Abstract
Integration of multisensor data provides the opportunity to explore benefits emanating from different data sources. A fusion between fraction images derived from spectral mixture analysis of Landsat-7 ETM+ and phased array L-band synthetic aperture radar (PALSAR) is introduced. The aim of this fusion is to improve the estimation accuracy of above-ground biomass (AGB) in lowland mixed dipterocarp forest. Spectral mixture analysis was applied to decompose a mixture of spectral components of Landsat-7 ETM+ into vegetation, soil, and shade fractions. These fraction images were integrated with PALSAR data using the discrete wavelet transform (DWT) and Brovey transform. As a comparison, spectral reflectance of Landsat-7 ETM+ was fused directly with PALSAR data. Backscatter of horizontal–horizontal and horizontal–vertical polarizations was also used to estimate AGB. Forest inventory was carried out in 77 randomly distributed plots, the data being used for either model development or validation. A local allometric equation was applied to calculate AGB per plot. Regression models were developed by integrating field measurements of 50 sample plots with remotely sensed data, e.g. fraction images, reflectance of Landsat-7 ETM+, and PALSAR data. The models developed were validated using 27 independent sample plots. The results showed that not all fused images significantly improved the accuracy of AGB estimation. The model based on Brovey transform using the reflectance of Landsat-7ETM+ and PALSAR produced an R2 of only 0.03–0.10. By contrast, fusion between PALSAR data and fraction images using Brovey transform improved the accuracy of R2 to 0.33–0.46. Further improvement in the accuracy of estimating AGB was observed when DWT was applied to integrate PALSAR with the reflectance of Landsat-7ETM+ (R2 = 0.69–0.72) and PALSAR with fraction images (R2 = 0.70–0.75).
Acknowledgements
We thank B.T. Sasongko, Bogor Agriculture University, for helpful discussions on PALSAR. The authors are grateful to Dr M.J. Canty for the DWT script, W. Nieuwenhuis for technical assistance, and my colleague, Claudia Pittiglio, for an in-depth discussion on wavelet theory. This research was supported by a grant from The Netherlands Fellowship Programme (NFP).