Swiss Ai Research Overview Platform
Observing the environmental trajectories of territories is essential for analysing their evolution in terms of nature conservation or resilience to climate change, and for developing future public policies in terms of the environment and land use planning. The TRACES project aims to enable the modelling and analysis of the environmental trajectories of territories, based on three areas of Artificial Intelligence. An environmental trajectory will be defined by a study territory, an observation period, and a set of indicators characterising this trajectory, which is multidimensional (space, time, theme) and multi-granular in nature. Several types of environmental trajectories will be proposed and considered. The indicator data used by the TRACES project will come from open and accessible official sources. The approach proposed here is centred on Knowledge Graphs (KGs) and the Semantic Web. Thus, in a first step, the TRACES project will generate, in the form of KGs, semantic environmental trajectories of territories (SETTs). These KGs will be developed from an ontological model of trajectories based on standard vocabularies, and on data extracted from a spatial data infrastructure dedicated to environmental indicators. The environmental observations that make up the trajectories and form the SETTs will be described in semantised data cubes, linked together over time. Web 3.0 standards will be used so that the SETTs are published in the Linked Open Data (LOD) Cloud, according to the FAIR (easy to Find, Accessible, Interoperable and Reusable) data principles. The Knowledge Graphs representing SETTs will then be enriched and completed by a targeted search in the open data sets made available by the LOD Cloud. Similarity measures adapted to the multidimensional nature of the SETTs will be proposed and integrated into Machine Learning techniques and algorithms in order to group similar SETTs into clusters, but also to extract frequent patterns, on which a completion and trajectory prediction phase based on recurrent neural networks will rely. Complementarily, a multi-agent system approach will develop explicit, agent-based models capturing the various behaviours of human actors and stakeholders, the dynamics between human actors and the environment and its impact on the environment, and the effects on the environment of policies implemented in the past. These models will be fed by the SETTs Knowledge Graphs, and by the information extracted from the LOD Cloud during their enrichment, as well as by the inferences produced by the Machine Learning algorithms. This knowledge will allow a better understanding of the mechanisms at work in the dynamics of the territories, and to simulate the effects of public policies, already tested or new. These different types of analysis will be completed by an interactive interface allowing the visualisation of these SETTs as graphs, using evolution curves, on maps showing the territorial dynamics, or through spatiotemporal cubes. All of these components will constitute an original and innovative processing chain, exploiting and extending current work on spatiotemporal Knowledge Graphs. The results of the TRACES project will provide a methodological and tool base for decision support for decision-makers and professionals in charge of territorial management, but also useful to inform citizens living in the territories studied. The cases studied by the TRACES project will concern territories in Switzerland, in France and on both sides of the border between the two countries from which the partner institutions of this project come.