%0 Thesis
%D 1999
%T Variability of Microwave Backscatter from Loblolly Pine Forest and the Implications for Forest Biomass Estimation with Imaging Radar
%A Day, J. L.
%X The microwave backscatter coefficient (sigma-0) of forest, as measured by imaging radar, varies depending on forest structure and biomass. It may, therefore, be possible to estimate forest biomass and other biophysical properties from radar data. Sigma-0 also varies in response to other forest and radar variables, including ground surface roughness and moisture, forest phenology and radar calibration. These extraneous sources of sigma-0 variation can interfere with biomass estimation. In this dissertation, I examine variability of sigma-0 for loblolly pine stands in Duke Forest, North Carolina, and assess the impact of variability on accuracy of biomass estimation. A microwave canopy backscatter model was used to study how sigma-0 of a forest changes as forest floor properties vary. L- and C-band backscatter was simulated at 3 polarizations and 3 radar incidence angles for pine stands at 3 biomass levels, while 5 ground surface parameters were varied over a range of realistic values, as determined from field data. The surface parameters are litter depth and moisture content, soil RMS height and correlation length, and soil moisture content. For incidence angles of 20-40 deg., L-HH varied by 5.3-9.6 dB as the surface parameters varied over their range, whereas L-VV varied by 3.7-4.5 dB. C-HH and C-VV were sensitive to the surface only at steep incidence (20-30 deg.) for the lowest biomass stand studied. L-HV and C-HV were relatively insensitive to the surface. Variation of actual, measured sigma-0 was examined for C- and L-band backscatter acquired over 21 loblolly stands in the biomass range of 3.5-44.5 kg/@ m sup 2@ during 10 passes of NASA's Shuttle Imaging Radar in April and October, 1994. Within any radar band-polarization combination and data take, the maximum sigma-0 range among the stands was 3.6 dB; in most cases it was 2-3 dB. RMS variation of mean forest sigma-0 for the 10 data takes (after correcting for incidence angle) was 0.4-0.7 dB, which is comparable to the standard deviation of sigma-0 among the 21 stands. Sigma-0 increased @ approx @1 dB when the canopy was wet, but variation of sigma-0 with surface moisture was not separable from other factors. Biomass to sigma-0 correlations were affected by radar look direction and appear unrelated to incidence angle or soil moisture. Multiple linear regressions of biomass versus sigma-0 (adjusted to equalize mean forest sigma-0 among data takes), yielded adjusted R-squared values up to 0.57. The regression models varied, the best model for each data take requiring a different combination of SIR-C bands. RMS error of biomass estimation decreased with the number of bands included in the model for estimates based on the regression data, but increased with the number of bands included for estimates made from data acquired in different shuttle passes. Analysis of the propagation of sigma-0 variance through the linear regression models confirms that estimation error increases with model size and sigma-0 variability.
%G eng
%9 phdPh.D. dissertation