Skip to content

ANR PRCI APPG2021 - Traces

Semantic Environmental Trajectories of Territories

Site Web

A propos

Human activities are the main drivers of the observed environmental megatrends such as climate change, biodiversity loss, use of resources, pollutions, …, which impact territories at all scales. Better study and understand the environmental trajectory of a territory, by selecting and using relevant, available, and comparable over time and space indicators, can help to master these megatrends, which appears as one of the main challenges regarding the sustainability of humankind. In this context, the TRACES project will adopt an Artificial Intelligence (AI) approach in order to define, model, implement, enrich, interrogate, analyze, forecast and visualize the environmental trajectory of a territory. To achieve such a complete processing chain, the TRACES project will rely on three domains of AI, Knowledge Representation, Machine Learning and Multi-Agent Systems.

Our first research hypothesis is that the semantic approach supported in Knowledge Representation through ontological models and languages, and Knowledge Graphs (KG) formalism and tools, can help, not only to build and implement semantic environmental trajectories of territories (SETT), but also to enrich them with complementary knowledge relying on the Linked (Open) Data available on the Web. Second, Machine Learning offer valuable techniques, algorithms and tools to group and categorize similar SETT in clusters, but also to extract frequent patterns in these SETT and then to complete then for forecasting future SETT. The third research hypothesis we make is that a Multi-Agent System approach helps to better understand the factors that draw territories trajectories, by modeling and simulating the way territories evolve and behave under systemic constraints. We will consider here that an environmental trajectory can be defined on the basis of a set of environmental indicators, built themselves using official data collected from surveys, various physical sensors, satellite images, etc. The environmental trajectory object produced and analyzed by the TRACES project will thus give an account of the dynamics of the studied territory, and in particular, convey its evolution in terms of biodiversity, nature conservation and resilience to climate change.

The models, algorithms, KG and agents designed by the TRACES project will contribute to a better exploration of environmental trajectories of territories as objects of study. Thus, the proposed processing chain is intended to assist professionals and experts in spatial planning in their longitudinal and comparative analyses, decision-makers in the elaboration of future environmental policies at different territorial levels, but also citizens in their understanding of the public policies implemented and the evolution of the territories they live in.

The first scientific and technical challenge for the TRACES project concerns the modeling of environmental trajectory of territory, which to the best of our knowledge, constitutes the first attempt for representing this notion by adopting a Semantic Web-based approach. The difficulty lies in both the definition of such an environmental trajectory in terms of indicators of possibly different kinds and their measures, and in the coupling with its intrinsic spatial and temporal dimensions. Therefore, as a whole, an environmental trajectory of a territory appears as a complex multidimensional and multigranular object whose modelling, visualization and analysis are difficult and yet unexplored tasks. An ontological model for such an object should also include standard vocabularies in order to facilitate data integration and ensure the reusability of the KG representing the SETT that will be created from both this conceptual description and the values of the indicators Exploring how Machine Learning clustering techniques can be applied to SETT faces three main challenges: 1) providing similarity measures adapted to the representations of multimodal data evolving along the temporal axis; 2) enabling multimodal clustering of trajectories for the definition of trajectory profiles; 3) enabling temporal prediction over multimodal data representations.

The main challenges raised by modeling environmental trajectories using an executable multi-agent system lie in: 1) leveraging SETT to derive, possibly in an automated manner, multi-agent based models reproducing past environmental trajectories; 2) combining and visualizing in the same hybrid models, spatially-explicit agents modeling human behavior and lattice-based territorial changes; 3) exploiting a series of scenarios highlighting relevant policies to provide useful prescriptive analysis. Lastly, several definitions of an environmental trajectory will be proposed to handle different perspectives on this study object. Indicators and data will be provided and elaborated by partner ISE, using the Swiss Data Cube platform. Taking advantage of the fact that TRACES is a joint project between Swiss and French researchers, we will choose three types of territory to conduct longitudinal and comparative analysis: 1) Swiss territories (at the state, canton and commune levels); 2) French territories (at the department, commune and metropolitan levels); 3) on both sides of the French-Swiss border (at commune and metropolitan levels). However, our approach intends to be generic and applicable to worldwide territories.

Universities

  • Université Grenoble Alpes (UGA-LIG)
  • Université de Bourgogne (UB - LIB)
  • Université de Genève (UG - ISE)
  • Université de Genève (UG - CUI)