Sunteți pe pagina 1din 42

Lecture Notes for BBE 8513: Hydrologic M odeling of Small W atersheds

CHAPTER ONE OVERVIEW OF MODELING PRINCIPLES Table of Contents

WATERSHED PROCESSES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1 Qualitative perspective.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1 Quantitative perspective.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2 DEFINITIONS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3 What is a model?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3 Empirical versus theoretical models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3 System terminology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-4 CLASSIFICATION OF HYDROLOGIC MODELS.. . . . . . . . . . . . . . . . . . . . . . . . . . . 1-6 Classification of modeling approaches. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-6 Continuous versus event models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-7 CONSERVATION OF MASS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-8 Definition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-8 Conceptual modeling approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-8 Process-based modeling approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-10 MODELING ISSUES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-14 Scale Issues. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-14 Modeling rigor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-17 Parameter issues.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-20 Hierarchy theory.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-21 Nonuniqueness.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-22 Linear versus nonlinear systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-22 MORE ON NONLINEAR SYSTEMS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-26 Lorenz system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-28 SELF-SIMILARITY AND FRACTALS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-30 Definitions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-30 NONLINEAR ALGEBRAIC EQUATIONS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-34 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-34 Bisector method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-35 Newton's/Secant method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-35 REFERENCES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-38 Assignment #1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-40

CHAPTER ONE1 OVERVIEW OF MODELING PRINCIPLES


The reductionist hypothesis does not by any means imply a "constructionist" one: The ability to reduce everything to simple fundamental laws does not imply the ability to start from these laws and reconstruct the universe. In fact, the more the elementary-particle physicists tell us about the nature of fundamental laws, the less relevance they seem to have to the very real problems of the rest of science, much less to those of society. The constructionist hypothesis breaks down when confronted with the twin difficulties of scale and complexity. Philip Anderson, Prominent Condensed Matter Physicist, 1972

WATERSHED PROCESSES Qualitative perspective The key watershed processes are shown below for a typical hillslope.

2011 Bruce N. W ilson and the Regents of the University of Minnesota. All Rights Reserved. 1-1

1-2 Quantitative perspective Hydrologic models are often built by conducting a water balance for individual watershed processes as shown below.

Each box utilizes a mass balance:

As an example, the inflow and outflows for the above soil storage box can be written as

1-3

DEFINITIONS What is a model? Definition

General types * Physical model (material) - Physical representation of the real world * Mathematical model (formal) - Mathematical representation of the real world Here we are only going to discuss mathematical hydrologic models. Empirical versus theoretical models Theoretical models Theoretical models represent observed data within a framework of general laws or principles. Lets consider the St. Venant equations for open channel flows.

Empirical model Empirical models represent observed data without a framework of general laws or principles. Lets consider the following regression analysis of annual runoff and annual precipitation.

1-4

System terminology Simple representation of a model Lets consider the following general representation of a model.

Definitions for systems Variables:

Parameters:

1-5 State: Values of the variables at an instant of time Memory: Length of time in which the input influences present state Linear:

Nonlinear:

Lumped: Operations are governed by ordinary differential equations Distributed: Operations are governed by partial differential equations Time-invariant: Coefficients of differential equations are independent of time Time-variant: Coefficients of differential equations vary with time Example Lets consider the following equation to represent the operation in the above diagram.

where

For this equation, we have:

1-6 CLASSIFICATION OF HYDROLOGIC MODELS Classification of modeling approaches Traditional classification Models are frequently classified into schemes similar to that shown below. Such classifications are useful in describing the key features of the models.

General characteristics of stochastic models

General characteristics of conceptual or parametric models

1-7 General characteristics of process-based models

Alternative Classification Comprehensive watershed models rarely fit neatly into the boxes. For example, all output from models are inherently stochastic in either the modeling of the process or in the parameter values.

