Deep learning for land use change

Why this workshop?

Predicting land use change is challenging, as it is a social-ecological process driven by complex, cross-scale interactions between social and ecological subsystems. This means predictors of land use change are context dependent, varying by location. The situation is further complicated by the changing nature of social-ecological interactions; past predictors of land use change may not represent current or future predictors 3. This makes it difficult to understand land use change dynamics through conventional statistical or process models.

Agent Based Models (ABMs) enable interactions and behavioural adaptations of ‘agents’ (e.g. decision makers) to be made explicit and socio-economic and biophysical processes to be linked, making them useful for modelling the emergent aspects of land use change. However, deciding the structure, agent behaviours and model parameters for ABMs across large spatial scales is a major challenge due to the lack of adequate data and the often complex nature of empirically-oriented ABMs. As such, ‘hybrid’ models which compare and combine ABMs with other approaches for predicting land use change may be a way forward, to leverage the advantages of multiple modelling paradigm.

DEEPLAND will address this challenge by developing and testing a novel hybrid causal, modelling framework for land use change that links structured machine learning (ML) approaches with domain-specific knowledge on land use change, with a particular focus on Rondônia in the Brazilian Amazon. The workshop builds on a previous workshop in September 2022 in Jena Germany by most of the same participants, as well as on-going modelling and conceptual development since this workshop.

The specific aims of this workshop are to test and update prototype ML approaches developed by consortium members, and improve these, as well as to test and improve methods for assessing performance of the models that capture systemic change in land use dynamics.

What will we do?

The workshop will bring together participants with different backgrounds, including AI/ML, land use change, and theoretical and quantitative ecology.

On the first day, provide a general overview of the project goals and achievements to date for new participants. We will also provide an overview of the project and results of preliminary models and methods for measuring systemic changes in land use dynamics by led by junior researchers on the project. We will then develop as a group an initial list of updates to existing models that we want to explore during the workshop (the models are designed to be updated quickly).

The second day will focus on updating and fully exploring existing modelling approaches to better understand the strengths and limitations of the models.

The final day (morning) will be used to come up with a specific list of post workshop improvements to the models, and a timetable for publishing the results in a group paper.

To keep discussions productive and focused, we aim for 12 to 15 participants.

Deliverables

We will facilitate the writing of a methods/agenda paper (with simple examples) to be published in a general science journal (e.g.Nature Sustainability, PNAS) outlining our finding. It will be co-authored by all interested workshop participants.

Interested?

The event itself is by invitation only and free of charge. Where possible, participants should expect to arrange their own transport and cover their accommodation costs. A few partially- and fully funded places will also be available, which will be allocated according to need, giving preference to early career researchers.

If you wish to participate, please get in touch with Ioannis Athanasiadi.

We hope to welcome you in Wageningen soon.