Radiometric slope correction for forest biomass estimation from SAR data in the Western Sayani Mountains, Siberia

Abstract

We investigated the possibility of using multiple polarization (SIR-C) L-band data to map forest biomass in a mountainous area in Siberia. The use of a digital elevation model (DEM) and a model-based method for reducing terrain effects was evaluated. We found that the available DEM data were not suitable to correct the topographic effects on the SIR-C radar images. A model-based slope correction was applied to an L-band cross-polarized (hv) backscattering image and found to reduce the topographic effect. A map of aboveground biomass was produced from the corrected image. The results indicated that multipolarization L-band synthetic aperture radar (SAR) data can be useful for estimation of total aboveground biomass of forest stands in mountainous areas.

Introduction

Several authors have developed methods and algorithms for mapping aboveground biomass in the boreal forest Beaudoin et al., 1994, Bergen et al., 1998, Dobson et al., 1992, Dobson et al., 1995, Kurvonen et al., 1999, Le Toan et al., 1992, Paloscia et al., 1999, Ranson et al., 1995, Ranson & Sun, 1997a, Rignot et al., 1994, Saatchi & Moghaddam, 2000, Saatchi et al., 1995. These studies concentrated on relatively flat areas, where terrain effects were not significant. Estimation of forest biomass using synthetic aperture radar (SAR) data can be complicated by topography that influences radar backscatter Bayer et al., 1991, Luckman, 1998, Rauste, 1990, van Zyl, 1993, particularly through local incidence angle, shadowing, and effects on radar backscattering can be complex. Changes in radar incidence angle caused by terrain slope can have several effects on radar image data. For example, radar backscattering varies with incidence angle, which varies with terrain slope and aspect. Foreshortening is also a terrain-induced effect where a smaller incidence angle results in more ground surface area being illuminated. If the terrain slope is larger than the radar incidence angle, layover occurs and the backscattering from the slope will mix with the signature from other targets. When a slope faces away from the radar and the slope is steeper than the incidence angle, shadowing occurs. There is no way to recover the signatures lost due to layover and shadowing. Another effect of terrain on the backscatter is the apparent change of the forest spatial structure in the radar field of view. For example, when trees of a relatively uniform stand grow on a slope, a portion of the sides of these trees will be directly exposed to the radar beam.

Terrain correction techniques are designed to reduce effects of incidence angle and illuminated target area. For correction of the illuminated pixel area, simple algorithms can be used if a suitable digital elevation model (DEM) exists Kellndorfer et al., 1998, Shi & Dozier, 1997. Correction of the backscattering dependence on incidence angle requires knowledge of the land cover type within a pixel. A few attempts have been made to correct terrain effects by using simple radar backscattering models and a DEM. For example, Goering, Chen, Hinzman, and Kane (1995) used a DEM and empirical radar backscatter models to reduce terrain effects from ERS-1 SAR images. However, Goyal, Seyfried, and O'Neill (1998) found that the small-scale topographic features resolved by SAR could not be resolved by a DEM in rugged terrain. Periodic artifacts due to the terrain model generation methodology were observed in the derived variables (e.g., slopes). Other methods, such as image ratios, were used to reduce the effects of radar incidence angle caused by topography Ranson et al., 1995, Ranson et al., 2000, Shi & Dozier, 1997, Wever & Bodechtel, 1998. Wever and Bodechtel (1998) proposed the use of L-band hv (Lhv) and X-band VV (Xvv) ratio or difference images for radiometric rectification. Ulander (1996) described a new equation for radiometric slope correction using the slope information derived from the SAR interferograms.

In the work reported herein, we discuss the correction for the dependence of illuminated pixel area on incidence angle using a DEM. We then present a method to correct for the backscatter dependence on terrain using a model to simulate radar backscattering of a forest stand on various slopes. The derived dependence of the L-band hh (Lhh) and Lhv backscattering on radar local incidence angle was used to remove the terrain effect from the Lhv data. Finally, a biomass map was produced from the corrected Lhv data.