We consider alternate formulations of recently proposed hierarchical Nearest Neighbor Gaussian Process (NNGP) models (Datta et al., 2016a) for improved convergence, faster computing time, and more robust and reproducible Bayesian inference. Algorithms are defined that improve CPU memory management and exploit existing highperformance numerical linear algebra libraries. Computational and inferential benefits are assessed for alternate NNGP specifications using simulated datasets and remotely sensed light detection and ranging (LiDAR) data collected over the US Forest Service Tanana Inventory Unit (TIU) in a remote portion of Interior Alaska. The resulting data product is the first statistically robust map of forest canopy for the TIU.

Globally, forests are a crucial natural resource, and their sound management is critical for human and ecosystem health and well-being. Efforts to manage forests depend upon reliable data on the status of and trends in forest resources. When these data come from well-designed natural resource monitoring (NRM) systems, decision makers can make science-informed decisions. National forest inventories (NFIs) are a cornerstone of NRM systems, but require capacity and skills to implement. Efficiencies can be gained by incorporating auxiliary information derived from remote sensing (RS) into ground-based forest inventories. However, it can be difficult for countries embarking on NFI development to choose among the various RS integration options, and to develop a harmonized vision of how NFI and RS data can work together to meet monitoring needs. The NFI of the United States, which has been conducted by the USDA Forest Service’s (USFS) Forest Inventory and Analysis (FIA) program for nearly a century, uses RS technology extensively. Here we review the history of the use of RS in FIA, beginning with general background on NFI, FIA, and sampling statistics, followed by a description of the evolution of RS technology usage, beginning with paper aerial photography and ending with present day applications and future directions. The goal of this review is to offer FIA’s experience with NFI-RS integration as a case study for other countries wishing to improve the efficiency of their NFI programs.

Large-area assessment of aboveground tree biomass (AGB) to inform regional or national forest monitoring programs can be efficiently carried out by combining remotely sensed data and field sample measurements through a generic statistical model, in contrast to site-specific models. We integrated forest inventory plot data with spatial predictors from Landsat time-series imagery and LiDAR strip samples at four sites across the eastern USA-Minnesota (MN), Maine (ME), Pennsylvania-New Jersey (PANJ) and South Carolina (SC)-in statistical modeling frameworks to analyze the performance of generic (all sites combined) versus site-specific models. The major objective was to evaluate the prediction accuracy of generic and site-specific models when applied to particular sites. Pixel-level polynomial model fitting was applied to the time-series of near-anniversary date Landsat variables to obtain projected metrics in the target year 2014 for which LiDAR strip samples were available. Two forms of models based on ordinary least-squares multiple linear regressions (MLR) and the random forest (RF) machine learning approach were developed for each site and for the pooled (i.e., generic) reference data frame. The models were evaluated using national forest inventory (NFI) data for the USA. We observed stronger fit statistics with the MLR than with RF for both the site-specific and the generic models. The proportions of variances explained (adjusted R 2 ) with the site-specific models were 0.86, 0.78, 0.82 and 0.92 for ME, MN, PANJ and SC, respectively while the generic model had adjusted R 2 = 0.85. A test of statistical equivalence of observed and predicted AGB for the NFI locations did not reveal equivalence with any of the models, possibly due to the different resolutions of the observed and predicted data. In contrast, predictions by the generic and site-specific models were equivalent. We conclude that a generic model provides accuracies comparable to the site-specific models for large-area AGB assessment across our study sites in the eastern USA.

Gathering information about forest variables is an expensive and arduous activity. As such, directly collecting the data required to produce high-resolution maps over large spatial domains is infeasible. Next generation collection initiatives of remotely sensed Light Detection and Ranging (LiDAR) data are specifically aimed at producing complete-coverage maps over large spatial domains. Given that LiDAR data and forest characteristics are often strongly correlated, it is possible to make use of the former to model, predict, and map forest variables over regions of interest. This entails dealing with the high-dimensional (∼10 2 ) spatially dependent LiDAR outcomes over a large number of locations (∼10 5 −10 6 ). With this in mind, we develop the Spatial Factor Nearest Neighbor Gaussian Process (SF-NNGP) model, and embed it in a two-stage approach that connects the spatial structure found in LiDAR signals with forest variables. We provide a simulation experiment that demonstrates inferential and predictive performance of the SF-NNGP, and use the two-stage modeling strategy to generate complete-coverage maps of forest variables with associated uncertainty over a large region of boreal forests in interior 1 arXiv:1801.02078v2 [stat.AP] 8 Nov 2018Alaska.