Cover- and density-based vegetation classifications of the Sonoran Desert using Landsat TM and ERS-1 SAR imagery

Scott M. Shupe, Stuart E. Marsh

Research output: Contribution to journalArticlepeer-review

44 Scopus citations


Arid lands are distinctive ecological zones that require vegetation maps for management and monitoring. The use of remote sensing for mapping desert vegetation is made difficult by the mixing of reflectance spectra of bright desert soils with the relatively weak spectral response of sparse vegetation. To investigate ways to improve desert vegetation mapping, a comparison of the effect of supervised classification using two contrasting measures of field vegetation data as reference data was performed. We took cover- and density-based field vegetation data that had been collected by the US Army on the US Yuma Proving Ground (USYPG) in southwest Arizona, converted them into cover- and density-based reference classification schemes and used them to train both maximum likelihood (ML) and artificial neural net (ANN) classifiers. The impact on the accuracy of cover- and density-based vegetation maps were further analyzed using different combinations of input data (i.e., Landsat Thematic Mapper (TM) imagery, ERS-1 C-band synthetic aperture radar (SAR) imagery, and elevation data). In spite of the fact that a cover-based plot classification is the logical training data for remote sensing classification, both cover- and density-based classified maps had similar accuracies for each data combination. The use of all data combinations gave the highest map classification accuracies, with the radar data improving the accuracy the most where the vegetation is dense. Classification accuracies of maps using the ML classifier were generally higher than those using the ANN classifier. ANN map classification accuracies improved significantly when the sigmoid transfer function was replaced with the hyperbolic tangent transfer function. Using the two contrasting measures for mapping proved complementary: the cover-based map located areas of significant tree presence that were not mapped on the density-based map and the density-based map located areas of significant cacti presence that were not mapped on the cover-based map. Creating both cover- and density-based vegetation maps may therefore better assist arid land management than creating only a cover-based vegetation map.

Original languageEnglish (US)
Pages (from-to)131-149
Number of pages19
JournalRemote Sensing of Environment
Issue number1-2
StatePublished - Oct 30 2004


  • Artificial neural network
  • ERS-1 SAR
  • Landsat TM
  • Maximum likelihood
  • Sonaran Desert
  • Vegetation cover
  • Vegetation density

ASJC Scopus subject areas

  • Soil Science
  • Geology
  • Computers in Earth Sciences


Dive into the research topics of 'Cover- and density-based vegetation classifications of the Sonoran Desert using Landsat TM and ERS-1 SAR imagery'. Together they form a unique fingerprint.

Cite this