paramecium didiniumjgrant3/cw/mcmfinal.pdf#...

9
A Review of Gary W. Harrison’s “Comparing PredatorPrey Models to Luckinbill’s Experiment with Didinium and Parameciumby Jason Grant, Bethany Herila, and Shant Mahserejian Final Project for ACMS 50730: Mathematical /Computational Modeling Fall 2012 – Dr. Alber Introduction Mathematical models describing predatorprey interaction often base their theory on the Lotka Volterra equations. They provide a good framework for describing the changes in populations between two groups, where one group consumes the other, such as wolves and sheep, lions and antelope, or even on smaller scales dealing with microbacteria. However, these mathematical equations often don’t provide the detailed circumstances that different predatorprey combinations could have in realworld circumstances. This was the case for Leo Luckinbill’s experiments in 1973, when he studied the behavior of two groups: Paramecium, and its natural predator Didinium. Due to their rapid growth and consumption rates, the changes in population dynamics of bacteria provide a convenient setting for studying predatorprey interplay. Although, from a mathematical point of view, there was more to be understood in order to accurately describe how the populations of these two groups changed when they were put together. Gary Harrison was interested in finding out which governing equations played a dominating role in characterizing the behavior of these two bacterial groups. This paper follows Harrison’s work as he searches for mathematical models that make a better fit to the data collected from Luckinbill’s experiments. 1. The Basic PredatorPrey Model LotkaVolterra’s Model The Lotka (1925) and Volterra (1931) equations provide a general model to predatorprey interactions. This basic model defines two rates, and population densities (number per cubic centimeter) over time for the prey and predatory, x ( t ) and y ( t ) respectively. dx dt = ρ 1 x K x ωf ( x ) y dx dt = σf ( x ) y γy where ωf ( x ) = ωx φ + x The parameters in this model are characterized as follows: ρ is the specific growth rate of the prey; K is the carrying capacity of the prey in the absence of predators; γ is the rate at which predators die in the absence of prey; ω is the maximum rate of prey consumed by a single predator; σ ω is the efficiency of converting prey to predator; φ is the halfsaturation constant; and ωf ( x ) , the rate at which prey are consumed per predator, also known as the functional response of the predator.

Upload: others

Post on 21-Jan-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

A  Review  of  Gary  W.  Harrison’s  “Comparing  Predator-­Prey  Models    to  Luckinbill’s  Experiment  with  Didinium  and  Paramecium”  

by    Jason  Grant,  Bethany  Herila,  and  Shant  Mahserejian  

Final  Project  for  ACMS  50730:  Mathematical  /Computational  Modeling  Fall  2012  –  Dr.  Alber  

   Introduction       Mathematical  models  describing  predator-­‐prey  interaction  often  base  their  theory  on  the  Lotka-­‐Volterra  equations.  They  provide  a  good  framework  for  describing  the  changes  in  populations  between  two  groups,  where  one  group  consumes  the  other,  such  as  wolves  and  sheep,  lions  and  antelope,  or  even  on  smaller  scales  dealing  with  micro-­‐bacteria.  However,  these  mathematical  equations  often  don’t  provide  the  detailed  circumstances  that  different  predator-­‐prey  combinations  could  have  in  real-­‐world  circumstances.  This  was  the  case  for  Leo  Luckinbill’s  experiments  in  1973,  when  he  studied  the  behavior  of  two  groups:  Paramecium,  and  its  natural  predator  Didinium.  Due  to  their  rapid  growth  and  consumption  rates,  the  changes  in  population  dynamics  of  bacteria  provide  a  convenient  setting  for  studying  predator-­‐prey  interplay.  Although,  from  a  mathematical  point  of  view,  there  was  more  to  be  understood  in  order  to  accurately  describe  how  the  populations  of  these  two  groups  changed  when  they  were  put  together.  Gary  Harrison  was  interested  in  finding  out  which  governing  equations  played  a  dominating  role  in  characterizing  the  behavior  of  these  two  bacterial  groups.  This  paper  follows  Harrison’s  work  as  he  searches  for  mathematical  models  that  make  a  better  fit  to  the  data  collected  from  Luckinbill’s  experiments.      1.  The  Basic  Predator-­Prey  Model    Lotka-­‐Volterra’s  Model     The  Lotka  (1925)  and  Volterra  (1931)  equations  provide  a  general  model  to  predator-­‐prey  interactions.  This  basic  model  defines  two  rates,  and  population  densities  (number  per  cubic  centimeter)  over  time  for  the  prey  and  predatory,  

x(t)  and  

y(t)  respectively.  