Continuous versus event models Continuous models * Operate over an extended period of time (many years) * Must keep track of wtsd conditions prior to rainfall * Model all components of hydrologic cycle * Usually less detail spent on any given component Event models * Operate only for single storms (hours) * Wtsd conditions specified prior to rain storm

1-8 * Groundwater and ET components neglected * Often greater detail is spent on individual components CONSERVATION OF MASS - THE PHYSICAL PRINCIPLE Definition Statement

Constant density formulation For a constant density fluid (such as liquid water), the volume of fluid is conserved, that is, differences in inflow and outflow volumes represent the change in stored volume. Conceptual modeling approach Channel routing example Lets consider a control volume for a given reach length (neglecting lateral inflow)

Conservation of mass

1-9 For a constant density fluid,

Features:

Conceptual approach Another equation is needed to close the equation set. Consider the following idealized view of storage in a river as conceptualized in the Muskingum's method

For this idealized representation, we have

Total storage in the reach is then defined as

1-10

S = KO + Kx(I-O) = (K-Kx) O + KxI where K and x are parameters that are assumed independent of time. A more general formulation for storage as a function of inflow and outflow can be written as

where Ki , Ko , n, and m are usually considered independent of time. Solution approach We now have two equations:

and therefore we can combine the two equation to obtain

or

For a known inflow hydrograph and known calibrated parameters Ki , Ko , n, and m, one obtains one equation and one unknown (outflow), which can be solved for outflow using analytical or numerical solution techniques. For the special case where m=1 and n=1, one obtains

which corresponds to a time-invariant, lumped linear system. Process-based modeling approach Channel routing example

1-11 Consider an infinitesimally small cross section for one-dimensional flow (neglecting lateral flow) as shown below.

Conservation of mass For a constant density fluid,

For flow rates defined using continuous functions,

1-12

and therefore

The conservation of mass can then be written as

which can be simplified by dividing by x. The approximation of the slope becomes exact at the limit of x 6 0. If we use Q=VA, the conservation of mass can be written as

Features:

Equation of motion Similar to the conceptual modeling approach, an additional equation(s) is (are) needed to close the equation set. As derived later, this equation can be obtained by using Newton's Second Law of Motion as,

1-13

where n=Manning's n is often considered to be a constant. The hydraulic radius (R) and area are functions of flow depth (h) for a given channel geometry, or

Solution approach We have the following equations to solve

for the following unknowns:

Features:

1-14 MODELING ISSUES Scale Issues Background Scale is used here as an indicator of the time duration or length dimension corresponding to a process, an observation or a model. Typical space and time scales for various hydrologic processes are shown below (from Sivapalan and Kalma, 1995). As discussed by Sivapalna and Kalma (1995), care is need if the observational scale is different than the modeling scale.

The above figure suggests an inter-relationship between spatial scale and temporal scale, that is, processes acting on a small spatial scale also have time scales that are small. The ratio of the characteristic length and time scales is called the characteristic velocity. The characteristic velocity is a single measure that can be used to summarize different hydrologic processes. For example, the characteristic velocity for atmospheric processes is on the order of 10 m/s, for channel flow of 1 m/s, and subsurface stream flow of 0.1 m/s (Sivapalan and Kalma, 1995).

1-15 Impacts on modeling Different scales between observations and models (and their parameters) can cause problems in representing the system. Considerable research has therefore been conducted to address the transfer of information across scales. Upscaling refers to taking information at smaller scale and aggregating it to represent processes at larger scales. Downscaling is the reverse approach. It takes information at the larger scales and disaggreates it to represent processes at smaller scales. These concepts are shown in the following schematic.

Scaling relationships The most widely used scaling relationship is the power law. Mathematically, it is defined as

where k is often called the scaling exponent. A powerful feature of this equation is that the functional response for a different scaled independent variable, say x (where is a constant), is directly proportional to the original function, that is,

where k is a proportionality constant. We therefore conclude that different values changes the proportionality constant, but not the shape of the function itself. Since can be any arbitrary, this results hold for all scales, and therefore the power law function has the property of scale invariance. Considerable research has been done in different fields to gain an understanding on why different processes have a particular scaling exponent.

1-16 An alternative viewpoint can be obtained by using logarithmic form of

where the slope of the function is constant and intercept value changes with the value of the constant . Simple scaling has been used to provide a theoretical structure for representing intra-storm rainfall intensity. A simple scaling model fo rainfall intensity can be written as

where i(t,Ts) is the rainfall intensity at time t for storm duration Ts, i(t,Ts) is the rainfall intensity for the a storm duration of Ts at time t, is the ratio of the two storm durations and k is the scaling exponent. The scaling of a hypothetical rainfall intensity corresponding to storm durations of 1 hour and 2 hour for k=-0.7 is shown below. The shape of the storm hyetograph is clearly the same for the two storms. Intensities differ by a proportionality constant of 2-0.7.

