Uta Berger - Institut de recherche pour le développement
Transcript of Uta Berger - Institut de recherche pour le développement
Is our world agent-based and can we model it?
Uta Berger
Faculty of Environmental Sciences, Department of Forest Sciences, Institute of Forest Growth and Computer Sciences
•! Faculty of Environmental Sciences
•! Department of Forest Sciences (Forest Faculty Tharandt)
•! Institute of Forest Growth and Computer Sciences
•! Professorship of Forest Biometry and Systems Analysis
Dresden
Tharandt
Forest Academy Tharandt • 2nd oldest in the world (est. by Heinrich Cotta 2011) • Term “sustainability” established
• ~ 700 students (~20.000 TU Dresden) • Forest Growth And Computer Sciences, Silviculture and
Forest Protection, Forest Botany and Forest Zoology, Soil Sciences, Forest Technics, Forest Economy, Plant Chemistry, ..
1763-184
4
Professorship of Forest Biometry and Systems Analysis !"#$#%&'(!)*+#&#$+!•! !"#$%&'()''#%$,#')#$!%-%,+%./*$*%/&0%*)$1)'!)*+#&%'./*$%+#,%!-#.%('/0%
1+2'/#3%"#%2)/-)3#&$)+4%+#,%$4$+#1$!-4*%1)'$!•! &#$+/&%./*!#'/5/04!)"(4%+%1+/567!+/%8-673%-#%-#0&%-#-!6/&#$+$%•! 6-#6'1(3%-8%'/*$#&3%./*%7#,'/%'(%*0)*0!#*3)&/*1#*+%5!'/*-)./*$!
7#+(/-/5/0)#$!8!9'/5/0)'%5!1/-#55)*0!•! "#,"9",7+!:&+3',%0-,'!!"#.%;<=>?%•! +.'#(:&+3',%0-,'!!"#.%;@=>?%•! 31+5+!%3(+53563%•! A%
:;&&#*+!$4$+#1$!<#!%&#!</&=)*0!/*!•! >+#./-9'3%;=/+B"!*%C-7(4%@8/"6+*%D4+"!+#,*%E"'(#+0?%•! F'#+(7/+5-#%-8%-+$%(/''3%"#%1"#'%,-0"#+(',%8-/'3(3%;G'/0+#H?%•! I'+(%3)+01%8-/'3(3%;<#,-#'3"+?%•! =+/$%&''(!'%"#9+3"-#3%;G'/0+#H?%•! I"#'%8-/'3(3%;CH/"+?%•! J+/0%3H3('03%"#%(4'%@#,'+#%/'."-#%;I'/7?%•! @#5&"-56+%/'3"3('#6'%-8%0"6/-&"+!%6-007#"5'3%;8/'34)+('/*%G'/0+#H?%•! C1"#H%!-&3('/%K34'/H%;L67+,-/?%•! M/"!!%;@#(+/656%N6'+#?%
Is our world agent-based and can we model it?
1.! Basic principles of IBM / ABM (example models: Behavioural Ecology, Plasmid Biology)
2.! IBM / ABM meets Machine Learning (example model: plant ecology) 3.! IBM merges ABM towards agent-based sciences
Faculty of Environmental Sciences, Department of Forest Sciences, Institute of Forest Growth and Computer Sciences
Same objectives •! understanding the functioning of complex
systems •! multiple linkages of social-economic-
ecological processes •! quantitative forecasts of effects of a
changing environment (climate, extreme events, !)
while considering •! variability of agents, organisms (plants
animals), microbes, .. (entities) •! interactions •! behaviour •! adaptation •! decision making
agent-based models (social sciences)
stand simulators (forest sciences)
individual-based models (ecology)
12th Summer School in IBM/ABM at the TU Dresden
agent-based models (social sciences)
stand simulators (forest sciences)
indiv.-based models (ecology)
1st Example Behavioural Ecology
Basic principles of ABM: interac.ons
10 (65)
Basic principles: interactions of organisms
a starling flocks at the sky of Termini (Italy) 11 (65)
!!
"
#
$$
%
&
'
''= (
n n
nseparation
pppp
nF 2
)(1!!
!!!pp
nF
nncohesion!!!
!"#
$%&
'= (
1
cohesion alignment separation
cohesion
alignment
separation
ffn
Fn
nalign
!!!!"#
$%&
'= (
1
alignment
ffn
Fn
nfnfalignFalignF!!!
