Project Title: Nationwide Forest Imputation Study (NaFIS)
Principal Investigators: Ken Brewer, Remote Sensing Applications Center, USDA Forest Service (FS); Andy Finley, Michigan State University; Ron McRoberts, Northern Research Station, USDA FS; Janet Ohmann, PNW Research Station, USDA FS
johmann[at]fs.fed.us (Western U.S.)
finleya[at]msu.edu (Eastern U.S.)
Collaborators: Mark Finco, Frank Sapio, Eric Smith, Jerry Beatty, Terry Shaw and Bill Smith USDA Forest Service; and Matt Gregory and Emilie Grossmann, Oregon State University.
Key Issues/Problems Addressed:
Spatially explicit digital data and map products displaying a large array of forest attributes are essential to providing the detailed information needed for the mapping of multiple forest threats. Imputation mapping is a promising technique, with potential for generating spatially explicit, border-to-border information on forest composition and structure across the U.S. The Nationwide Forest Imputation Study (NaFIS) was conducted with the intent of serving as a pilot project to further assess that potential.
Setting and Approach:
The process of imputation mapping was studied within several parts of the continental U.S. The process involved integrating Forest Inventory and Analysis’s Annual Inventory plots with spatially explicit information on climate and topography using Landsat TM5 image data. Four distance metric choices (referred to as model types) were compared including Euclidean (EUC), most similar neighbor (MSN), Gradient Nearest Neighbor (GNN), and random forest nearest neighbor (RAN), along with five different values of k: 1, 2, 5, 10 and 20 (the number of neighbors integrated to make each model prediction). The effects of changing model type and k values on model accuracy were analyzed in a variety of dimensions including: a) plot-level accuracy (root mean square difference and kappa measures), b) regional-scale accuracy (assessing for areal bias in the mapped model predictions), and c) plant community-scale accuracy (summarized from multivariate species-abundance predictions at the plot level).
- Accuracy varied little across the four model types, although RAN was slightly more effective for categorical predictions than the other methods.
- Predictions were most accurate when data was summarized within the forest portion of each plot, and whole-plot summaries were second in accuracy.
- Accuracy varied greatly across values of k. Higher values of k resulted in a slight increase in plot-level accuracy of core variables, but also led to a variety of problems for species abundance predictions (i.e., increasing errors of commission for species predictions and generating maps that contained unrealistic species combinations).
- Choice of k value appeared more critical than choice of model type in generating a map that will be appropriate for multiple uses.
A sampling grain using whole plots is recommended for national implementation, at least in the context of the mountainous west forests (where minor location errors in reference plot data can lead to large mismatches with the spatial data). Nationwide Forest Imputation Study map data and software information are available on the NaFIS project website: http://blue.for.msu.edu/NAFIS/.
Grossmann, E., J. Ohmann, M. Gregory and H. May. 2009. Nationwide Forest Imputation Study (NaFIS) – Western Team Final Report. Oregon State University, Corvallis, OR. On-line report. 53pp. (PDF, 2.78 MB)
Pierce, K.B., J. L. Ohmann, Wimberly, M. C., Gregory, M. J. and J. S. Fried. 2009. Mapping wildland fuels and forest structure for land management: a comparison of nearest neighbor imputation and other methods. Canadian Journal of Forest Research 39: 1901-1916. (PDF, 900 KB)
WWETAC Project ID: FY07JB35
Map of coterminous US showing pilot mapping zones (yellow) and National Forests (in green).