dxdt

= ρ 1− xK

⎝ ⎜

⎠ ⎟ x −ωf (x)y

dxdt

=σf (x)y − γy

⎪ ⎪

⎪ ⎪

where ωf (x) =ωxφ + x

 

The  parameters  in  this  model  are  characterized  as  follows:  

ρ  is  the  specific  growth  rate  of  the  prey;  

K  is  the  carrying  capacity  of  the  prey  in  the  absence  of  predators;  

γ  is  the  rate  at  which  predators  die  in  the  absence  of  prey;  

ω  is  the  maximum  rate  of  prey  consumed  by  a  single  predator;  

σω  is  the  efficiency  of  

converting  prey  to  predator;  

φ  is  the  half-­‐saturation  constant;  and  

ωf (x) ,  the  rate  at  which  prey  are  consumed  per  predator,  also  known  as  the  functional  response  of  the  predator.    

Page 2: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

  This  initial  model  is  quite  fundamental  in  nature.    It  is  assumed  that  the  predator's  growth  rate  is  strictly  dependent  and  proportional  to  the  rate  of  prey  consumption.  The  preliminary  extension  to  the  model  begins  with  the  functional  response  of  the  predator,  

ωf (x) .  Evidence  from  Holling  (1959)  and  Murdock  and  Oaten  (1975)  suggest  that  predators  often  exhibit  a  saturation  effect  in  their  functional  rate  when  prey  is  abundant,  and  gave  rise  to  the  Holling  Type  2  functional  response  formula.    Parameter  dependence     In  order  to  display  the  importance  of  these  parameters  in  the  behavior  of  the  population  dynamics  simulated  by  this  model,  the  ODE45  solver  in  MATLAB  was  used  to  solve  this  system  of  equations  for  a  given  set  of  non-­‐zero  parameters.  The  resulting  simulation  demonstrates  the  oscillatory  behavior  of  a  predator-­‐prey  scenario  and  their  stabilizing  behavior  as  time  goes  on  in  Figure  1.  This  behavior  is  a  typical  characteristic  of  the  Lotka-­‐Volterra  Model,  and  therefore  a  favored  choice  for  describing  a  scenario  exhibiting  this  sort  of  phenomena.  Changing  the  values  of  each  parameter  one  by  one  to  a  non-­‐significant  value  of  0.1  demonstrates  how  this  characterization  is  lost,  as  displayed  in  Figure  2.    

 Figure  1:  A  simulation  generated  by  solving  the  Lotka-­Volterra  Model  using  a  set  of  non-­zero  parameters  

Page 3: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

 Figure  2:  A  simulation  generated  by  solving  the  Lotka-­Volterra  Model,  but  setting  each  parameter  to  the  insignificant  value  of  0.1  

one  by  one  is  the  following  order:  (top  row,  left  to  right)  

γ ,  

K ,  

φ ,  and  (bottom  row,  left  to  right)  

ρ ,  

σ ,  

ω .  

Due  to  the  drastic  changes  in  the  profiles  of  the  simulated  population  densities  for  both  predator  and  prey,  it  is  clear  why  each  parameter  serves  an  important  role  in  creating  the  characteristic  behavior  that  the  Lotka-­‐Volterra  Model  is  known  for.  So,  in  studying  to  improve  this  model,  merely  changing  the  parameter  values  isn’t  enough  to  have  it  fit  observed  experimental  data.  Rather,  other  considerations  and  alterations  must  be  made  to  better  describe  the  governing  forces  behind  a  certain  real  world  scenario.    Stability     Steady-­‐state  equilibrium  of  the  predator-­‐prey  model  occurs  when  oscillations  between  the  species'  density  have  dampened  and  equilibrium  values  are  reached  by  both  prey,  

x* ,  and  predator,  

y* .    The  equilibrium  point  

E1 = (x*,y*)  may  be  considered  stable  or  unstable  within  the  system,  and  it  is  

found  by  setting  

dxdt

=dydt

= 0,  and  then  solving  the  equations  for  

x  and  

y .  Doing  so  gives  the  following  

points  for  stability:  

