Assessing the uncertainty of biomass change estimates obtained using multi-temporal field, lidar sampling, and satellite imagery on the Kenai Peninsula of Alaska (Invited)

There is increasing interest in the development of statistical sampling designs for aboveground biomass (and carbon) inventory and monitoring programs that can make efficient use of a variety of available data sources, including field plots, airborne lidar sampling, and satellite imagery. While the use of multiple sources, or levels, of remote sensing data can significantly increase the precision of biomass change estimates, especially in remote areas (such as interior Alaska) where it is extremely expensive to establish field plots, it can be challenging to accurately characterize the uncertainty (i.e. variance and bias) of the estimates obtained from these complex multi-level designs. In this study we evaluate a model-based approach to estimate changes in biomass over the western lowlands of the Kenai Peninsula of Alaska during the period 2004-2009 using a combination of field plots, lidar sampling, and satellite imagery. The model-based approach -- where all inferences are conditioned on the model relating the remote-sensing measurements to the inventory parameter of interest (e.g. biomass) - is appropriate for cases where it is cost-prohibitive, or infeasible, to establish a probability sample of field plots that are both spatially and temporally coincident with each remote sensing data set. For example, a model-based approach can be used to obtain biomass estimates over a period of time, even when field data is only available for the current time period. In this study, lidar data were collected in 2004 and 2009 over single swaths that covered 130 Forest Inventory and Analysis (FIA) plots distributed on a regular grid over the entire western Kenai. Field measurements on FIA plots were initially acquired over the period 1999-2003 and fifty-percent of these plots were remeasured in the period 2004-2009. In addition, high-accuracy coordinates (< 1 meter error) were obtained for these FIA plots using survey-grade GLONASS-enabled GPS equipment. Changes in biomass (and associated uncertainty) estimated from field remeasurements alone were compared to changes estimated using 1) multi-temporal lidar data, 2) multi-temporal lidar and Landsat TM data, and 3) multi-temporal Landsat TM data. Both analytical and resampling (i.e. bootstrapping) approaches to variance estimation were presented and evaluated.