Vegetation Community Classification and Mapping of the INL Site




Jeremy P. Shive, Amy D. Forman, Aurora Bayless-Edwards, Ken Aho, Kristin N. Kaser, Jackie R. Hafla and Kurt T. Edwards

June 2019

The INL Site vegetation map is a dynamic dataset that may be updated periodically. It is the user's responsibility to insure the most recent dataset is being used. Please contact Jeremy Shive ( to inquire about any recent updates to the dataset.
Note: The INL Site vegetation map was completed before the Sheep Fire burned in 2019. This region of the map will need to be updated when appropriate field data and recent imagery are available.

The Idaho National Laboratory (INL) Site is located in southeast Idaho and occupies 2,300 km2 (890 mi2) of sagebrush steppe. The INL Site is managed by the U.S. Department of Energy (DOE) and serves as a science-based, applied engineering national laboratory that supports the DOE missions in nuclear and energy research, science, and national defense.

The most recent vegetation mapping effort at the INL Site was completed in 2011, and in terms of resolution, accuracy, and statistical rigor, this vegetation classification and map represented a substantial improvement over previous mapping efforts. The vegetation class descriptions and the map have been used extensively to support inventory and monitoring of ecological resources on the INL Site. However, it is important to update the classification and map periodically to ensure that both the vegetation classes identified on the INL Site and the mapped boundaries of those classes remain accurate.

Three main factors justify updating the vegetation classification and map. First, four large wildland fires burned approximately 23% of the INL Site leaving the map outdated in those regions. Second, there were numerous map polygons assigned to two-class complexes which can overestimate the area of some individual classes and can make it more difficult to directly target sampling or monitoring in one specific vegetation class. Finally, field observations, especially within recently burned areas, showed that vegetation communities have begun to shift in composition, and in some regions non-native annual grass and forb abundance has increased considerably.

The goal of this project was to develop an updated vegetation classification and map of the current distribution of plant communities on the INL Site. Our specific objectives included: 1) characterize the vegetation community types present on the INL Site; 2) define the spatial distribution of those community types; and 3) conduct a quantitative accuracy assessment of the resulting map.

Objective 1 – Plant Community Classification An update to the vegetation classification was the first step in the process of updating the vegetation map for the INL Site. The primary objective of the plant community classification was to sample a representative range of plant communities across the INL Site and organize them into meaningful vegetation classes. Our approach for the previous classification effort relied heavily on quantitative methodologies. Overall, the technique worked well; each class was readily defined by a few dominant or co-dominant species and similarity scores between vegetation class pairwise comparisons were typically below 50%. Because the prior classification approach yielded vegetation classes that were meaningful with respect to local plant community dynamics and were useful from a mapping standpoint, we used the same approach for the classification update. However, we made changes to increase plot sampling efficiency. The new plot sampling methodology better addressed plot-to-polygon scale issues, and improved characterization of underrepresented classes.

During the summer of 2017, we collected vegetation data on 333 plots to support the updated vegetation classification. Plots were selected according to a stratified random design using Geographic Information System (GIS) data layers including the previous vegetation map updated with current wildland fire boundaries. Cover data were collected using point interception frames located along a 50 m transect within a conceptualized linear plot. We completed a quantitative classification using cover by species data from each plot. For the classification update, we compared eight classification methods using seven evaluators. The seven evaluators used to determine the most appropriate classification method were also used as criteria to assess the optimal number of clusters, or vegetation classes for that classification method. For both the model selection and evaluation of the optimal number of classes, we considered twenty-nine possible classification solutions.

We determined that beta-flexible (β = -0.25) was the best classification method and the optimal solution contained 16 clusters. The update to the classification resulted in 10 fewer classes than the prior classification. The reduction in the number of vegetation classes from the classification update is a consequence of some of the localized, patchy classes from the initial classification being enveloped into fewer, more comprehensive classes that are more interpretable at the targeted mapping scale. Therefore, the linear plot design did appear to yield a better overall classification, resulting in vegetation classes that were more reasonable for their intended use.

As with the results from the previous classification effort, we organized and interpreted the updated vegetation class list within the context of the National Vegetation Classification (NVC). The NVC is a hierarchical framework under which standardized vegetation classes, or species associations, are organized. Of the 16 vegetation classes identified in the INL Site classification update, 12 are natural vegetation classes and four are ruderal classes, or classes dominated by non-native species. Within the native classes, there was one woodland class, six shrubland classes, two shrub grasslands, and three grasslands. Within the ruderal classes, there was one shrubland, two grasslands, and a class characterized by mixed weedy forbs that tend to dominate areas with a specific hydrologic regime, namely playas. All the vegetation classes identified for the INL Site in the classification update were classified at hierarchical levels comparable to an Association in the NVC.