Prey isocline for dxdt

= 0 : x* = 0 and y* =ρω

1− xK

⎝ ⎜

⎠ ⎟ φ + x( ) which is a concave - down parabola

Predator isocline for dydt

= 0 : x* =γφ

σ − γ and y* = 0 which is a vertical line

 

 Using  the  graphical  stability  condition  of  Rosenzwieg  and  MacArthur,  

E1  is  stable  if  the  vertical  predator  isocline  is  to  the  right  of  the  vertex  of  the  parabolic  prey  isocline,  or:  

γφσ −γ

>K −φ2

 

  If  the  previous  inequality  is  not  satisfied,  then  

E1  is  unstable  and  oscillations  will  spiral  out  toward  a  stable  limit  cycle.    Analysis  shows  that  the  further  

E1  begins  to  the  left  of  the  vertex  of  the  prey  

Page 4: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

isocline,  the  faster  the  oscillations  will  diverge.    Whether  the  predator  and  prey  coexist  depends  on  how  close  the  oscillations  come  to  the  boundaries  

x = 0 and y = 0 .    Theoretically,  any  real-­‐value  above  zero  implies  the  presence  of  predators  or  prey;  however,  in  a  realistic  scenario,  a  population  density  below  1  is  considered  to  be  extinct.      Luckinbill's  Experiments     Leo  Luckinbill  conducted  an  experiment  in  1973,  which  consisted  of  three  predator-­‐prey  interactive  models.  Paramecium  aurelia  was  grown  together  with  its  predator,  Didinium  nasatum  in  6  mL  of  standard  Cerophyl  medium.  In  the  first  run,  experiment  A,  the  Didinium  consumed  all  the  Paramecium  within  a  few  hours.    During  the  second  phase,  expriment  B,  the  medium  was  thickened  using  Methyl  Cellulose,  and  the  populations  went  threw  two  to  three  cycles  before  the  predator  group  became  extinct.    The  primary  effect  of  adding  Methyl  Cellulose  was  to  reduce  the  swimming  speed  of  both  species,  although  it  turned  out  to  have  a  greater  effect  on  the  “hunting  ability”  of  the  predator.  In  the  third  and  final  experiment  C,  the  medium  was  changed  to  half-­‐strength  Cerophyl,  but  also  thickened  with  Methyl  Cellulose.  By  doing  so,  the  carrying  capacity  of  the  Paramecium  was  reduced  and  again  allowed  the  Didinium  to  move  at  as  slower  rate.  Luckinbill  showed  that  reducing  the  nutrient  supply  for  the  paramecium  by  half  reduced  only  decreased  the  growth  rate  slightly  while  mostly  decreasing  the  carrying  capacity  from  approximately  900  individuals/mL  to  400  individuals/mL.  This  resulted  in  the  amplitude  of  the  stable  limit  cycle  decreasing  to  a  smaller  radius,  and  produced  oscillating  populations,  which  were  sustained  over  several  cycles  (33  days)  without  either  population  going  extinct.         Though  the  final  experiment  lasted  for  33  days,  a  constant  population  density  was  never  reached.  Qualitatively,  it  is  suggested  that  by  further  thickening  the  medium  and/or  using  a  weaker  Cerophyl  medium,  a  stable  system  could  be  produced  with  the  population  settling  to  constant  density  values.  Luckinbill's  experiments  showed  that  the  basic  predator-­‐prey  model  is  a  good  start  for  a  qualitative  match,  however,  it  performed  poorly  in  computing  the  numerical  results  to  make  a  qualitative  match.  Therefore,  finding  a  mathematical  model  that  accurately  described  Luckinbill’s  experiments  was  desired.                2.  Error  Estimates,  Solution  Method,  and  Relevant  Values       In  order  to  quantify  values  pertaining  to  the  mathematical  model,  an  error  estimation  system  is  established  to  provide  a  measure  for  comparing  various  types  of  models,  but  also  to  find  appropriate  values  for  parameters  and  initial  conditions  that  would  fit  the  real  observed  data  values.  As  for  the  experimental  data,  it  is  reported  that  the  numbers  for  experiments  A  &  B  where  only  available  in  the  form  of  graphs  from  Luckinbill’s  report,  while  the  numbers  for  experiment  C  were  provided  by  Luckinbill  directly  via  personal  communication.  So,  in  order  to  retrieve  values  for  experiments  A  &  B,  digital  analysis  was  applied  to  their  respective  graphs,  and  therefore  they  suffered  an  additional  source  of  error.  Furthermore,  the  real  data  from  experiments  A  &  C  proved  to  be  difficult  to  fit,  so  therefore  the  preliminary  error  was  first  computed  against  the  data  from  experiment  B.  Once  a  match  was  found  for  this  set  of  data,  the  necessary  parameters  were  adjusted  appropriate  to  make  comparisons  for  the  data  sets  form  experiments  A  &  C.        Methods  and  Error  Criteria     The  Marquardt-­‐Levenberg  method  was  implemented  to  estimate  the  parameters  in  the  differential  equations,  which  were  then  solved  using  a  variable  step  Runge-­‐Kutta  method.  These  simulated  results  were  matched  to  the  experimental  data  in  the  least  squared  sense  in  order  to  find  the  best  fit.  This  meant  parameter  values  were  chosen  based  on  minimizing  the  error  term  S2  defined  by:  

