Methods and Approaches

The research conducted and scientific output of the TRACES project focus on five major interrelated issues or axes of research.

Axis 1) Preparation of case studies, indicators and environmental data.
Three study areas in Switzerland and France were selected: the Canton of Fribourg (127 Swiss municipalities), Greater Geneva (209 Swiss and French municipalities) and the Pays d’Évian (45 French municipalities), for a total of 317 municipalities. The TRACES project mainly uses indices (measurements) that assess various environmental characteristics for each pixel of a satellite image. This raw data is available via the Swiss Data Cube for all municipalities in the three study areas. The TRACES project focused on Landsat satellite indices (LIS) relating to vegetation, urbanisation, snow and water, the Landsat LST index relating to surface temperature, data from the five European and Swiss CLC (Corine Land Cover) campaigns (1990, 2000, 2006, 2012, 2018), digital elevation model (DEM) data and regional auxiliary data. The observation period chosen is the one for which LIS, LST and CLC data were available at the start of the project, i.e. from 1985 to 2022, with approximately one measurement per week for LIS and LST. The satellite index measurements were pre-processed to ensure their consistency and high quality. Finally, five satellite indices representing vegetation (NDVI), urbanisation (NDBI), snow (NDSI), water (NDWI) and ground temperature (LST) were studied in particular, focusing on how these indices could be combined or correlated to form several global environmental indicators.

Axis 2) Production and visualisation of a knowledge graph (KG).
The KG produced by this axis represents the environmental semantic trajectories (SETTs) of the 317 municipalities studied. An ontological model of SETTs has been developed. It describes three levels of trajectory: 1) the raw trajectory; 2) the structured trajectory; 3) the thematic trajectory. Based on the pre-processed time series produced by axis 1) of the five selected satellite indices, a knowledge graph was developed using the RDF Data Cube vocabulary. This RDF Data Cube is obtained by transforming the pre-processed raw data available in csv files into RDF triples. An initial enrichment process increased the semantics of the triplets obtained. This cube makes it possible to extract and explore, in the form of a KG, the raw trajectory of an index for a given period and municipality in three main dimensions (time, space and index or indicator). The structured trajectory is itself a KG obtained from the KG of the raw trajectory by applying a segmentation method (studied by task T4) that aims to break down the time series of index measurements into segments. Each segment is annotated with information that characterises it (measurements at the ends, duration, slope, etc.). The segmented trajectory is linked to the raw trajectory and also contains information on the segmentation method used. The thematic trajectory is a semantic trajectory that describes the evolution of an environmental index (related to vegetation, urbanisation, snow, water, or surface temperature) for a given period and municipality. It is composed of the same segments as the structured trajectory, annotated using terms from a vocabulary adapted to the audience to whom the SETTs will be disseminated. The environmental semantic trajectory (SETT) for each municipality studied is composed of five thematic trajectories relating to changes in vegetation, urbanisation, snow, water, and surface temperature. Finally, a user interface has been designed and developed to enable the selection and interactive spatio-temporal visualisation of the information contained in the SETT KGs.

Axis 3) Enrichment of environmental knowledge graphs.
The first challenge here was to enrich the knowledge graphs produced by axis 2. To do this, a comparative analysis of thesauri dedicated to environmental sciences was conducted. Controlled vocabularies and thesauri are essential elements of the TRACES project because they ensure consistent data structuring, facilitate data integration and analysis, and enable effective communication between the various stakeholders involved in territorial management, whether they be decision-makers, professionals or citizens. The emergence of the semantic web and associated technologies offers new opportunities to improve access to and sharing of environmental knowledge, while the harmonisation of vocabularies and the interconnection of data are becoming major challenges. The thesauri studied were evaluated in terms of the semantic relationships proposed and their technical and linguistic aspects.
Analysing interactions between society and the environment requires conceptual tools capable of representing the complexity of the relationships involved. The DPSIR (Driving forces, Pressures, State, Impacts, Responses) model meets this objective by providing a structured framework that highlights the causal links between human activities, their effects on the environment and the responses implemented. Developed by the European Environment Agency in the late 1990s, its main contribution lies in the explicit introduction of driving forces and impacts, which enrich the analysis of environmental pressures and make it possible to better link root causes to observed consequences. As part of the TRACES project, a DPSIR ontology was designed and developed, enabling the articulation of various elements: socio-economic determinants, measurable pressures, environmental quality indicators, ecological or health impacts, as well as institutional and societal responses. Finally, large language models (LLMs), which emerged during the TRACES project, open up new perspectives for the extraction, formalisation and networking of knowledge. One of the major challenges identified concerns the ability to generate ontologies from texts in order to feed and enrich environmental knowledge graphs. With this in mind, he studied the possibilities offered by LLMs and agentic AI to go beyond their linguistic processing power and coordinate specialised agents to automate concept detection, relationship extraction and knowledge structuring in a standardised ontological format, thus ensuring better integration with existing environmental vocabularies.

Axis 4) Methods for detecting changes, classification, and prediction for time series of environmental data.
The raw data extracted from satellite images (axis 1) are time series of satellite environmental index measurements. In order to achieve a semantically rich synthesis of these measurement series in the form of semantic environmental trajectories (SETT), it was necessary to study methods for detecting changes in these series. The determination of change detection methods is one of the major scientific challenges of the TRACES project in view of its main objective: understanding climate change and its effects on vegetation and temperature dynamics. The ability to detect and quantify these changes over time is crucial to establishing environmental trajectories as decision-making tools for developing targeted interventions and ensuring the long-term sustainability of natural resources. To this end, an in-depth analysis of the main existing change detection methods (Rupture, BFAST, BEAST, hidden Markov models, etc.) was conducted on time series of key satellite index measurements (such as NDVI or LST). Another study was conducted on machine learning approaches dedicated to prediction in time series, with a view to applying them to raw data consisting of time series of satellite index measurements.

Axis 5) Agent-based modeling of environmental semantic trajectories and prescriptive analysis.
This axis focuses more specifically on the evolution of urban areas. Assessing the sustainability of cities requires the development of tools dedicated to holistically capturing the complexity of urban systems. It is not enough to analyze each component separately (urban mobility, land use change, etc.), but also their interactions, as these are of paramount importance for the evolution of the system as a whole. To this end, agent-based models are capable of integrating the key characteristics of these complex systems and are of proven value for public policy development due to their ability to provide information for decision-making processes. In this context, three agent-based models dedicated respectively to modeling, simulation, and impact assessment: 1) commuting along the Cornavin-Meyrin-CERN axis; 2) DPSIR analysis of residential choices based on public transport provision in Geneva, and 3) the development and evolution of the public transport network in the Greater Geneva region. This work has highlighted the advantages of agent-based models for capturing the complexity and emerging phenomena of urban systems. These models incorporate population heterogeneity and individual behavior, and allow for the exploration of various hypothetical scenarios, providing valuable information for public policy development. The aim was also to explore the contribution of a DPSIR framework (axis 3) and knowledge graphs (axes 2 and 3) to these agent-based models dedicated to environmental semantic trajectories.