4.03.02 - Machine learning and computational ecology
Technological developments in data acquisition have shifted the focus in forest modelling from sampling-based approaches to population-based methods. In these new settings, the equation-based models are supplemented with statistical learning techniques. Irrespective of the modelling approach, the computational aspects of the modelling process adds a new dimension to the complex relationships to be represented with equations or algorithms. These data offer new challenges to modelling spatially explicit ecological processes not well addressed in the broader statistical learning community.
The objective of the 4.03.02 unit is to foster conversation among scientists interested in complex modelling using advanced statistical and computational methods and to provide a platform for researchers and practitioners of forest modelling.
The transition from sample to population models is accompanied by an increase in the number of tools used to answer forest research questions. The traditional equation-based models that relate a set of predictors to a dependent variable, whose epitome are the taper and site index equations, is now accompanied by a plethora of machine learning and heuristic techniques. Furthermore, the approaches focused on one aspect of forest have given way to holistic approaches aimed at understanding ecosystem-level pattern-process linkages. Therefore, hierarchical or simultaneous models are now defining the representation of forest dynamics. Furthermore, stand-level approaches are now either integrated within higher-order landscape models or broken down into smaller modelling units, such as individual trees to inform process-based landscape simulation models.
In the current social and scientific settings, the hot topics addressed by the 4.03.02 unit are:
- Development of parsimonious multiscale models
- Spatially explicit models of the development of tree components as a part of the forest ecosystem
- Integration of remote sensing data into landscape modelling
- Machine learning modelling for the exploration and prediction of landscape disturbance patterns
- Spatial decision support tools to integrate model predictions into transparent and quantitative management decision making.
- Application of landscape simulation models to facilitate hypothesis testing of small-scale ecological processes