Page 5: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

S2 =x(ti) − xobserved (ti)

sxi

⎣ ⎢

⎦ ⎥

2

+i=1

N

∑ y(ti) − yobserved (ti)syi

⎣ ⎢

⎦ ⎥

2

i=1

N

∑  

The  terms  

x(ti)  and  

y(ti)  are  the  simulated  values  for  prey  and  predator  densities,  respectively,  at  time  steps  

ti .  The  terms  

xobserved (ti)  and  

yobserved (ti)  are  values  from  experimental  data.  The  terms  

sxi  and  

syi  are  weights  representing  the  relative  measurement  errors  in  the  observation.  Though  several  options  for  these  weights  were  considered,  the  best  fitting  model  showed  no  difference  between  utilizing  a  weight,  versus  calculating  unweighted  error  values.  Also,  other  error  values  were  calculated  using  a  fixed  prey  growth  rate,  and  altering  intial  conditions,  but  they  did  not  pose  much  of  a  difference  in  the  error  analysis,  especial  in  the  case  of  the  best  fitting  model.  For  the  sake  of  simplicity,  the  unweighted  error  values  (where  

sxi = syi =1)  while  allowing  movement  in  all  of  the  parameters,  represented  by  S12,  is  the  only  one  focused  on  for  this  report.      Initial  Conditions     Luckinbill  reported  that  the  initial  number  of  organisms  at  the  start  of  his  experiments:  

x(0) =15 Paramecium/mL and y(0) = 5.833 Didinium/mL .  However,  it  is  also  reported  that  the  first  0.5  day  of  measurements  do  not  reflect  the  predator-­‐prey  interactions  predicted  by  the  suggested  mathematical  models,  because  as  admitted  by  Luckinbill,  “well  fed  Didinium”  were  placed  in  the  experimental  medium.  The  error  value  computed  by  taking  this  time  delay  into  consideration  did  not  differ  significantly  from  the  S12  value  for  the  best  fitting  model;  therefore,  it  is  not  discussed  in  more  detail  in  this  report.    Parameter  Values     Along  with  the  experimental  data  for  the  cases  described  by  experiments  A,  B,  and  C,  measurements  of  Paramecium  grown  alone  were  taken  for  both  regular  and  weaker  solutions  of  Cerophyl.  This  data  was  used  to  compute  the  prey’s  carrying  capacity  in  both  cases  as  K=898  and  400  respectively.    The  rest  of  the  parameters  (for  each  respective  model)  were  found  by  numerically  finding  values  that  provided  a  best  fit  in  the  least  squares  sense  to  the  real  data  from  experiment  B.     In  addition,  it  was  reported  that  a  case  was  never  observed  where  two  very  different  sets  of  parameters  both  giving  good  fits  to  the  observed  data.  Therefore,  it  seemed  practical  to  only  estimate  parameters  to  a  degree  of  accuracy  no  higher  than  5%.  After  all,  the  goal  of  the  study  was  not  to  determine  parameters,  but  what  types  of  behavior  take  place  in  this  predator-­‐prey  interaction.      3.  Finding  a  Quantitative  Match       I.  Basic  Lotka-­‐Volterra  Model:  When  using  the  basic  model  utilizing  reasonably  sized  parameters,  it  predicts  the  correct  number  of  oscillations  in  the  population  densities,  hence  providing  a  good  qualitative  match.  Though  this  sets  a  solid  place  to  start,  the  quantitative  fit  was  described  as  “disappointing”.  The  resulting  error  value  was  S12=236,137.  Due  to  this  large  error,  there  was  great  interest  to  alter  the  model  with  possible  considerations  that  would  improve  the  error,  and  match  the  conditions  theoretically  describing  the  governing  equations  of  the  experiment  more  closely.       II.  Predator  Mutual  Interference:  At  first  glance,  the  most  obvious  mismatch  is  that  the  predators  seem  to  grow  too  rapidly.  This  led  to  attempting  to  introduce  competition  amongst  the  predators  for  space  or  some  resource  other  than  the  prey.  This  was  done  by  including  a  new  function  

