Development of a Multi-Temporal Remote Sensing Classification Methodology for Nature Classes in the Dutch Land-use Database: A Phenology-Based Approach

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

18 Dec 2008 09:00 - 18 Dec 2008 09:30
Unit: Laboratory of Geo-Information Science and Remote Sensing
Location: Gaia 2
Organisation: Wageningen University

 By Hillebrand Dalstra
Abstract:
A multi-temporal, phenology based, land cover classification methodology was developed for the LGN forest and heath nature classes using both phenological MODIS data and Landsat TM satellite imagery. The objective of the research was to assess the opportunities for a multi-temporal phenological based remote sensing classification. Furthermore, the aim of this study was to investigate which classification approach would achieve an overall classification accuracy at acceptable level and to evaluate the accuracy of the multi-temporal phenological based classification.
VI time series and phenological indicators were analyzed to identify important stages in the selected forest and heath classes’ phenological cycles. The analysis of VI time series and the VI ratio based phenological indicators revealed differences between the two VIs, whereas the main difference being their sensitivity to vegetation changes. The EVI based indicators appear to be more sensitive for variances in the vegetation and performed better with the limiting temporal factor. The results supported the classification phase, where The Landsat TM images were used to further differentiate and classify the selected nature classes.
The supervised, EVI multi-temporal, maximum likelihood, classification was determined to be satisfactory with an overall accuracy of 71 % for heath with the kappa coefficient 0.62 and the forest accuracy of 90% with the kappa coefficient of 0,78 and thereby outperforming the single EVI classification and the single 6 band Landsat TM classification. The comparison of the supervised, EVI multi-temporal, maximum likelihood, classification with the original LGN 5 database demonstrated an overall accuracy increase of 5%. The use of multi-temporal imagery captured the phenological differences of the nature classes and this information was used to produce superior classification results.
Keywords: LGN, land cover, multi-temporal, remote sensing, phenology, MODIS, Landsat, EVI, NDVI, ratio method, forest, heath.
Print this activity