| Dealing
with multiple scales is often a key question in renewable resource
management. In some case, the system dynamics are intrinsically
linked to a specific spatial entity, which should obviously be taken
into account in the model. In other case, the decision to incorporate
a spatial entity is influenced by the fact that information is available
at this level. Nevertheless, it is important to have the possibility
to manipulate and to incorporate into the same model spatial entities
defined at different hierarchical levels.
Originated
from the field of Distributed Artificial Intelligence, Multi-Agent
Systems (MAS) are potentially suitable for linking several hierarchical
levels. In a MAS, an agent is a computerised autonomous entity that
is able to act locally in response to stimuli from the environment
or to communication with other agents. Cormas (Common-pool Resources
and Multi-Agent Systems) is a multi-agent stimulation platform specially
designed for renewable resource management. It provides the framework
for building models of interactions between individuals and groups
sharing natural resources. With Cormas, the design of the spatial
support rests on spatial entities, which are themselves a category
of agents. When these entities yield resources, they are competent
to arbitrate their allocation, according to pre-defined protocols,
between the concurrent demands formulated by other agents exploiting
their resources. The way agents are exploiting resources may depend
on their own partial representation of the environment, which are
based on these same spatial entities.
Following
a general overview of the Cormas simulation platform, three models
built using this toolkit are presented, by emphasising the overlapping
of their multiple hierarchical scales.
- In
the first model, the spatial grid is made up of 100 x 100 cells.
Three spatial scales are taken into account. An ecological dynamic
is defined at the cell level, where tree agents are located.
The plant growth and seed scattering are the basic biological
processes of the tree agents, which are also characterised by
a genotype. Plots are defined as 10 x 10 cells. Farmer agents
own 10 plots, and have to decide their land use (either crop,
pasture, or forestry). This agricultural dynamic is finally
influenced by biodiversity subsidies awarded to the owners of
forestry plots that are part of a forest, a forest being defined
as an aggregation of continuous forestry plots and characterised
by a global biodiversity index calculated from the genotypes
of the tree agents.
-
In northern Thailand, an on-farm diagnostic survey analysed
the influence of the main cropping systems on the risk of soil
erosion under various slope and climatic conditions at the field
level. A typology of rapidly diversifying household-based farming
systems was built to understand farmers' differentiated strategies
and their degree of susceptibility to erosion at this level.
Whereas, as GIS-based analysis of landuse changes was carried
out at the village watershed level. To go beyond the site specificity
of this empirical multi-scale study and to better understand
and model the interactions between agro-ecological and socio-economic
dynamics, a multi-agent model is built to facilitate knowledge
integration across scales and disciplines.
-
The objective of the third model is to understand the interactions
between fuel wood consumption and landscape dynamics. The hypothesis
is that the evolution of the landscape in the Kayanza region
in Burundi can be explained by fuel wood consumption. An initial
map is defined and agents collect fuel wood, have access to
different parts of the territory and can perform exchanges.
The size of the population increases and the migration of agents
from over populated parts of the territory to unoccupied patches
is simulated. The effects of changing rules about foraging,
exchange and access are evaluated at various scales, including
the forest and landscape levels. Finally, the use of such models
and multi-agent systems to represent knowledge on processes
at various level of complexity and to simulate their interactions
according to a bottom up approach for understanding landscape
dynamics are discussed.
cormas
computer session
To
develop the model, it is essential to know the language and
write the code - this is called 'small talk'. In a simple model,
step one is to define the entites:
-
these
can be spatial, and there can be a hierarchical organisation
of these entities, something that is unique to Cormas
-
these may also be social, such as farmers and animals
Furthermore,
the size of the environment can be defined. The predator and
prey can be created as agents, and their attributes and preocedures
can then be defined. These can be selected fomr a list, or it
is possible to create your own.
Step
two: the model will then schedule the entities created.
Step
three: it is then possible to choose a way to look at your entities;
for example to choose the colour and shape to give them. This
is particularly useful with more complicated models.
Step
four: now that predators and prey are on the map, it is possible
to choose a time scale.
Further
models show the negotiation between farmers about plots of land.
In such a model, the entities are also designed: the plots of
land and the farmers. With regard to the farmers, it is a generic
quality of Cormas that it is possible for them to have a communication
attribute. It is possible to have two grides, on as before, and
the other to watch the communication between the farmers.
These
two very simple models illustrate the principles of Cormas, and
were used in Francios' model.
|