g(y)  into  the  model  as  follows:  

Page 6: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

dxdt

= ρ 1− xK

⎝ ⎜

⎠ ⎟ x −ωf (x)g(y)

dydt

=σf (x)g(y) − γy

⎪ ⎪

⎪ ⎪

where ωf (x) =ωxφ + x

and g(y) =y

1− βy

 

Here,  

β > 0  is  the  rate  of  reduction  in  predators  at  high  predator  densities,  hence  controlling  the  population  of  predators  when  too  many  of  them  came  to  be.       This  alteration  in  the  model  produced  simulations  that  returned  an  error  of  S12=120,313,  which  is  about  50%  of  the  basic  model  error,  a  tremendous  improvement  for  a  first  attempt.  Although,  as  far  as  the  numbers  were  concerned,  the  predator  density’s  peaks  occurred  too  soon,  and  the  population  levels  dropped  too  low  to  correctly  represent  the  experimental  data.         III.  Ratio-­‐Dependent  Functional  Response:  This  consideration  was  based  on  the  fact  that  the  predator  functional  response  does  not  depend  on  the  prey  density  alone,  but  on  the  ratio  of  prey  to  predators  

xy .  So,  the  functional  response  term  was  redefined  and  gave  us  the  following  model:  

dxdt

= ρ 1− xK

⎝ ⎜

⎠ ⎟ x −ωf (x,y)y

dydt

=σf (x,y)y − γy

⎪ ⎪

⎪ ⎪

where ωf (x,y) =ω( xy )φ + ( xy )

 

Unfortunately,  using  this  ratio-­‐dependent  functional  response  based  model  was  far  too  stable  to  represent  Luckinbill’s  experiments,  especially  for  the  set  of  parameter  values  chosen.  Therefore,  this  type  of  correction  was  abandoned.       IV.  Leslie-­‐type  model:  In  order  to  control  the  predator  population,  an  attempt  was  made  to  introduce  a  carrying  capacity  for  the  predator  that  is  proportional  to  the  prey  density.  This  required  redefining  the  differential  equation  for  the  predator’s  population  rate  of  change.  

dxdt

= ρ 1− xK

⎝ ⎜

⎠ ⎟ x −ωf (x)y

dydt

=σ 1− γyx

⎝ ⎜

⎠ ⎟ y

⎪ ⎪

⎪ ⎪

where ωf (x) =ωxφ + x

 

Making  this  change  to  the  basic  model  resulted  in  an  error  of  S12=143,491,  about  60%  of  basic  model  error.  Note  that  this  is  not  as  good  as  the  improvement  by  considering  predator  mutual  interference.  Furthermore,  the  Leslie-­‐type  model  still  contained  systematic  errors,  including  incorrectly  predicting  the  addition  of  Methyl  Cellulose  to  the  medium.  This  consideration  was  also  abandoned.    

