GNNFire Map Products
This section describes the four kinds of models (Table 1) distributed as JFSP final products, and summarizes the plot data (Table 2) and satellite imagery (Table 3) used in model development.
Download GNNFire Imputation Maps
Four kinds of GNN models
The use of multivariate statistics and imputation by GNN results in unique spatial modeling properties. As with most predictive models, alternative model forms can be specified to optimize for different objectives and outcomes. Rather than provide one GNN model and map for each study area, we've developed four GNN models that illustrate major variations in model form as a function of spatial pattern and emphasis on species composition vs. forest structure. We expect each kind of model to have advantages for certain applications, and we seek feedback from map users on the alternative model forms as they exercise the maps (e.g., via feedback received on the web site and from follow-up surveys of users who download or request GNN data). The four kinds of models are summarized in Table 1 and described below.
Table 1. -- Summary of four kinds of GNN models, as determined by spatial filtering of Landsat-derived explanatory variables and specification of response variables that emphasize species composition vs. forest structure.
|
Spatial filtering of
Landsat-derived explanatory variables |
Response variables
Species <=====> Structure |
||
|
Median-filtered
|
Species model (no imagery
in model) |
Species-size model,
median-filtered |
Structure model,
median-filtered |
|
Unfiltered
|
--
|
Structure model,
unfiltered |
|
The appearance (spatial patterning) of the final GNN maps are strongly influenced by tuning the spatial resolution of the independent variables - particularly those derived from Landsat TM imagery. Median-filtering of the raw Landsat imagery has the effect of reducing the fine-scale heterogeneity, or salt-and-peppering, in the final map, while maintaining boundaries between contrasting vegetation conditions (e.g., of clearcuts or stand-replacing fires). The median filtering consists of moving a nine-pixel window across the image, and assigning the median value of nine pixels to the center pixel. Grids for individual bands, ratios, and transformations are filtered independently. In general, overall accuracy in resulting GNN predictions appears to be little affected by the filtering, so decisions on which model to use are largely subjective based on appearance - at least until more experience is gained on the sensitivity of map applications to different spatial grains.
Species model: Response variables used in model development are basal area by tree species. Landsat TM, disturbance, and ownership variables are not included as explanatory variables. This model provides the most accurate spatial predictions of distributions of individual species and of community types that are defined based on species composition. Stand structure variables are not attached to this grid.
Species-size model: Response variables used in model development are basal area by species and size-class (but not all size-classes were recognized for all species). Explanatory variables derived from Landsat TM imagery are median-filtered, which reduces fine-scale heterogeneity or "salt-and-peppering" in the GNN map. This model is a "hybrid" between the species and structure models, and may be a good compromise model for applications where elements of both species composition and stand structure are needed, and covariance among these elements must be maintained (e.g., if tree lists are to be input into simulation models such as the Forest Vegetation Simulator). Accuracy for species variables in this model was intermediate between the species and structure models. Accuracy for structure variables was comparable or slightly worse than the structure model.
Structure model, median-filtered: Response variables are basal area by species group (conifer or hardwood) and size-class, total canopy cover, snag density by size class, and total down wood volume. This model provides slightly better overall accuracy of structure and fuels variables compared to the species-size model, but less accurate depiction of species distributions compared to the species and species-size models. We are distributing structure models developed at both coarse (median-filtered) and fine (unfiltered imagery) resolutions. Maps developed from median-filtered imagery contain less fine-scale variability than the unfiltered versions.
Structure model, unfiltered: Same as above but developed from unfiltered satellite imagery, which results in a map with much more fine-scale heterogeneity or "salt-and-pepper" effect.
Plot data used in GNN models
Table 2. -- Summary of plot data used in GNN models.
|
Study area
|
Data source
|
Ownerships
sampled |
Years measured
|
Number of plots
|
|
|
Species
model
|
Species-size
and structure models
|
||||
|
California
|
FIACA
|
Nonfederal
|
1993-1994
|
306
|
200
|
|
R5
|
National Forest
|
1995-2000
|
1,407
|
1,288
|
|
|
YOSE
|
Yosemite National Park
|
1988-1989
|
236
|
347
|
|
|
All plots
|
1,949
|
1,835
|
|||
|
Oregon
|
FIAWO
|
Nonfederal
|
1995-1997
|
572
|
385
|
|
BLMWO
|
BLM
|
1997
|
115
|
99
|
|
|
R6 (CVS)
|
National Forest
|
1993-1996
|
316
|
279
|
|
|
All plots
|
1,003
|
763
|
|||
|
Washington
|
FIAEW
|
Nonfederal
|
1991
|
475
|
468
|
|
R6 (CVS)
|
National Forest
|
1993-1997
|
1,856
|
1,808
|
|
|
NCNP
|
North Cascades National Park
|
2000
|
43
|
49
|
|
|
All plots
|
2,374
|
2,325
|
|||
Satellite imagery used in the GNN models
Table 3. -- Satellite imagery used in GNN models.
Study area |
Model | Landsat imagery used |
| California |
ca_spp | no imagery |
| ca_sppsz_fil |
2000 Landsat ETM+, median-filtered, tasseled cap transformations | |
| ca_struct_fil |
1992 Landsat TM and 2000 Landsat ETM+, median-filtered, raw bands | |
| ca_struct_unf | 1992 Landsat TM and 2000 Landsat ETM+, unfiltered, raw bands | |
| Oregon |
or_spp |
no imagery |
| or_sppsz_fil |
1996 Landsat TM, median-filtered, tasseled cap | |
| or_struct_fil |
1996 Landsat TM, median-filtered, raw bands | |
| or_struct_unf | 1996 Landsat TM, unfiltered, raw bands | |
| Washington | wa_spp |
no imagery |
| wa_sppsz_fil | 2000 Landsat ETM+, median-filtered, raw bands | |
| wa_struct_fil | 1992 Landsat TM and 2000 Landsat ETM+, median-filtered, raw bands | |
| wa_struct_unf | 1992 Landsat TM and 2000 Landsat ETM+, unfiltered, raw bands |
Download Imputation Maps
These files are ArcInfo workspaces with one GRID and have been zipped using WinZip. Users should be aware that they may take a long time to download on slower connections.
| Study Area | GNNFIRE Model | Download link (file size) |
| California | GNN species map |  (16.5 MB) | GNN species-size map (using filtered imagery) | (30.0 MB) | GNN structure model (using filtered imagery) | (35.6 MB) | GNN structure model (using unfiltered imagery) | (46.2 MB) | Plot database (MS Access format) | (1.7 MB) |
| Oregon | GNN species map | (12.0 MB) | GNN species-size map (using filtered imagery) | (22.4 MB) | GNN structure model (using filtered imagery) | (28.9 MB) | GNN structure model (using unfiltered imagery) | (33.1 MB) | Plot database (MS Access format) | (1.0 MB) |
| Washington | GNN species map | (21.3 MB) | GNN species-size map (using filtered imagery) | (37.7 MB) | GNN structure model (using filtered imagery) | (38.7 MB) | GNN structure model (using unfiltered imagery) | (45.7 MB) | Plot database (MS Access format) | (2.1 MB) |