Potential of multisensor SAR for land use/land cover mapping in Sweden

Md Latifur Rahman Sarker, Yifang Ban, Janet Elizabeth Nichol

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review


The idea of multisensor data application has introduced a new dimension in the field of remote sensing in recent years. Very few attempts, however, have been made to use a multisensor approach, particularly multisensor Synthetic Aperture Radar (SAR) data for land use/land cover mapping. This research investigates the capability of spacebome multisensor SAR data, including RADARSAT fine-beam, RADARSAT standard-beam, ERS-2, and JERS-1, for extracting land use/land cover information in Sweden considering different sensor combinations (single sensor, double sensor, triple sensor and multisensor combinations), different image processing techniques (Raw, texture measures, filtered and combination of texture & filtered measures) and performance of various classification algorithms (MLC, ANN, k-NN, Sequential Masking and Object based classifier e-Cognition). The results demonstrate that despite the potential of multi-temporal single sensor SAR, the double, triple and multisensor SAR has a greater potential for land use/land cover mapping. But the potential of multisensor SAR for land use/land cover mapping depends on the characteristics of the combined sensors themselves, as well as the number of images. Among the different combinations, the best results was achieved using triple sensor combination (RADARSAT fine-beam, ERS-2, and JERS-1) because of its capability of providing the best complementary information, while the second and the third best results were obtained from multisensor combination (RADARSAT fine-beam, ERS-2, JERS-1 and RADARSAT standard-beam) and double sensor combination (ERS-2 and JERS-1) respectively. It is also revealed that the raw images produced very poor results using all combinations and all classifiers due to speckle noise, while the mean texture measures produced the best results using almost all combinations and classifiers. Remarkable variation was found among the performance of different classifiers with respect to different sensor combinations and different image processing techniques but ANN was clearly superior to the other classifiers with respect to all combinations and image processing techniques as a result of its non-parametric nature and its high mathematical base. The results indicate that the pixel-based classifier namely ANN is more accurate (around 90% overall accuracy and 0.90 Kappa coefficient) compared with object-based classification for extracting land use and land cover information from multiple sensor SAR. Overall it was found that the best performance (more than 90% overall accuracy and more than.90 Kappa coefficient) can be achieved using a sequential masking approach because of its step by step classification technique.
Original languageEnglish
Title of host publicationProceedings of the 1st International Postgraduate Conference on Infrastructure and Environment, IPCIE 2009
Number of pages8
Publication statusPublished - 1 Dec 2009
Event1st International Postgraduate Conference on Infrastructure and Environment, IPCIE 2009 - Hong Kong, Hong Kong
Duration: 5 Jun 20096 Jun 2009


Conference1st International Postgraduate Conference on Infrastructure and Environment, IPCIE 2009
Country/TerritoryHong Kong
CityHong Kong


  • Multiple sensors SAR
  • Object-based
  • Pixel-based
  • Sequential masking and ANN

ASJC Scopus subject areas

  • Building and Construction
  • Environmental Science(all)

Cite this