Methodological aspects of agent-‐based modeling and simula5on of complex systems ! !
Benoit Gaudou
step 1: define the modeling questions
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 4: define the agents (characteristics, dynamics)
UMR 5505 IRIT, CNRS / Université de Toulouse -‐ benoit.gaudou@ut-‐capitole.fr
1
«A model is a simplified representa5on of a reference system that helps to answer a ques5on on this system.» [Drogoul]
A map is a static model. Here of urban area of Can Tho (Vietnam) in 2010. [Konings, 2012] [Can Tho on google maps]
Meteorologic dynamic models allow to predict a typhoon evolution over 48h. Neoguri t yphoon, J uly 2 014 http://lesbrindherbes.org/2014/07/06/l-enorme-typhon-neoguri-se-dirige-japon/
2
A simula5on is an execu5on of a computer dynamic model.
Simulation can be used for: Validation, Evaluation, Verification Simulation aim: test or verify an hypothesis on the reference system; validate a theory. ! Understanding, Exploration
Simulation aim: rebuild the system to understand it, producing an object that we can experiment ! Communication, Training, Visualization
Simulation aim: «show» and share the model of the reference system dynamic. ! Control, Action, Decision-‐Support System
Simulation aim: simulation as Decision-‐Support System to a policy or actors’ decisions (that could be assessed on the simulation) ! Prevision, Prediction, Anticipation
Simulation aim: predict possible evolutions of the reference system, given specific Alexis Drogoul perturbations.
3
Three model examples will be presented today. Understanding and exploring urban spatial dynamics Case study: Can Tho (Vietnam)
!
Assessment of the social, economic and environmental impacts of the various alternative of definition and management of (new) water Volume Available for Agriculture? Case study: Adour-‐Garonne basin (France)
!
Reproducing and exploring past events using agent-‐based geo-‐historical models Case study: floods of 1926, Hanoi (Vietnam) !
4
Current changes in worldwide urbaniza5on paFerns at all geographic scales are quite likely the most significant since the advent of humanity !
Coloniza5on of remote rural and wild-‐land spaces; !
Emergence of new types of urban specializa5ons (e.g., technopoles); Increasing spa5al segrega5on of social groups; !
Spa5al fragmenta5on of labor markets; !
Increasing threats from natural hazards as urban structures become more dispersed, resource-‐demanding and complex; !
Exposure to new kinds of vulnerabili5es given their increasing dependence on technology;
Slide by Alexis Drogoul
5
To address the challenges raised by these changes, there is a need for tools and models that can support deciders in: !
Providing knowledge on and understanding the dynamics of the urban system, An5cipa5ng and forecas5ng future changes or trends of development, Describing and assessing impacts of future development, Exploring different policies and op5mizing planning and management of sustainable urban areas. !
Focus on designing and exploiDng urban morphogenesis (or growth) models, i.e. models that represent the evoluDon(s) of the spaDal inscripDon of the city. Slide by Alexis Drogoul
6
The data requirements for designing urban models are intense due to the complexity of their object and the variety of their objec5ves
Satellite & aerial images -‐ Surveys -‐ Census -‐ Field studies -‐ Sociological studies
Topography, hydrography Economical data
Slide by Alexis Drogoul
Transport infrastructures
Environmental data
Social data
Land cover, land use data
7
Example: Can Tho city (Vietnamese Mekong Delta)
Slide by Alexis Drogoul
8
Why Can Tho ? It is the largest and fastest growing city in the Mekong Delta. Its evoluDon underlines a number of the challenges raised by urban growth. A comprehensive dataset on the evoluDon of the city shape has been gathered and built by researchers and pracDDoners.
Existing settlem
Strategic locatio
! s
!
1750
1950
2010 Can Tho,
Source: Can Tho, How to Grow ?,Vera Konings,
Master thesis, 28th June 2012, Univ. of Delft, the Netherlands
Slide by Alexis Drogoul
9
Modeling is a mul5-‐step cycle.
step 1: define the modeling questions
step 7: explore the model Cycle proposed by Patrick Taillandier
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
«Every model has to start from a clear ques5on, problem, or hypothesis.» [Grimm, et al. 2010]
step 1: define the modeling questions
step 7: explore the model
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
Which ques5ons for which models ? Among dynamic models we can distinguish: -‐ Reproduction models: that try, given data, to reproduce as precisely as possible these data by simulation. -‐-‐-‐> we cannot understand the way the phenomenon is generated.
But it could «predict the future», e.g. weather forecast. !
-‐ Understanding models: that try to understand processes that generate phenomena. They are focused on the processes and try to get the phenomenon by simulation. -‐-‐-‐> allow to understand, explain and generate the phenomenon. Consequence: we can « ask questions to the model» via What-‐If scenario. !
We cannot tackle the same questions with all kinds of models.
12
We focus on a par5cular phenomenon: the urban growth of Can Tho.
[Konings, 2012]
We have chosen to build an understanding model: -‐ we will look at mechanisms able to reproduce the urban growth !
