Coupling Forest In-Situ and Spaced-Based Lidar Samples to Improve National-Scale Forest Inventory: a Joint Spatial Modeling Framework for Forest and Lidar Variable Prediction Lever

• Activity Number: 296 - Advances in Inference for Massive Spatio-Temporal Environmental Data with Applications in Remote Sensing
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2018 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Chad Babcock* and Andrew Oliver Finley and Hans-Erik Andersen and Bruce Douglas Cook and Douglas C Morton
• Companies: University of Washington and Michigan State University and USDA Forest Service and NASA Goddard Space Flight Center and NASA Goddard Space Flight Center
• Keywords: lidar; Nearest Neighbor Gaussian process; Gaussian process; forest aboveground biomass; geostatistics

Abstract

Future satellite lidar missions will collect data for narrow bands along orbital tracts, resulting in lidar metric sets of incomplete spatial coverage. This lack of coverage means traditional regression approaches that consider lidar data as predictors cannot be used to generate maps of forest variables. We implement a coregionalization framework to jointly model sampled lidar and forest aboveground biomass (AGB) field data to create maps. We inform the model with USFS-FIA forest measurements and GLAS spaceborne lidar to predict AGB across the contiguous US. We cast our model within a Bayesian hierarchical framework to better model complex space-varying correlation structures among the lidar metrics and FIA data. To circumvent computational difficulties that arise when fitting geostatistical models to massive datasets, we use a Nearest Neighbor Gaussian process prior. The coregionalization framework examined here is directly applicable to future spaceborne lidar missions. Pairing space-based lidar with the extensive FIA forest plot network using a joint prediction framework can improve forest AGB accounting and provide maps for analysis of the spatial distribution of AGB.