You are on page 1of 37

BSC 417/517 Environmental Modeling Introduction to Oscillations

Oscillations are Common


Oscillatory behavior is common in all types
of natural (physical, chemical, biological) and human (engineering, industry, economic) systems Systems dynamics modeling is a powerful tool to help understand the basis for and influence of oscillations on environmental systems

First Example: Influence of Variable Rainfall on Flower Growth


Flower growth model of S-shaped growth
from Chapter 6:
g rowth area of flowers decay

intr insic g r owth rate decay rate actual g rowth r ate ~ g rowth r ate multipli er fr action occupi ed sui table area

actual_growth_rate = intrinsic_growth_rate*growth_rate_multiplier

growth_rate_multiplier = GRAPH(fraction_occupied)

Growth Rate Multiplier for Modeling S-Shaped Growth


1

Growth rate multiplier

0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 Fraction occupied

Analogy Between Logistic Growth Equation and Growth Rate Multiplier Approach
Logistic equation:
dN/dt = r N f(N) f(N) = (1 N/K) K = carrying capacity

Growth rate multiplier approach


dN/dt = r N GRAPH(fraction_occupied) fraction_occupied = area_of_flowers/suitable_area

If GRAPH(fraction_occupied) is linear with slope of negative


one, then we have recovered precisely the logistic growth equation

Analogy Between Logistic Growth Equation and Growth Rate Multiplier Approach
Growth rate multiplier approach
dN/dt = r N (1 area_of_flowers/suitable_area) Logistic equation: dN/dt = r N (1 N/K) The two equations are identical because N/K = area_of_flowers/suitable_area

Growth Rate Multiplier Approach is More Flexible Than the Classical Logistic Equation
Logistic equation has an analytical solution:
Nt = N0ert/(1 + N0(ert 1))/K However, no simple analytical solution exists if growth rate multiplier is a nonlinear function of N In contrast, its easy to numerically simulate such a system using the graphical function approach

Growth Rate Multiplier Approach is More Flexible Than the Classical Logistic Equation
1

Growth rate multiplier

0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 Fraction occupied

First Example: Influence of Variable Rainfall on Flower Growth


Assume rainfall varies sinusoidally around a mean
of 20 inches/yr with an amplitude of 15 inches/yr and a periodicity of 5 years:
Rainfall = 20 + SINWAVE(15,5) Rainfall = 20 + 15*SIN(2*PI/5*TIME)

Assume optimal rainfall for flower growth is 20


inches per year Define relationship between intrinsic growth rate and rainfall using a nonlinear graphical function

Relationship Between Intrinsic Growth Rate and Rainfall


Intrinsice growth rate (1/yr)
1 0.8 0.6 0.4 0.2 0 0 5 10 15 20 25 30 35 40 Rainfall (inches/year)

Flower Model With Variable Rainfall


r ai nfal l g rowth area of flowers decay

~ intr insic g r owth rate decay rate actual g rowth r ate ~ g rowth rate multipli er fr action occupi ed sui table area

Flower Model With Variable Rainfall


1: r ainfall 1: 2: 3: 60. 2.D 2: i ntri nsi c g rowth r ate 3: actual g r owth rate

Period = 5 yr Period = 2.5 yr


1: 2: 3: 30. 1.D 1 3 1 3 3 2 1: 2: 3: Pag e 1 0 0 0.00 5.00 10.00 Year s Unti tled 15.00 20.00 10:05 PM Sun, Oct 27, 2002 2 2 2 3 1 1

Flower Model With Variable Rainfall


1: ar ea of fl ower s 1: 2: 3: 700 200 300 3 2 1: 2: 3: 350 100 150 2: decay 1 3: g r owth 1

3 2

3 1: 2: 3: Pag e 1 0 0 0 1 0.00 2 10.00 20.00 Year s Unti tled 30.00 40.00 10:10 PM Sun, Oct 27, 2002

Flower Model With Variable Rainfall


Sinusoidal changes in rainfall causes large
swings in growth rate but only minor swings in area and decay General pattern of growth is S-shaped, with a superimposed cycle of 2.5 year (compared to 5 years for rainfall) Equilibrium flower area is lower than that obtained with model employing constant optimal intrinsic growth rate