A reason is the possibility to do urban planning prospection using this kind of models.
13
Examples of ques5ons that can be asked to the Can Tho urban model growth. ! !
Influence of the transport network on urban growth: Construction or destruction of a road Improvement of the transport network (e.g. larger roads... ) Influence of the network on the ratio residential, services buildings. !
Influence of the building of a shopping center, or other services. !
Influence of the urban growth on the traffic and on public transport. !
What services (hospital, school, water network...) are needed to add taking into account urban growth ?
14
Modeling is a mul5-‐step cycle.
step 1: define the modeling questions
step 7: explore the model
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
There is a huge number of languages to write a dynamic model. Consider 2 main approaches: -‐ equation-‐based models: describe the evolution at macroscopic level !
-‐ agent-‐based models: describe microscopic entities, their behaviors and interactions between these entities and generate a macroscopic behavior
LWR macroscopic model
NaSch microscopic model
!
Example: traffic !
We focus on agent-‐based model here. Example of epidemic equation-based model 16
Example: macro-‐simula5on model of traffic
Based on an analogy between the traffic flow and the fluid flow in a pipe.
The vehicle flow is characterized by macroscopic attributes. Individual behaviors are averaged in a global equation.
17
Example: agent-‐based simula5on model of traffic
Behaviors (rules) are associated to individual vehicles.
Allows a variety of behaviors and stochasticity. Analytical results "impossible" to get: obligation to go through the simulation.
18
En55es and dynamics
!
«An entity is a distinct, separate object or actor that behaves as a unit and may interact with other entities or be affected by external environmental factors. »
Ferber 89
!
«Its current state is characterized by its state variables or attributes. » !
Processes/dynamics are change factors in the system. They can be performed by the entities or have an environmental source (e.g. weather ... ) ! ! !
Definition by [Grimm et al, 2010] 19
Model 1: en55es, dynamic and scales •Spatial scale •Area that covers Can Tho city and the nearby suburbs !
•Time scale •Duration : 15 years from 1999 to 2014
MeKong Delta
! !
70 km
•Entities to take into account •Buildings •Roads •Rivers
!
•Dynamic to take into account •The construction of buildings
45 km
Considered area
20
Model 2: en55es, dynamic and scales •Spatial scale •The An Binh ward of Can Tho !
•Time scale •Duration : 5 years from 2005 to 2010 ! ! !
Can Tho 3 km
•Entities to take into account •Buildings •Roads •Rivers
!
•Dynamic to take into account
3 km
•The construction of buildings Considered area
21
Modeling is a mul5-‐step cycle.
step 1: define the modeling questions
step 7: explore the model
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
Data collec5on, cleaning and management are a huge and hard work in a modeling and simula5on project! First: Get data -‐ collect yourself, -‐ ask/buy from government agency, -‐ get open source data (Open Data), e.g. OpenStreetMap, IGN (France) -‐> unequal quality
co-‐evolution in the MAELIA Project between question/data/ representation
Data
Data’
Data’’
!
E.g. : to get data from Can Tho, we need to ask the Department for Natural Ressources and Environment (Sở tài nguyên môi trường Cần Thơ)
Representation
Representation’
Stable Representation
Question’
Question’’
!
E.g. : in France, IGN, INSEE, Water Agency...
Initial question
Then: process and clean them. 23
Data need to be cleaned and processed, depending on the aim and need of the model. Aim: building of a common GIS merging data from all the various sources. !
Example on the MAELIA project of data integration.
Example of issues to tackle: -‐ convert all the data formats in a common one -‐ clean data (e.g. correct invalid polygons, overlapping or disjoints polygons...) -‐ transform data to be used in the model (e.g. , switch from polygons to polylines for roads...) -‐ correct/create attribute table for each geographic objects -‐ integrate / cross various data sources 24
Modeling is a mul5-‐step cycle.
step 1: define the modeling questions
step 7: explore the model
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
Given the chosen en55es and the data collected, we have to choose what will be an agent in our model (i.e. an en5ty with a behavior) What will be an agent depends mainly on the chosen scale (and data & question). -‐ If we want to simulate a city, do we need to take into account each inhabitant, each household, each building ... ?
Can Tho - Ninh Kieu - 2010
!
First attempt: -‐ as we are interested in the new buildings, we can chose to make buildings as agents. -‐ They have a shape, location and type as attributes. !
We have concrete agents:
1 agent = 1 physical entity !
Dynamics should then be described at the level of the building.
! This choice leads to a vector model 26
Given the chosen en55es and the data collected, we have to choose what will be an agent in our model (i.e. an en5ty with a behavior) The previous choice is relevant to work at the district scale (not a too high number of buildings). To deal with the whole city, we have to switch to something else...
Can Tho 1999
!
Second attempt: -‐ we choose to discretize to whole space on a regular grid. Agents will be the cells. -‐ they are characterized by a rate of building on the cell area. !
We have abstract agents:
1 agent = 1 entity aggregating buildings. !
