Estimation of change in forest variables using synthetic aperture radar

Abstract: Large scale mapping of changes in forest variables is needed for both environmental monitoring, planning of climate actions and sustainable forest management. Remote sensing can be used in conjunction with field data to produce wall-to-wall estimates that are practically impossible to produce using traditional field surveys. Synthetic aperture radar (SAR) can observe the forest independent of sunlight, clouds, snow, or rain, providing reliable high frequency coverage. Its wavelength determines the interaction with the forest, where longer wavelengths interact with larger structures of the trees, and shorter wavelengths interact mainly with the top part of the canopy, meaning that it can be chosen to fit specific applications. This thesis contains five studies conducted on the Remningstorp test site in southern Sweden. Studies I – III predicted above ground biomass (AGB) change using long wavelength polarimetric P- (in I) and L-band (in I – III) SAR data. The differences between the bands were small in terms of prediction quality, and the HV polarization, just as for AGB state prediction, was the polarization channel most correlated with AGB change. A moisture correction for L-band data was proposed and evaluated, and it was found that certain polarimetric measures were better for predicting AGB change than all of the polarization channels together. Study IV assessed the detectability of silvicultural treatments in short wavelength TanDEM-X interferometric phase heights. In line with earlier studies, only clear cuts were unambiguously distinguishable. Study V predicted site index and stand age by fitting height development curves to time series of TanDEM-X data. Site index and age were unbiasedly predicted for untreated plots, and the RMSE would likely decrease with longer time series. When stand age was known, SI was predicted with an RMSE comparable to that of the field based measurements. In conclusion, this thesis underscores SAR data's potential for generalizable methods for estimation of forest variable changes.

  This dissertation MIGHT be available in PDF-format. Check this page to see if it is available for download.