Système multi-agent pour la modélisation des écoulements de surface sur un petit bassin versant viticole du Layon Agent-based modelling for hydrological surface processes on a small watershed (Layon, France)

Les catastrophes naturelles sont fréquentes dans la région Sud Ouest de Madagascar, en particulier en ce qui concerne les inondations. Évaluer la vulnérabilité des habitants face à ces catastrophes se révèle être d’une importance majeure. La vulnérabilité étant un concept théorique, il n’est pas aisé de lui assigner une valeur chiffrée. Il existe cependant plusieurs méthodes pour « mesurer » la vulnérabilité aux inondations. La plus classique est l’utilisation d’un indicateur qui se base sur l’agrégation de plusieurs critères pondérés aboutissant sur un indice synthétique. Dans ce travail, nous voulons ajouter un caractère dynamique à ce type d’indicateur en utilisant une approche modélisatrice : les systèmes multi-agents (SMA). Ainsi les comportements de toutes les entités mises en jeu (habitants, habitations, flux d’eau) lors d’un évènement catastrophique seront simulés par un modèle qui met en avant l’utilisation de l’agent : individu autonome qui agit selon des règles spécifiques. Plusieurs scénarios sont pris en compte et comparés afin d’aboutir à un indicateur dynamique à l’échelle de l’habitation. Cette approche permet non seulement une observation intuitive de chaque évènement catastrophique mais aussi de pouvoir ajouter une dimension supplémentaire qu’un indicateur statique classique ne peut pas prendre en compte, l’aspect prospectif de la vulnérabilité. À long terme, ce type d’approche pourrait aboutir à un modèle de prédiction.

Land use and landscape structure play an important role in the functioning of flash floods on the arable plateaus of northern France. Landscape structures have changed considerably over the last decades with an increase in runoff-producing surfaces and an enlargement of the plots. This evolution results in an increase in runoff volumes accompanied by an increasingly easy diffusion of runoff on the slopes. There is therefore potentially an increasingly frequent and important spatial connectivity from the plots to the thalwegs, which can lead to the formation of flash floods. This study proposes to model this diachronic evolution of connectivity in a small agricultural basin of Seine Maritime using a multiagent system (MAS) and to develop synthetic indicators characterising these spatial links in the flow processes. The model outputs show that spatial connectivity has been steadily increasing over the past 70 years due to the enlargement of the parcel grid and the growth of runoff surfaces. For example, for the same 20 mm/h rainfall, the connectivity indicator increases from 40.99% (in 1947) to 78.33% (in 2015). This observation is observed for all levels of rainfall intensity, including the lowest. This modelling, carried out for a 116 ha basin in arable farming, can be transposed to all small agricultural basins.

Agent-based simulation has gained interest in geography in recent years. Several platforms facilitate the development of spatially focused multi-agent models. However, each platform and language has its own specificities, and the same conceptual model can lead to different behaviors and results depending on the type of implementation used. Reproducing the same model on multiple platforms is the best way to test its reliability, while eliminating biases related to implementation. This "good practice" is still very rare, yet essential for making a more robust model. In this paper we present the implementation of the same conceptual model on two distinct platforms (Netlogo and Gama). Despite some overall similarities in the results, differences are observed in the outputs and functioning. These variabilities come from various factors, but a significant proportion results from the different ways of integrating geographical data, which differ between the two platforms. To highlight this geographical aspect, we use a conceptual model dealing with a flood evacuation.

Références de la thèseRakotoarisoa M. M. (2017). Les risques hydrologiques dans les bassins versants sous contrôle anthropique : modélisation de l'aléa, de la vulnérabilité et des conséquences sur les sociétés : Cas de la région Sud-ouest de Madagascar. Thèse de doctorat en géographie, Université d'Angers ; Université de Tuléar. Madagascar est un pays qui se trouve usuellement confronté à divers aléas climatiques. Ils se sont manifestés ces dernières années sous diverses formes, entrainant d...

The south-west of Madagascar faces various climate hazards, especially rising water levels. Each year, the city of Toliara is subjected to quick and devastating floods, which appear to have increased in intensity and frequency. In addition, this region is one that represents completely the “north/south fracture”, that is to say a “...technologic fracture between northern countries and southern countries concerning the availability of data and the model deployment constraints” (Payet, 2015). Thus, all the hydrometric data available for this region are incomplete. Based on this issue of missing data, this study is divided into two parts: the first is the reconstruction of the data and the second is the modeling of these data to obtain a rigorous representation of the chronicles. First, the reconstruction part must fulfill the following two conditions: (i) represent the overall dynamics of the series and (ii) be simple enough to easily take them into account in the analysis and interpretation at the end of the modeling. We offer a way to reconstruct these missing data by combining seasonal decomposition methods and simple linear regression. After that, we use classic statistical methods to adjust the model to our chronicles. We obtain two types of model: a SARIMA model, which is a Seasonally Auto-Regressive Integrated Moving Average model, and an ARIMAX model (Auto-Regressive Integrated Moving Average with eXogeneous inputs), which is a combination of a linear model and our SARIMA model. The two models are quite functional as we tested their quality and the statistical significance of their coefficients. This method can help scientists working in southern countries not to discard incomplete data but to exploit them as much as possible because they convey important and usable information, as long as we consider the uncertainty of their conditions of creation.