjusting the sampling time step (such that all time points are evenly spaced over the modulation period) a totally different set of time sampling points is generated. If the results of the LSQ analysis procedure are the same, then the number of time sampling points is sufficient. As has become apparent in the preceding paragraphs, two categories of frequencies may be defined. The first category consists of those frequencies which have sufficiently large associated forcing/response -95- amplitudes that they should be included in the full non linear analysis in order for the computation to be consistent to a given order of accuracy. At these full non linear analysis frequencies, responses are calculated (and hence the linear core model is run) allowing full interaction between all frequencies in this category. The second category of frequencies consists of frequencies which do not have significant enough levels of forcing/response amplitudes to be included as full non linear analysis frequencies, but are used to allow the LSQ procedure to more efficiently extract accurate information at the full non linear analysis frequencies. These frequencies do not affect the interaction occuring between the full non linear analysis frequencies. As seen earlier, the number of time sampling points required is very case dependent. Furthermore a re-analysis of the signal with a slightly higher number of time sampling points allows us to assess if the time sampling density is sufficiently high for the LSQ procedure to accurately analyze the signal. This allowed the convenient implementation of an automatic time step selection feature in the computer code TEA-NL (Non Linear Tidal Embayment Analysis), which chooses the minimum time sampling rate to achieve a specified level of accuracy for the non linear frequency domain pseudo forcing amplitudes at each of the LSQ analysis frequencies. This time step selection process need only be applied at a few representative nodes in the grid at each cycle. If the number of time sampling points required exceeds a user specified maximum number (the number at which it is more efficient to include more frequencies for the LSQ analysis), TEA-NL stops execution and allows the user to input additional LSQ frequencies thereby premitting the LSQ analysis to be performed efficiently. -96- The number of frequencies generated for a typical tidal problem with several significant astronomical forcing frequencies is quite large. Although the tabular method of Chapter 2 gives a rough idea of the importance of frequencies, we won't a priori know precisely which frequencies have sufficiently large forcing/response amplitudes to be included in the overall non linear analysis. TEA-NL has been set up to select those frequencies which should be included as full non linear analysis frequencies based on a user specified non linear pseudo forcing threshold. The threshold is a percentage of the maximum spatially averaged non linear forcings of all the frequencies. This automatic selection procedure of full non linear analysis frequencies allows the convenient and economical use of TEA-NL while ensuring that the order of accuracy of the harmonic analysis (and hence the non linear interaction) is consistent providing the user does not neglect to input certain important frequencies. Frequencies with spatially average forcing amplitudes less than the specified percentage of the maximum will be excluded as full non linear analysis frequencies and are used only as LSQ analysis frequencies. This percentage is taken as a fraction of the expected smallest quantity of interest. 5.2 Iterative Convergence The iterative stability of every scheme which solves a set of non linear algebraic equations through direct iteration is dependent on the magnitude of the right hand side term being smaller than the magnitude of the linear terms on the left hand side of the equations. Therefore the success of the fully non linear scheme used in TEA-NL is dependent on the non linearities being of sufficiently small magnitude with -97- respect to the linear terms so that the non linear solution is perturbed linear solution. only a For most tidal embayments, the non linear terms in the governing equations do not dominate the linear terms. However, the magnitude of the non linear friction term may become quite substantial in shallow embayments with rapid velocities. Fortunately, as is indicated by Eq. 2.20, for the case of a single astronomical forcing tide, the major portion of the harmonically decomposed friction term is distributed to the main astronomical forcing frequency itself. It can also be shown that for the case of several astronomical forcing tides, the dominant tidal frequency (usually M2 ) will still have the largest pseudo forcing contribution from the non linear friction term. Furthermore the magnitude of the coefficient of the forcing at the dominant frequency will be close to the case where only the dominant astronomical forcing exists. We note that this dominant harmonic friction term may be approximated as a linearized friction term and incorporated with the other linear terms on the left hand side of the equations. This led to the use of the linearized friction term on both sides of the momentum equations in Chapter 3. Hence the iteration now occurs about a right hand side loading term which equals the difference between a linearized friction and the fully quadratic friction term. the linearized friction factor is properly estimated, this can reduce the right hand side loadings by an order of magnitude. For the fully non linear scheme this increases the rate of convergence substantially and in cases where friction dominates, makes an otherwise divergent iteration converge. We note that the value of the non linear friction coefficient, cf, is dependent on the bottom surface, whereas the linearized -98- If friction coefficient (see Eq. 2.6) is a property of both the non linear friction coefficient and the local flow. The effectiveness of including linear friction in decreasing the magnitude of the right hand side loading term is dependent on how closely the linearized friction term approximates the loading term of the harmonic expansion for non linear friction (in Eq. 2.20). The most convenient scheme to obtain a good local estimate for linear friction X is to update the user prescribed value at the beginning of the second cycle of iteration using nodal values of cf and the results of the first iteration for the nodal magnitude of velocity for the dominant frequency. Hence, program TEA-NL only requires that a reasonable global linearized friction factor (based on some global cf and global estimate for velocity) be specified in order to start the iterative process. The updated nodal values for X obtained in the second iteration are not only helpful in speeding convergence rates, but also allow an improved fully linear solution (if TEA-NL is run in the linear mode) due to the much improved local values for X. Finally we note that iterative convergence rates may be improved by including the computation of finite amplitude and convective The acceleration effects only beyond the second cycle of iteration. reason for this is that friction is usually the dominant term and once the repsonses in elevation and velocity have adjusted for it, the effects of the other terms are more accurately assessed. if For example, the user specifies a linearized friction factor which is too low, the elevation amplitude in the first cycle would be overestimated. This in turn results in the over-adjustment for finite amplitude effects in the second cycle, while the elevation amplitudes being calculated in the -99- second cycle would have adjusted better to the actual non-linear friction. However, including finite amplitude effects only beyond the third cycle of iteration does not allow this type of overcompensation. So far we have seen how convergence rates could be improved. However, the question that remains is at what point we can consider the solution to have converged. TEA-NL is Obviously the computational effort of directly related to the total number of iterative cycles that need to be run. When determining the degree of accuracy which the iteration process should achieve we should consider the following points: (i) There is an order of accuracy associated with the finite element method which was used for the spatial discretization of the governing differential equations. The accuracy depends on the grid size and the types of gradients (relative to the grid size) for both the flow field and the depth variation. Furthermore, the linear core solution exhibits a certain degree of node to node oscillation in for elevation and velocity. the solution calculated This oscillation may be typically quantified to the order of several percent of the magnitude of the solution at a given node. The oscillation is somewhat greater for velocity than for elevation. The accuracy of the computation is not improved by carrying the iterative accuracy beyond the estimated percentage of the node to node oscillation. This then indicates that achieving a relative accuracy of several percent at each frequency is sufficient for iterative convergence. -100- (ii) We note that non linear pseudo forcings are generated with elevations and velocities which contain a certain amount of node to node oscillation. Hence we expect some deterioration in the solution achieved at each of the various levels of frequencies described in Chapter 2 (i.e., more node to node oscillation). The degree of deterioration depends on the magnitude of oscillation relative to the overall magnitude of the forcing (signal to noise ratio). Furthermore, it depends on which of the non linear terms are included in the analysis and their relative importance. For example, the finite amplitude forcing term is due to a gradient of the product of elevation and velocity. If the gradients in elevation and velocity are smaller than the relative node to node oscillation (which depends on the magnitude of the terms), we would expect a meaningless set of harmonic pseudo forcings to be generated. In general, the forcing signal to noise ratio will be such that it will allow the meaningful calculation of significant response frequencies. For the finite amplitude and convective acceleration pseudo forcing terms, this level of deterioration increases as energy is cascaded down to the various levels of frequencies. For the friction pseudo forcings this signal/noise effect is less pronounced due to the fact that the forcing term does not include any derivatives and furthermore energy is cascaded from the major astronomical forcing levels to all the compound and overtide frequencies at once. We conclude that this noise in the -101- pseudo forcings must be considered when determining the convergence achievable at each of the frequencies. (iii) There is an order of accuracy associated with the truncation of the harmonic series used for the time discretization of the governing equations. Hence the frequencies taken into consideration are effected to a certain degree by the lack of interaction with the missing harmonics. As previously mentioned the overall accuracy of the computation by not considering this interaction