Page 7: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

  V.  Sigmoid  functional  response:  So  far,  all  previous  variations  predicted  that  the  prey  densities  drop  to  values  much  lower  than  the  real  data  measurements.  This  suggests  that  the  predator  reduces  its  searching  rate  when  prey  populations  become  scarce.  The  suggested  correction  was  to  apply  a  sigmoid  functional  response  to  predator  density  by  redefining  

f (x)  in  the  model  defined  by  the  predator  mutual  interference.  

dxdt

= ρ 1− xK

⎝ ⎜

⎠ ⎟ x −ωf (x)g(y)

dydt

=σf (x)g(y) − γy

⎪ ⎪

⎪ ⎪

where f (x) =x

φ + x1− 1+νx( )e−νx[ ] and g(y) =

y1− βy  

The  resulting  error  for  this  model  configuration  is  S12=93,546  (40%  of  basic  model  error)  by  using  a  rate  of  predator  reduction  of  

β = 0.00057  ,  which  is  only  slightly  better  when  the  rate  was  set  to  

β = 0.  Hence,  including  the  predator  mutual  interference  effect  is  unnecessary  for  its  added  parameter  and  complexity.    Also,  the  visual  fit  is  not  as  good  as  the  predator  mutual  interference  alone,  in  particular  the  predator  densities.  It  makes  the  predation  rates  too  low  and  drives  prey  to  extinction,  and  therefore  leaving  more  room  for  improvement.         VI.  Delayed  Predator  numerical  response:  Another  offset  made  by  previous  versions  of  the  model  predicted  that  the  predator  density  respond  instantaneously  with  changes  in  prey  densities.  They  showed  it  to  be  quicker  than  what  the  real  data  suggests,  resulting  in  peaks  in  the  density  graphs  occurring  too  early.  To  fix  this  issue,  a  time  lag  applied  in  the  predator’s  numerical  response  by  assuming  (1)  prey  consumption  creates  an  inflow  of  energy  stored,  and  (2)  overall  predator  growth  rate  is  an  increasing  function  of  average  energy  per  predator.  So,  a  new  variable  z  is  introduced  representing  the  total  energy  storage  of  predators.    The  model  is  rescaled  for  simplification  and  combined  with  a  sigmoid  functional  response  to  be  stated  as  follows:  

dxdt

= ρ 1− xK

⎝ ⎜

⎠ ⎟ x −ωf (x)y

dzdt

=σf (x)y −δz

dydt

= z − γy

⎪ ⎪ ⎪ ⎪

⎪ ⎪ ⎪ ⎪

where f (x) =x

φ + x1− 1+νx( )e−νx[ ]

 

Here  the  term  

δz  represents  the  predator’s  reproductive  rate.  This  configuration  of  the  model  gives  the  best  fit  to  the  experimental  data  with  an  error  value  of  S12=30,084,  only  13%  of  basic  model  error,  which  is  a  tremendous  improvement.  Though  other  combinations  using  the  standard  functional  response  and  the  predator  mutual  interference  were  attempted,  they  did  not  yield  the  optimal  results  of  this  version  of  delayed  predator  response  with  the  sigmoid  functional  response.  Continuing  in  this  manner  by  introducing  a  delay  in  prey  growth  only  made  slight  improvements,  hence  not  justifying  the  complexity  of  the  involvement  of  additional  parameters  associated  with  it.    

Page 8: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

4.  Conclusion     Since  the  standard  Lotka-­‐Volterra  model  utilizing  Holling’s  functional  response  equation  only  provided  a  qualitative  match  the  results  from  Luckinbill’s  experiments,  several  alterations  were  considered  in  an  attempt  to  find  a  quantitative  fit  to  the  observed  data.  The  optimal  model  to  describe  the  predator-­‐prey  recorded  interplay  was  based  on  the  criteria  not  only  for  the  best  fit  in  the  least  squares  sense,  but  also  for  simplicity  as  to  avoid  the  complexity  introduced  by  minimally  beneficial  parameters.  Therefore,  using  a  Delayed  Predator  Numerical  Response  along  with  a  Sigmoid  Functional  Response  was  decided  to  be  the  best  model  to  describe  the  population  dynamics  between  Paramecium  aurelia  and  its  natural  predator  Didinium  nasatum.  The  resulting  simulation  compared  with  the  real  data  is  shown  in  Figure  3.    

 

Figure  3:  Best  fit  of  the  model  with  Delayed  Predator  Response  and  Sigmoid  Functional  Response  for  (solid)  vs.  observed  data  from  Luckinbill’s  experiments  (dashed)  for  experiment  B.  