Much research has been done to investigate these scaling concepts for stochastic processes. Application to stochastic processes is discussed in Chapter 2. The term power-law is often used to refer to a particular class of probability density functions used to describe extreme

1-17 events. Modeling rigor Constructionist approach Closely associated with upscaling is the constructionist approach to modeling. It follows from the reductionist approach used in most scientific investigations. With this approach, a complex process is understood by subdividing the system into smaller and smaller components until it can be studied effectively and fundamental principles established. This concept is shown below.

A constructionist modeling approach builds models simply reversing the results of reductionist hypothesis. It builds models by linking principles established for subcomponent (or even sub-sub components) and components. This concept is shown below.

The constructionist approach generally presumes as modeling techniques become more fundamentally-based, confidence in predicted values increases. This concept is shown below.

1-18

Although this belief may seem reasonable, the modeler needs to be wary about using models that have good theoretical definitions but are not easily applied to field conditions. For example, some soil parameters are well defined theoretically but can vary greatly with time and space. As discussed more later, small errors in parameter values can result in considerably different results for nonlinear systems. Role of reliable parameter values is discussed in greater detail. An alternative viewpoint to the constructionist approach is to develop model for a particular spatial (or maybe temporal) scale. Parameter values are limited to a particular scale. From this perspective, the application of the model by upscaling or downscaling to a different scale results in less confidence in predictions because the parameters are not define for those scales. Uniform Slop Approach The uniform slop approach in model development presumes that the model is only accurate as the weakest component and therefore the level of sophistication among modeling components should be uniform. Lets reconsider the flow chart given previously.

1-19

IF all of the above components have the same relative impact on the solution, it is unwise to have a very rigorous for one of the components (say, overland flow) and have crude models for the other components. The overall accuracy of the model will be determined by the accuracy of the crude models. Relative Importance Approach With the relative-importance approach, modeling efforts are focused on those components that have the greatest impact on the solution or are of greatest interest in the study. IF the solution is more sensitive to one of the components (say, infiltration), then efforts should be placed on modeling this component more precisely. IF you are interested in examining the relative impact of human activities (say, global

1-20 warming), then you may want to focus modeling efforts on those components influenced by those activities. Parameter issues The selection of modeling techniques is also dependent on (1) the availability of parameters and (2) the sensitivity of results to errors in parameter values. Insight in the role of these two factors on the confidence-in-prediction versus modeling-complexity curve (previously given) can be obtained using a first-order analysis discussed in detail in Chapter 11. For this type of analysis, the uncertainty in predictions is defined as

where the uncertainty in prediction is represented by the variance and is equal to the sum of the product of the sensitivity of the model to the parameter (i.e., sensitivity coefficient, Si) and the variance of the parameter itself. By increasing the number of parameter, n, with model complexity, the overall uncertainty in predictions can actually increase with more complex models. Modeling uncertainty can decrease with a large number of parameters if the sensitivity coefficient(s) and/or the variance(s) of the parameters are smaller for the more complex model. Regardless of the validity of the theoretical approach, a modeler will likely not use a method that users are unable to determine its parameter values. Outdated theoretical techniques are often used because parameters are readily available. It is usually expensive and tedious to obtain a reliable parameter set for alternative methods. The prediction of infiltration will be used to illustrate parameter issues. Infiltration is a particularly important type of porous media flow that is discussed in detail in Chapter 4. A rigorous approach for modeling infiltration is to use Darcys equation for unsaturated flow as shown below.

where q is the specific discharge, hc is the capillary pressure head (we are neglecting gravity head), K() is the unsaturated hydraulic conductivity, and is the soil moisture content. In Darcys equation, K() and =Mhc/M can be viewed as parameters of the soil system. Although considerable research has been done to determine K() and , these parameters are difficult to define for infiltration processes. Not only can values vary greatly with hillslope location, but these parameters are a function of dynamic processes such as tillage practices

1-21 and subsequent reconsolidation, root growth, surface sealing and cracking, and earthworm activities. Because of these complexities, values for K() and are poorly defined for most infiltration problems. In contrast, the SCS curve number is also used to predict infiltration processes. This method has obvious theoretical limitation (as shown in Chapter 4), but it does have a reliable set of parameters that have been successfully used in hydrologic design for more than 40 years. The infiltration modeler can then choose between using the good theoretical model of Darcys equation with a questionable set of parameters or using the questionable theoretical model of the curve number method with a good set of parameters. Hierarchy theory Overview Hierarchy theory has been developed from ecosystem modeling to address modeling processes at different scales. Only a brief introduction to the concepts is given here. Lets introduce the approach by considering the growth of algae as shown below. It has been speculated that the growth rate can be limited by a nutrient depletion zone around the cells and not the average nutrient concentration that can exceed those necessary for sustained growth.

