WWETAC Projects
Project Title: Nationwide Forest Imputation Study (NaFIS)
Status: Ongoing
Principal Investigators: Ken Brewer, Remote Sensing Applications Center (RSAC), USDA Forest Service (FS); Andy Finley, Michigan State University; Ron McRoberts, Northern Research Station, USDA FS; Janet Ohmann, PNW Research Station, USDA FS
Co-Principal Investigators: Mark Finco, RSAC, USDA FS; Matt Gregory, Oregon State University (OSU); Emilie Grossmann, OSU
E-mail Contacts: Andy Finley, finleya[at]msu.edu (eastern US); Janet Ohmann, johmann[at]fs.fed.us (western US)
Web: http://blue.for.msu.edu/NAFIS/
Collaborators: Forest Health Technology Enterprise Team, Washington Office, USDA FS (Frank Sapio and Eric Smith); Forest Inventory and Analysis Program, Washington Office, USDA FS; Western Wildland Environmental Threat Assessment Center (WWETAC), USDA FS (Jerry Beatty and Terry Shaw); Eastern Forest Environmental Threat Assessment Center (EFETAC), USDA FS (Bill Smith)
Key Issues/Problems Addressed: Forest resource managers, policymakers, and researchers increasingly desire spatially explicit, border-to-border digital data and map products for a large array of forest attributes. Nationwide digital forest map products currently are available for only a few forest attributes, and generally lack useful measures of statistical validity. These data layers are needed to support applications ranging from scenario modeling (e.g., fire, insect, pathogens, wildlife habitat) at the mid- or regional-scale, to broad ecosystem modeling (e.g., carbon sources/sinks, climate change, and ecosystem services). The National Forest Inventory and Analysis (FIA) is an important source of comprehensive, quantitative information about forests. However, because FIA surveys are sample-based, estimates often are not statistically reliable for small geographic areas due to insufficient sample sizes, and they are unable to depict spatial distributions of the attributes. Therefore, model-based approaches for creating digital maps are needed.
Nearest neighbor techniques have been used in forestry not only to improve inventory estimates for small areas, but also to produce multi-attribute map data. These techniques produce estimates of un-sampled areas by relying on the relationship between sampled areas and spatially extensive, statistically correlated data such as physiographic, environmental, and remotely sensed spectral data. A number of groups in the Forest Service and elsewhere have developed, tested, and applied various nearest neighbor approaches, but a rigorous evaluation of alternatives and a strategy for potential national implementation has not yet emerged.
Study Goal and Objectives: This project evaluates alternative nearest neighbor techniques with the ultimate goal of recommending an approach for nationwide implementation. Study objectives are to: (1) evaluate alternative nearest neighbor algorithms, (2) develop computing systems for efficient implementation, and (3) produce map products, including variance estimators and accuracy assessments, for multiple forest attributes. The vision for a national nearest neighbor application is to rely on FIA data as the primary source of forest data, and Landsat as the primary remotely sensed data. Other forest, environmental, and physiographic data will be tested for applicability, and ultimately these data inputs could vary by region.
General Description: We are investigating nearest neighbor techniques through a pilot study focused on up to seven mapping zones across the US (see figure). The pilot mapping zones represent a range of ecological conditions and vary in terms of FIA data availability. The analyses will evaluate efficient nearest neighbor algorithms, variance estimators, and data processing techniques for broad-scale mapping. Lessons learned from the pilot study will provide operational guidance for efficient implementation nationwide. Although we will provide a core set of methods and data that are consistent nationally, there may be some regional variation in specific components of nearest neighbor techniques.
Status: The pilot study began in fall 2007 and is expected to span two years. As of August 2008, spatial data development is nearly complete for all mapping zones. Initial application of nearest neighbor techniques is currently underway in mapzones 7 (Oregon) and 51 (Michigan).
Deliverables: The pilot will test proof-of-concept for nationwide nearest-neighbor mapping, and spatial data products based on these concepts for up to seven large pilot regions. Spatial data products will depict a national core set of forest variables at moderate spatial resolution (30-m by 30-m pixels), based on Landsat Thematic Mapper (TM) satellite imagery. All map products will include associated accuracy assessments (confusion matrices and root mean square errors), pixel-level (spatial) estimates of uncertainty, and metadata.
Background Citation: Additional information on nearest neighbor applications in the Pacific Northwest, including publications and reports, can be found at http://www.fsl.orst.edu/lemma.
Map of coterminous US showing pilot mapping zones (yellow) and National Forests (in green).
Project ID: FY07JB35
