Validation of land use change models - A case study on the Environment Explorer

Laboratory of Geo-Information Science and Remote Sensing
  Education
  Research
  Publications
  Models
  News & Calendar
  News
  Calendar
  Archive
  News
  Calendar
  2011
  2010
  2009
  2008
  2007
  2006
  2005
  2012
  Staff
  Equipment
  Contact details
  Workshops

8 Mar 2006 09:30 - 8 Mar 2006 10:00
Unit: Wageningen UR
Location: room Gaia 1, Droevendaalsesteeg

 By: Jasper van Vliet  

Abstract 
Land use and land cover change models are tools to analyze, simulate or predict the dynamics in land use and land cover. They are widely used in planning and policy making to explore scenarios. Technical issues of these models have been receiving much research effort since computer power has been increasing rapidly last decades. Model analysis and evaluation however received less attention. Due to a lack of proper techniques and methods, there is no standard or guideline to evaluate these models. In this thesis a guideline for operational validation of these models is proposed. A goodness of fit measure as well as a measure for patterns and complexity was applied on the case study model, the Environment Explorer.
The development of a model is an iterative process. In this process the following phases can be distinguished: the set-up phase, the conceptualization phase, the model construction phase, the evaluation phase, the use phase, and finally the reporting phase. While evaluation is a part of the model development cycle, distinct parts of the modeling process can again be evaluated. In this thesis only the validation of model results, operational validation, is considered. As proposed guideline, the model should perform better than the null model and a random model. As well, the sources of errors should be investigated. In what way it should perform better is depending on the desired functionality.
The Environment Explorer is a discrete model in both space and time. It aims at exploring possible future land use for a period up to 30 years ahead. For this research the model was calibrated in two ways, automated and by hand, with land use data of 1960 and 1980. To validate, the 2000 land use cover was simulated starting from the 1980 land use cover. Results were compared to the 2000 land use cover map.
Goodness of fit was measured with Kappa statistics. Because of the model’s exploring character, two other methods were applied as well: Fuzzy Kappa statistics and landscape metrics. Four landscape metrics were selected: fractal dimension, shape index, perimeter area scaling and Shannon’s diversity index. Metrics were compared globally and focally.
Fuzzy Kappa statistics showed only minor differences as compared to the standard Kappa statistic. Landscape metrics did reveal different results. It was concluded that landscape metrics are a useful addition to traditional validation based on accuracy measurements when applied on explorative models. Case study results on the Environment Explored showed as well, that is hard to outperform the null model when only a very limited amount of information was used. This was concluded for both simulations and both comparison methods. The automatically calibrated model performed slightly better on the Kappa statistics, whereas the model calibrated by hand performed better on the landscape metrics.
Print this activity