is no worse than that caused by not considering these terms in the first place. However, performing a calculation of a given frequency beyond the estimated percent of the missing non linear interaction would be meaningless. (iv) Bottom friction and bottom topography can only be described to within a certain degree of accuracy. Therefore the iterative accuracy being sought should also take into account the variability in response associated with the uncertainty in parameters. All these points should be taken into consideration when determining the level of accuracy which the iteration process should achieve. This level is case dependent and also varies for each of the frequencies for which the calculations are being performed. Program TEA-NL allows the determination of the level of accuracy achieved and the rate of convergence by calculating the following -102- ~'-L-~'mrrrrrYr~-lir~jYBrr~ a~rr~. -rrc~ parameters at each cycle and for each freuqency under consideration: Dn = maximum global difference between magnitude of elevation calculated at consecutive cycles DU = maximum global difference between magnitude of velocity calculated at consecutive cycles Dr = average global difference between magnitude of elevation calculated at consecutive cycles DU = average global difference between magnitude of velocity calculated at consecutive cycles n average global value of elevation amplitude U = average global value of velocity ampltiude RP = convergence rate for elevation amplitude RU - L = relative convergence level for elevation convergence rate for velocity amplitude LU = relative convergence level for velocity The relative convergence levels are defined as: D L (5.7a) DU LU (5.7b) -- U It is these values that should be used to determine convergence in TEA-NL. Typically they will be on the order of several percent. We shall discuss values for these parameters in more detail in Chapter Finally, we note that R, 6. and RU are not only indicative of the improvement in accuracy with each cycle of iteration, but may also be indicative of the level of signal to noise in the pseudo forcings. -103- CHAPTER 6. APPLICATION Program TEA-NL is a very flexible computer code which allows the general calculation of non linear tides. ability to compute compound tides. TEA-NL is unique in its As an example application, TEA-NL will be used to investigate tidal circulation within the Bight of Abaco in the Bahamas. This is an ideal application not only because significant non linear tides are generated, but also because extensive field data collection and analysis have been performed for this shallow semi-enclosed basin (Filloux and Snyder, 1979). Furthermore, the basin is such that it allows simple boundary conditions to be applied for the non linear tides. 6.1 Description of Bight of Abaco and Its Tides The Bight of Abaco (Figure 6.1) is a shallow embayment with land boundaries consisting of the Island of Abaco along the southern and eastern parts of the embayment and the islands of Little Abaco and Grand Bahama along the northern parts. Although the northwestern part of the embayment does have a number of shallow connections to the open ocean, data taken by Filloux and Snyder (1979) showed that these openings were relatively opaque to the tides and could therefore be treated as land boundaries. They reached this conclusion by comparing the amplitudes of major astronomical tides at several locations lying within the bight along the northwestern boundary to those at a nearby location in the open ocean. -104- --- EICILI^* -- ~-*r--rrs(ul --I- ~*axmrrr^-- Figure 6.1 Geography of Bight of Abaco, Bahamas. represents 200 m contour. -105- ~-lrr (lre~~ ~ ---- l-~(~ Dotted line The western edge of the bight is connected to the open ocean and is characterized by an extremely sharp discontinuity in depth. The 200 meter depth contour lies between 1 to 3 kilometers from the 5 meter contour and the 1000 meter contour lies between 3 to 15 kilometers from the 5 meter contour. Ultimately depths drop to between 1500 and 2000 meters. Hence depths on the ocean side of the boundary drop by a factor of between 200 and 800. reflected. As a wave passes over a step of this size, it is largely This may be shown by considering the reflection and transmission coefficients of an incoming long wave passing over a step from depth h1 to depth h 2 which are expressed as (Ippen, 1966): /h I /h K r = K = t -1- 1 (6.1a) /h 2 + 1 1 2 /h/h 1 (6.1b) + 1 2 Values for various depth ratios h1 /h2 are shown in Table 6.1. These equations, which were derived for a vertical step, will apply to the Bight of Abaco case since even the shortest possible overtide wavelength (M6 at a depth of 2 m has a wavelength of 66 km) is many times greater than the distance over which the most substantial portion of the depth drop occurs. For the Bight of Abaco h /h2 has a range of between 0.005 and 0.001 and it is seen that the reflected wave has an amplitude between 0.87 and 0.94 of the incoming wave and is reflected out of phase with respect to the incoming wave. The transmitted wave has an amplitude between 0.13 and 0.06 of the incoming wave and is in phase with the -106- _ _____1_LI1LCIIYlliI~~i -I Ue~-~I~ -I~__~IXI -(~~ Table 6.1 Reflection and Transmission Coefficients for a Long Wave Passing Over a Step from Depth h1 to Depth h2 for Various Depth Ratios h 1 h Kh r K t 0.1 -0.52 0.48 0.01 -0.82 0.18 0.005 -0.87 0.13 0.001 -0.94 0.06 -1.00 0.00 0 -107- -^X incoming wave. When considering the lateral expansion that occurs, the amount of reflected energy increases and the transmitted wave becomes even smaller. Hence it is a reasonable approximation in this case to assume that the wave is totally reflected. This allows the very convenient treatment of the boundary conditions for the non linear tides. Since depths on the open ocean side of the boundary are very deep, there will be essentially no non linear tides generated there. The non linear tides generated within the shallow bight, however, will be reflected back into the bight due to this severe depth discontinuity. As a result no non linear tidal species will exist in the open ocean and their amplitudes should be specified as zero along the ocean boundary. Figure 6.2 shows the bathymetry within the bight. In the region along the open ocean boundary, depths vary between 2 and 5 meters. This region actually forms a sill since depths increase again in the interior of the bight. In the northern half of the bight, a 7 - 8 meter depression is the dominant feature. Depths become very shallow along the northern edge. Bottom characteristics also vary somewhat within the bight. The sill region has a bottom surface characterized by numerous sand waves with heights between 1 and 3 meters. represented in the depth distribution. These sand bores are not The northern depression region contains muddy mounds with a relief of 10 cm and horizontal scale of several meters. The relatively flat southern portion of the bight has a bottom surface consisting of a thin sediment cover over rock, punctuated in patches by sea fans and corals (Filloux and Snyder, 1979; Snyder, Sidjabat and Filloux, 1979). -108- Figure 6.2 Bathymetry of the Bight of Abaco, Depth Contour in Meters. -109- Bahamas. Filloux and Snyder (1979) have run a series of three field experiments, each lasting approximately one month, which measured elevation at 15 locations within the bight. At each location bottom mounted tide gauges with a sensitivity of 1 cm collected a time history of bottom pressure. The bandpass characteristics of the data are such that steady motion, surface wave motion and turbulence are excluded from the data records collected. These time history records were then harmonically decomposed using a least squares analysis procedure which uses 5 astronomical frequencies (01, K 1 , N 2 , M2, S2) and two overtide frequencies (M4 and M 6 ) for the analysis series. A time point sampling rate of 4 data points per hour was used from the available recorded 40 readings per hour (Filloux, private communication). Atmospheric pressure records were also collected and harmonically analyzed in the same manner as the tidal records. This allowed the bottom pressure amplitudes to be adjusted to reflect only water pressure variation. Hence surface elevation amplitude and phase data were obtained for the M 2 , N 2 , S2, 01 and K1 astronomical tides and the M 4 and M 6 overtides at 25 points throughout the bight. The M 2 tide is the major component having an amplitude of 40 cm along the open ocean boundary. The amplitudes and phases along the open ocean boundary of the five astronomical components are summarized in Table 6.2. The amplitudes and phase lags for surface elevation, obtained at the various measurement points in the bight for the M 2 , M 4 , M 6 and N2 tides are shown in Figures 6.3 through 6.6. These figures only reflect data which Filloux and Snyder (1979) considered to yield consistent and stable estimates. -110- eru~irarrari-pr---,~--uLIrruu,~,~yiyaF1 Table 6.2 Summary of Measured Astronomical Tides Along the Open Ocean Boundary Tide Amplitude (cm) Phase Lag (radians relative to M 2 ) K1 9. 3.5 01 7. 3.7 N2 10. 5.9 M2 40. 0.0 S2 6. -111- 0.9 __~~~~~_~V +39.2 Measured Elevation Amplitude + 21.E 419.6 16.4 +17.2 37.4 +39.0 17.5 18.5 +29.2 +14.6 15.6 13.5 16.9 +18.3 15. 3 +14 .3 1I. 26 9 +38.8 + 12.2 + 39.4 \.15 .5 15.9 14.5 27.9+ +12.7 15.8+ 14.1 16.7 + 16.7 15.9 +16.6 +16.3 +18.1 17.617.9 17.1 1 .1 Figure 6.3 19.1 19.2+ 18.8 Field Data for M2 Astronomical Constituent (after Filloux & Snyder, 1979) (a) Amplitude in centimeters -112- *L(LI*_IIY~ _1_I1IILI__II~ + -0.07 +0.56 +0.87 Measured Elevation Phase Lag 0.96 +0.89 0.00 +0.00 1.15 +0.96 0.87+ 0.86 1.29 1.33 0.82 1.32 +1.10 +0.17 0.35+ + 0.19 -0.02 40.94 + 0.07 1.50 1l.52 1.52 0.17+ +1.70 1.94+ 1.83 +1.68 1.80 + 1.85 + 1.73 1.92 88 Figure 6.3 2.01 1.95+ 1.921.76 Field Data for M2 Astronomical Constituent (after Filloux & Snyder, 1979) (b) Phase lag in radians (relative to the M2 tide) -113- -_I~--~- I M +0.2 0F Measured Elevation Amplitude +0.8 +2.1 1.2 1.4 +0.2 0.2 ,1.2 +1.2 + 1 +0.3 0.9 0.8 1.2 1.1 0.8 +0.7 0.6+ 40.2 +0.3 +0.2 0.3+ 0.9 + 0.5 0.5+ 0 .2+ 0.5 0.3 + 0.5 +0.5 0.3 0.5 + 0.7 + 0.6 0.5 0.5 + 7 0.6 0.9+ Figure 6.4 .0+ 0.7 0.9 Field Data for M4 Overtide Constituent (after Filloux & Snyder, 1979) (a) Amplitude in centimeters -114- , I, -r*--a~ Y1-Ya~ul---rrrcP;~~iIIYTPll)lli- +1.57 Measured Elevation Phase Lag + 0.33 + 1.22 1.26 1.33 A+ 1.33 1.06 +; +6.02 0.38+ 1.74 1. 06 +2.39 +0.38 +1 i.59 +1.52 44.92 + 1.95 + 1. 5.65+ + 3.56 4.05+ 4.05+ 3.91 2.22 2.44 2.88 +2.91 +2.73 2.72 2.60 +2.56 2.88 2.90+ Figure 6.4 2.60+ 2.95 3.04 Field Data for M4 Overtide Constituent (after Filloux & Snyder, 1979) (b) Phase lag in radians M2 tide) -115- (relative to the +0.3 r Measured Elevation Amplitude +0.5 +0.1 0.2 0.3 0.2 0.2 0.8 1.6 +1.3 0 0.8 0.7 0.8 +1.0 +0.3 0.6 + + 0.4 +0.2 +0.7 +0.2 +0.7 0.7 0.7 0.3+ +0.3 0.4+ 0.4 0.2 +0.4 0.4 + 0.3 +0.6 0.5 0.3 + 0.4 0.5 0.5+ Figure 6.5 07 0.5+ 0.6 Field Data for M Overtide Constituent (after Filloux &6 Snyder, 1979) (a) Amplitude in centimeters -116- ~II-----"II~-I~L~L~" -T~ yl ~~_ +3.80 "" Measured Elevation Phase Lag +1.71 +0.23 4.40 4.69 + 5.67 4.63 +4.40 4.82 4 3.79 + 3.16 3.7 3.63 3.32+ 5.10 +5.15 + 2.37 +5.25 +3.18 44.28 4.92 +5.95 2.34+ 4.97 +4.76 4.94 3.91+ 1.74 1.90 5.58 5.29 5.67 +0.45 +5.64 0.38 6. 