Objective 2 – Vegetation Class Delineations and Mapping We used the 2017 Idaho National Agricultural Imaging Program (NAIP) color-infrared multispectral imagery as the primary base map layer for map delineations. The 2015 Idaho NAIP imagery was also utilized in regions where standing water was present in the 2017 imagery and obscured the ground. To assist with the vegetation class delineations, we calculated two vegetation indices (i.e. the Normalized Difference Vegetation Index and the Soil-adjusted Vegetation Index), as well as a statistical texture layer (i.e. 3x3 Range) from the baseman imagery. We also used ancillary GIS data layers (e.g. digital elevation model) during the image delineation process to help identify patterns on the landscape.

Based on previous mapping experience, we understood the limitations of applying automated image classification methods in a semi-arid sagebrush steppe environment and relied on manual photointerpretation of digital imagery directly within a GIS. The map delineations were produced through manual interpretation and digitizing at a 1:6,000 mapping scale using a suite of GIS editing tools.

After reviewing the vegetation class list resulting from statistical clustering, it was apparent that several vegetation classes were unlikely to be recognizable in multispectral imagery. Consequently, there were two sets of the original 16 vegetation classes that were combined into a single map class resulting in a total of 14 map classes. To capture the fine-scale details of five non-vegetation classes (e.g. paved roads and borrow sources) and one agricultural class, we digitized at approximately a 1:2,000 mapping scale.

Once the map delineations were completed, we implemented spatial topology to perform the final Quality Assurance/Quality Control of the map polygons. Topology rules test whether polygons erroneously overlap one another or have small gaps between adjacent polygons that should share a common edge. We manually edited all vector errors and topology validation was rerun to verify all geometric errors were fixed.

The updated INL Site vegetation map contains 7,637 polygons, of which 7,265 (95.1%) represent vegetation classes. The remaining 372 (4.9%) polygons were assigned to nonvegetation special classes that accounted for only 30.3 km2 (7,478.8 acres) of the total mapped area. The Big Sagebrush – Green Rabbitbrush (Threetip Sagebrush) Shrubland class contained the largest amount of total area mapped with 851.2 km2 (210,330.9 acres). The second largest class mapped was the Green Rabbitbrush / Thickspike Wheatgrass Shrub Grassland and Needle and Thread Grassland class with 570.8 km2 (141,035 acres). The three largest map classes cover 73.2% of the vegetated area on the INL Site, suggesting the majority of vegetation communities are dominated by big sagebrush (Artemisia tridentata) or species most commonly associated with post-fire communities where big sagebrush was previously present.

The Big Sagebrush – Green Rabbitbrush (Threetip Sagebrush) Shrubland class also had the greatest number of map polygons with 2,388 and an average polygon area of 0.36 km2 (88.1 acres). The class containing the second largest number of polygons was the Cheatgrass Ruderal Grassland class with 1,435 polygons. However, the mean area of Cheatgrass Ruderal Grassland class was much smaller at 0.06 km2 (15.9 acres) and many of the polygons mapped were isolated individual patches rather than larger contiguous areas.

Objective 3 – Vegetation Map Accuracy Assessment During the summer of 2018, a total of 453 independent validation plots were collected and used to support the accuracy assessment of the final vegetation map. We used a standard error matrix to calculate map accuracy metrics including user’s/producer’s accuracy, overall accuracy and the Kappa statistic.

Initially, we maintained the two big sagebrush classes [i.e. Big Sagebrush – Green Rabbitbrush (Threetip Sagebrush) Shrubland and Big Sagebrush Shrubland] as distinct classes that were each allocated the appropriate number of random field validation plots. However, upon reviewing two instances where independent field crews sampled the same plot location at different times, we found that field crews confused big sagebrush classes (i.e. Class 6 and Class 8) in both cases. Consequently, whenever either class was recorded in the field or assigned to map polygons, they were combined prior to the accuracy assessment calculations. Combining these two vegetation classes resulted in 13 total map classes considered for the accuracy assessment. The accuracy assessment results showed an overall map accuracy of 77.3% and a Kappa value of 0.75. The map accuracy result values were higher than three of the four methods used to validate the previous vegetation map. The Kappa value is close to the 0.8 threshold which can be interpreted as strong agreement and is also higher than three of the four error matrix results from the previous vegetation map accuracy assessment.

The Juniper Woodland had the highest user’s and producer’s accuracy at 100% with no documented mapping errors. The map class with the next highest user’s accuracy was the combined Big Sagebrush – Green Rabbitbrush (Threetip Sagebrush) Shrubland and Big Sagebrush Shrubland class at 93.9%. There were five other classes that all had a user’s accuracy above 80%. The second highest producer’s accuracy was the Black Sagebrush Shrubland class at 94.7%. The Shadscale Saltbush – Winterfat Shrubland class was also very high with a producer’s accuracy of 93.3%. There were four additional classes that had producer’s accuracy above 80%.

Vegetation Community Classification and Mapping of the Idaho National Laboratory Site Report 2011


STOLLER-ESER-126 - The Idaho National Laboratory Site Long-Term Vegetation Transects: A Comprehensive Review - Amy D. Forman, Roger D. Blew, Jackie R. Hafla - June 2010