We can choose the number of cells, so the discretization rate, and so the precision.
! This choice leads to a raster model. 27
The UML language is commonly used as a modeling tool to describe the kinds of agents, their aFributes and opera5ons. UML is a graphical language allowing to specify, visualize, build and document elements of a system or a model. !
We focus only on class diagrams = the part of UML allowing us to represent the types of agents (species), their attributes and relations between these types. !
Example from the raster model: for agents cell, with it set of attributes !
Name of species attributes
operations 28
Determina5on of the processes that can generate the phenomenon. Introduc5on example. Which processes/rules an we introduce to reproduce the phenomenon ? !
E.g. : What will be x? and y? ?
1
2
3
5
8
13
x?
y?
29
Determina5on of the processes that can generate the phenomenon. Introduc5on example. Which processes/rules an we introduce to !reproduce the phenomenon ? !
!
E.g. : What will be x? and y? ?
!
!
!
!
!
!
!
! !
1
2
3
5
!
8
13
x?
y?
!
By observing the number series, we can observe ! a regularity: -‐ 3 = 1 + 2 ! -‐ 5 = 2 + 3 ! -‐ 8 = 3 + 5 So we can predict next values: -‐ 13 = 5 + 8 -‐ x? = 8 + 13 = 21 -‐ y? = 13 + x? = 13 +21 = 34
30
Similarly, to generate the urban growth, you can look at this picture and propose rule(s) to locate the next buildings...
[Konings, 2012]
31
Modeling is a mul5-‐step cycle.
step 1: define the modeling questions
step 7: explore the model
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
Models can be implemented and simulated using any generic programming language, but also with some dedicated (open-‐source) pla]orms. ! NetLogo http://ccl.northwestern.edu/netlogo/ Pedagogical tool with a specific modeling language. Numerous examples in various domains. Ideal for beginners and prototypes, but somehow limited for large models.
! ! Repast Simphony http://repast.sourceforge.net/ Toolbox for agent-‐based modeling in Java, numerous libraries of connected technologies (statistical analysis, GIS, 3D models, etc.). Very powerful but requires very good programming skills.
! ! GAMA http://gama-‐platform.googlecode.com A compromise between the two platforms above. Using a dedicated modeling language, this platform pays a great deal of attention to the modeling of the environment and the use of GIS data, while staying open to new technologies through plug-‐ins.
33
Highlight on the GAMA pla]orm
for its management of GIS data
for its user-‐friendly and powerful IDE ! for its modeling language
https://code.google.com/p/gama-‐platform/
34
Modeling is a mul5-‐step cycle.
step 1: define the modeling questions
step 7: explore the model
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
A model is an object that can be manipulated or explored via simula5on and on which we can launch experiments. Parameters
w1
between 0 and 1 w2
between 0 and 1 w3
between 0 and 1
Outputs
Experiments
Model Inputs
Data
36
O`en there are at least 2 ways to run simula5ons.
! !
GUI experiment: Gui experiments allow the modeler to observe indicators evolution during a simulation !
batch experiment: batch experiments allow model to observe the evolution of indicators of the simulation (computed at the end of the simulation) over the change of parameters.
37
GUI experiments allow to test various parameters values and observe the output.... which can be long and fas5dious at hand ... w1 = 0.2
w2 = 0.0
w3 = 1.0
Experiments
Error = 0.2869
w1 = 0.7
w2 = 0.5
w3 = 0.2
Experiments
w1 = 0.5
w2 = 0.3
w3 = 0.8
Experiments
Error = 0.5093
Error = 0.4344
......
38
The batch mode allows to automatically execute parameter space exploration.
Exploring the parameters space of a model can be used to: •
Correct the model,
•
Do sensibility analysis,
•
Calibrate it,
•
Discover emerging properties,
•
…
A necessary step!
Explore the model behavior through the explora5on of the parameters space. Aim: to have a deep knowledge of the behavior of the model (and in particular its sensitivity to parameters variations) ==> the range of values to which the simulation is very sensitive -‐> discretization need (step= 1, 0.1 or 0.0001 ...)
Mean size of aggregates depending on the 2 parameters:
- number of different neighbors accepted
- population density
(Daudé E. & Langlois P. 2006)
!
Example on the Schelling’s model: !
Population
density
! ! ! ! !
Number of different neighbors that can be accepted by an agent
«Exhaustive» exploration of the Schelling’s model 40
The calibra5on allows to find parameters values making the simula5on results fiang to real data. Objective: What are the values of parameters:
-‐ that allow to obtain a «realist» model ?
-‐ that minimize the difference between real and simulated data ? Real data (at the time of the simulation end) error = difference between simulated data and real data parameter1 = ??
p2 = ??
p3 = ??
Real data at initial time
Experim
Simulated data at the end of the simulation
41
Issues linked to the calibra5on. Find a good indicator (often named fitness) to represent the fact that a simulation is «good» (e.g. the difference between real and simulated data): -‐ in the case of urban growth model, this indicator should represent the distance between real and simulated maps.