25 0.10 6. 21 ' 0 .28 0.70+ 1.17 0,45 Figure 6.5 Field Data for M6 Overtide Constituent (after Filloux & Snyder, 1979) (b) Phase lag in radians (relative to the M 2 tide) -117- _ +9.7 Measured Elevation Amplitude -,o + 3.7 +3.2 3.4 +10.3 9.5 4.2 4.1 +7.6 2.7 3.8 3.4 4.3 4.0 + +7.0 +10.3 3.3 2.6 + 2.9 2t 9 +10.0 7.7+ 2.2 + 3.8 3.3 +2.0 3.6+ 3.4 2.2 2.7 3.5 +2.3 +2.4 2.7 2.1 1.7 +4.3 4.0 3.6 3.2 + 2.3 Figure 6.6 Field Data for N2 Astronomical Constituent (after Filloux & Snyder, (a) Amplitude in centimeters -118- 1979) ~~-X aar~rs-- I.aar~ i)iWII~~~PIYILc--arx~ +0 ' + 5.78 Measured Elevation Phase Lag '.19 o061 +0.61 0.52 0.70 +0.79 5.85 +6.02 4+ 6.00 6.14 + 0.87 +~1.13 0. 61 +1. 10 +6.005.85 +0.68 -.0.00 6.02 0.93 +1.03 + 1.33 0.98 + 1.41 1.41 + 1.95 0.91 + 1.20 1.69 + 1.22 41.06 1.31 0.98 1.48 1.43+ 1.82 Figure 6.6 1.50+ 1.85 +1.50 Field Data for N2 Astronomical Constituent (after Filloux & Snyder, 1979) (b) Phase lag in radians (relative to the M2 tide) -119- Filloux and Snyder (1979) predicted the theoretical statistical error associated with their data analysis procedure. Their actual errors, determined by the variability in results at a point from experiment to experiment, generally substantially exceeded the computed theoretical statistical error. They suggest that this may be the result of not having included more harmonic components in their least squares analysis procedure. As we shall see later in this chapter, not including more harmonic components could very well have led to increased errors associated with certain of the harmonics which Filloux and Snyder were able to extract from the time history records. The variability in the analyzed measurement data may be quantified by calculating the proportional variance for each harmonic. The proportional variance is computed by summing the square of the difference between the elevation amplitude value obtained from experiment to experiment at a particular location and the average value at that location and then dividing by the sum of the squares of all the measured values. For the computation of proportional variance of measurement data only locations with more than one data point are used. The proportional variance of measurement data for each harmonic j is expressed as: ' K L Vm= £= K IKm nj (x, k) - 1 Z 2 k n. (x' k) (6.2) k k=1 k=l L SK £ Snj(x 2 , k) £=l k=1 where m n. = measured elevation amplitude component for jth harmonic xk measurement location within the bight = -120- L = total number of measurement locations with multiple data points K = total number of measurement data points at location Z The reduced data presented by Filloux and Snyder (1979) was used to compute the proportional variances for each of the seven harmonics used in their data analysis. These proportional measurement variances, Vm, were computed for the bight as a whole and are presented in Table 6.3a. Furthermore proportional measurement variances are presented for the three distinct regions (the sill region, the northern bight and the eastern bight) into which the bight may be divided. Each of these three regions contain an equal number of multiple data point measurement locations. The proportional variances are generally equal or somewhat lower in the sill region and eastern part of the bight (when compared to values for the entire bight) and are generally substantially greater in the northern part of the bight. A somewhat easier way to interpret these proportional variances is to examine the proportional standard deviation which is equal to the square root of the proportional variance: Sm j = i (6.3) j The proportional standard deviation may be viewed as the standard deviation in terms of a fraction of a global representative measure of amplitude. Table 6.3b presents the associated proportional standard deviations for measurement data. It is noted that the M 2 , 01 and K1 tides (K1 measurement errors are consistent with M 2 and 01 in the northern and eastern bight and K1 is therefore included) all show very -121- Table 6.3a Measurement Error for Each Frequency in Terms of Proportional Variance, Vm J m V Tide Entire Sill Bight Region Northern Bight K1 0.0063 0.0015 0.0154 0.0006 01 0.0008 0.0007 0.0013 0.0005 N2 0.0171 0.0055 0.0450 0.0126 M2 0.0009 0.0012 0.0007 0.0013 S2 0.0205 0.0134 0.0530 0.0009 M4 0.0105 0.0134 0.0285 0.0014 M6 0.0144 0.0054 0.0294 0.0120 -122- Eastern Bight ~YI~ Table 6.3b Measurement Error for Each Frequency in Terms of Proportional Standard Deviation, m S Tide Entire Sill Northern Bight Eastern Bight Bight Region K1 0.08 0.04 0.12 0.02 01 0.03 0.03 0.04 0.02 N2 0.13 0.07 0.21 0.11 M2 0.03 0.03 0.03 0.04 S2 0.14 0.12 0.23 0.03 M4 0.10 0.12 0.17 0.04 M6 0.12 0.07 0.17 0.11 -123- modest measurement errors while the N2, S2, M 4 and M 6 tides all exhibit somewhat higher errors. There is no correlation between average amplitude in a region and the error incurred in the reduced measurements. As previously mentioned, we will later show that the N 2 , S2 , M 4 and M 6 tides display greater measurement error due to the existence of significant (relative to each of these tides) closely spaced compound tides which were not used in the data reduction procedure. In the following two sections we shall examine the circulation patterns predicted by TEA-NL for the bight. First, only the M 2 astronomical tide and the steady, M4 and M 6 overtides it generates will be considered. Then the much more complex interaction of the closely spaced M 2 and N 2 astronomical tides and their associated compound and overtides will be investigated. 6.2 Overtide Computations for the Bight of Abaco TEA-NL requires both a description of the geometry of the embayment, in the form of a finite element grid, and a set of boundary conditions for each of the various harmonic components being considered. The finite element grid discretization for the Bight of Abaco is shown in Figure 6.7. The grid has been refined in the area surrounding the small island which lies along the ocean boundary and also in the region adjacent to the Island of Grand Bahama. In this section we shall examine the circulation resulting from the main astronomical (boundary forcing) component and its most significant overtides. Along the ocean boundary the M 2 amplitude of 40 cm measured by Filloux and Snyder (1979) is used as the elevation prescribed boundary -124- ~---~~--L-ieu ---l----- Figure 6.7 Finite Element Grid Discretization for Bight of Abaco, Bahamas. -125- condition. Furthermore for the steady, M 4 and M 6 overtides, amplitudes of zero are prescribed along the ocean boundary due to its reflective nature. Land boundaries are all specified with zero normal flux for all frequencies. In order to determine the sensitivity of variations in bottom friction, three sets of overtide runs were performed with TEA-NL corresponding to non linear friction factors of 0.003, 0.006 and 0.009. Results for both elevation amplitude and phase of the M2 component are shown in Figures 6.8 through 6.10. The effects of the increase in friction factor cf are illustrated in Figure 6.11b which shows M 2 elevation amplitudes along the trajectory defined in Figure 6.11a. This trajectory passes through three representative areas in the bight; the sill region, the central bight and the northern bight. The increases in bottom friction correspond to decreases in elevation amplitude and increases in phase shifts although the overall pattern of the distributions remain similar. The sill region, which corresponds to a very shallow region which has rapid flow, is a high gradient region and shows the most substantial reduction (damping) in elevation amplitude and the largest increases in phase lag. in the Elevation amplitudes are the smallest center part of the basin and increase again somewhat towards the eastern and northern land boundaries. The increase in amplitude is the most significant along the northern edge of the bight where depths decrease very rapidly (from 8 m to 1 m). We note that for a non linear friction coefficient of cf = 0.009 there is excellent agreement between the M 2 elevations predicted by TEA-NL (Figure 6.10) and the measurements by Filloux and Snyder (1979) (Figure 6.3). -126- This value for cf is consistent Irr~. -L--~-------cPm~i~ r~.*~.~g(. -u~~ 32- 528 Figure 6.8 Results of TEA with Full Non Linear Friction Effects (cf = 0.003) for M2 Astronomical Constituent (a) Amplitude in centimeters. -127- O.N 0.o( \ run D2 PHASE 1.0 2 1.5 Figure 6.8 Results of TEA with Full Non Linear Friction Effects (cf = 0.003) for M2 Astronomical Constituent (b) Phase lag in tide -128- radians (relative to M2 run el AMPLITUDE 2'/ 32 Figure 6.9 Results of TEA with Full Non Linear Friction Effects (cf - 0.006) for M 2 Astronomical Constituent (a) Amplitude in centimeters. -129- 0.0 10 PHASE 1... 1,5 Figure 6.9 Results of TEA with Full Non Linear Friction Effects (cf = 0.006) for M2 Astronomical Constituent (b) Phase lag in radians tide -130- (relative to M2 20 2q Figure 6.10 2 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) for M2 Astronomical Constituent (a) Amplitude in -131- centimeters. __;II_~Y _XI~_;~ 1.1 5,43 Figure 6.10 Results of TEA with Full Non Linear Friction Effects (cf - 0.009) for M2 Astronomical Constituent (b) Phase lag in tide) -132- radians (relative to M2 Figure 6.11a Trajectory along which M Elevation Amplitudes are Compared for Varying Friction Factor in Figure 6.11b -133- i 50 _ Sill Region 'II I . Northern Depression Central Bight 45 40 35 nMI (cm) 30 25 0.003 c 20 - cf = 0.006 15 Cf = 0.009 10 5 0" I I I I I I " I I I S Figure 6.11b Comparison of M 2 Elevation Amplitudes for Varying Friction Factor, cf, along Trajectory S with what would be expected for a water depth of 2 - 7 m with large bedforms (dunes of 1 m) as is shown in Table 6.4. Figures 6.12 through 6.14 show overtide (steady, M 4 and M 6 ) elevation amplitudes and phases corresponding to the M 2 forcing tide with a friction factor cf = 0.009. These computations did not include the effects of finite amplitude in the continuity equation or convective acceleration in the momentum equation and hence the overtide responses shown are generated by the non linear friction term. The most significant overtide is the M 6 tide (Figure 6.14) since as was discussed in Chapter 2, the harmonically decomposed friction pseudo forcing term is distributed mainly to the M 2 and M 6 frequencies. Of secondary importance is the finite amplitude effect in the bottom friction term which generates pseudo forcings/responses at even harmonics. This is reflected in the weaker responses at the steady and M 4 overtides (as is shown in Figures 6.12 and 6.13). We note that TEA-NL requires the specification of a zero reference at some point in the domain with respect to which all elevations are computed. For this case a point along the ocean boundary was used for convenience. Figures 6.15 through 6.18 illustrate the effects of including in the computation both the finite amplitude term from the continuity equation and the friction term (again with a friction factor of cf = 0.009). As was shown in Chapter 2, the most significant pseudo forcings due to the finite amplitude term in the continuity equation are distributed to the steady and M 4 overtides and are a result of the responses at M 2 . This is reflected in the substantial increase in response for the steady and M 4 overtides (compared to computations -135- ________L_ (l___jllIC--~-~L-~-~ -~I~I Table 6.4 Bottom Values for Friction Factor cf in Terms of Depth and Bottom Roughness (from Wang and Connor, 1975) H H [m] 1 2 5 10 20 30 40 50 100 roughness k[] [sec k [m] Stones 0.07 0.025 0.0061 0.0049 0.0036 0.0028 0.0023 0.0020 0.0018 0.0017 0.0013 Small rocks 0.20 0.030 0.0088 0.0070 0.0052 0.0041 0.0033 0.0028 0.0026 0.0024 0.0019 7unes 0.50 1.10 0.035 0.0095 0.0070 0.0056 0.004410.003910.0035 0.0033 0.0026 0.040 0.0092 0.0073 0.0058 0.0051 0.0046 0.0043 0.0034 "' -136- Figure 6.12 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) for Steady State Constituent. Amplitude in centimeters. -137- I~L __LLI1___~____1~1_~_C--II Y~U.LII ~tl -^-.