Perception-Based Maps as input data for Agent-Based Modeling: application to the colonization of the Ecuadorian Amazonia - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

Perception-Based Maps as input data for Agent-Based Modeling: application to the colonization of the Ecuadorian Amazonia

Résumé

The PASHAMAMA project aims at modeling the installation of settlers in the northern part of the Amazonian region of Ecuador under pressure from oil exploration and agrarian ressources from 1960 to 2016. Particularly, the model takes into consideration the intensive oil extraction touching the zone which has induced a high rise of population, pollution, agricultural work and deforestation and aims at reproducing these phenomena. One of the main difficulties in the construction of such a model is the complexity to model the agents representing human beings, and in particular their behaviors, their representation of the world of themselves, and of the other agents, as well as their decision making process.The traditional approach in Artificial Intelligence is to look at psychological or sociological theories to build formal models of human behaviors rather than directly considering real data. But this approach often proves unsatisfactory in interdisciplinary work where it appears too disconnected from the real system to be accepted. We therefore adopt in this work the opposite approach: our starting point is real field data, quantitative (territory,practices) but also qualitative data from PBRM (Perception-Based Regional Mapping), that are semi-directed surveys letting local inhabitants produce subjective maps of the environment they know. One of the principal advantages of the use of quali-quantitative tools is the possibility to highlight the subjective and biased representation of the questioned person in its agro-ecological and socio-economic environment. To this purpose, we rely on data collected during field studies organized between 2013 and 2016 to collect perception-based maps on the three modeled provinces in the PASHAMAMA model: Dayuma, Pacayacu and Sacha provinces of the Amazonian area of Ecuador. A key component of the model is the installation of new colons in the studied area over time and in particular their choice of finca (i.e. parcel). This process is particularly important because it has a strong impact on the simulation results and it is particularly relevant forour study because we assume that its result comes from a settlers decision-making process. In the current version of the model, the finca is chosen thanks to the minimization of a function of the distance and the type of roadsThe key idea of this work is to introduce subjective data in this decision-making process: we thus extend the previous function by introducing the settlers’ desire to maximize the quality of the soil in their choice. We run experiments to compare results without the use of soil quality in the choice (original model), with the use of the real soil quality map (fully-informed model), and with the use of the subjective soil quality data coming from PBRM (subjective model). We assess simulation results given actual deforestation datareliminary results show that the introduction of PBRM data improves the model results in terms of settlement areas of the colons. Another interesting result is this result is confirmed on the three studied provinces, in the limit of a well-differentiated area: improvements on a province with a uniform soil quality are not significant
Fichier non déposé

Dates et versions

hal-03416320 , version 1 (05-11-2021)

Identifiants

  • HAL Id : hal-03416320 , version 1

Citer

Clément Nicolle, Doryan Kaced, Benoit Gaudou, Mehdi Saqalli. Perception-Based Maps as input data for Agent-Based Modeling: application to the colonization of the Ecuadorian Amazonia. GAMA Days 2021, Frédéric Amblard; Kevin Chapuis; Alexis Drogoul; Benoit Gaudou; Dominique Longin; Nicolas Verstaevel, Jun 2021, Online, France. ⟨hal-03416320⟩
26 Consultations
2 Téléchargements

Partager

Gmail Facebook X LinkedIn More