Population
density
Number of different neighbors that can be accepted by an agent
!
To find the parameters is closed to an optimization problem (minimize a value): -‐ lot of methods exist (exhaustive, Genetic algorithms, ... )
http://nickmalleson.co.uk/wp-content/uploads/2013/04/parameter_space.png
42
Once the model has been calibrated, we can explore the result of the simula5on to various scenario. We can define scenario that we want to explore: -‐ building of a new road, of new houses, services...
What-If water channels
are improved ?
?????????? 43
Steps in an experiment design.
General method : •
Define an initial situation and a stop criterion (of 1 simulation)
•
Select the set of parameters to explore and the indicators
•
Define a strategy of exploration
•
Execute, collect and save the data
•
Analyze the collected data
Can Tho Case study • Objective: reproduce the urban growth of the agglomeration of Can Tho from 1999 to 2014 Empty space Water space Built space Road City center Can Tho 1999
For each free cell (empty space): evaluation of their constructability value based on 3 indicators:
-‐ density (C1): ! -‐ distance to road (C2): ! -‐ distance to city center (C3):
45
Can Tho Case study: first simula5on with only the distance to road criterion
Construction along the roads 46
Can Tho Case study: first simula5on with only the distance to city center criterion
Expansion of the city center 47
Can Tho Case study: first simula5on with only the (low) density criterion
Lot of small «building blocks» covering uniformly the space C1 : density = 1 - (number of built neighbor cells / number of neighbor cells)
48
Can Tho Case study: first simula5on with only the (high) density criterion
Expansion from high density areas C1 : density = (number of built neighbor cells / number of neighbor cells)
49
Can Tho Case study: calibra5on of the model to reproduce the real urban growth
Empty space Water space Built space Road
How to evaluate the error between the simulation and the observed data?
Use of the Fuzzy Kappa Sim indicator 50
Modeling is a mul5-‐step cycle.
step 1: define the modeling questions
step 7: explore the model
step 2: identify the elements (entities, dynamics) to model step 3: collect data
step 6: calibrate the model
step 4: define the agents (characteristics, dynamics)
step 5: implement the model
We can also use the model to assess What-‐If? scenario on the model.
! !
From an initial state
What-If no change is done
in terms of urban planning ?
??????????
Modeling can be useful: -‐ to build Decision-‐ Support Systems (DSS) -‐> to help urban planning. [Konings, 2012]
What-If water channels
are improved ?
??????????
52
References [Konings, 2012] Konings V., « Can Tho, how to grow? Flood proof expansion in rapidly urbanising delta cities in the Mekong delta: the case of Can Tho.», Master thesis, University of Delft, 2012.
http://repository.tudelft.nl/view/ir/uuid%3A84c58546-‐e387-‐44f7-‐b8ff-‐0958b018ae83/ !
[Grimm et al., 2010] Volker Grimm, Uta Berger, Donald L. DeAngelis, J. Gary Polhill, Jarl Giske, Steven F. Railsback. «The ODD protocol: A review and first update», Ecological Modelling 221, 2760–2768, 2010.
53
References: Web Some websites: ! JASSS : Journal of Artificial Societies and Social Simulation http://jasss.soc.surrey.ac.uk/
! ACE : Agent-‐based Computational Economics http://www.econ.iastate.edu/tesfatsi/ace.htm
! GisAgents : GIS and Agent-‐Based Modelling http://www.gisagents.blogspot.com/
! OpenABM: Open Agent-‐Based Modeling Consortium http://www.openabm.org/
!
Slide by Alexis Drogoul
54
References: some books Four books: ! GILBERT N and TROITZSCH K. G. (2005) Simulation for the Social Scientist. Milton Keynes: Open University Press, Second Edition. GILBERT N (2006) Agent-‐based Models. Series: Quantitative applications in the Social Sciences. Sage Publications. AMBLARD F et PHAN D (2007) Modélisation et Simulation multi-‐agents: applications pour les Sciences de l’Homme et de la Société. Hermès Editions. TREUIL JP, DROGOUL A et ZUCKER JD (2008) Modélisation et Simulation à base d’agents. Dunod & IRD Editions.
Slide by Alexis Drogoul
55
Les ou5ls pour la modélisa5on et simula5on de systèmes complexes ! !
Benoit Gaudou
UMR 5505 IRIT, CNRS / Université de Toulouse -‐ benoit.gaudou@ut-‐capitole.fr
56
Etapes de modélisa5on
Etape 1: définir les questions de modélisation
Etape 7: explorer le modèle
Etape 2: identifier les éléments (entités, dynamiques) à modéliser Etape 3: collecter les données
Etape 6: calibrer le modèle
Etape 4: définir les agents (caractéristiques, dynamiques)
Etape 5: implémenter le modèle 57
Etapes de modélisa5on Etape 2: identifier les UML... Approches participatives
éléments (entités, Etape 4: définir les dynamiques) à modéliser Etape 1: définir les Méthode ARDI agents questions de (caractéristiques, modélisation dynamiques) EtapeGénération 3: population
Manipulation SIG
collecter les OpenJUMP
données synthétique
QuantumGIS
Etape 7: explorer le modèle
Etape 6: calibrer le modèle
YANG
Etape 5: implémenter le modèle 58
Exemple de modélisa5on par5cipa5ve
Serge Bobo Kadiri, Dschang University, Cameroon {
[email protected]}! Christophe Le Page, CIRAD Montpellier, France {
[email protected]}
59
Exemple de modélisa5on par5cipa5ve
60
La méthode ARDI Une méthodologie de co-‐design d’un modèle conceptuel des interactions socio-‐environmentales dans un système socio-‐écologique. !
