Tropical forest mapping using polarimetric and interferometric SAR data

K.U. Prakoso

Research output: Thesisinternal PhD, WU

Abstract

A study was made of the potential of the combined use of C-, L- and P-band polarimetric and C- and L-band interferometric airborne SAR data for tropical forest mapping. These data were collected with the NASA/JPL AirSAR during the PacRim-2 2000 campaign in Indonesia. The Sungai Wain forest reserve and its surrounding area near Balikpapan city, located in the province of East-Kalimantan, Indonesia, was chosen as study site. This site covers an area of 10 x 60 km2, which consists of a wide variety and complex mosaic of vegetation and land cover types. Extensive and detailed ground data measurements were made.

The approach presented in this thesis includes new elements such as (1) slope correction, using InSAR, (2) mapping, using a new reversible transform technique and (3) Iterated Conditional Modes (ICM), using prior knowledge such as height and texture. The C-band InSAR DEM allowed for a correction of the disturbing effects of relief on the backscatter level. The new method based on a reversible transform of the covariance matrix was introduced to describe the full polarimetric information of land cover type target properties, allowing for the development of simple and robust classifiers. The ICM approach was extended using additional information such as 3-D textural information derived from the InSAR DEM. This approach was demonstrated using a substantial ground truth data set of land cover observations; the result was then validated using a large independent data set with a different legend structure. The results show that most land cover types can be accurately mapped. The land cover classification result reached 88.9% accuracy for a commonly used legend, while for the independent data set, using a 'radar' legend with more classes, the result improved to 93.8%. Validation of the results by inter-comparison provided large consistency. The best results of land cover type classification are obtained for the C- and P-band and for the C-, L- and P-band fully polarimetric combinations, for which the additional use of relief correction and texture had no noticeable effects. Biomass data were collected for a large number of forests transects and several non-forest plots. The linear relationships between biomass and the radar responses in C-, L- and P-band were not strong. It is evident that this result is caused by the complexity of vegetation, frequent forest fires and the effect of radar saturation at a certain biomass level.

his study has provided the first experiences with this kind of new technology and methodology in Indonesia. It may offer a substantia) contribution to the developments of similar approaches for tropical forest regions in general, especially in those areas where it is difficult to obtain data from optical sensors.

Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • Wageningen University
Supervisors/Advisors
  • Feddes, Reinder, Promotor
  • Hoekman, Dirk, Co-promotor
Award date3 Jan 2006
Place of PublicationWageningen
Publisher
Print ISBNs9789085043140
Publication statusPublished - 2006

Keywords

  • mapping
  • forest inventories
  • remote sensing
  • imagery
  • indonesia
  • polarimetry
  • interferometry
  • classification

Fingerprint

Dive into the research topics of 'Tropical forest mapping using polarimetric and interferometric SAR data'. Together they form a unique fingerprint.

Cite this