~I_ 0.4 run e2 AMPLITUDE 0.T 0- OO.LL 0. L Figure 6.13 Results of TEA with Full Non Linear Friction Effects (cf - 0.009) for M4 Overtide Constituent (a) Amplitude in -138- centimeters. i. o M4 run e2 /5 PHASE 3.O Figure 6.13 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) for M4 Overtide Constituent (b) Phase lag in radians (relative to M 2 tide) -139- -~ -L-*~ II~I~L~i~~^---~~--I1I~-~-IIY..P~--- Figure 6.14 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) for M6 Overtide Constituent (a) Amplitude in centimeters. -140- Figure 6.14 Results of TEA with Full Non Linear Friction Effects (cf - 0.009) for M6 Overtide Constituent (b) Phase lag in radians (relative tide) -141- to M2 I I 20' Figure 6.15 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for M 2 Astronomical Constituent (a) Amplitude in centimeters. -142- 0.7L I.A Ii 1.8 Figure 6.15 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for M2 Astronomical Constituent (b) Phase lag in radians tide) -143- (relative to M2 ~I=-IIOL~I~-Y.. ~------~ 0.0 STEADY run e3 AMPLITUDE .2.5 2,25 12.o 2.5 2.25 2,0 2.5 Figure 6.16 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for Steady State Amplitude in centimeters. Constituent. -144- Figure 6.17 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for M4 Overtide Constituent (a) Amplitude in centimeters. -145- ..-.. mlux.-r~---^--r*llra~-*IY"eY"Lnar~~.-- run e3 PHASE 1.5 2.5 Figure 6.17 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for M4 Overtide Constituent (b) Phase lag in radians tide) -146- (relative to M2 Figure 6.18 Results of TEA with Full Non Linear Friction Effects (cf = 0.009) and Finite Amplitude Effects for M6 Overtide Constituent (a) Amplitude in -147- centimeters. / Figure 6.18 Results of TEA with Full Non Linear Friction Effects (cf a 0.009) and Finite Amplitude Effects for M6 Overtide Constituent (b) Phase lag in tide) -148- radians (relative to M2 III__IUY__~___I*__CI__LII___, with friction effects alone) as is shown in Figures 6.16 and 6.17. Contributions of the finite amplitude pseudo forcing to other frequencies (M2, M6, etc.) are of secondary importance since these are generated by either overtide responses (which are much smaller than the M2 response) or through the interaction of an overtide response with the main M2 tide. This is unlike the friction term for which all overtides may be directly generated by the astronomical tides. The limited effect of the finite amplitude term on frequencies other than steady and M 4 is demonstrated in Figures 6.15 and 6.18 which show that responses at M2 and M 6 remain essentially unchanged (compared to computation with friction effects alone). The changes resulting from including convective acceleration in the computation in addition to friction and finite amplitude may be seen by examining Figures 6.19 through 6.22. As expected responses at M 2 and M 6 remain essentially unchanged while there are very slight changes for the steady state and M4 elevation amplitude distributions. All three overtides calculated have responses of approximately equal importance. In general, patterns of elevation amplitude vary somewhat for these overtides. in the sill region. However they all exhibit high gradients High gradients for both elevation and flux are also prominent for the M2 astronomical tide in this region. The high gradients of the main tide, together with higher M2 velocities and elevations and (furthermore) shallower depths, result in much greater non linear pseudo forcings in this region relative to the rest of the -149- Figure 6.19 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M2 Astronomical Constituent (a) Amplitude in centimeters. -150- ~--(irrrrr~-~i-i 3 -li L-yCyrsUlpl lrrYI*~C'_r--_- __ 0.6 _-O,L) " o. I~1e Figure 6.19 1 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 2 Astronomical Constituent (b) Phase lag in radians (relative to M 2 tide) -151- s~l-~_qii~L~LBgirCLirY-n~-~ --.~^.-~X STEADY 0.0 run e4 AMPLITUDE 2.7 5 Z.5 2.25 Figure 6.20 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for Steady State Constituent. Amplitude in centimeters. -152- ---~'*siPII-YLXlr~i 0., Figure 6.21 LI.0 - - ' Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M4 Overtide Constituent (a) Amplitude in -153- centimeters. _ _ run e4 PHASE 3.0 Figure 6.21 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M4 Overtide Constituent (b) Phase lag in tide) -154- radians (relative to M2 _III____YII__LYiUI~_CIIP.I.XI~1__X^~. 1.0 o.5 Run e4 AMPLITUDE 2.5 ,.5 2.5 2.0 I.0 I. Figure 6.22 ~ c Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M6 Overtide Constituent (a) Amplitude in -155- centimeters. 3.0 Run e4 PHASE 4.5 4-S so 6.0 Figure 6.22 Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M6 Overtide Constituent (b) Phase lag in radians (relative to M2 tide) -156- bight. The sill region is therefore responsible for a substantial portion of the generation of the overtides. We note that for the steady state component, the high elevation gradients in the sill region correspond to a seaward flushing current which is shown in Figure 6.23a. This steady residual current, which is the result of the finite amplitude term in the continuity equation, exists only in the sill region since the gradients in elevation are small in the rest of the bight. It should be stressed that the steady residual current computed and shown in Figure 6.23a is a residual velocity current and is equivalent to the time averaged velocity expressed as: uR = u(t) = (6.4) uw= 0 This steady Eulerian residual velocity is associated with the net drift of a particle traveling with the velocity of the fluid. It is distinct from the time averaged flux which gives the following residual flux current: QR = u(t)(h + n(t)) = uRh + u(t)n(t) (6.5) We note that mass is not conserved when considering the steady velocity currents by themselves since, as was noted in Chapter 3, the harmonic continuity equation for the steady component is coupled with harmonic continuity equations for other frequency components. For the residual flux currents, however, mass is conserved for each individual frequency constituent. This is due to the fact that the continuity equation in terms of flux is a linear differential equation and therefore leads to -157- Figure 6.23a Velocity Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for Steady State Component. -158- Figure 6.23b Velocity Results of TEA with Full Non Linear Friction Effects (c = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 2 Component at Time of Maximum Ebb for the M2 Component Relative to the Ocean Boundary -159- Figure 6.23c Velocity Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M4 Component at Time of Maximum Ebb for the M 2 Component Relative to the Ocean Boundary -160- Figure 6.23d Velocity Results of TEA with Full Non Linear Friction Effects (cf = 0.009), Finite Amplitude Effects and Convective Acceleration Effects for M 6 Component at Time of Maximum Ebb for the M2 Component Relative to the Ocean Boundary -161- uncoupled harmonic continuity equations for all the various frequencies. Hence residual flux currents indicate net mass flushing patterns. For our particular case, while net steady velocity currents are quite significant (maximum velocities in the bight for the steady residual component are approximately 10% of the maximum velocities in the bight for the main M 2 component at maximum ebb), the net steady flux currents will be insignificant. This is due to the fact that the residual velocity current, uR, is generated by the time averaged finite amplitude term, u(t)n(t). Hence the two terms in Eq. 6.5 balance to yield no net residual flux, QR. However, in general, net steady flux currents can exist and will depend on the type of ocean connections and forcings, the depth and bottom friction factor distributions, as well as the geometry of the embayment. Figure 6.23b shows the predicted velocities at maximum ebb (relative to the ocean boundary) for the main component (M2 ). We note that the scaling is greater by a factor of 10 relative to Figure 6.23a. Figures 6.23c and d show the predicted velocities associated with the M 4 and M 6 overtide components at the time of maximum ebb for the M 2 component (relative to the ocean boundary). varies for these figures. We note again that the velocity scaling The actual total velocity is obtained by adding all four components at any one time. Finally we note that at some locations along land boundaries, non-zero normal velocities exist. These are especially severe along the northern boundary where very sharp depth gradients exist. Hence the actual amount of flux leakage is limited due to the shallow depths in these areas. Furthermore these non-zero velocities can be eliminated by refinement of the grid in any problem -162- areas since normal fluxes are natural boundary conditions and are thus satisfied exactly in the limit. So far we have seen that the overtides generated are all of about equal magnitude. The friction term is the most important in that it is responsible for the responses of the main astronomical tide (M2 ) in addition to the generation of the M 6 overtide. The finite amplitude term in the continuity equation generates for the most part the M 4 and steady tides, while the convective acceleration term has little effect compared to the other non linear terms. Let us now compare the overtides computed by TEA-NL, for a friction factor of 0.009 and with all non linearities included, to those experimentally obtained by Filloux and Snyder (1979). As was previously noted, agreement between TEA-NL predictions and measurements by Filloux and Snyder (1979) for both elevation amplitude and phase of the M 2 tide was excellent. Since the measurements by Filloux and Snyder do not reflect steady state circulation, no comparisons can be made. For the M4 overtide, TEA-NL predictions (Figure 6.21) and measurements (Figure 6.4) show good agreement. However, when comparing TEA-NL results for the M6 tide (Figure 6.22) with measurements (Figure 6.5), there is some discrepancy. The numerical predictions for the M 6 amplitude exceed measurements by a factor of about 2.0. Phase errors are less pronounced although agreement is not as good as for the M 4 tide. The variability between the TEA-NL overtide predictions