INTEGRATION OF CANOPY HEIGHT INFORMATION DERIVED FROM STEREO IMAGERY WITH SAR BACKSCATTER DATA TO IMPROVE BIOMASS MAPPING

Abstract:

Accurate forest biomass estimation over large areas is important for studies of global climate change and the carbon cycle. Synthetic Aperture Radar (SAR) is known to be effective for assessing forest biomass. SAR penetrates farther into forest canopies than optical sensors, so SAR data from forested areas can be related to standing woody biomass, especially at longer L and P bands wavelength. The effect of forest structure on radar signature reduces its sensitivity to biomass when the biomass reaches a threshold level (e.g. ~100Mg/ha at L-band). Therefore the ability for forest biomass mapping using only backscattering coefficients is limited. However, including height data in forest biomass mapping using SAR data will improve the sensitivity beyond saturation levels.

There are many ways to get information related to forest canopy height including: 1) Lidar, a direct measurement of canopy height; 2) Height of scattering phase center (HSPC) from InSAR; 3) HSPC difference from two bands of InSAR, and 4) Polarimetric Interferometric SAR, which employs the polarization-dependent coherences. Photogrammetry (or stereo imagery) is another technique for quantifying forest vertical structure and is a traditional technique for the extraction of a digital surface model. The launch of spaceborne sensors, the application of digital cameras, the maturation of photogrammetry theory and the development of fully digital and automatic image processing make the application of photogrammetric methods feasible. Our previous studies using ALOS PRISM data have shown that the canopy height derived from PRISM stereo data were highly correlated with LVIS RH50 data.

In this study we have integrated this canopy height with L-band SAR imagery data to map forest biomass in our test site in Howland, Maine. The point cloud data from multi-pair stereo imageries of five PRISM scenes were co-registered and used along with the USGS NED data to calculate the mean canopy height at 30m pixels. Multi-polarization L-band SAR backscattering data and LANDSAT NDVI and VCF data were processed into the same pixel size. Field biomass data were then used in both multi-variable regression and random forest models to map forest biomass commonly covered by these imagery data. The results were compared with the biomass map generated using airborne LVIS lidar data.