This project, "A novel approach to regional fuel mapping: linking inventory plots with satellite imagery and GIS databases using the Gradient Nearest Neighbor method" (referred to as "GNNFire") is Project 01-1-4-09 funded by the Joint Fire Science Program ( http://jfsp.nifc.gov/).
The GNNFire project investigated use of the Gradient Nearest Neighbor (GNN) method for mapping vegetation and fuels in three contrasting ecoregions in the Western US. The GNN method uses multivariate direct gradient analysis to link field plot data, satellite imagery, and maps of environmental variables in a raster GIS database. Individual pixels are associated with forest inventory plots that have the most similar spectral and environmental characteristics. A suite of detailed plot variables is then imputed to each pixel, allowing simultaneous and consistent predicting of a wide range of vegetation attributes. Because the plot-level variables are attached to the GIS database, a wide array of summary variables and classifications can be portrayed to meet different objectives.
Prior to the GNNFire project, GNN had been successfully used to generate forest vegetation maps suitable for detailed, stand-level modeling across large multi-ownership provinces in coastal Oregon and central Oregon. However, the method had not been tested in other ecoregions, nor specifically for mapping fuels. Accurate regional maps of vegetation and fuels are increasingly needed for assessing fire hazard, planning fuel management, and modeling the behavior and effects of prescribed burns and wildfires. In order for such maps to be useful to land managers, they must accurately predict a large number of vegetation and fuel attributes across heterogeneous, multi-ownership landscapes. We hypothesized that the flexibility of GNN, combined with its capacity for multivariate spatial predictions, would make it particularly useful for developing regional fuels maps.
This study examined the utility of GNN for predicting fuel patterns in three prototype landscapes in coastal Oregon, northeastern Washington, and the California Sierra Nevada, which encompass vegetation from dense forests to rangelands in a mosaic of natural and human-dominated environments. Our two primary objectives were: (1) Develop a methodology for using multivariate statistical models and imputation to simultaneously map fuel characteristics, species composition, and forest structure as continuous variables across environmentally heterogeneous, multi-ownership landscapes using inventory plot data, remote sensing imagery, and environmental GIS data layers; and (2) apply this methodology to generate vegetation and fuel maps for pilot landscapes located in three distinctive ecoregion divisions in the western US.