!"#"#"$"$"%
&%&%'%'%= (
1!!!!!"!"!
#!#!
$$$$$%$%$
&$&$
'
''= (
n n
nseparation
pppp
nFseparationFseparation 2
)(1!!
!!!
separation
ppn
Fn
ncohesionFcohesionF !!!!"#"#"$"$"%
&%&%'%'%= (
1
cohesion
steering rules according Craig Reynolds (1987, boids) http://www.red3d.com/cwr/boids/
pictures: Hanno Hildenbrandt
!!
"
#
$$
%
&
'
''= (
n n
nseparation
pppp
nF 2
)(1!!
!!!pp
nF
nncohesion!!!
!"#
$%&
'= (
1
cohesion alignment separation
cohesion
alignment
separation
ffn
Fn
nalign
!!!!"#
$%&
'= (
1
alignment
ffn
Fn
nfnfalignFalignF!!!
!"#"#"$"$"%
&%&%'%'%= (
1!!!!!"!"!
#!#!
$$$$$%$%$
&$&$
'
''= (
n n
nseparation
pppp
nFseparationFseparation 2
)(1!!
!!!
separation
steering rules according Craig Reynolds (1987, boids) http://www.red3d.com/cwr/boids/
pictures: Hanno Hildenbrandt
!!
"
#
$$
%
&
'
''= (
n n
nseparation
pppp
nF 2
)(1!!
!!!pp
nF
nncohesion!!!
!"#
$%&
'= (
1
cohesion alignment separation
cohesion
alignment
separation
ffn
Fn
nalign
!!!!"#
$%&
'= (
1!!!!!"!"!
#!#!
$$$$$%$%$
&$&$
'
''= (
n n
nseparation
pppp
nFseparationFseparation 2
)(1!!
!!!
separation
ppn
Fn
ncohesionFcohesionF !!!!"#"#"$"$"%
&%&%'%'%= (
1
cohesion
steering rules according Craig Reynolds (1987, boids) http://www.red3d.com/cwr/boids/
pictures: Hanno Hildenbrandt
!!
"
#
$$
%
&
'
''= (
n n
nseparation
pppp
nF 2
)(1!!
!!!pp
nF
nncohesion!!!
!"#
$%&
'= (
1
cohesion alignment separation
cohesion
alignment
separation
ffn
Fn
nalign
!!!!"#
$%&
'= (
1
alignment
ffn
Fn
nfnfalignFalignF!!!
!"#"#"$"$"%
&%&%'%'%= (
1ppn
Fn
ncohesionFcohesionF !!!!"#"#"$"$"%
&%&%'%'%= (
1
cohesion
steering rules according Craig Reynolds (1987, boids) http://www.red3d.com/cwr/boids/
pictures: Hanno Hildenbrandt
screenshots of an ABM developed by Hanno Hildenbrandt
(Uni Groningen, Netherlands)
http://news.nationalgeographic.com/news/2009/01/photogalleries/locust-swarm-theory-serotonin/photo5.html
Locust Swarms - Subsahara
DeAngelis D and Zhang B: F1000Prime Recommendation of [Dkhili J et al., Ecol Modell 2017, 361:26-40].
2nd Example Antibiotica Resistance
(Plasmid Biology)
Basic principles of ABM: interac5on & variability
Vertical component (cell fission) persistence from one generation to the next
all figures: Wikimedia commons
Horizontal component (conjugation) persistence from one generation to the next
non transmissible transmissible •! repressed •! derepressed
Theory: replication and plasmid burden lead to a loss without positive selection.
Reality
Concept for the persistence of plasmids without positive selection.
0.0
0.2
0.4
0.6
0.8
1.0
vert
ical
tran
smis
sion
fitn
ess
1!"
0.0 0.2 0.4 0.6 0.8 1.0
horizontal transmission fitness#
plasmid persistence requiresbeneficial plasmid genes
" > #
plasmids potentially persistin the absence of selection
" < #
"= # isoline
compensatoryadaptation
A
P
Zwanzig et al. in prep.