and measurements may be quantified by calculating the proportional variance for each of the harmonics at which reduced measurement data are -163- This proportional variance is computed by summing the available. square of the difference between each of the experimental elevation amplitude values available at each measurement location and the value computed by TEA-NL at that location and then dividing by the sum of the squares of all the measured values (Snyder, Sidjabat and Filloux, 1979). Hence the proportional variance evaluating the error between predictions and measurements for each harmonic j is expressed as: L K E In (x,, k) - n (x) 2 (6.6) £=l k=l Vp L K1 z E £=l k=l where measured elevation amplitude component for the jth harmonic n~ 3 TEA-NL predicted elevation amplitude component for the jth harmonic L = total number of measurement locations K = total number of measurement data points at location k The proportional prediction variances, V3, are calculated for the entire bight and the three sub-regions previously defined and are presented in Table 6.5a. It is noted that regional values for V are about the same or less in the northern and eastern bight compared to values for the entire bight while in the sill region they are somewhat higher. These proportional prediction variances V uncertainties in the reduced measurement data Vm. include the Hence if the average measured value at each point were correct, the net error between predicted -164- Table 6.5a Overtide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of Proportional Variance, V j P Tide Entire Sill Bight Region Northern Bight M2 0.0114 0.0142 0.0038 0.0099 M4 0.0907 0.1489 0.0387 0.0767 M6 0.9198 0.9774 1.2368 0.7892 -165- Eastern Bight Table 6.5b Overtide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of Proportional Standard Deviation, sP i Sp Tide Entire Bight Sill Region Northern Bight Eastern Bight M2 0.11 0.12 0.06 0.10 M4 0.30 0.39 0.20 0.28 M6 0.96 0.99 1.11 0.89 -166- and measured elevation amplitudes would be obtained by subtracting VM. 3 from V . Values for proportional prediction standard deviations are 3 defined by: S j = (6.7) j and are shown in Table 6.5b. All values for the predicted M 6 tide are greater than the measured amplitude. Table 6.5b shows that this amplitude excess is equal to approximately one. M 6 values are too high by a factor of 2.0. Hence the predicted Snyder, Sidjabat and Filloux (1979) had the same overprediction problems for the M 6 tide with their numerical model. The numerical model applied to the Bight of Abaco by Snyder, Sidjabat and Filloux (1979) is a frequency domain model which uses finite differences to resolve the spatial dependence of the governing equations. As for TEA-NL, the non linear harmonic coupling in their model is handled with an iterative scheme which cycles through the various sets of harmonic equations. However, their model is based on an analytical harmonic separation of the governing equations and uses a number of approximate expansions for the various terms. Furthermore, their model only performs computations for 5 astronomical tides (K , 1 01, N 2 , M 2 and S2 ) and two overtides (M4 and M 6 ) and does not consider any compound type interactions. Snyder, Sidjabat and Filloux found that their optimal overall solution was obtained at a friction factor, cf, equal to 0.007 and that this resulted in proportional prediction variances of VM2 = 0.07, Vp = 0.33 and VM = 0.79. These to proportional standard deviation 2prediction values of 4 6 values of deviation standard prediction proportional correspond to -167- S M2 = 0.89. = 0.56 and S S M6 M4 0.26 Hence TEA-NL predictions are somewhat better than their predictions for the M 2 and M4 tides while the error for the M 6 tide is about the same for both models. Snyder, Sidjabat and Filloux (1979) were able to obtain better agreement between their numerical model predictions and their field measurements by deviating from the standard quadratic law by either including a linear friction component or by allowing for significant The first mechanism assumes bottom friction to be nontidal currents. composed of a linear and a quadratic part of the form: b T p = c fl (6.8) u + cf2 uIu f2 For this two parameter friction law, the quadratic friction coefficient, cf2 , is substantially reduced from that used for the fully quadratic one parameter law. The reasons for their improved agreement may be readily explained as follows. The friction forcing felt by the M 2 tidal component is about the same as the one parameter friction law if the linear friction factor cfl is sufficiently large to compensate for the reduction in the quadratic friction factor, cf2 (recall that the largest portion of the quadratic friction term acted as a linear term at the actual forcing frequency). This then allows the response at M 2 to be the same as that calculated when the one parameter fully quadratic law was used. The M 4 overtide response will be largely unaffected since this is generated by the finite amplitude term. Hence, since the M 2 response remains the same, the finite amplitude forcing at M4 will remain the same. The M 6 overtide response, however, will be reduced depending directly on how much the quadratic coefficient cf 2 -168- has been reduced from the one parameter law. This is due to the fact that the M 6 tide is now generated by the reduced quadratic term cf21ulu (and the dominant M2 response velocities have remained the same for both friction laws). Snyder, Sidjabat and Filloux found that with friction factors of cfl = 0.00086 m/sec and cf2 = 0.0033 they were able to reduce the proportional prediction variance for the M 6 tide to VP = 0.178 (SP = 0.42). The second mechanism by which they were M6 M6 able to reduce the error for the M 6 tide was to include an rms nontidal current of 0.28 m/sec. This nontidal current is about equal to the maximum tidal velocity in the bight. The proportional prediction = 0.231 (SP = 0.48) in variance for the M6 tide was reduced to V' M6 M6 this manner. However the deviations required from the standard quadratic law for the first mechanism and the large nontidal current required for the second mechanism in order to significantly improve the results are not supported by the flow conditions which exist in the bight. As this point we note that Filloux and Snyder (1979) and Snyder, Sidjabat and Filloux (1979) did not consider steady state in either their analysis of the experimental data or in their numerical model. However, as was seen from TEA-NL results, the steady state overtide response was of the same order of magnitude as other velocities considered making it inconsistent to not consider this steady term. Furthermore, Snyder, Sidjabat and Filloux (1979) did not consider any compound tides generated through the non linear interactions between the various astronomical tides. In the next section we shall establish the importance of compound tides in the bight by examining the interaction of the M2 and N 2 tides. -169- Compound Tide Computations for the Bight of Abaco 6.3 In this section we shall study the compound tidal interactions in the bight. In particular, we are interested in examining the significance of compound tides with frequencies in the neighborhood of and M 6 tides. the M From Table 6.2 we note that the N 2 tide is roughly of the same magnitude as the diurnal constituents K 1 and 01 . However, the importance of the compound tides generated by these diurnal constituents which lie close to the M 4 and M 6 tides is much less than the compound tides generated by semi-diurnal constituents. Furthermore, Table 6.2 shows that the N2 forcing tide is roughly twice as large as the S2 tide. Therefore looking at the compound tides generated by the M 2 and N 2 tides will give us a good understanding of the compound tidal interaction. In addition it will allow us to assess the importance of compound tides in the vicinity of the M 4 and M 6 overtides. The frequencies most likely to be of importance for the M 2 - N 2 interaction are readily obtained by using the response-forcing tables discussed in Chapter 2. The frequencies produced with this technique after the second cycle of the procedure are listed in Table 6.6. We note that these compound tides separate into five frequency clusters. The first cluster consists of the steady zero frequency response and two long period (28 days and 14 days) residual tidal components. The remaining response clusters are grouped around 12, 6, 4 and 3 hours. The synodic period for these frequencies is 28 days. The frequencies listed in Table 6.6 were used in the application of TEA-NL. A time sampling rate of 132 points (spread over 28 days) was -170- I--L--L*nin~~_~ Table 6.6 Tides of Possible Interest for M2 and N2 Interaction -1 ) (rad/sec _________I___ 28.05 0.0 0 2.5927x10-6 MN 2MN (days) ,________ ____________ steady ~T -T Freq. Freq. Comp.* Tide ~.;~~~a*--*l~l~l~i~P---~ --~n~m~~-l----- .~~x.,._ -6 5.1854x106 -4 1.3538x10- 2w1 -2w2 2 28.05 days 28.05 14.02 days 0.56 12.89 hrs 28.05 4 12.65 hr s 28.05 1.4056x10-4 12.42 hrs 28.05 1.4316x10-4 12.19 hrs 0.56 3NM4 2.7335x10- 6.38 hrs 28.05 N 2.7594x10- 6.33 hrs 28.05 -4 2.7853x10-4 6.27 hrs 28.05 2.8113x10 4 6.21 hrs 28.05 2.8372x10 4 6.15 hr s 28.05 wl-2w2 2NM 2 N2 1.3797x10 "'2 1 2 M2 2w2 2MN 2 4 MN4 1 2 3 2w 1-3w 2 M4 3MN 4 132 21+2 2w+2w2 1 N6 -4 4.1391x10 - 4.22 hrs 0.56 2NM 6 4.1650x10 - 4.19 hrs 28.05 4.1909x10 - 4 4.16 hrs 28.05 4.14 hrs 28.05 5.5189x10 -4 3.16 hrs 0.56 5.5447x10-4 3.15 hrs 28.05 5.5707x10-4 5.5966x10- 3.13 hrs 28.05 3.12 hrs 28.05 w 2MN 2 6 4.217x10 - M6 2w 1 +2w 2 N8 3NM 8 1 2 +2 2 2MN 8 3MN 8 5.6225X10 M 8 = = -4 3.10 hrs I I *wl W2 4 WM 2 UN 2 -171- a required in order to obtain accurate harmonic analysis results up to and including the 4 hour period cluster. Friction and finite amplitude effects were considered while convective acceleration was neglected in the computations due to its limited importance. The boundary conditions are specified such that the M 2 and N 2 astronomical constituents have amplitudes and phases set equal to values measured by Filloux and Snyder (1979). For all overtide and compound tides, zero elevation is specified along the ocean boundary. Globally averaged harmonic pseudo forcing amplitude distributions for the continuity and momentum equations are shown in Figures 6.24a,b. These figures show the ratios of the harmonic forcing at each frequency to the maximum harmonic forcing of all frequencies considered. Furthermore these figures only reflect tides with a pseudo forcing greater than 1% of the maximum harmonic continuity or momentum pseudo forcing. We note that the tides with the most prominent forcings correspond to frequencies in the first two rows of Table 2.5b. Figure 6.24a shows the finite amplitude pseudo forcings being distributed mainly to the steady cluster and the M 4 cluster. The forcings are distributed to the cluster in a similar way as for the overtide case. However, now not only are certain overtides generated but compound interactions of relative importance also exist. Besides the steady pseudo forcing, a 28 day period finite amplitude pseudo forcing exists. now significant. Furthermore both an M 4 and MN 4 pseudo forcing are The N 2 overtides