Actors! Resources! Dynamics! Interactions
De l’identification d’un problème… ... à une question ! !
Inspiré des slides de Christophe LE PAGE, MISS-‐ABMS, Montpellier, September 2013
61
Co-‐construc5on d’une représenta5on partagée en 4 étapes A = Acteurs/ intervenants R = Ressources D = Dynamiques/ processus I = Interactions !
Chaque étape correspond à un type particulier de question. !
1. Quels sont les acteurs principaux qui interagissent avec la rivière et son débit ? 2. Quelles sont les ressources en relation avec le débit ? 3. Quels sont les principaux processus intervenant dans the bassin Adour-‐Garonne et qui affecte le débit en eau ? 4. Comment chaque acteur utilise la ressource ? 62
Etapes de modélisa5on Etape 2: identifier les UML... Approches participatives
éléments (entités, Etape 4: définir les dynamiques) à modéliser Etape 1: définir les Méthode ARDI agents questions de (caractéristiques, modélisation dynamiques) EtapeGénération 3: population
Manipulation SIG
collecter les OpenJUMP
données synthétique
QuantumGIS
Etape 7: explorer le modèle
Etape 6: calibrer le modèle
YANG
Etape 5: implémenter le modèle 63
Yang : Yet Another Network Generator YANG (Yet Another Network Generator) est un outil dédié à la reconstruction de réseaux plausibles à partir de règles locales. !
Il a été appliqué à la reconstruction de réseaux sociaux au Kenya, de réseaux sociaux des solidarités économiques sur l’île de Mayotte et d’un réseau neuronal à partir d’études neuro-‐ biologiques.
Yang : http://sourceforge.net/projects/yang-‐j/
64
Yang : exemple de réseau produit sur l’exemple du Kenya
65
Etapes de modélisa5on Etape 2: identifier les UML... Approches participatives
éléments (entités, Etape 4: définir les dynamiques) à modéliser Etape 1: définir les Méthode ARDI agents questions de (caractéristiques, modélisation dynamiques) EtapeGénération 3: population
Manipulation SIG
collecter les OpenJUMP
données synthétique
QuantumGIS
Etape 7: explorer le modèle
Etape 6: calibrer le modèle
YANG
Etape 5: implémenter le modèle 66
Références
•GAMA : https://code.google.com/p/gama-‐platform/ •Netlogo : http://ccl.northwestern.edu/netlogo/ •Cormas: http://cormas.cirad.fr/ !
•Méthode ARDI: http://cormas.cirad.fr/pdf/guideARDI.pdf !
•Yang : http://sourceforge.net/projects/yang-‐j/ !
•SIG: OpenJUMP : http://www.openjump.org/ •SIG : Quantum GIS : http://www.qgis.org/ !
•Gephi : http://gephi.org/ •OpenMole : http://www.openmole.org/ 67
ARDI : références Etienne M., 2009. Co-‐construction d'un modèle d'accompagnement selon la méthode ARDI : guide méthodologique. Cardère éditeur, Laudun, 71 p. !
Etienne M., Du Toit D., Pollard S., 2011. ARDI: a co-‐construction method for participatory modelling in natural resources management. Ecology and Society 16(1): 44. [online] URL: http://www.ecologyandsociety.org/vol16/iss1/art44/ ! !
•Méthode ARDI:
http://cormas.cirad.fr/pdf/guideARDI.pdf 68
Reproducing and exploring
past events using
agent-‐based geo-‐historical models Nasser Gasmi(a,c,e), Arnaud Grignard(a), Alexis Drogoul(a,d), Benoit Gaudou(b),
Patrick Taillandier(c), Olivier Tessier(e) & Vo Duc An(a) Prelim inary proof of con cept of the ARCH IVES projec t
(a) UMI 209 UMMISCO, IRD / UPMC
(b) UMR 5505 IRIT, CNRS / University of Toulouse
(c) UMR 6266 IDEES, CNRS / University of Rouen
(d) JEAI DREAM, IRD / University of Can Tho
(e) EFEO, French School of Asian Studies
69
The analysis and transmission of past disasters is an integral part of disaster management The experience of past disasters allows local knowledge to be used to develop community responses that both help to raise awareness of risks and also help prepare for improved future disaster response and reconstrucDon
Prevention! • Land use planning ! • Learning from events! • Technical measures
Inspired by Integral Risk Management Cycle, FOCP 2012
70
However, being able to learn from a past disaster event requires that 3 different issues are solved Issue 1: The availability and accessibility of the data concerning this event Issue 2: The construcDon of relevant informaDon from these data Issue 3: The reconstrucDon of coherent «stories» from these informaDon ! ! ! ! ! ! ! ! !
