nt 
id 
  
  
Felkner, John 
A SPATIAL ECONOMIC AND ENVIRONMENTAL COMPUTER PREDICTIVE MODEL OF LAND USE 
CHANGE IN THAILAND USING A STATISTICAL TREE CLASSIFICATION APPROACH 
John FELKNER 
Harvard University 
Graduate School of Design 
gsd97jjf @ gsd.harvard.edu 
Working Group VI/4 
KEY WORDS: Spatial, Land Use Change, Tree Classification, GIS, Remote Sensing, Thailand, Predictive. 
ABSTRACT 
This approach integrates the tools of Geographic Information Systems (GIS) and remotely sensed imagery to map land 
use change in two Provinces of Thailand from 1979 to 1999, and then uses a statistical tree classification model to 
predict future change. 
1 INTRODUCTION 
As population growth, conversion of land to agricultural uses, urbanization and deforestation have continued to grow at 
every increasing rates — especially in developing countries — interest in and the need for quantitative empirical models 
of deforestation and urbanization has grown in recent years (Kaimowitz and Angelsen 1998). 
The two Provinces used in this study are Chachoengsao and Sisaket, which were chosen because they are representative 
of both economic and environmental variation within Thailand as a whole. Measured land use change from 1979 to 
1989 — mapped in a raster GIS environment — is used as the dependent variable, and eight spatial GIS models 
representing socioeconomic and environmental factors are used as the independent variable predictors. The validity of 
the resulting statistical models to predict future change are evaluated by GIS comparison with existing 1999 land use, 
derived from Landsat satellite imagery. 
This approach complements other statistical approaches in spatial deforestation and urbanization models by using a 
statistical tree classification approach (Kaimowitz and Angelsen 1998). The tree approach has a number of distinct 
advantages over more traditional statistical approaches, primarily in that it does not require normalization of input 
variables and provides an automated approach both for distinguishing existing homogeneous partitions and creating a 
hierarchy of predictive variable priorities in large disparate datasets (Venables and Ripley 1994; O'Connor, Jones et al. 
1996). 
2 DESCRIPTION OF RESEARCH 
2.1 Economic and Environmental Models 
This predictive land use model uses a tree classification analysis performed in Splus software using maps of land use 
change from 1979-1989 as the dependent variable and eight socioeconomic and environmental models as independent 
input factors. The basic hypothesis is that future land use change can be explained best by a combination of economic 
and environmental factors rather than by either economic or environmental factors alone. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 433