themselves are not of importance. Figure 6.24b shows the friction pseudo forcing being distributed mainly to the astronomical frequencies themselves. -172- Besides the M 2 -YIYL6~-~I"--"-~-CYrrr-~~I~I~ P 4 () .10 .0908 .07.06- N2 M6 N 2MN 12MN 2 .04 .03.02 - .01 0.00 1.00 2.00 3.00 4.00 -4 (rad/sec) w x 10 Figure 6.24a Continuity Equation Pseudo Forcing Vector Ratios Due to M 2 - N 2 Interaction -173- P M2.10 2MN2 2 .08 .07 .06 STEADY .05.04 M 6 H M4 MN MN4 2N 6 .03.02 - .010.00 1.00 2.00 3.00 4.00 x 10- 4 (rad/sec) Figure 6.24b Momentum Equation Pseudo Forcing Vector Ratios Due to M 2 - N 2 Interaction -174- 5.00 ~"----~~--L-"L~~~L-~^-l*l ll~-at~--TY r~-L~~---i~r~a~*~il_ 4~,, overtides, MN, 2MN 2 , MN4 and 2MN 6 compound tides are now of significance. We note that for both the continuity equation and momentum equation pseudo loadings, the compound pseudo forcings shown are typically 40-50% of the magnitude of the adjacent M 2 overtides. Hence to be consistent in the order of approximation of the analysis we must take these compound tides into consideration. Figures 6.25 through 6.32 show the most significant tides associated with the non linear interaction of the M 2 and N 2 tides. The steady state (Figure 6.27), M 2 (Figure 6.25) and M4 (Figure 6.30) constituents are essentially the same as for the M 2 overtide computations. The M 6 constituent (Figure 6.32) does show some reduction in amplitudes but predicted values still substantially exceed measured values. The N 2 astronomical constituent (Figure 6.26) shows very good agreement with measurements (Figure 6.6). The variability between the TEA-NL compound tide predictions and measurements are again quantified by calculating the proportional prediction variance, VP. 3 Values for V J are shown in Table 6.7a. Proportional prediction variances V? are again about the same or J less in both the northern and eastern bight compared to values for the entire bight with the exception of the values for V in the northern bight. for the N 2 tide However as was noted from Table 6.3a, the proportional measurement variances, Vm, in the northern bight were in general substantially greater than values for Vm for the bight as a whole. This was especially true for the N 2 tide. Hence the net error between TEA-NL predictions and measurements will be substantially reduced. We conclude that in general the numerically predicted -175- _ __ 20' Figure 6.25 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf 0.009) and Finite Amplitude Effects for M2 Astronomical Constituent (a) Amplitude in centimeters -176- r~~~-I"-~a-C -r -~nu~rr~ ~9~ 0.O I.2- .5 Figure 6.25 0'(0 Results of TEA with M2-N2 Interaction and with Full Non Linear Friction (cf - 0.009) and Finite Amplitude Effects for M2 Astronomical Constituent. (b) Phase lag in radians (relative to M2 tide) -177- Figure 6.26 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for N2 Astronomical Constituent. (a) Amplitude in -178- centimeters ---~I Figure 6.26 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf 0.009) and Finite Amplitude Effects for N2 Astronomical Constituent. (b) Phase lag in tide) -179- radians (relative to M2 Figure 6.27 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for Steady State Constituent. Amplitude in centimeters. -180- Figure 6.28 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (Cf 0.009) and Finite Amplitude Effects for MN Compound Constituent Amplitude in centimeters -181- MN4 run h3 0, AMPLITUDE 0.2 0.L Figure 6.29 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for MN4 Compound Constituent (a) Amplitude in -182- centimeters. rrl--~i~i-P.-~ly Clli~iY~-~L~ C1~yl~ s Figure 6.29 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for MN4 Compound Constituent (b) Phase lag in radians (relative to M 2 tide) -183- Figure 6.30 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for M4 Overtide Constituent (a) Amplitude in -184- centimeters. ILCILLL~mltl IIII111 Figure 6.30 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for M4 Overtide Constituent (b) Phase lag in radians (relative to M2 tide) -185- ~ 00. 0*6 6.6 Figure 6.31 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for 2MN 6 Compound Constituent. (a) Amplitude in centimeters -186- ~ Figure 6.31 .I-.IXII-UIL ~I~CI*^ L I~---%~IIItl~._ Results of TEA with M2-N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for 21iN4 6 Compound Constituent. (b) Phase lag in radians (relative to H2 tide) -187- ~.1II___ run h3 0.5, AMPLITUDE 1.0 Figure 6.32 Results of TEA with M2 -N 2 Interaction and with Full Non Linear Friction (cf = 0.009) and Finite Amplitude Effects for M6 Overtide Constituent (a) Amplitude in -188- centimeters. _I_____LPL__ILLI____~I ~1IY-~-~.-..l.^~ Figure 6.32 Results of TEA with M2-N2 Interaction and with Full Non Linear Friction (cf 0.009) and Finite Amplitude Effects for M6 Overtide Constituent (b) Phase lag in tide) -189- radians (relative to M2 I_ Table 6 .7a Compound Tide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of Proportional Variance, V 3 P Vj Tide Entire Sill Northern Eastern Bight Region Bight Bight N2 0.0253 0.0187 0.0680 0.0305 M2 0.0114 0.0138 0.0066 0.0077 M4 0.0799 0.1800 0.0384 0.0304 M6 0.4693 0.6011 0.5356 0.3964 -190- Table 6.7b Compound Tide Computation Errors Expressed as Error Between Measurements and TEA Predictions in Terms of Proportional Standard Deviation, SI 3 -191- distributions are better for the northern bight than for the bight as a whole. For the sill region proportional prediction variances, VP , for the M 4 and M 6 overtides exceed values for the bight as a whole. Values for the proportional prediction standard deviation are shown in Table 6.7b. Hence agreement between predictions and measurements for both the M 2 and N 2 tides overall is excellent (Snyder, Sidjabat and Filloux (1979) obtained a proportional prediction variance for the N2 tide of V~ = 0.02). N2 Agreement is good for the M tide and has improved for the M 6 tide when compared to the overtide computations. For the M 2 tide about 75% of the locations have predicted amplitude values which exceed the average measured values at a location while for the N 2 tide the fraction is only 60%. For the M 4 tide only about 50% of the locations have overpredicted amplitudes while 25% of the locations have predicted values equal to the average of the measured values. Finally for the M 6 tide all locations have overpredicted amplitudes with the exception of locations actually on the ocean boundary. As may be deduced from Table 6.7b, the overprediction factor for the M 6 tide has been reduced to about 1.7 for these compound tide computations. As would be expected from our examination of pseudo forcing values, there are now also significant compound responses. There is a monthly varying MN compound tide (Figure 6.28), a MN 4 compound tide (Figure 6.29) adjacent to the M 4 and a 2MN 6 compount tide (Figure 6.31) adjacent to the M 6 . Although these compound tides are somewhat smaller than their adjacent overtides (by a factor of approximately 2 to 3), -192- . they are important to the dynamics of the bight. ................. iJ --- i--- i Furthermore, it is noted that patterns for both the elevation amplitude and phase shift distributions of adjacent compound tides and overtides are very similar. 6.4 Discussion In the previous section it was ascertained that agreement between the reduced experimental data and the TEA-NL numerical predictions which included the full non linear interaction of the M 2 and N 2 astronomical tides was excellent for the astronomical tides themselves and good for the M 4 overtide. However M 6 predictions exceeded measured values by a factor of about 1.7. In this section we shall explore some of the various possibilities that might explain and/or improve the discrepancy which exists between measurements and predictions for the M 6 overtide. Let us first determine what effect neglecting compound tides has had on the measurement data reduction procedure used by Filloux and Snyder (1979). As was discussed in Chapter 5, the least squares harmonic analysis procedure is much more sensitive to the neglect of frequencies of relative importance within a cluster than when a frequency is dropped outside of a cluster. Hence the procedure may have trouble resolving a tide if a closely spaced adjacent tide exists and is not included as an analysis tide. The associated error in the results will depend on both the relative significance of the two tides and the time point sampling density. However the error introduced into the reduced measurement data under consideration should be about the same for both the M 4 and M 6 tides since as the results of the compound tide -193- - computations showed, the MN 4 and 2MN 6 compound tides are of about the same relative importance with respect to their adjacent M and M6 4 overtides. Furthermore the spacing between the MN 4 and M 4 tides and between the 2MN 6 and M6 tides is equal (Ts - 28 days). confirms that reduced measurement the M 4 and M 6 tides. Table 6.3 errors are about the same for both Hence we conclude that data errors for the M 6 tide are not responsible for the large discrepancy that exists between predictions and measurements for that tide. However the grouping of measurement data errors discussed in Section 6.1 may be readily explained by examining the importance of closely spaced compound tides which were neglected in the data reduction. Recall that the K1 , 01 and M 2 tides generally had very low measurement errors (S ~ 0.03) while the N2 , S2, M 4 and M 6 tides all had larger errors (Sm - 0.10-0.14). The tides with higher measurement errors all have relatively important compound tides in their vicinity which were neglected in the measurement data analysis. As was seen in Figures 6.24a and b, the 2MN 2 tide is of relative importance with respect to the N 2 tide and will also certainly be important with respect to the S2 tide. Furthermore as was previously mentioned, the MN and 2MN 4 6 tides are proportionally significant with respect to the M and M6 4 tides. However the tides with lower measurement errors do not have significant closely spaced compound tides. While the 2MN 2 tide is significant with respect to the N 2 tide, it is not significant with respect to the much larger M 2 tide. The 01 and K1 tides will have no important compound tides located in their vicinity even when all five major astronomical tides are included in the computations. -194- I__IYL_____lill*I_~ll~LlI~-~ We note from Table 2.6 that the K1 and 01 tides are extremely closely spaced (Ts - 208 days). However the error associated with the data reduction procedure is still very low. Hence we conclude that the proportional measurement variances for the N 2 , S2, M 4 and M 6 tides can be reduced to the levels achieved for the 01, K1 and M 2 tides if select compound tides (2MN2 , MN 4 , 2MN and most likely the 2MS 6 2, MS4 and 2MS 6 tides) are included in the least squares analysis procedure used to reduce the measured elevation time history records to harmonic amplitudes. The time sampling rate could be kept about the same as that used by Filloux and Snyder (1979) which may be deduced from the fact that the K1 and 01 reduced measurements showed very low error. Thus far we