This is what historians do, but it would be helpful to do it in a more systema5c way as this potenDally concerns hundreds of thousands of past disasters.
71
The quan5ty of digital informa5on about past risk events is strongly dependent on when in history they have happened
Digi5za5on of physical documents
Produc5on of digital documents
Future
Past 500
1000
1500 1600 1700 1800
1900
2000
today
72
A first step is to make more informa5on available through the exploita5on and automated analysis of available digi5zed contents Ins5tu5onal analysis
Social network analysis
(Web)mapping Analysis of digital informa5on Digi5za5on of physical documents Produc5on of digital documents
Future
Past 500
1000
1500 1600 1700 1800
1900
2000
today
73
Geohistorical modeling is one solu5on, which extrapolates the digital informa5on available in order to produce new digital informa5on through simula5on
Geohistorical modeling is a way to tell «stories» and to make these «stories» accessible to anyone. Simula5on of digital models Analysis of digital informa5on Digi5za5on of physical documents Produc5on of digital documents
Past
Future 74
We focus on one specific event: July 1926 floods in and around Hanoi and its management by authori5es.
Context
Method
Model
75
1
Context
76
Context: July 1926 floods in Hanoi (Vietnam, Red River Delta) Map of dykes (1905)
77 http://www.icem.com.au/maps/high_res/3d_slope.jpg
The loca5on: Hanoi in 1926
Contour lines (brown)! Lakes (blue)!
Buildings(red)!
Red River (blue)!
6"
78
The event: floods and dikes break in Summer 1926 Chronology:
! - 25th to 30th of July: increase of water level and main dyke breaches
Breach at Gia Quất
Dykes Breaches
28th, evening (old dyke)
29th, at 9 AM (new dyke)
! - 31th of July to November: plugging of dykes
Hà Nội downtown
Breach at Ái Mộ
29th, at 4 PM
Breach at Lâm Du
29th, between 4 PM and 5PM
79
2
Method
80
Source: inspired by David Sheeren’s presentation at MAPS5
Issue 1: Building of a comprehensive and navigable corpus from the heterogeneous set of archives. v First: delimitation of the system: Ø event: floods of Red River in July 1926 Ø study area: Gia Lâm (district of Hanoi)
v Identification of sources:
Contour lines (brown)!
Ø The NaDonal Archives Center #1 (Ha Nôi, Viêt Nam) Ø The French mapping agency -‐ IGN (Paris, France), Ø The French School of Asian Studies -‐ EFEO (Ha Nôi, Viêt Nam).
Lakes (blue)!
Buildings(red)!
Red River (blue)!
6"
81
Issue 2: Representa5on of this corpus in its mul5ple (spa5al, temporal, social) dimensions so that it can be manipulated and explored freely by users. v to create accurate GIS data from the disparate maps available in the corpus v to integrate temporal informaDon and to offer the same querying and navigaDon faciliDes in Dme than the ones exisDng for 2D and 3D spaDal data.
82
Issue 3: The reconstruc5on of coherent «stories» from these informa5on Building of models of the reality depicted by the corpus so that they can offer an account of what happened but also become the objects of experiments: Ø hydrological sub-‐model, Ø crisis management sub-‐model.
Experimental approach based on hypotheDcal reasoning («what would have happened if ...», «what effect this decision could have had on ...»)...
https://code.google.com/p/gama-platform/ 83
3
Modeling spa5o-‐temporal dynamics
of the event
84
Hydrological model: collect data v Data available: Ø Digital Elevation Model (DEM) Ø Shapefile of the dykes Ø Shapefile of the buildings Ø Shapefile of the Red river Ø Shapefile of the lakes
Hydrological model: floods as diffusion on a regular grid v Advantages Ø Simple to understand and to implement Ø Require little data Ø Allows simple integration of shapefiles and DEM data
v Drawbacks Ø The Digital Elevation model plays a key role (high resolution -‐> time consuming, low resolution -‐> imprecise)
v Calibrations using historical data
water height of the highest dykes/ buildings located
altitude
height
The crisis management: from textual documents to agent-‐based model
Hierarchical organization
Set of documents
from Archives
87
The crisis management sub-‐model Given:
-‐ the hydraulic model
-‐ the description of the relationship between actors !
Aim of the model Introduction of:
-‐ actors’ actions against floods
-‐ how orders and information are transmitted through the actor network !
Question
How orders and information transmissions induce the building of small dykes ? 88
2 separated models linked by dykes
89
Build s mall d Simplifica5ons ykes!
We ca nnot d o anyt >
-‐ we focus on the reaction of actors
to the floods in first days
!
On actions:
-‐ as physical actions, we focus only on the building of small dykes.
