Challenges and State of the Art

Human activities are the main drivers of observed environmental megatrends such as climate change, biodiversity loss, resource use, pollution, etc., which have an impact on territories at all scales.

The main hypothesis of the TRACES project is that studying and understanding the environmental trajectory of a territory, by selecting and using relevant indicators that are available and comparable over time and space, can help to control these megatrends, which appear to be one of the main challenges in terms of humanity’s sustainability.

To support this hypothesis, the TRACES project draws on three areas of artificial intelligence (AI):
1) knowledge representation, particularly ontologies and knowledge graphs (KG), in order to model and construct semantic environmental trajectories of territories (SETT), but also to enrich them with complementary knowledge based on open and linked data available on the web;

2) machine learning, which provides techniques, algorithms and tools to construct these SETTs, compare them and group them by similarity;

3) Multi-agent systems whose models and associated simulation processes provide a better understanding of the factors that determine the trajectories of territories, and how territories evolve and behave under systemic constraints.

In Computer Science, the concept of semantic trajectory has mainly focused on the trajectories of moving objects (cars, planes, boats) or pedestrians [Alvares 2007, Parent 2013]. The semantics expressed mainly describe a moving object or person, as well as the activities characterising the sequence of movements and stops that make up the trajectory. With regard to environmental trajectories, Bryan et al. [Bryan 2016] propose a land use trade-off model (LUTO) to support a comprehensive, detailed, integrated and quantitative scenario analysis of land use and sustainability in Australian agricultural land between 2013 and 2050. The LUTO model integrates several spatio-temporal models and data layers from various sources, but without any explicit semantics. Trajectories [Anquetin 2018] is an interdisciplinary project dedicated to the processes of co-evolution between human societies and the environment. Although the notion of environmental trajectories is at the very heart of this project, no semantic modelling is proposed and the trajectory must be extracted manually from time series of relevant indicators. To our knowledge, the observation of environmental dynamics through trajectories has not yet been addressed using a semantic approach, as reported in [Giuliani 2020a].

In the field of Knowledge Representation, three approaches appear relevant for modelling SETTs:
1) the 4D-fluent approach of [Welty 2006] for representing four-dimensional objects that evolve over time;
2) the Continuum model and the TSN ontological model for representing spatio-temporal relationships [Harbelot 2013] and territorial changes [Bernard 2018];
3) the design model for modelling life trajectories in the semantic web [Noël 2017].
On the other hand, standard ontologies have been defined to describe indicator observations in the Web of Data, notably RDF Data Cube (QB) [Cyganiak 2014], O&M and SSN.
Knowledge graphs (KGs) make it possible to describe real-world entities and their interrelationships in a graph, and to define the concepts of entities and the relationships between these concepts. Based on RDF triples as elementary units, KGs naturally provide a framework for data integration, unification, analysis and sharing [Hogan 2020].
In the context of environmental trajectories, enriching KGs with linked data is essential for data exchange and sharing, thereby expanding navigation, discovery and query capabilities [Nativi 2020].

Producing trajectories from time series of Earth observation data requires the use of advanced statistical analysis methods [Krispin, 2019]. Once obtained, environmental trajectories can be compared, classified and extended using machine learning techniques [Huang 2019].
Agent-based models, meanwhile, capture the behaviour of human actors and their impact on land use change across different spatial, temporal and dynamic dimensions [Millington 2016]. These systems provide a representation at different scales in space and time [Giuliani 2020b] of the micro-behaviour of individuals in specific areas or environments to meso- or macro-behaviour at the regional level, for example, and seem well suited to the study of public policy use [Le 2008] at multiple scales.

Bibliographical references

[Alvares 2007] L. O. Alvares et al. Towards semantic trajectory knowledge discovery. Data Mining and Knowledge Discovery, 2007.
[Anquetin 2018] S. Anquetin et al. TRAJECTORIES: Social-ecological trajectories of French alpine valleys under climate variability”, EGU2018, European GeoSciences Union, 2018.
[Bernard 2018] C. Bernard et al. Modeling changes in territorial partitions over time: Ontologies TSN and TSN-Change. In Proceedings of the 33rd Annual ACM Symposium on Applied Computing, SAC’18, pp. 866–875. ACM, 2018.
[Bryan 2016] B. A. Bryan et al. Land-use and sustainability under intersecting global change and domestic policy scenarios: Trajectories for Australia to 2050, Global Environmental Change, Volume 38, pp. 130-152, 2016.
[Cyganiak 2014] R. Cyganiak and D. Reynolds. The RDF Data Cube Vocabulary, 2014, https://www.w3.org/TR/vocab-data-cube/.
[Giuliani 2020a] G. Giuliani et al. Knowledge generation using satellite Earth Observations to support Sustainable Development Goals (SDG): a use case on Land Degradation. International Journal of Applied Earth Observation and Geoinformation, 2020.
[Giuliani 2020b] G. Giuliani et al. Essential Variables for environmental monitoring: What are the possible contributions of Earth Observation Data Cubes, Data 5(4):100, 2020.
[Harbelot 2013] B. Harbelot et al. Continuum: A spatiotemporal data model to represent and qualify filiation relationships. 4th ACM SIGSPATIAL International Workshop on GeoStreaming, pp. 76–85. ACM, 2013.
[Hogan 2020] A. Hogan et al. Knowledge Graphs. CoRR 2020.
[Huang 2019] C. Huang et al. MiST: A Multiview and Multimodal Spatial-Temporal Learning Framework for Citywide Abnormal Event Forecasting. In The World Wide Web Conference (WWW ’19)
[Krispin, 2019] R. Krispin. Hands-On Time Series Analysis with R: Perform time series analysis and forecasting using R. Packt Publishing Ltd.
[Le 2008] Q Le  et al. Land-Use Dynamic Simulator (LUDAS): A multi-agent system model for simulating spatio-temporal dynamics of coupled human–landscape system. I. Structure and theoretical specification, Ecol. Informatics 3(2), pp. 135-153, 2008.
[Millington 2016] J. D.A Millington et al. (Guest Editors). Special Issue “Agent-Based Modelling and Landscape Change”, Land 2016.
[Nativi 2020] S. Nativi et al. Towards a knowledge base to support global change policy goals, International Journal of Digital Earth, 13:2, 188-216. 2020.
[Noël 2017] D. Noël et al. Design patterns for modelling life trajectories in the semantic web. In International Symposium on Web and Wireless Geographical Information Systems, pp.51-65. Springer, 2017.[
Parent 2013] C. Parent et al. Semantic trajectories modeling and analysis. ACM Computing Surveys, vol. 45, no. 4, pp. 1-42, 2013.
[Welty 2006] C. Welty et al.  A reusable ontology for fluents in OWL. In FOIS, volume 150, pp. 226–236, 2006.