Characterizing Vegetation 3D structure Globally using Spaceborne Lidar and Radar.

We characterized global vegetation 3D structure using ICEsat-I/Geoscience Laser Altimeter (GLAS) and improved spatial resolution using ALOS/Phased Array L-band Synthetic Aperture radar (PALSAR) data over 3 sites in the United States. GLAS is a 70m footprint lidar altimeter sampling the ground along-track every 170m with a track separation near the equator around 30km. Forest type classes were initially defined according to the Global Land Cover 2000 map (GLC2000), and 5-degree latitude intervals. This strategy enabled analysis of canopy structure as a function of land cover type and latitude. This produced an irregular grid geographically consistant with GLC2000. To estimate canopy height we removed the ground component from the lidar waveform and computed the centroid of the component due to the forest canopy. Canopy height within a grid cell was produced by computing the weighted mean of the GLAS estimates contained within that cell. The weights were used to reduce the impact of slope on Lidar height estimation errors. Slope is the single most significant source of error when estimating height with a large footprint lidar. It stretches the waveform and causes false estimates of canopy height. The Shuttle Radar Topography Mission (SRTM) elevation data was used to derive slope and weights. Thus, data points located in flat areas were assigned a higher weight than points located in slopes. For each forest type, we modeled the relationship between Lidar-estimated canopy height and five environmental variables: temperature, precipitation, slope, elevation, and anthropogenic disturbance. This ecological model was constructed using the machine learning method Random Forest, due to its flexibility and non-parametric nature. Model accuracy was calculated by subsampling the Lidar data set: using 75% of the data set to produce the map previously described and the remaining 25% for validation. This approach was chosen to characterize individual forest canopy types and their structure changes as a function of latitude. Finally, we used PALSAR data to improve spatial resolution over 3 test sites in the United States. We processed 25 PALSAR interferometric pairs. Although the impact of temporal decorrelation (46 days repeat cycle) on the PALSAR interferometric pairs was significant, we derived an empirical regression between GLAS estimates, the interferometric correlation and polarimetric ratios. This regression was used to improve spatial resolution in Bartlett, Howland and Sierra Nevada forests. This study is directly related to the DESDynI mission as recommended by the National Research Council Decadal Survey.