In terms of hierarchy theory, there are two scales of interest. There is the small scale corresponding to the region around cells. This scale ultimately determines the growth rate of algae. The second scale corresponds to the larger scale of the average nutrient concentration. This scale can indirectly influence algal growth by its interactions with the depletion zone. Nonetheless, the growth of algae is determined by the characteristics of the small scales surrounding the algae and not the average nutrient concentration of the tank. In Chapter 11, the use of an average parameter value to represent process is discussed in greater detail. Key concepts of hierarchy theory

1-22 Hierarchy theory assumes that a complex system can be divided into natural groups of different spatial (and/or temporal) scales. Interactions within group are complex, whereas interactions between groups are relatively simple. The groups can be organized into hierarchical systems where the higher position indicates control over lower positions. Frequently the high frequency process of the lower level group is controlled by a lower frequency input from a higher-level group. A simplified representation of these concepts is shown below. Additional details are given by Allen and Star (1984).

Nonuniqueness Deterministic models are frequently viewed as theoretical representation of processes. It is common for different models, using substantially different representations, to predict the response with equal accuracy. A unique representation is elusive: the problem is underdetermined. This is particularly an issue if the parameters are obtained using calibration. Nonuniqueness challenges the assumptions that models can be used as theoretical framework for studying processes. Nonuniqueness is best illustrated for watershed models involving several processes. Linear versus nonlinear systems Properties of linear systems In general, linear differential equations are relatively easy (at least compared to nonlinear equation to solve. Analytical solutions are often possible. Lets review the results obtained from channel routing for assumptions of a linear model.

For the special case of Ki = 0, one obtains

1-23

This linear differential equation can be easily solved using specified initial and boundary conditions. Lets multiple both sides by the integration constant et/Ko to obtain

From basic calculus, we know that

We therefore obtain

which can be integrated as

where C is the integration constant. For the special case of an upstream boundary condition of I=Io =constant, we obtain

For the special case of the initial condition of O=0 at t=0, we obtain that C=-Io. The solution to this problem is the following simple relationship

Note that for the special case of I=Io =constant, a simpler solution can be obtained by direct integration. Properties of nonlinear systems In general, relatively few analytical solutions are available for nonlinear differential equations and they are more difficult to solve numerically. In addition, they can be sensitive to initial and boundary conditions resulting in deterministic chaos. As an example, we will consider the logistic growth rate model. With this model, growth rate is defined relative to the mass or size of the system. A simple growth rate for this type of

1-24 system is defined as

or by rearranging terms

where dy/dt is the rate of change in growth rate, is a rate constant, ymax is the maximum possible mass or other indicator of size. The above function is referred to as autocatalytic growth function. It is governed by a nonlinear differential equation. Lets examine the logistic model using the following simple (perhaps too simple) finitedifference approximation to the derivative (and using the previous time step values for x):

which can be rearranged as

If we multiply both sides by the constant of t/(1+tymax), we obtain

To simplify computations, we will define variables of

which vary from y only by the constant of t/(1+tymax). We then obtain the well-known finite-difference equation of

where = 1+tymax. The above simple recursive relationship is widely cited in textbook, usually independent of the logistic growth model. For the numerical approximation of the logistic model, the value of can be made close to one by simply reducing the time step. A computer algorithm for the solution of this equation is simply: Enter value for alpha Enter value for xold Loop for time (say i=0 to 100) Print time, xold xnew=alpha*(1 - xold)*xold

1-25 xold = xnew end of loop The solution for a=4.0 is shown below. The values are never the same and are irregular (at least visually). It is called deterministic chaos, that is, although the values appear to be random, they are in fact generated by a simple deterministic algorithm.