The  model  parameters  numerically  determined  for  the  best  fit  of  this  version  of  the  model  are  as  follows:  

K = 898 ρ = 3.02 ω = 9.74φ = 54.3 σ = 9.15 γ =1.78ν = 0.0983 δ =1.78 z(0) = 90.45

 

    Reporting  on  the  success  of  this  study,  the  researchers  made  a  significant  improvement  from  the  standard  Lotka-­‐Volterra  predator-­‐prey  model,  to  the  final  version  by  eliminating  87%  of  the  error  in  the  least  squares  fit  to  Luckinbill’s  observed  experimental  data.  By  including  the  delaying  the  response  of  a  predator  consuming  prey,  and  converting  it  to  energy  for  reproduction,  as  well  as  redefining  the  functional  response  to  a  Sigmoid  type,  the  real  world  scenario  was  described  more  accurately.     As  successful  as  their  efforts  were,  there  were  some  shortcomings  in  this  study  that  could  have  improved  the  rigor  and  trueness  of  the  numbers  represented.  First,  attention  is  turned  to  the  data  values  used  from  experiments  A  &  B.  The  physical  printed  graphs  from  the  1973  study  published  by  Luckinbill  were  actually  used  via  digital  analysis,  as  opposed  to  having  the  actual  measured  values  representing  the  true  data.  This  digitization  in  itself  introduced  errors  that  were  not  discussed,  and  could  have  been  avoided  had  the  actual  numbers  been  used,  as  in  the  case  for  experiment  C.  The  discrepancy  in  the  data  values,  though  probably  small,  could  have  effected  the  values  in  sharp  peaks  appearing  in  Luckinbill’s  graphs,  and  therefore  compromising  the  error  estimates  used  when  fitting  the  model.    

Page 9: Paramecium Didiniumjgrant3/cw/mcmfinal.pdf# This#initial#model#is#quite#fundamental#in#nature.##It#is#assumed#that#the#predator'sgrowthrateis# strictlydependentandproportionalto#the#rate#of#prey#consumption.#

  Second,  there  was  some  lack  of  confidence  in  the  values  of  some  parameters  that  were  chosen.  In  particular,  the  prey’s  growth  rate,  

ρ ,  was  left  variable  while  fitting  with  the  S12  error,  whereas  it  was  fixed  to  the  value  of  3.3  when  calculating  the  S22  error.  The  reasons  for  this  decision  were  left  vague,  though  a  greater  emphasis  was  put  onto  the  S12  value  through  the  research.       Finally,  no  details  on  how  population  measurements  were  reported.  Thus,  the  order  of  accuracy  in  the  population  measuring  process  was  not  declared,  leaving  the  “goodness”  of  the  measured  data  unclear.  This  was  considered  an  important  fact  to  skip,  because  the  searching  for  a  better  fitting  model  used  error  analysis  criteria,  which  ended  in  the  S12=30,000  range.  An  open  question  of  “Why  did  we  stop  here?”  is  not  addressed,  leaving  uncertainty  about  the  best  fitting  model  that  was  settled  upon.  Perhaps  there  are  other  considerations  that  would  have  improved  the  mode  even  further,  up  to  the  level  of  accuracy  of  the  errors  incurred  during  the  population  measurement  process.  Providing  more  information  from  this  angle  would  have  strengthened  the  argument  that  the  final  model  chosen  was  indeed  the  best  model  to  decribe  Luckinbill’s  experiments.     As  discussed  by  the  author,  there  is  also  more  room  for  continuing  the  study  to  the  next  level.    It  is  a  good  idea  to  repeat  the  experiment  to  retrieve  better  data  using  measurements  with  more  frequent  time  increments.  The  current  data  used  time  steps  of  0.5  days.  The  initial  conditions  were  also  questioned,  since  well-­‐fed  Didinium  was  used,  and  therefore  delaying  the  anticipated  predator-­‐prey  interaction.  This  new  set  of  experiments  could  also  use  an  attempt  to  achieve  the  stable  conditions  claimed  by  finding  the  right  Cerophyl  and  Methyl  Cellulose  combination  as  predicted  by  the  stability  analysis.  Of  course,  this  research  made  a  final  claim  that  there  is  a  best  fitting  model  to  the  Paramecium  vs.  Didinium  scenario,  which  is  begging  for  new  data  to  conduct  verification.    Reference    Harrison,  G.  W.  1995.  Comparing  Predator-­‐Prey  Models  to  Luckinbill's  Experiment  with  Didinium  and  Paramecium.  Ecology  76:  357-­‐374.