have seen that experimental data error levels can In not be responsible for the poor fit of the predicted M 6 tide. fact the measurement error levels calculated in Section 6.1 are in general quite modest compared to TEA-NL prediction-measurement error levels. Therefore let us now examine some possible ways in which the fit for the M 6 tide could be improved. The first issue to be examined is the correctness of the boundary conditions which were applied with TEA-NL. A good indication which justifies treating the shallow connections to the open ocean in the northwestern part of the bight as land boundaries is the low prediction-measurement errors in the northern bight. In fact as was seen in the previous section the net prediction-measurement errors are in general substantially less in the northern part of the bight when compared to those for the bight as a whole. This confirms Filloux and Snyders' (1979) conclusion regarding this boundary. -195- _._._ ~i ii.llLI-LI -XLII L^ Let us now assess whether the treatment of the ocean boundary along the western edge of the bight as being totally reflective for overtides is justifiable. In the previous section we noted that prediction-measurement error levels for the M and especially the M 6 4 tides were substantially greater than those for the M 2 and N2 tides. In addition, regional error levels for both the M 4 and M 6 tides were the highest in the sill region. We also note that although measured values along the ocean boundary are small for both the M and M6 4 tides, they certainly are not negligibly small compared with overall bight values for these tides as would be the case for a totally reflective boundary. These facts possibly indicate that either the assumption of the boundary being totally reflective is not entirely correct and/or the location of the reflective boundary is incorrect. As was seen in Section 6.1, the actual reflection coefficient was only about 0.90. Hence a certain amount of leakage of overtide energy into the open ocean does occur. In fact if the ocean boundary were totally reflective as was assumed, then no astronomical tides would be allowed to enter the bight either. The question is whether the somewhat inflated reflection coefficient of 1.0 which was used contributes significantly to overpredicted overtides. tide almost all predicted values are too high. For the M 6 For the M 4 tide only 50% of the comparison locations were overpredicted with another 25% of the comparison locations having equal predicted and measured values. However it may be shown that the overall contribution to the proportional variance for the M 4 tide from overpredicted points far outweighs that from underpredicted points not only due to there -196- being more overpredicted points but also due to the fact that the prediction-measurement differences were on the average much greater for the overpredicted points. This indicates that both M4 and M 6 computations suffer mainly from overprediction which is consistent with a reflection coefficient which is too high. Hence accounting for the correct degree of reflection will definitely improve prediction-measurement error levels for both the M4 and M 6 overtides. However since an approximately equal influence of reflection coefficients would be expected on both the M4 and M 6 tides there will still remain a substantial discrepancy between the M and M 4 6 error levels. The other possible problem with the open ocean boundary condition applied for the overtides in the computations could be the location of the reflection boundary. The grid used (Figure 6.7) has the ocean boundary located at the beginning of the sharp depth drop as if a vertical step were located there. However it would seem more logical to place this boundary somewhere in the middle of the range of the most substantial depth drop in order to better simulate a would be reflective boundary. We recall that the 1000 meter contour was between 3 and 15 kilometers from the present boundary. Hence placing the reflective boundary halfway between the 5 and 1000 meter contours would put it between 1 to 7 kilometers (depending on where along the boundary) away from the boundary used in Figure 6.7. This adjustment distance can be significant when compared to the overall scale of the sill region. We note that defining an ocean boundary in the manner just described would create difficulties in this case since there are no astronomical -197- tide measurements at that location. We conclude that allouing for some overtide transmission out of the bight and accounting for the fact that the reflection does not totally occur at the upper edge of the depth drop will contribute towards improving the M 4 and M 6 distributions but are not the dominating physical mechanisms which explain the much larger prediction-measurement errors of the M 6 overtide. Let us now assess whether the use of one equal value for friction factor, cf, for the entire bight contributed significantly to the large error level of the M 6 tide. The use of spatially dependent friction factor values is certainly physically well motivated due to the significant variation in bottom surface characteristics within the bight. The sill region has a substantially greater bottom roughness (dunes of 1 to 3 m in a depth of 2 - 5 m) than other areas in the bight and, as Table 6.4 shows, a value of cf greater or equal to 0.009 would be expected in the sill region. Table 6.4 also indicates that a value of cf = 0.009 is somewhat too high in other regions of the bight. Hence in the sill region cf could be greater or equal to the value used in the computations while in the remainder of the bight a lower value should be used. If the value of cf used in the computations were significantly below the actual value for the sill region, then the higher than actual values in other parts of the bight might compensate for this. However this hypothesis which favors the use of localized friction factors is not supported by the error distributions for the M 2 and N2 tides. The M 2 tide was dominantly overpredicted due to both the number of overpredicted points and the -198- Illlll__l~i L~-- *~*~llb~ fact that the average overpredicted differences far exceed underpredicted differences. For the N 2 tide, overprediction only slightly dominated underprediction. As has been previously discussed, the sill region is the most important region in terms of both the effect of friction on the main tides and the generation of non linear tides. That the most significant impact of friction on all distributions is in the sill region is demonstrated by the high gradients in elevation amplitude and phase which exist there. Furthermore since the non linearities are the most significant in the sill region (due to high elevation amplitudes, high velocities and shallow depths relative to other parts of the bight), the non linear overtides and compound tides are largely generated there. Hence a spatially varying friction factor will not drastically effect any of the computed distributions if values for cf in the sill region are kept the same. The limited impact of spatially varying friction factor is confirmed by the findings of Snyder, Sidjabat and Filloux (1979) who performed a limited number of computations to check for the sensitivity of this effect. In Section 6.2 the effects of variations in global friction factor were checked with intervals of cf equal to 0.003. Given the general dominance of the overpredictions it is likely that more refined increases in cf beyond the value of 0.009 will have some effect in reducing general error levels. As was state earlier an increase of the value for cf in the sill region would be justifiable due to the bottom roughness there. Furthermore values for cf in the -199- remainder of the bight could most likely be reduced to physically more realistic values without effecting the tidal distributions in the bight. An increase in cf in the sill region not only causes decreases in the amplitudes of the main tide distribution but also generally decreases the M 6 amplitudes. This is due to the fact that for this case the effect of the reduction in M 2 velocities due to increased friction proportionally outweighs the actual increases in the friction factor itself. and hence a reduced M 6 tide. This results in a reduced M 6 pseudo forcing However extrapolating the effects of the reduction in amplitudes of the M and M tides for changes in cf 2 6 of 0.003, we conclude that this fine tuning process will not have a major impact in reducing M 6 error levels. Finally let us consider the effects of only including the M 2 and N2 astronomical tides and neglecting the 01, K 1 and S2 tides in the computations performed. As stated in Section 6.3, after the M 2 and N 2 tides, the S2 will probably be the most influential to the M6 . A very significant improvement in M 6 errors was achieved by including the N 2 tide in the computation (recall VP 0.9198 to 0.4693 and Sp M6 dropped from dropped from 0.959 to 0.685). However as was seen from Table 6.2, the S2 tide is only about half as large as the N2 tide. Assuming that the improvement in the M 6 solution due to considering the S2 tide (in addition to the M 2 and N 2 ) is proportionally (to the amplitude of the tide) the same as that achieved when the N2 tide was included, the error for the M 6 could be brought down to about VP = 0.24 (Sp = 0.50). However this M6 M6 assumes that the processes are linear, which of course they are not. -200- -~I-r_--rr~unr~r*nluBhrl--dOYIPYYt;P~__ - ~-iiu-.ru~,y--- Hence any actual improvement could be more or less than this. We note that we do not expect any modification to the M4 tide due to the inclusion of the S2 tide, in the same way that the M 4 remained essentially unchanged when the N 2 tide was added. We have examined a number of possibilities for their effectiveness in improving the fit between the predicted and measured overtides. No single mechanism seems capable of reducing overtide error levels to those of the astronomical tides. However a combination of these mechanisms may achieve significantly better overtide fits. The largest reduction in M 6 error will most likely be brought about by the inclusion of the S 2 astronomical tide in addition to the M 2 and N 2 tides in the computation. As was seen this could very well lead to reducing M 6 error levels close to those presently achieved for the M4 tide. Improved treatment of the main ocean boundary would bring about improvements in fit for both the M4 and M 6 tides. Finally, the fine tuning of the friction coefficient, cf, could produce minor improvements for the error levels of all tides. -201- CHAPTER 7. CONCLUSIONS A computer model, TEA-NL, which computes tidally driven circulation in coastal embayments, has been developed. The finite element method was used to resolve the spatial dependence in the governing equations while a hybrid time domain - frequency domain approach was used to resolve the time dependence. With this approach the non linear terms are iteratively updated in the time domain to produce time histories which are then harmonically decomposed with the least squares method. The least squares method was extremely well suited for this purpose since it allows the resolution of very