Spatial Modeling of Lidar-Derived Woody Biomass Estimates Collected Along Transects in a Heterogeneous Savanna Landscape

Waveforms from the Ice, Cloud and land Elevation Satellite have successfully estimated footprint-level canopy height and aboveground biomass even in structurally complex savanna ecosystems. However, at the landscape level wall-to-wall maps are preferred since they are more easily integrated with other geospatial data products. We evaluated and compared the utility of inverse distance weighting, cokriging, regression kriging and image segmentation methods to create wall-to-wall maps from footprint-level estimates of biomass across a 13 600-Ha Oak savanna landscape in Santa Clara county, California. The four methods estimated biomass with between 39% (inverse distance weighting) and 66% (image segmentation) of variance explained and RMSE of 42% and 32% of the mean, respectively. When more waveforms were available across or along track to characterize patch biomass with the image segmentation method, 78% of variance in biomass was explained (RMSE = 21% of the mean). Overall, the mean biomass estimated by the four methods did not differ significantly but a visual inspection of the output maps showed marked differences in the ability of each model to mimic the primary variable's landscape-level trend. We conclude that transects of lidar data can be used to create wall-to-wall biomass maps in savannas but the methods require a higher sampling intensity and informative auxiliary data to reproduce the variability of the biomass across the landscape. We recommend that future satellite lidar missions increase the sampling intensity across track so that biomass observations are made and characterized at the scale at which they vary.