WP4: Agent-based model development
In WP 4, agent-based modelling will be used to explore path-dependency and [other] mechanisms which can create lock-ins. The modelling will be part of the effort to integrate qualitative and quantitative knowledge from WPs 1, 2, 3 and 6, in this case, providing a formal context in which to analyse the logical consequences of combinations of observations under different scenarios.
The first task in this work package will be to arrange a workshop in which researchers from different disciplines together will identify the important actors in the case study, how actors of the same type differ from each other and how the decisions of one actor can influence the opportunities and decisions of others. The types of agents in the model (representing actors in the case study) could include farmers (e.g. beef, dairy and mixed production systems), retailers, supporting businesses, and consumers. However, precise specification of the model will depend on the outcome of this workshop, and on interactions with other researchers in the project as it progresses.
For each type of agent, a simple rule-based decision model will be developed using both quantitative data and expert knowledge. Examples of quantitative data can be econometric data from the Norwegian farm accounting records and quantitative data on strategies and behaviour of farmers e.g. from the “Trender i norsk landbruk”-project. Statistics Norway (SSB) and the Norwegian Agricultural Authority (SLF) provide data on land-use, subsidies, land-use, production
of beef and milk and the distribution of the production on different farming systems.
Farm level data on natural resources, including soil and climatic conditions, are available for some areas (see Skjelvåg et al. 2012) and these data will be used as a basis for estimating productivity and climate gas emissions in the different farming systems. For the different farming systems in the model, GHG emissions will be estimated based on the already developed Holos model adapted for Norwegian beef and dairy farming (Kröbel et al. 2012, Bonesmo et al. 2013).
The Holos model will be integrated so that it can run seamlessly within the agent-based model.
The farming system and the farming activities on individual farms will be described in the model so that it can be changed or adjusted based on decisions made by the farmers. The description of the farming system and activities has to be specific about what technologies are used, required quantities of different input factors (e.g. fertilizer, feed concentrates and fuel for machinery) and the economic costs associated with these choices. The technical description of farming activities in the model will be based on previous farm-level models developed for beef and dairy farming in Norway (e.g. Hovstad 2012, Roer et al. 2013).
Calibration and validation of agent-based models is an on-going area of research, and depends on the type of model built. Examples include Moss and Edmonds (2005) ‘cross-validation’ approach, in which the micro level is validated descriptively and the macro level quantitatively using statistical tests. For agent-based models, the descriptive representation is as, if not more important than quantitative fit, especially if using stylised models, or exploring scenarios for which no data are available (Polhill et al. 2010).
Based on input from the other work packages in the project, the model will be used to run different scenario experiments that explore possible lock-in situations and transformations towards a system with a smaller carbon footprint.
WP Leader, NIBIO