closely spaced and narrowly banded energy in an extremely efficient manner. With harmonic forcings on the system and the harmonically decomposed non linear terms (pseudo-forcings), the governing equations separated into sets of linear equations in the frequency domain. This led to the development of a linear core solver which solved each set of linear equations at a given frequency in an extremely efficient manner. The linear core solver yields accurate solutions (not overdamped) while it shows very low spurious oscillations. TEA-NL allows the general investigation of the effects of the non linear interaction between tidal components in shallow estuaries. Hence not only can overtides be computed but compound tides can also be assessed. The importance of compound tides was seen in the application of TEA-NL to the Bight of Abaco where certain of the compound tides generated through the interaction of the M 2 and N 2 astronomical tides had responses equal to about 50% of the corresponding adjacent M 2 overtides. We note that these compound tides can be -202- a ;__)~ I __X~ ~_~ especially important in the assessment of long period residual fluctuations which exist in addition to any steady state residual currents. There are several aspects of TEA-NL which could be improved. The first aspect concerns the fact that at present the TEA-NL user must specify all frequencies of possible interest. However use of an FFT at several selected locations (typifying various parts) in the embayment would allow the identification of all important frequencies. These frequencies could then be used for the much more economical least squares harmonic analysis procedure for all points in the embayment. This would make the model more user friendly and also ensure that no important frequencies are neglected. Furthermore this will simplify the simulation of complex wind histories. A further aspect which would improve TEA-NL would be the use of higher order finite elements because of their increased accuracy per number of total nodes and the convenience of the larger elements with which the embayment may be discretized. We note that the size of the largest element in the domain must reflect the size of the smallest wavelength present (e.g. Mg) (so that wave shape can be adequately represented). In very shallow embayments wavelengths decrease while the importance of higher harmonics increases, possibly requiring a very fine grid. The higher the order of the element, however, the larger the minimum element size which can be used. Furthermore higher order elements would be more convenient to accomodate rapid changes in geometry and depth. -203- ___ _i 1_~~1 _I__ j/*~~_ Finally the treatment of flux boundary conditions could be improved. TEA-NL presently treats them as natural which, unless boundaries are sufficiently refined, could lead to leakage. The specific refinement of boundary areas however is often inconvenient. Hence TEA-NL would be improved by re-formulating such that fluxes were treated as essential boundary conditions and as such were more strictly enforced (i.e., no errors were allowed regardless of element sizes along the boundary). Despite these minor inconveniences in usage, TEA-NL is an effective model for simulating both short term (1 day) and long term (1 month and more) tidally driven circulations in embayments. Its most important feature is that it allows an accurate assessment of compound tides which include long term periodically fluctuating residual circulations. p -204- REFERENCES Studies on finite amplitude waves in Askar, A. and A.S. Cakmak, 1978. Water Resources, 4:229-246. in Advances bounded waterbodies. Benque, J.P., B. Ibler, G. Labadie and B. Latteux, 1981. Finite element method for incompressible viscous flows. GAMM Conference on Numerical Methods In Fluid Mechanics. (4th), (ed. Viviand), 10-20. Bonnefille, R., 1978. Residual phenomena in estuaries, application to the Gironde Estuary. Hydrodynamics of Estuaries and Fjords, (ed. J. Nihoul), 187-195. Connor, J.J. and C.A. Brebbia, 1976. Finite Element Techniques for Fluid Flow, Newnes-Butterworth, London. Daily, J.W. and D.R.F. Harleman, 1966. Defant, A., 1961. Fluid Dynamics, Addison Wesley. Physical Oceanography, Vol. 2, Pergamon Press, Oxford. Doodson, A.T., 1921. The harmonic development of the tide-generating potential. Proceedings of the Royal Society of London, A100:305-328. Dronkers, J.J., 1964. Tidal Computations in Rivers and Coastal Waters, North Holland Publ. Co., Amsterdam. Filloux, J.H. and R.L. Snyder, 1979. A study of tides, setup and bottom friction in a shallow semi-enclosed basin. Part I: Field experiment and harmonic analysis. Journal of Physical Oceanography, 9:158-169. Gray, W.G. and D.R. Lynch, 1977. Time-stepping schemes for finite element tidal model computations. Advances in Water Resources, 1(2):83-95. Gray, W.G. and D.R. Lynch, 1979. On the control of noise in finite element tidal computations: a semi-implicit approach. Computers and Fluids, 7:47-67. Gray, W.G., 1982. Some inadequacies of finite element models as simulators of two-dimensional circulation. Advances in Water Resources, 5:171-177. Finite element analysis of long-period water waves. Grotkop, G., 1973. Computer Methods in Applied Mechanics and Engineering, 2:147-157. Hald A., 1952. Statistical Theory with Engineering Applications, John Wiley and Sons, Inc., London. Ianniello, J.P., 1977. Tidally induced residual currents in estuaries of constant breadth and depth. Journal of Marine Research, 35(4):755-786. Ippen, A.T., 1966. Estuary and Coastline Hydrodynamics, McGraw-Hill Book Co., Inc., New York. -205- Jamart, B.M. and D.F. Winter, 1980. Finite element solution of the shallow water wave equations in Fourier space, with application to Knight Inlet, British Columbia. Proceedings of 3rd International Conference on Finite Elements in Flow Problems, Vol. 2, Banff, Alberta, (ed. D.H. Norrie), 103-112. Kawahara, M., K. Hasegawa and Y. Kawanago, 1977. Periodic tidal flow analysis by finite element perturbation method. Computers and Fluids, 5:175-189. Kawahara, M., 1978. Steady and unsteady finite element analysis of incompressible viscous fluid. Finite Elements in Fluids, Vol. 3, (ed. Gallegher et al.). Kawahara, M. and K. Hasegawa, 1978. Periodic Galerkin finite element method of tidal flow. International Journal for Numerical Methods in Engineering, 12:115-127. Kawahara, M., H. Hirano, K. Tsubota and K. Inagaki, 1982. Selective lumping finite element method for shallow water flow. International Journal for Numerical Methods in Fluids, 2:89-112. Ketter, R.L. and S.P. Prawel, 1969. Modern Methods of Engineering Computation, McGraw Hill Book Co., Inc., New York. King, I.P., W.R. Norton and K.R. Iceman, 1974. A finite element model for two-dimensional flow. Finite Element Method in Flow Problems, (ed. Oden et al.). Lamb, H., 1932. York. Hydrodynamics, 6th edition, Dover Publications, New Le Provost, C. and A. Poncet, 1978. Finite element method for spectral modeling of tides. International Journal for Numerical Methods in Engineering, 12:853-871. Le Provost, C., G. Rougier and A. Poncet, 1981. Numerical modeling of the harmonic constituents of the tides, with application to the English Channel. Journal of Physical Oceanography, 11:1123-1138. Lynch, D.R. and W.G. Gray, 1979. A wave equation model for finite element tidal computations. Computers and Fluids, 7(3):207-228. Lynch, D.R. 1981. Comparison of spectral and time-stepping approaches for finite element circulation problems. Proc. Oceans '81, IEEE Pub., 810-814. Lynch, D.R. 1983. Progress in hydrodynamic modeling; review of U.S. contributions, 1979-1982. Reviews of Geophysics and Space Physics, 21(3): 741-754. Munk, W.H. and K. Hasselmann, 1964. Super-resolution of tides. Studies in Oceanography, University of Washington Press, 339-344. -206- r Munk, W.H. and D.E. Cartwright, 1966. Tidal spectroscopy and prediction. Phil. Trans. Royal Society of London, 259:533-581. Nakazawa, S., D.W. Kelly, O.C. Zienkiewicz, I. Christie, and M. Kawahara, 1980. An analysis of explicit finite element approximations for the shallow water equations. Proceedings of 3rd International Conference on Finite Elements in Flow Problems, Vol. 2, Banff, Alberta, (ed. D.H. Norrie), 1-12. Newland, D.E., 1980. An Introduction to Random Vibrations and Spectral Analysis, Longman, London. Niemeyer, G., 1979. Long wave model independent of stability criteria. Journal of the Waterway, Port, Coastal and Ocean Division, ASCE, 101:WW1:51-65. Oppenheim, A.V. and R.W. Schafer, 1975. Digital Signal Processing, Prentice Hall Inc., Englewood Cliffs, N.J. On the calculation of tidal Pearson, C.E. and D.F. Winter, 1977. currents in homogeneous estuaries. 7(4):520-531. Platzman, G.W., 1981. tidal models. Journal of Physical Oceanography, Some response characteristics of finite element Journal of Computational Physics, 40(1):36-63. Proudman, J., 1953. Strand WC2. Dynamical Oceanography, Methuen and Co. Ltd., Sani, R.L., P.M. Gresho and R.L. Lee, 1980. On the spurious pressures generated by certain GFEM solutions of the incompressible Navier-Stokes equations. Proceedings of the 3rd International Conference on Finite Elements in Flow Problems, Vol. 1, Banff, Alberta, (ed. D.H. Norrie). Snyder, R.L., M. Sidjabat and J.H. Filloux, 1979. A study of tides, setup and bottom friction in a shallow semi-enclosed basin. Part II: Tidal model and comparison with data. Journal of Physical Oceanography, 9:170-188. Speer, P., 1984. Tidal distortion in shallow estuaries. PhD Thesis, Massachusetts Institute of Technology/Woods Hole Oceanographic Institution. Tee, K.T., 1981. A three-dimensional model for tidal and residual currents in bays. in Transport Models for Inland and Coastal Waters, (ed. Fisher), 284-309. Taylor, C. and J. Davis, 1975. Tidal and long wave propagation - a finite element approach. Computers and Fluids, 3:125-148. van de Kreeke, J. and A.A. Chiu, 1981. Tide-induced residual flow in shallow bays. Journal of Hydraulic Research, 19(3):231-249. Van Dorn, W., 1953. Wind stress on an artificial pond. Journal of Marine Research, 12(3). -207- Walters, R.A. and R.T. Cheng, 1980. Calculations of estuarine residual Proceedings of the 3rd currents using the finite element method. in Flow Problems, Vol. 2, Elements Finite on Conference International Norrie). D.H. (ed. Alberta, Banff, Walters, R.A. and R.T. Cheng, 1980. Accuracy of an estuarine hydrodynamic model using smooth elements. Water Resources Research 16(1):187-195. Walters, R.A. and G.F. Carey, 1983. Analysis of spurious oscillation modes for the shallow water and Navier-Stokes equations. Computers and Fluids, 11(1):51-68. Wang, J.D. and J.J. Connor, 1975. Mathematical modeling of near coastal circulation. Ralph M. Parsons Laboratory for Water Resources and Hydrodynamics, T.R.# 200, Massachusetts Institute of Technology. Improved finite element Williams, R.T. and O.C. Zienkiewicz, 1981. forms for the shallow water wave equations. International Journal for Numerical Methods in Fluids, 1:81-97. Wu, J., 1969. Wind stress and surface roughness at air-sea interface. Journal of Geophysical Research, 74(2). Zienkiewicz, O.C., 1977. The Finite Element Method, Third Edition, McGraw-Hill Book Co., Ltd., London. 1 -208-