Search

Because of the significance of C resident in forest trees, efforts are being directed to better quantify the amount and distribution of this component of the global C cycle. The need for information over large areas has lead to the implementation of remote sensing instruments such as synthetic aperture radar (SAR). This dissertation addresses all of the stages needed to derive the parameters fundamental to assessing C storage and change using a primarily SAR-based methodology. These include: (1) ground reference data, (2) forest classification, (3) biomass estimation, and (4) C estimation. The imagery used is multi-temporal SIR-C/X-SAR imagery of the NASA Michigan Forests Test Site (MFTS) in the Hiawatha National Forest in northern Michigan. Ground data was collected and analyzed for 70 forest test stands. These data include stand structural and biomass statistics used for image classification and construction of biomass, C, and ANPP (above-ground net primary production) estimation functions. A hierarchical structural methodology is used to classify SIR-C/X-SAR scenes to Level II (forest community). Overall unbiased accuracies are compared and results show that multi-temporal (97%) may show improvement over single-date (April 90%, Sept. 95%, and Oct. 98%), and definitely allows good improvement over pooled (April 90%, Sept. 77%) classifications. Empirical biomass estimation algorithms are developed for each structural type and when combined with the classified image, above-ground biomass is mapped in the image domain at 5.79 $\times$ 10$\sp9$ kg in 51,448 ha's. Finally, three components of C storage and change are estimated: (1) C stored in above- and below-ground living vegetation (3.22 $\times$ 10$\sp9$ kg), (2) C gain from ANPP (0.75 to 1.31 kg/m$\sp2$/yr), and (3) C removed by forest harvesting (6.02 $\times$ 10$\sp6$ kg). Results confirm that it is possible to successfully develop a multi-stage approach to estimate C storage and change using SIR-C/X-SAR. imagery as the major data set. This provides a new method for estimating these parameters at the landscape level. The method improves on previous ones by incorporating actual (as opposed to potential) vegetation cover, plus the amount of vegetation, and results in qualifications which are significantly closer to measured calibration values.