Wik
imed
ia c
omm
ons
Wikimedia commons
Werisch et al. (2017)
Wer
isch
et a
l. 20
17
Zwanzig et al. submitted
Genetic Inspiration: Human Genome Project (MIT, early 2000) Computer Science: Bioinformatics, Data Mining, Simulation Experiments (last decade) Material Science: currently
3rd Example Plant Ecology
ABM meets Machine Learning
Competition
Symmetric
Asymmetric
Lin et al. 2013
Symmetric
benign harsh
Asymmetric
Facilitation
stress gradient
Schwinning & Weiner 1998, Lin et al. 2012, 2014
Plant-Interaction (PI) Modell
!E!!!!!!$!!!!!!!!!!!!!!!!!!F
dm/dt = fpfqcA[ 1 – ( m / M)1/4 ] m .. Current biomass M .. Maximum biomass C .. Initial growth rate A .. Area within the plant exploits resources fp .. competition among neighbouring plants fq … facilitation among neighbouring plants (in case of stress)
f p = (Ano +vimi
p
v jmjp
j=1
n j!k=1
no! Ao,k ) / A
fq =1! SAf +1
=1! S
(1! wimiq
wjmjq
j=1
n j"k=1
no" )Ao,k +1
Ano area of plant i without overlap Ao,k overlapping area with neighbouring plant j p, q parameter of symmetry – asymmetry !, " species-specific weightS stress level
The two-layer version of the PI model including both above- and belowground compartments connected via metabolic constraints.
!m!t
=
2"!AGR = 2" fR,a f p,aca Aa[1# (m / Ma )1/4 ], !AGR < !BGR
2"!BGR = 2" fR,b f p,bcbAb[1# (m / Mb )1/4 ], !AGR > !BGR
!AGR+!BGR, !AGR = !BGR
$
%
&&
'
&&
!AGR, !BGR .. above- and belowground growth rate a, b .. above- and belowground compartment
May et al. (2009), Lin et al. (2013)
Lin
et a
l. (2
013)
Plant-Interaction (PI) Modell
”I have a dream …”
Observations above-ground Dynamic (Mortality, Spatial distribution …)
.. can be used to reveal mechanisms below-ground (competition mode, resource limitation ..)
Dynamic observations at stand level: •! Mortality •! Spatial distribution (Clark Evans Index)
Mechanisms to be detected: •! Below-ground competition mode •! Resource limitation
The self-organizing map (SOM = Kohonen layer) links the factors without defining dependencies.
Peters & Berger (2016)
Machine-Learning
Peters & Berger (2016)
SOM training SOM layer after training
Peters & Berger (2016)
Peters & Berger (2016)
Peters & Berger (2016)
IBM / ABM
towards agent-‐based science?
Full-fledged
Discrete Individuals
Local Interactions Finite
numbers of individuals
Variation among Individuals
Moderate life history structure and behaviors
Complex life histories including memory and other cumulatively changing internal variables
Adaptive behavior
Age , size, stage structure models; pde and matrix
Spatial Models; e.g., pdes, metapopulation models
Classical Progenitors
Level 1 Level 2 Level 3 Level 4 Level 5
Evolutionary change
Physiology
See also: DeAngelis & Mooij (2005). Ann. Rev. Ecol. Syst. 36 42 (30) IBM
agent-based models (social sciences)
stand simulators (forest sciences)
indiv.-based models (ecology)
Vinc
enot
(201
8) P
roc.
R. S
oc. B
285
: 201
7236
0
Vinc
enot
(201
8) P
roc.
R. S
oc. B
285
: 201
7236
0
Conclusions • tendency to full-fledged models (mechanistic, physiology
based) • IBM / ABM toolbox (standardization, ODD, TRACE) • merging with formerly separated modelling techniques • ready to go towards agent-based sciences
Acknowledgements • Cyril Piou (CIRAD, UMR CBGP Montpellier, France) • Vanlerberghe Flavie (CIRAD, UMR CBGP Montpellier, France) • Jamila Dhkili (Université Ibn Zohr, Agadir, Morocco) • Martin Zwanzig (alias Werisch, TU Dresden, Germany) • Yue Lin (Scion, Rotura, New Zealand) • Ronny Peters (TU Dresden, Germany) • Hanno Hildenbrandt (University of Groningen, The Netherlands) • Christian Vincenot (Kyoto University, Japan)
https://tu-dresden.de/bu/umwelt/forst/ww/bsa/forschung/projekte?set_language=en