Modelling dynamic landscapes with Spatial, Temporal And Multiscale Primitives – STAMP

Computer modelling of systems in space and time is common practice in many scientific disciplines. It allows for instance the verification of the knowledge we have of a system, and helps to better understand how the system works while aiming at predicting its evolution according to different scenarios. When the system considered is a landscape, a scene where plant, animal and human life interplays, modelling can be applied to help analyse a variety of important issues facing society today, such as the degradation of natural ecosystems (biodiversity loss), the emergence and spread of new diseases (environmental constraints, climatic changes) or uncontrolled urban growth (economic migrations). However, problems like these are difficult to tackle because of hidden complexities resulting from multiple sources that include processes interacting at different spatial and temporal scales. The overall objective of the present project is to overcome present limitations of dynamic landscape modelling. We hypothesize that many of these problems (scaling, time, field/entity, data quality, and fuzziness) are interlinked and should not be dealt with independently from each other, as it is currently being done. We are also proposing that a key to these problems is to better design spatial, temporal and multiscale primitives (STAMP), providing them with a number of basic features that address the main limitations met in existing studies. For example, one such feature would be that a geographic entity is able to contain – and be contained in – other entities, as suggested by the Hierarchy Theory applied in Landscape Ecology, but also other such features as derived from topology, time sequencing, local or domain-centric arrangements, statistical behaviour, etc. In this project, our objectives are: 1) to assemble the basic features that the primitives must possess in order to better model landscape dynamics, 2) to formalise the concepts and develop the practical tools to manipulate them, and 3) to verify their applicability in a variety of real situations. We deliberately choose not to restrict to one specific modelling application, so as to avoid any bias towards the issues relevant to that particular application, but rather consider broadening a trans-disciplinary approach to excerpt, from various ecological models, or various landscape description, what can be generic enough to become a reusable pattern. To reach these objectives, we rely on a close collaboration between modellers, computer scientists specialised in language theory and researchers studying patterns and processes at landscape scale. Our approach consists first in bringing up the basic features, by analysing on one hand various works published in the scientific literature, and on the other hand, four real cases in which our teams are already involved through European and International projects: 1) dynamics of the mangrove ecosystem in Guiana, 2) forest landscapes under human pressure in South India, 3) impact of climate change and variability on small agriculture in West Africa (AMMA), and 4) risk of emergence (avian flu) and spread (Rift Valley fever) of infectious diseases (EDEN, ECOFLU). The internal structure of the primitives unifying these basic features will then be designed by capitalising on the recent advances in the field of information science such as meta-modelling, interoperability and Model-Driven Engineering. This second step will result in a domain-specific language (DSL) capable of carrying appropriate semantics to manipulate the new spatio-temporal modelling primitives. Finally, different application prototypes will be developed using the DSL, inspired by the abovementioned real situations. This work will inevitably involve a to-and-fro exchange between the applications and the design phase, under the sanction of software feasibility. This project should be considered as exploratory, where the focus is on building the modelling formalism. Its full scale application on the four real cases, and possibly others, would be carried out in future larger projects. Overall, we expect this work to make a positive contribution towards providing modellers with a more appropriate formalism together with accessible tools to manipulate the spatio-temporal and multi-scale concepts that are necessary to model challenging landscape dynamics.