Two solutions are shown in the above figure for a=4: one for an initial value of x=0.01 and one for an initial value of x = 0.01001. Sensitivity of initial conditions is a very important characteristics of deterministic chaos. Chaos in the logistic growth rate model was the result of the numerical approximation. This nonlinear differential equation does indeed have an analytical solution. This is easily shown by rearranging and integrating the original model as

where partial fractions were used to divide the first integrand into its two components. By integrating, we obtain

where C is an integration constant, which is related to the initial growth size at t=0. Another example of nonlinear system is one governed by Duffings equation. Duffing's nonlinear equation is defined as

1-26

The solutions for two different initial conditions are shown below

The above solutions were obtained using Runge-Kutta methods. To investigate the inherent sensitivity of the system (as opposed to the numerical solution), the time step was reduced by a factor of 10 until no perceived changes in the curve could be seen. The sensitivity to initial condition is therefore an inherent characteristics of Duffings nonlinear equation. As discussed by Baker and Gollub (1990), two necessary conditions for chaos are (1) systems governed by at least three independent dynamic variables (see example given by Baker and Gollub to understand how Duffings equation satisfies this condition) and (2) the governing equations contains one or more nonlinear terms that couples variables. Systems represented by nonlinear equations that exhibit features of deterministic chaos are very difficult to model. Very small errors in the specification of initial conditions can results in greatly different prediction results. MORE ON NONLINEAR SYSTEMS Period-doubling to chaos Lets reconsider the simple finite-difference equation developed in the previous section as a numerical approximation to the logistic growth model, that is

where in this section no physical significance is given to the relationship. The solution to the

1-27 equation is shown below for = 0.9, = 1.5, =3.3, and =3.5.

For =0.9 and = 1.5, the solution converged to a single solution. For =3.3, the solution oscillates between two values. This is called a period-2 oscillation. For =3.5, the solution oscillates between four values. This is called a period-4 oscillation. The period-doubling is useful in understanding the dynamics of nonlinear systems. Bifurcation diagrams are useful in visualizing this process. The bifurcation diagram for the simple finite-difference equations is shown below. Here the solution is obtained for different value of . Values for the y-axis correspond to values between 200 and 400 iterations (initial start-up effects have been removed).

1-28

As shown by the bifurcation diagram, there is only one stable solution for < 3. For 3<<3.5, there is a period-2 oscillation. As increases, there is a period-4 oscillation, and eventually to period-8, period-16 and so forth to chaos. For nonlinear systems that approach chaos by a period doubling bifurcation the change in the values corresponding period approaches a asymptotic limit known as the Feigenbaum number. It is defined mathematically as

where NF is the Feigenbaum number.

Lorenz system Background The Lorenz system of nonlinear equations was originally developed by a meteorologist (Lorenz) to forecast weather conditions. It is now widely cited as a seminal work in deterministic chaos. His equations were developed using an idealized representation of thermally driven convection. By substituting a Fourier series expansion into the partial differential equations of the convection system, and by considering only the low-frequency modes, Lorenz obtained the following set of differential equations: , and

1-29 The numerical solution for x obtained using the Runge-Kutta method is shown below for two different initial conditions for x. Clearly the solution is sensitive to the initial conditions. This sensitivity suggest that the weather forecasting will always be difficult because small changes in initial conditions result in large differences in predicted weather patterns.

The phase space for the Lorenz system is shown below. This is an example of a chaotic attractor. Regardless of initial values (within certain limits), the solution to the Lorenz system has this characteristic shape. However, the orbit never converges exactly to a previous state. Chaotic attractors typically have structures that resembles the overall structure when different magnification scales are applied to them. Such features are called self-similar. Fractal numbers are very useful in describing these features. Fractals are discussed in the next section.

1-30

SELF-SIMILARITY AND FRACTALS Definitions Background The geometric structures of chaotic attractors are often extremely complex. Fractal geometry has proved useful in describing processes that exhibit similar features over a range of scales associated with chaotic attractors. In addition to chaotic attractors, fractals have been used in the study of soil characteristics, stream networks, plant form, turbulence, and river flows. Fractal geometry can be used as a descriptive tool for characterizing natural phenomena or as a predictive tool for modeling them. Lets start by reviewing the use of dimensions in Euclidean geometry. A point is zerodimensional, a line is one-dimensional, an area is two-dimensional and a volume is threedimensional. A point on a line requires only one coordinate (e.g., x for the equation y=mx+b); a point on a plane requires two coordinates to describe its position; and three coordinates are necessary to specify the position of a point in three dimensional space. Dimension can also be viewed as a measure of the size or bulk of the system. For example, a point has no real size, corresponding to L0, a line has a length or L1, a surface on a plane has an area (L2) and so forth. Lets generalize the definition of bulk of the system as a length scale raised to the power D, that is, LD, where D is the fractal dimension. Before introducing the fractal dimension, we will first considering an approach for measuring the length of curve shown below by using dividers of step length g=g1 and g= g2. An

1-31 approximate length is obtained by the product of the number steps and the step length. The approximation becomes more accurate as the step length becomes smaller. For a standard curve in Euclidean geometry, the exact length is obtained as g approaches zero, or

where L is the length, g is the divider length and N is the number of steps.