General Conclusions
Cycles imposed from outside the system
can be transformed as their affects pass through the system Periodicity can be modified as a result of system dynamics Quantitative effect of external variations can be moderated at the stocks in the system

Oscillations From Inside the System


Consider oscillations that arise from structure
within the system New version of flower model in which in the impact of the spreading area on growth is lagged in time, i.e. there is a time lag (2 years) before a change in fraction occupied translates into a change in growth rate
lagged_value_of_fraction = smth1(fraction_occupied,lag_time)

Structure of First-Order Exponential Smoothing Process


change in fraction occupied

0.0
lagged value of fr action occupied

fr action occupied

1.0

lag time

2.0

change_in_fraction_occupied = (fraction_occupied-lagged_value_of_fraction_occupied)/lag_time

Structure of First-Order Exponential Smoothing Process


1: l agg ed val ue of fr action occupi ed 1: 1.00 1 1 1

1:

0.50

1 1: Pag e 1 0.00 0.00 2.50 5.00 Time Unti tled 7.50 10.00 9:26 PM Mon, Oct 28, 2002

Flower Model With Lagged Effect of Area Coverage


g rowth area of flowers decay

intr insic g r owth rate fr action occupied actual g rowth rate ~ g rowth rate multipli er

decay rate area avail abl e lag time

lag g ed value of fr acti on

Flower Model With First Order Lagged Effect of Area Coverage


1: ar ea of fl ower s 1: 2: 3: 1300 2: decay 3: g r owth

1 1: 2: 3:

650

3 1: 2: 3: Pag e 1 2 0 1 0.00 2 10.00 20.00 Year s Unti tled 30.00 40.00 9:37 PM Mon, Oct 28, 2002 3 2 3 2 3

Flower Model With First Order Lagged Effect of Area Coverage


Area of flowers overshoots maximum
available area, which causes a major decline in growth so that decay exceeds growth by 8th year of simulation Area declines, which frees up space, which eventually results in an increase in growth Variations in growth and decay eventually fade away as the system approaches dynamic equilibrium = damped oscillation

Higher Order Lags are Possible


STELLA has built-in function for 1st, 3rd,
and nth order smoothing, which can be used to produced any desired order of lag The higher the order of the lag, the longer the delay in impact Example = third order lag

Structure of Third Order Exponential Smoothing Process


chang e in fraction occupied 1 lag g ed value of fr action occupied 1 fr action occupied lag g ed value of fr action occupied 2 chang e in fraction occupied 2

lag g ed value of fr action occupied 3 lag time chang e in fraction occupied 3

Structure of Third Order Exponential Smoothing Process


1: l agg ed val ue of fr action occup 1: 2: 3: 1.00 0.90 1 2 3 2: l agg ed val ue of fr action occup 3: l agg ed val ue of fr action occup 1 2 3

1: 2: 3:

0.50 0.45

1: 2: 3: Pag e 1

1 0.00 0.00 0.00 2 3 2.50 5.00 Time Unti tled 7.50 10.00 9:50 PM Mon, Oct 28, 2002

Flower Model With First vs. Third Order Lagged Effect of Area Coverage
1: ar ea of fl ower s fir rst order l ag 1: 2: 1500 2: ar ea of fl ower s thir d or der l ag

1 1: 2: 750 2

1 2

1: 2: Pag e 1

1 0.00

2 10.00 20.00 Year s Unti tled 30.00 40.00 9:54 PM Mon, Oct 28, 2002

Flower Model With First vs. Third Order Lagged Effect of Area Coverage
Third order lag shows more volatility Flower area shoots farther past the carrying
capacity of 1000 acres and goes through large oscillations before dynamic equilibrium is achieved Increased volatility arises because of the longer lag implicit in the third order smoothing

Further Examination of Lag Time Effect


Compare simulations with third order
smoothing and lag times of 1, 2, or 3 years Longer lags lead to greater volatility Flower area in simulation with 3 year lag time shoots up to greater than 2X the carrying capacity