-‐ we do not consider actors that do not follow orders. !
On dynamics:
-‐ top-‐down order chain (orders to local authorities to builds small dyke)
-‐ bottom-‐up information chain (local authorities transmitting information of help need) Technically: use of FIPA-‐ACL to manage communication between agents
90
Issue 3: The reconstruc5on of coherent «stories» from these informa5on Model integraDng: Ø hydrological sub-‐model Ø crisis management sub-‐model
Experimental approach based on hypotheDcal reasoning («what would have happened if ...», «what effect this decision could have had on ...»)...
https://code.google.com/p/gama-platform/ 91
MABS 2014, Paris, France, May 5th-‐6th, 2014.
Conclusion
92
Conclusion: first achievements and perspec5ves v Still a preliminary proof of concept. v What we have: Ø A huge dataset (maps, reports, ... ) about a parDcular case Ø A first model that has everything to reproduce the crisis and its management (during first days) Ø A very interesDng case study (from a social point of view)
!
v Main perspective: automatic generation of actors and their behavior from textual documents (e.g.: using process-‐mining tools, SNA...) v Main «results»: Ø «Formal» model of the organisation (which lets appear difference between theoric and actual organization) Ø a reflexion with Historian about the role and the interest of simulaDon for historical research and Digital HumaniDes in general.
Ø the interest to not reproduce the actual event (e.g. not the same breach locaDon) in order to be able to test hypotheses
93
Floods in Hanoi (2008) ...
94
The MAELIA project: ! A multi-‐agent platform t o d eal w ith ! water scarcity problems Frédéric Amblard, Yves Auda, Jean-‐Paul Arcangeli, Maud Balestrat, Marie-‐Hélène Charron-‐Moirez, Benoit Gaudou, EDenne Gondet, Yi Hong, Romain Lardy, Thomas Louail, Eunate Mayor, Pierre Mazzega, Clément Murgue, David Panzoli, Sabine Sauvage, José-‐Miguel Sanchez-‐Perez, Christophe SiberDn-‐Blanc, Patrick Taillandier, Olivier Therond, Nguyen Van Bai, Maroussia Vavasseur
95
Context
96
The Adour-‐Garonne Basin: a structural water deficit
Hydroelectricity production Storage capacity
Irrigated area
Fernandez et Verdier Since 1950-‐60 development of water storage in dams for hydroelectricity : water releases in winter !
Since 1980 development of important irrigated areas: agricultural water consumption in summer is up to 80 % of total consumption during the low-‐flow period !
Current water storage capacity is inadequate to meet temporal and spatial distributions of water needs
97
Water crises During low-‐flow periods in some watersheds, river flows are regularly measured under the regulatory threshold (DOE) which guarantees a normal functioning of the aquatic ecosystems
Watersheds with recurrent river flow < the regulatory thresh
During low water period, irrigation = 75% of water
The ways to manage water • Low water period management – Water releases from dams – Withdrawal restrictions (issue of drought decree) • Long-‐term policies – Building of new resources – Withdrawal regulations
Evolution of the water management gouvernance !
Since 2009, to avoid regular water crisis French government is instituting a new regulation of agricultural water withdrawals Determination of new: -‐ water Volumes Available for Agriculture (VAA) Yellow watersheds: VAA < water withdrawn in d at watershed level Red watersheds: VAA vehement protests of farmers
Main ques5ons to be handled with the MAELIA pla]orm
What are the social, economic and environmental impacts of the various alternative of definition and management of (new) water Volume Available for Agriculture? !
Robustness to climate variability? !
Technical feasibility and social acceptability?
101
Objec5ves
Build a platform able to simulate the evolution of water resources management on the Adour-‐Garonne basin Assess impacts of alternative water management policies during low water crisis by simulation. Under various scenarios of climate change !
Space and time scale : draining basin (Adour-‐Garonne basin:116.000 km²) at the field scale over 30 years at the day scale
102
Approach
Approach: agent-‐based modeling of (complex) Social-‐Ecological Systems
« gestion spatiale de l’eau » (Narcy & Mermet, 2007)
6
Paradigm: agent-‐based modeling and simulation: agent-‐based models can be versatile and heterogeneous.
SimPop model http:// www.simpop.parisgeo.cnrs.fr/
Alexis
First steps of an interdisciplinary project • Partners who do not know each other, with their own methods, conceptual frameworks, expressions (words and semantic) !
• Fields: hydrology, computer science, geohydrology, agronomy, linguistics, sociology, history, software engineering… !
• A model, a simulator, a multi-‐agent system ?????? !
• A project dynamic to build !
• 18 months to find a way… • How to conduct the interdisciplinary modeling process?