The divider length can be viewed as a measure of the scale of measurements. In the above example, the length of the arc was independent of scale. An useful application of fractal dimension is for surface that the length changes with scale. This concept will be used to define the fractal dimension in the next section. Fractal dimension from Richardsons approach Richardson examined the effects of scale on the length of coastlines. In contrast to the simple arc example, this length varies with one map scale as more bays and/or peninsulas become noticeable at smaller scales (as defined using the length of the divider). At very small scales, the distances around individual rocks and pebbles would further increase the length of coastlines. The observed lengths of the Australian Coast and the West Coast of Britain obtained by Richardson for different measurement scales (i.e., g) are shown below. These coastline lengths are well-approximated by the relationship where (1-D) and F are parameters that can be obtained from the slope and intercept of a log-log plot. In contrast to Euclidean length, the number of steps,

1-32

is now dependent on the step size.

For comparison, it is clearly desirable to have a measure of coastline length that is independent of g. Lets consider a coastline constructed from small polygons of side length g. If the side length are raised to the power D, Mandelbrot (1983) asserts that a bulk measure of coastline length can be obtained "in the dimension D" that is independent of g. By defining the coastline as the number of steps multiplied by its bulk measurement, we obtain

where =F is a measure of the bulk of coastline length and is independent of g and D is called the fractal dimension. Any dimension smaller than D results in 64 as g 60, and any dimension larger than D results in 6 0 as g 60. The ratio of coastline length from two different scales (g1 and g2) follows directly from Richardsons equation, that is,

where =g2/g1 is a scale ratio. The fractal dimension is then readily defined as

An alternative formula can be obtained from the ratio of the number of steps (i.e., N((g1) and N(g2)) for two different scales. The fractal dimension is then defined as

1-33

Self-Similarity Self-similarity is an important concept in most fractal geometries. It implies that patterns repeat themselves at all scales of observation. If a coastline is self-similar, the pattern of bays and peninsulas at a large scale is statistically similar to that observed for only a portion of the coastline examined at a smaller scale. Precise definitions of self-similarity are given by Mandelbrot (1983). The construction of self-similar shapes is useful in improving our understanding of fractals. Lets consider the construction of the self-similar the triadic Koch curve. We start with an original line (called initiator by Mendelbrot, 1983) to the far left. For the middle figure, the originally unbroken line is divided into 4 segments (N2=4), each of which are 1/3 of the length of the whole line (g2=g1/3). This process is continued for the far-right figure. Here each of the middle line segments is subdivided into 4 segment that are 1/3 of length of the middle figure. By using the fractal dimension previously given, the fractal dimension for the triadic Koch curve is obtained as

which is also obtained using the definition from the number of line segments. This approach is used directly in the box-counting method of the next section.

Statistically self-similar is also often interest in the representation of watershed processes.

1-34 Here the probability density functions are invariant with respect to scale. This is discussed in greater detail in Chapter 2. More information on statistically self-similarity is also given by Mandelbrot (1983).

NONLINEAR ALGEBRAIC EQUATIONS Introduction Background Hydrologic models frequently require the solution of nonlinear algebraic equations (in addition to possible nonlinear differential equations). An example is determining the flow depth in a channel from Manning's equation for a known flow rate. For a rectangular channel, this problem can be written as

If flow rate (Q), Manning's n, channel width (b), and bed slope (S) are known, we have one equation and one unknown h. It is not possible to rearrange the equation and solve for h directly. General formulation A more general formulation can be written as where for Manning's equation

where K is a function of Manning's n, bed slope, and channel width (K = 1.49 b5/3S1/2). The goal is to determine x such that f(x) is zero as shown below.

1-35

Bisector method Approach The key computational steps for the bisector method are shown below. 1. Bracket function with negative (xn ) and positive (xp ) x corresponding to f(x=xn) < 0 < f(x=xp) 2. Select acceptable tolerance: XTOL or/and FNTOL 3. Determine midpoint: xm = (xn + xp)/2 4. Determine function value: f(xm) 5. Decide: If ABS( f(xm) < FNTOL then quit If f(xm) < 0 then xn = xm If f(xm) > 0 then xp = xm Discussion The bisector is a reliable but relatively slow iterative technique. Newton's/Secant method Theory Lets consider the Taylor's series expansion about some known point x=a.