Flower Model With Third Order Lagged Effect of Area Coverage and Variable Lag Time
1: ar ea of fl ower s 1 year thir d or 1: 2: 3: 2500 2: ar ea of fl ower s 2 year thir d or 3: ar ea of fl ower s 3 year thir d or

1: 2: 3:

1250 3 1 2 3 1 2 1 2

1: 2: 3: Pag e 1

1 0.00

2 10.00 20.00 Year s Unti tled 30.00 40.00 10:01 PM Mon, Oct 28, 2002

Effects of Volatility Illustrated


Plot growth and decay together with flower area for simulation

with 3 year time lag Flower area and growth rate increase in parallel even after carrying capacity is reached; flowers do not feel the effect of space limitation due to the time lag Once effect of space limitation kicks in, growth rate drops rapidly to zero Active growth does not resume until ca. year 15, meanwhile decay continues on New growth spurt occurs at around year 20, utilizing space freed-up during previous period of decline Magnitude of oscillations does not decline over time = sustained oscillation

Effects of Volatility Illustrated


1: ar ea of fl ower s 3 year thir d or 1: 2: 3: 2500 2: g r owth 3 3: decay 3

1 1: 2: 3: 1250

1 1 2 1: 2: 3: Pag e 1 1 0.00 2 3 10.00 3 2 20.00 Year s Unti tled 30.00 40.00 10:16 PM Mon, Oct 28, 2002 3 2 3

Effects of Volatility Illustrated


Key reason for sustained volatility of the model with

long time lag is the high intrinsic growth rate To illustrate, repeat simulation with different values of the intrinsic growth rate and a 2 year lag time Sustained oscillation (volatility) occurs with intrinsic growth rate of 1.5/yr With intrinsic growth rate of 1.0/yr, oscillations dampen over time With intrinsic growth rate of 0.5/yr, no oscillations occur (system is overdamped)

Influence of Intrinsic Growth Rate on Volatility


area of flowers 2 year thi rd order lag 2: 1: 3000 1- 2- 3-

r = 1.5/yr

3 1: 1500

r = 1.0/yr
3 3 3 1 2 2 1

2 1 10.00

r = 0.5/yr
1: Pag e 1 0 1 0.00 2 20.00 Year s Unti tled 30.00 40.00 10:28 PM Mon, Oct 28, 2002

Summary of Oscillatory Tendencies


Simple flower model gives rise to three
basic patterns of oscillatory behavior: Overdamped Damped Sustained depending on the values for lag time and intrinsic growth rate Can summarize the observed effects with a parameter space diagram

Oscillatory Behavior: Parameter Space Diagram


3

+
Overdamped

Sustained

Lag time (yr)

Damped

Sustained

Overdamped

Critical dampening curve

0.5 1.0 1.5 Intrinsic growth rate (yr-1)

Critical Dampening Curve


Hastings (1997) analyzed a logistic growth model
with lags, and found that oscillations occurred only when the product of the intrinsic growth rate and time lag (a dimensionless parameter) was greater than 1.57 Flower model is not identical to Hastingss model, but there is sufficient similarity to warrant using his findings as a working hypothesis for position of the critical dampening curve Define FMVI = Flower Model Volatility Index as the product of the time lag and the intrinsic growth rate in the flower model
FMVI = intrinsic growth rate x lag time

Curve For Critical Dampening


Curve in our parameter space diagram was drawn
so that FMVI is 1.5 everywhere along the curve Assuming that the FMVI of 1.5 is analogous to Hastingss value of 1.57, hypothesize that oscillations will appear only whenever the parameter values land above the curve Results of the six simulations discussed previously support this hypothesis

The Volatility Index


The dimensionless parameter FMVI is a plausible
index of volatility because it reflects the tendency of the system to overshoot its limit Can be interpreted as the fractional growth of the flowers during the time interval required for information to feed back into the simulation
FMVI = growth rate (1/year) x lag time (year)

The higher the index, the greater the tendency to


overshoot

You might also like