Impact of Similarity Metric Selectionfor Multiclass Scenario Discovery ofLand-Use Change Models

Impact of Similarity Metric Selectionfor Multiclass Scenario Discovery ofLand-Use Change Models

Quispel, Omar (TU Delft Technology, Policy and Management)

Kwakkel, J.H. (mentor)
Comes, M. (graduation committee)
Bramka Arga Jafino, Bramka (graduation committee)
Aydin, N.Y. (graduation committee)

Delft University of Technology

Engineering and Policy Analysis

2021-03-11

Land-use change models are often used to explore future land-use. Currently, most land-use change models run a small number of predetermined scenarios. To better address the multidimensional nature of uncertainty about the future, previous studies have argued for covering a wider range of the uncertainty space than is possible with existing scenario approaches. One way to approach this, is by using exploratory modelling. Instead of running the model on a small number of pre-defined scenarios, one uses sampling techniques over a plausible range of the uncertainty space to generate large-scale simulation experiments. With respect to land-use change models, this results in a large number of future land-use maps. To better make sense of the resulting maps, the next step necessitates the identification of plausible distinctive land-use patterns through clustering algorithms. Previous studies regarding quantification of land-use maps (dis)similarity focus only on comparing maps on a one-to-one basis or in small numbers for validation and calibration purposes. In this study the use of different map of similarity metrics on the resulting clusters of land-use patterns is systematically investigated. Specifically, the implications of using various cell-by-cell similarity metrics and landscape structure similarity metrics to cluster the resulting land-use maps are tested. The Land Use Scanner model is used for this purpose. It was found that the choice of (dis)similarity metric plays a significant role in the formed clusters of maps. If exploratory modelling is to be applied to land-use change models, it is thus important that great care is taken in the selection of the proper clustering algorithm.

Land use change modelling
Scenario Discovery
Exploratory Modelling and Analysis
Similarity Metrics
Multiclass Scenario Discovery
Clustering analysis

http://resolver.tudelft.nl/uuid:f5f7fd17-93cb-43cd-9890-92d0541f5935

Student theses

master thesis

© 2021 Omar Quispel