Moderate Resolution Data and Gradient Nearest Neighbor Imputation for Regional-National Risk Assessment
This study was designed to test the feasibility of combining a method designed to populate pixels with inventory plot data at the 30-m scale with a new national predictor dataset. The new national predictor dataset was developed by the USDA Forest Service Remote Sensing Applications Center (hereafter RSAC) at the 250-m scale. Gradient Nearest Neighbor (GNN) imputation was designed by the USDA Forest Service Pacific Northwest Research Station (hereafter PNW) to assign a plot identifier, and, therefore, a link to associated plot data, to each pixel within a target raster. GNN was implemented at 30-m resolution in three separate multi-million-hectare regions of the Western United States (Pierce and others, in review). Concurrently, RSAC developed a set of spatial predictor surfaces at 250-m resolution for use in producing nationally consistent data products. These data have been used for modeling forest types and forest biomass for the conterminous United States and Alaska (Blackard and others, in press; Ruefenacht and others, in press). These predictor data have also been used for large regional applications.
In this study, we substituted the 250-meter predictor data for the 30-meter predictor data used thus far in GNN. Our objective was to quantify the difference in performance using the lower spatial resolution predictors. We remodeled the same three regions that were mapped at 30 m with the 250-m data set and compared the error structure of the two modeling efforts. For species presence/absence models in the two areas with large environmental gradients, the Sierra Nevada and northeastern Washington, the species models performed substantially the same at the two resolutions. For the region with reduced environmental heterogeneity and moderate environmental gradients, coastal Oregon, species models did not work well with either the 30-m or 250-m studies. Models geared towards mapping forest structure did not perform as well as the 30-m models and may be insufficient for risk-assessment use.
- Introduction : A great wealth of resources has been expended to inventory our Nation’s forests, and an equally substantial amount of effort has gone into acquiring remotely sensed data.
- Methods : Subsections found in Methods include synopses of the study area, vegetation and moderate resolution predictor data, satellite imagery, and biophysical environment data.
- Results : Analysis results of Gradients in Species Composition, Species Mapping Performance and Structure Mapping Performance of GNN.
- Discussion- Accuracy of GNN Vegetation Maps : For the Washington and California study sites, species distributions were modeled equally well with the 250-m data and the 30-m data.
Encyclopedia ID: p3443