1-36

This concept is shown below.

We are interested in a solution for x where f(x) = 0. By neglecting higher order terms, we obtain

If f'(a) can be computed directly, then the procedure is called Newton's method. If f'(a) is computed numerically, the procedure is called the Secant method. Usually, we will need to use the numerical estimate of slope. Approach The key computational steps for the secant method are shown below. 1. Select acceptable tolerance: XTOL or/and FNTOL 2. Determine initial estimate of the function for two points f(x=xi-1) and f(x=xi). From these values, the slope can be estimated as

3. Estimate new x value:

1-37

4. Check for acceptable tolerance:

5. Repeat step 3 using xi =xi+1 and xi-1 = xi. Discussion The secant method does not always converge on a solution. However, when it does converge, the secant method usually converges more rapidly than the bisector method.

1-38 REFERENCES Atkinson, 1978. An Introduction to Numerical Analysis. John Wiley & Sons, New York. Allen, T.F.H. and T.B. Starr. 1982. Hierarchy: Perspectives for Ecological Complexity. The University of Chicago Press, Chicago, IL. Baker, G.L. and J.P. Gollub. 1990. Chaotic Dynamics: An Introduction. Cambridge University Press, New York, NY. Chow, V.T., D. R. Maidment, and L.W. Mays. 1987. Applied Hydrology. McGraw Hill. DeAngelis, D.L. 1992. Dynamics of Nutrient Cycling and Food Webs. Chapman & Hall, New York. Dooge, J.C.I. 1973. Linear theory of hydrologic systems. USDA-ARS, Technical Bulletin No. 1468. Hale, J. and H. Kocak. 1991. Dynamics and Bifurcations. Springer-Verlag. New York. Kaplan, D. and L. Glass. 1991. Understanding Nonlinear Dynamics. Springer-Verlag, New York. NY. Mandelbrot, B.B. 1983. The Fractal Geometry of Nature. W.H. Freeman and Company, New York. Overton, D.E. and M.E. Meodows. 1976. Stormwater Modeling. Academic Press. Press, W.H., B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling. 1986 Art of Scientific Computing. Cambridge Press. Rainville, E.D. and P.E. Bedient. 1974. Elementary Differential Equations. Macmillan Publishing, New York. Rieu, M. and G. Sposito. 1991. Fractal fragmentation, soil porosity, and soil water properties: I. Theory. Soil Sci. Soc. Am. J. Vol. 55: 1231-1238. Sakai, K. 2001. Nonlinear Dynamics and Chaos in Agricultural Systems. Elsevier, New York, NY. Tarboton, D.G., R.L. Bras, and I. Rodriguez-Iturbe. 1988. The fractal nature of river networks. Water Resources Research, Vol. 26 (4): 2243-2244.

1-39 Thompson, J.M.T. and H.B. Stewart. 1986. Nonlinear Dynamics and Chaos. John Wiley and Sons. Tyler, S.W. and S.W. Wheatcraft. 1990. Fractal processes in soil water retention. Water Resources Research, Vol. 26 (5): 1047-1054. Vicsek, T. 1992. Fractal Growth Phenomena. World Scientific Publishing Co., River Edge, NJ.

1-40 Assignment #1

Due Date: Problem #1 (20 points) You wish to compute the outflow hydrograph for a channel reach that is defined by the following differential equation.

At t=0, outflow from the channel is zero. The inflow hydrograph is defined by the following equation:

where Io is the inflow at t=0 and corresponds to 200 cfs for this problem. Derive the analytical equation that determines the outflow hydrograph for these conditions. Problem #2 (30 points) Write a computer program to solve the finite-difference equation given in the lecture notes. For an initial conditions of x = 0.01, obtain the solutions for a=3.3, a=3.5, and a=4.0. Plot the results as done in the lecture notes. For a=4.0, also solve the finite-difference equation for x=0.01001. Problem #3 (30 points) Write a computer program using the bisection method and the secant method to compute the flow depth from Manning's equation, as discussed in class. Select three combinations of the parameters of flow rate, Manning's n, slope, and channel width. Compare the flow computed with the depths obtained with the bisection and secant methods with that specified in your parameter set.

S-ar putea să vă placă și