Communication Dans Un Congrès Année : 2025

Coupling GAMA, R software and the OpenMOLE platform to explore an agent-based model simulating interactions between farms

Résumé

Crop-livestock systems at landscape level encompasses interactions between specialized agricultural systems. Materialized by biomass flow between farms, these interactions can increase farm self-sufficiency (SS) and local resource circularity (manure, forage, grain, straw) [2]. However, it remains unclear how these interactions occurs and to which extent the resulting farm network enhances performance and resilience towards external hazards at both farm and landscape levels. To better understand these processes, we built an agent-based model that simulates individual farms able to take independent decisions on interactions through biomass flows while considering social behaviors [1]. We implemented it in the GAMA modeling platform. To efficiently perform model explorations for its parametrization, verification and validation, we coupled GAMA simulations to the R software using the workflow facilities of OpenMOLE (https://openmole.org/). In the ABM, agents are farms (either cereal, livestock or crop-livestock) and the cooperative localized in the Ariège region (France). At each time step (year), farms produce biomass that can be exchanged with other farms, the cooperative or the global market. Exchange decisions are based on a score that considers i) exchange distance, ii) exchangeable biomass quantity iii) trust between farms, which reflects farm exchanging habits and evolves based on past exchange experiences. Farm individual preferences give more or less weight to one or other of these variables. At initialization, we use random farm locations (according to regional farm type distribution) and compute farm distance in order to quicken the scoring process for exchanges at each cycle. Farm distance computation is time-consuming when farm number is high (10 minutes for 300 agents) in GAMA. Thus, we produce the farm localization dataset with GIS and network packages (igraph, sf, sfnetworks) from the R software. GAMA simulation outputs are also analyzed with the R software to compute performance, network and resilience indicators. Using OpenMOLE, we upscaled the experiments on the large powerful computing environment of the European Grid Infrastructure (EGI) (https://www.egi.eu/). We explored our model with the Sobol sampling technique to define the number of cycles (8) and the number of farm agents (300) to simulate. We also proved that multiple farm location datasets are not necessary to get a diversity of simulation outputs. Pattern Space Explorations (PSE) helped us to choose the indicators that would show a wide diversity in the model output space ; it also indicated us which model mechanisms was worth to keep and which was not. Profile explorations revealed clear relations (sigmoidal curve) between the livestock SS and the farm preference for farm exchanges. The coupling of the GAMA model with R software using the OpenMOLE (with three different methods) guided and sup- ported the model parametrization to establish a reference in the Ariège region. The pipeline (structure, scripts and coupling) will be presented at the conference. Next steps are i) to decompose the reference into multiple baselines with different farm network ‘strength’, ii) apply scenarios to show if the network ‘strength’ can influence its performance and resilience.
Fichier principal
Vignette du fichier
Gama_days_2024_Peter.pdf (338.49 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04887937 , version 1 (15-01-2025)

Identifiants

  • HAL Id : hal-04887937 , version 1

Citer

Aurélien Peter, Romain Reuillon, Myriam Grillot. Coupling GAMA, R software and the OpenMOLE platform to explore an agent-based model simulating interactions between farms. GAMA Days 2024, Nov 2024, Online, France. ⟨hal-04887937⟩
0 Consultations
0 Téléchargements

Partager

More