UCSB Biogeography Lab

UCSB Biogeography Lab Publications Abstracts

MODELING ERRORS IN DIGITAL LANDUSE/LANDCOVER MAPS

PAINHO, MARCO OCTAVIO TRINDADE, 1992
PhD Dissertation, Department of Geography, University of California, Santa Barbara. 171 pp.

Landuse/landcover maps and in particular vegetation maps play a fundamental role in Geographic Information Systems (GIS) that are oriented towards natural resource management. It is essential that vegetation representations in a database be adequate and accurate for any particular GIS application. The research presented in this dissertation addresses the problem of attribute accuracy of vegetation maps in its spatial and non-spatial components. The methodology used intensively analyzes and intensively dissects a small-scale vegetation map in order to show which and to what degree the different factors contribute to map error. Positional error is addressed only in terms of digitizing error and its possible effect on attribute accuracy. The research addresses the relationships among error, generalization, polygon homogeneity, and map resolution for a small-scale vegetation map (CALVEG). The spatial distribution of map errors is related to production and processing errors (such as digitizing), boundary effects, spatial distribution and spatial autocorrelation and their association with different vegetation formations and environments. The results show that polygon mis-identification is the major source of attribute error in CALVEG. Use of a simpler classification scheme derived by class aggregation greatly improves classification accuracy. While class aggregation generally increases attribute accuracy, spatial generalization of ground reference data does not increase agreement of these data with CALVEG, suggesting that the error source is polygon mis-labeling, rather than excessive generalization. The study of generalization shows that the number of vegetation types rapidly decreases as the minimum mapping unit increases. Analysis of spatial dependence shows map errors to be positively autocorrelated up to distances of 7 km and random for distances greater than 7 km. Map errors are mainly associated with Conifer and Hardwood Forest and Woodland types on the east and west sides of the Sierra Nevada. Boundary effects don't play a very significant role in map error. Although digitizing error exceeds in some cases the normally accepted error, it is not very significant given the undefined nature of most vegetation boundaries and the small scale of the analyzed map.

Go to UCSB Biogeography Lab home page



What's New Research & Projects Remote Sensing & GIS Landscape Ecology Conservation Planning Publications People Data Links	\| Home \| Contact \| UCSB \| Bren \| ICESS \|

	UCSB Biogeography Lab Publications Abstracts MODELING ERRORS IN DIGITAL LANDUSE/LANDCOVER MAPS PAINHO, MARCO OCTAVIO TRINDADE, 1992 PhD Dissertation, Department of Geography, University of California, Santa Barbara. 171 pp. Landuse/landcover maps and in particular vegetation maps play a fundamental role in Geographic Information Systems (GIS) that are oriented towards natural resource management. It is essential that vegetation representations in a database be adequate and accurate for any particular GIS application. The research presented in this dissertation addresses the problem of attribute accuracy of vegetation maps in its spatial and non-spatial components. The methodology used intensively analyzes and intensively dissects a small-scale vegetation map in order to show which and to what degree the different factors contribute to map error. Positional error is addressed only in terms of digitizing error and its possible effect on attribute accuracy. The research addresses the relationships among error, generalization, polygon homogeneity, and map resolution for a small-scale vegetation map (CALVEG). The spatial distribution of map errors is related to production and processing errors (such as digitizing), boundary effects, spatial distribution and spatial autocorrelation and their association with different vegetation formations and environments. The results show that polygon mis-identification is the major source of attribute error in CALVEG. Use of a simpler classification scheme derived by class aggregation greatly improves classification accuracy. While class aggregation generally increases attribute accuracy, spatial generalization of ground reference data does not increase agreement of these data with CALVEG, suggesting that the error source is polygon mis-labeling, rather than excessive generalization. The study of generalization shows that the number of vegetation types rapidly decreases as the minimum mapping unit increases. Analysis of spatial dependence shows map errors to be positively autocorrelated up to distances of 7 km and random for distances greater than 7 km. Map errors are mainly associated with Conifer and Hardwood Forest and Woodland types on the east and west sides of the Sierra Nevada. Boundary effects don't play a very significant role in map error. Although digitizing error exceeds in some cases the normally accepted error, it is not very significant given the undefined nature of most vegetation boundaries and the small scale of the analyzed map. Go to UCSB Biogeography Lab home page
	Email stoms@bren.ucsb.edu