Mapping residential density patterns using multi-temporal landsat data and a decision-tree classifier

S. McCauley, S. J. Goetz

Research output: Contribution to journalArticlepeer-review

41 Scopus citations

Abstract

We examined the utility of Landsat Thematic Mapper (TM) imagery for mapping residential land use in Montgomery County, Maryland, USA. The study area was chosen partly because of the availability of a unique parcel-level database of land use attributes and an associated digital map of parcel boundaries. These data were used to develop a series of land use classifications from a combination of leaf-on and leaf-off TM image derivatives and an algorithm based on ‘decision tree’ theory. Results suggest potential utility of the approach, particularly to state and local governments for land use mapping and planning applications, but greater accuracies are needed for broad practical application. In general, it was possible to discriminate different densities of residential development, and to separate these from commercial/industrial and agricultural areas. Difficulties arose in the discrimination of low-density residential areas due to the range of land cover types within this specific land use, and their associated spatial variability. The greater classification errors associated with these low-density developed areas were not unexpected. We found that these errors could be mitigated somewhat with techniques that consider the mode of training data selection and by incorporation of methods that account for the presence and amount of impervious surfaces (e.g. pavement and rooftops).

Original languageEnglish (US)
Pages (from-to)1077-1094
Number of pages18
JournalInternational Journal of Remote Sensing
Volume25
Issue number6
DOIs
StatePublished - Mar 2004
Externally publishedYes

ASJC Scopus subject areas

  • General Earth and Planetary Sciences

Fingerprint

Dive into the research topics of 'Mapping residential density patterns using multi-temporal landsat data and a decision-tree classifier'. Together they form a unique fingerprint.

Cite this