106
The MAELIA interdisciplinary modeling method ARDI Ambiguity Knowledge
K & Data
SES Rep
Question
Q’
Data Model
Data Model’
Data Model
SES structure
SES structure and processes
Sensitivity analysis Qualibra.-validation
Q’’
Q’’’
Q
Collective learning (methodology/various fields/ teamwork) 107
The integrated MAELIA model
MAELIA: spatial processes Ecological processes • • •
Soil-‐crop model Hydrologic model Weather
Human activities (Decision process) • Farmer decision
– crop allocation plan – crop management ! • State services decision: Socio-‐economic processes – decree of water-‐use restrictions (phenomena) (severity & spatial extension) • Demographic changes (INSEE, • Dam Manager decision: municipality level) – water releases • Land Cover changes (Corine Land Cover database) • Drinking Water Consumption • Industrial Water Consumption Processes interact in space (field, farm, sub-watershed) and time (day) ➔ coupling of social and ecological processes
MAELIA: the structure of socio-hydrosystems for lowflow management Agricultural model
Hydrologic model
Normative model Cognitive Resources
Other water uses model Material Resources
Actors
MAELIA: the structure of socio-hydrosystems for lowflow management Agricultural model
Each farm with each irrigable and non-‐irrigable islet
Normative Each d am and modelregulation sectors
Hydrologic model
Water ressources: River segments, lakes, pounds, groundwaters
Each domestic and industrial Other aw ater uses m odel withdrawal nd discharge point
MAELIA: the structure of socio-hydrosystems for lowflow management Agricultural model
Hydrologic model
An integrated representation of the structure of water management situations in the Adour-‐Garonne basin
Normative model
Other water uses model
The MAELIA case study First application of MAELIA to the up-‐stream part of the Garonne river (from Pyrénées mountains to Toulouse) !
Drainage area: 6000 km2
Length: 140 km !
104 Elementary Watersheds !
2532 farm(er)s 27 000 active field plots (with crops)
Integration principles
Integration principles
• Spatial discretization in sub-‐basins / watersheds (Zone hydrographique (ZH)).
104 Zones
Hydrographiques in the MAELIA area.
Example on a ZH
source : Carthage
Processes at the day scale 1
Weather: Rain, Snow 5
Withdrawals (irrigation, domestic,
industrial) 6
Water Flow
to Downstream
2 Farmers daily activity
3
4
Water Flow from Upstream
Plant growth
7
Normative model (prefect, police …118 )
Processes at the year scale
2 Update Economic market
3
1
4
Water volume allocation
Urban sprawl: -‐ Land Cover change -‐ Water consumption
Farmer yearly decision
(cropping plan)
119
20
MAELIA : Farmer decision model A Belief-‐Desire-‐Intention architecture to represent the farmer’s:
– multi-criteria choice of the cropping plan according the belief theory on 4 criteria : profit, variability of profit, workload and similarity to the last Strategic decision cropping plan (can be switch off) (year) – Memorize yields, prices, water available, workload
Operational decision (day)
Different operations (some just for workload constraints) : – Tillage – Sowing a field – Protect cultures – Surface tilling – Fertilise a filed – Irrigation of a field – Harvesting of a field
MAELIA : soil-‐crop model Priority to simulate realistic crop yields over the Adour-‐Garonne Basin without great and often problematic calibration work !
➔ Empiric crop model developed by INRA in Toulouse called “Jeu d’O” !
Developed step by step according results of agronomic experiments in the AGB during the last 20 years !
It represents effect of -‐ climate, soil and cropping system on water soil dynamics and yield for the eleven main crops of the AGB -‐ cropping system: tillage, sowing date and irrigation
MAELIA : the hydrologic model Selection of the SWAT model : -‐ applied to various contrasted situations -‐ a great user/developer community -‐ open-‐source -‐ possibility to define the size of sub-‐ watersheds to adapt the model to the investigated agro-‐hydrological issues !
Re-‐implementation into MAELIA of the SWAT formalisms representing : -‐ The snow accumulation and smelt -‐ The land phase -‐ The routing phase -‐ The hydrology (input) of dam and pound !
Elementary Watersheds used as “sub-‐ watershed”
Elementary watersheds used as SWAT’ watershed
Conclusion
Project agenda and perspective • Sensitivity analysis: • over 30 (hydrolic model), 7 (farmers, with forcing cropping plan) or 37 parameters (both processes) • Sensitivity over 14 outputs • Spatial Sensitivity (of hydrologic model) • Positive consequences: • bugs detection • question precision of some forcing data (e.g. refine the number of slope and altitude classes in inputs) • Calibration: multi-‐objective: reproduce • water flows and • anthropic dynamics
Application example: Clément’s Murgue thesis • Objectives: Propose method and tools to design and assess alternative uses of the agricultural space. Imagine possible alterna5ves:
….#
Cropping plans: Crops, spaDal distribuDon Water management strategies
Simulate in 5me/space: !
Modeling & simulaDon (using the MAELIA plaworm)
Assess each alterna5ves
Par$cipa$on*
Needs in irrigaDon: how many, when, where? Consequences on hydrology Conséquences on farms
Variabilité)clima,que:)
Modélisa)on+
Evalua&on)
Hydrology in MAELIA
http://code.google.com/p/gama-platform/
MAELIA